Modulating pseudokinase conformation from the ATP-binding site

SUMMARY

Pseudokinase domains (PsKDs) mediate key signaling processes involved in human disease. Despite transformative success in targeting catalytically active protein kinases, drug development for pseudokinases has lagged behind. Although they lack catalytic activity, pseudokinases are conformationally dynamic and are thought to function by switching between conformations that bind different signaling partners. This conformational switching model suggests the possibility of modulating pseudokinase signaling using small molecules that alter or bias conformational transitions. Focusing on the three PsKDs in receptor tyrosine kinases (RTKs) that retain nucleotide binding (ErbB3, EphB6, and EphA10), we used hydrogen deuterium exchange mass spectrometry (HDX-MS) to ask whether different small molecules can promote distinct changes in conformation or dynamics. We detected distinct conformational effects in different locations within the PsKD when different small molecules bind the ATP-binding site. Our results suggest that these small molecules will be useful probes for understanding pseudokinase function and ultimately for developing possible therapeutics.

Graphical Abstract

eTOC Blurb

Pseudokinases are catalytically inactive but conformationally dynamic. Using hydrogen-deuterium exchange mass spectrometry, Sheetz et al. find that different kinase inhibitors can induce distinct conformational changes upon binding to the same ATP-binding site of a given pseudokinase. This raises the possibility of specifically modulating allosteric signaling functions of pseudokinases.

INTRODUCTION

Protein kinases are among the most important classes of drug target,^1–4 with over 100 small molecule kinase inhibitors now clinically approved by the FDA.⁵ One important subgroup within this family – the pseudokinases^6–8 – has been relatively neglected, however, and recent successes^9,10 suggest that this represents a missed opportunity. Pseudokinases comprise around 10% of all human protein kinase homologues.¹¹ They lack one or more conserved catalytic residues^8,11, impairing or abolishing their capacity to catalyze phosphotransfer, and appear to be divided in their abilities to bind ATP.^7,12,13 The absence of significant catalytic activity has led to pseudokinases being passed over as drug targets until relatively recently,¹² despite their importance in signaling and disease.^14,15

Where functionally understood, pseudokinases appear to allosterically regulate key interaction partners like many catalytically active protein kinases,^7,16,17 and prospects for drugging them have focused on conformational disruption of this regulation.^12,18,19 We previously showed how this can be achieved for the pseudokinase domain from ROR1, a member of the receptor tyrosine kinase (RTK) superfamily²⁰ that has lost the ability to bind ATP.²¹ Ponatinib and GZD824 (both BCR-ABL tyrosine kinase inhibitors) bind selectively to the pseudokinase domain from ROR1 (but not ROR2) and cause conformational changes.²¹ Others have now expanded the repertoire of available ROR1 ligands,^22,23 and the discovery of small molecules likely to target other sites in the PTK7 pseudokinase domain have also been reported.²⁴

The only pseudokinase currently targeted by an FDA-approved small molecule is the Janus kinase (JAK) family member TYK2, with deucravacitinib approved for treating plaque psoriasis and psoriatic arthritis.^9,25 JAKs all have a tandem pair of kinase-like domains,²⁶ with a pseudokinase domain (PsKD) domain that retains ATP binding ability^13,27 followed by a conventional tyrosine kinase domain (TKD). JAKs are activated by dimerization, which results from their binding to cytokine receptors that dimerize upon extracellular ligand binding. JAK dimerization promotes intermolecular interactions – including key contacts that involve the PsKD – that activate the kinase and replace autoinhibitory intramolecular interactions seen in the monomer.^28–30 Myeloproliferative neoplasms frequently hijack this mechanism through gain-of-function mutations in the JAK2 PsKD that cause ligand-independent JAK-STAT signaling.³¹ Deucravacitinib occupies the PsKD ATP-binding site in TYK2,¹⁰ and is thought to destabilize intermolecular PsKD/PsKD interactions involved in TYK2 dimerization²⁸ (possibly also stabilizing autoinhibitory interactions). Interestingly, potential PsKD-binding TYK2 activators have also recently been reported.³²

Small molecules that bind the ATP-binding sites of other pseudokinases have also been evaluated as potential inhibitors.⁷ These include conformation-modulating agents that bind KSR³³ and STRADα,³⁴ as well as repurposed covalent EGFR inhibitors that bind to and destabilize the Tribbles pseudokinases.³⁵ Our previous hydrogen-deuterium exchange mass spectrometry (HDX-MS) demonstrate that pseudokinase domains are just as dynamic and capable of binding small molecule inhibitors as conventional kinase domains^21,36 – despite initial suggestions to the contrary.³⁷ With these precedents, we set out to understand how small molecule inhibitors might modulate the conformation (and thus potentially function) of RTK pseudokinases that retain ATP binding capability. Three RTK pseudokinases retain ATP-binding ability and are thought to function as scaffolds or allosteric regulators: ErbB3/HER3,^38–40 EphA10,⁴¹ and EphB6.^41,42 Each binds ATP with micromolar affinity, suggesting that the nucleotide-binding site will always be saturated at the millimolar levels of ATP found in cells. Using HDX-MS, we found that a subset of tyrosine kinase inhibitors (TKIs) that bind ErbB3, EphA10 or EphB6 significantly alter dynamics or local structure in regions of the PsKD thought to play key functional roles. Our results have important implications for targeting this neglected subfamily of protein kinases, potentially building on the success of deucravacitinib.

RESULTS AND DISCUSSION

Conformational dynamics of EGFR and ErbB3 kinase-like domains

We began our studies with the ErbB3 pseudokinase domain (ErbB3-PsKD) and first asked how it compares in its conformational dynamics with the closely related conventional tyrosine kinase domain (TKD) from EGFR – following our earlier studies of RTK pseudokinases with occluded ATP-binding sites.²¹ ErbB3-PsKD adopts a structure reminiscent of an inactive kinase in all available crystal structures, with the αC helix in an outwardly-displaced position (αC-’out’: Figure 1A) and a distorted activation loop.^37–39 By contrast, the conventional EGFR-TKD has been crystallized in both inactive-like (αC-’out’) and active-like (αC-’in’) conformations (Figure 1B).^43–47 This distinction led Campbell et al.³⁷ to suggest that the ErbB3 PsKD may sample fewer conformational states than EGFR-TKD. To assess this, we used HDX-MS to compare conformational fluctuations in the unliganded ErbB3 PsKD with those in EGFR-TKD (Figure 1C,D).^36,48 HDX-MS provides information on region-specific structural dynamics in proteins by measuring the rates of backbone amide hydrogen atom exchange with deuterium present as D₂O in the solvent used for the experiment.^49,50 More rapid exchange is observed for backbone amide hydrogens in structurally flexible or solvent-accessible regions than for those in structurally rigid regions, allowing analysis of time-dependent deuterium uptake to give an assessment of local backbone flexibility. To compare HDX in these two different domains, we used the PyHDX tool⁵¹ to infer residue-level protection values and free energies (ΔG_open) as described in STAR Methods. As shown in Figure 1C, unliganded ErbB3-PsKD and EGFR-TKD have broadly similar exchange profiles when plotted across sequence-aligned residues (Pearson correlation coefficient of 0.74), consistent with them both sharing the conserved protein kinase core with similar dynamics as summarized for ErbB3-PsKD in Figure 1D. There are differences, however. Approximately 30% of aligned residues differ significantly (by >2σ) – giving a noise-corrected root mean square difference in estimated ΔG of 1.36 kcal/mol (a protection factor difference of 10-fold). This reflects differences concentrated in certain areas, with no significant difference overall (p = 0.71). Notably, the region after helix αC shows less protection in ErbB3-PsKD than EGFR-TKD (Figure 1C), consistent with previous reports,⁵² and with ErbB3 having a shorter αC helix (~10 residues rather than ~15 in EGFR) – therefore likely to be subject to greater local unfolding.⁵³ Several regions in αF and αG also appear to show more H/D exchange in ErbB3-PsKD than EGFR-TKD, whereas the αH/αI region appears to show less. Overall, however, these data indicate that the conformational and dynamic architectures of ErbB3-PsKD resemble those of conventional TKDs²¹ such as EGFR-TKD, but with quantifiable local divergence in regions that may be associated with ErbB3 regulation.

Figure 1. The ErbB3 PsKD resembles conventional TKDs in conformational flexibility.

A. Structure of ErbB3-PsKD (red) bound to AMP-PNP/Mg²⁺ (shown as sticks and green sphere for Mg²⁺) determined by X-ray crystallography³⁹ (PDB ID: 3LMG). The relatively short ErbB3 αC helix adopts an αC-’out’, inactive-like position, stabilized by interactions with its distorted activation-loop helix.

B. Crystal structures of EGFR-TKD in inactive/αC-’out’ (left) and active/αC-’in’ (right) conformations, bound to AMP-PNP/Mg²⁺ (left, PDB ID: 2GS7⁴⁷) and compound 2 (right, PDB ID: 5JEB⁴⁴) respectively, labeled and shown as sticks (with green sphere for Mg²⁺).

C. ErbB3-PsKD displays similar dynamics to EGFR-TKD. Free energies (ΔG_open) of exchange were calculated from HDX-MS data using the PyHDX software package⁵¹ as described in STAR Methods, where a large positive ΔG_open value corresponds to strong protection, and a value close to zero less protection. Estimated ΔG values for unliganded ErbB3-PsKD are plotted in red (± standard deviation/SD), and values determined for unliganded EGFR-TKD from van Alderwerelt van Rosenburgh et al.⁷⁸ are plotted in grey, with connecting dashed grey lines at ± standard deviation. Values on the x axis represent the sequence-aligned residue number of the two proteins, referenced to ErbB3-PsKD. Secondary structure elements are indicated above the plot, with locations of the glycine-rich loop (G-loop), VAIK motif in β3, αC helix, HRD motif, DFG motif, and activation loop (A-loop) also noted. For EGFR data, n = 9 from three independent labeling campaigns; for ErbB3 data, n = 6 from two independent labeling campaigns. Peptide exchange data for ErbB3-PsKD are also shown in Figure S1.

D. HDX-MS data from the 10 min time point for ErbB3-PsKD were mapped onto its crystal structure (PDB ID: 3LMG) to give a sense of which structural elements show lesser (black) or greater (orange) exchange. The percent exchange of each amino acid was assigned using the DynamX software package (Waters) as described in STAR Methods, and is color coded on a scale from orange (100% exchange) to dark grey (no exchange). Regions are represented as transparent ribbons rather than PyMol cartoon (e.g. in helix αD) where there was no coverage.

E. DSF analysis of melting temperatures for ErbB3-PsKD in the absence (red dashed line) or presence (red solid line) of 5 mM AMP-PNP and 10 mM MgCl₂. AMP-PNP/Mg²⁺ causes a rightward shift (ΔT_M of 15°C) in the melting curve, reflecting increased stability upon binding. Curves represent means of three technical replicates.

F. HDX-MS analysis of ErbB3-PsKD in the presence and absence of AMP-PNP/Mg²⁺ reveals regions most influenced by ATP-site occupancy. A butterfly plot⁵⁵ is shown, in which the difference in percent backbone amide proton exchange (Δ% exchange) is plotted against median residue number (in ErbB3 sequence) for peptides seen in datasets for both unliganded and AMP-PNP/Mg²⁺-bound ErbB3-PsKD. Lower Δ% exchange represents protection upon AMP-PNP/Mg²⁺ binding. Significantly protected peptides were determined using the hdxstats software package⁶⁰, with criteria described in STAR Methods, and are denoted with asterisks at the bottom of the graph and listed in Table S2. Locations of ErbB3-PsKD secondary structure elements are indicated. For AMP-PNP/Mg²⁺ data, n = 9 from three independent labeling campaigns; for unliganded data, n = 6 from two independent labeling campaigns. Woods plots for all time points are shown in Figure S2.

See also Figures S1 and S2.

Effect of ATP binding on ErbB3 pseudokinase dynamics

The ErbB3 PsKD binds ATP/Mg²⁺ with a dissociation constant (K_D) of approximately 1 μM,³⁹ so will always be saturated at cellular ATP concentrations. Indeed, occupancy of the ATP-binding site was previously reported to be crucial for ErbB3 to perform its signaling function in neuregulin-induced ErbB3/ErbB2 heterodimers.^37,54 Binding of the non-hydrolysable ATP analogue AMP-PNP (plus Mg²⁺), which we utilized to prevent ATP hydrolysis by ErbB3’s residual activity,³⁹ stabilizes ErbB3-PsKD to denaturation by 15°C (to a T_M of 45°C) in differential scanning fluorimetry (DSF) studies (Figure 1E), consistent with previous findings.¹³ This stabilization is clearly evident when comparing HDX-MS profiles of ErbB3-PsKD with- and without bound AMP-PNP/Mg²⁺. A differential butterfly plot⁵⁵ shows that AMP-PNP/Mg²⁺ binding reduces H/D exchange broadly across the ErbB3 pseudokinase domain (Figure 1F). Particularly protected regions include strands β1-β3, the αC-β4 region in the N-lobe, strand β8, and the N-terminal part of the activation-loop that includes the DFG motif (Figures 1F, 2, and S1) – all regions adjacent to the ATP-binding site. These regions also include residues that line a hydrophobic ‘AP-2 pocket’ in the N-lobe of ErbB3-PsKD (above β3 in Figures 1A and 2A) that was focused upon by Campbell et al.³⁷ and includes residues F704, L709, W728, and V786. Campbell et al.³⁷ argued that this pocket is coupled allosterically to the ErbB3 ATP-binding site based on analysis of computational molecular dynamics (MD) simulations with- and without bound nucleotide. We saw no nucleotide-dependent changes in HDX in the region of F704 (Figure 2B). The β1 and β2 peptides, which include L709 and W728 respectively, did show small nucleotide-dependent HDX changes (Figures 2C,D), but several shorter peptides that include L709 showed no such change (Figure S1). No peptides containing V786 (in β5) were captured in our study, but adjacent regions that include β4 (Figures 2G and S1) did show nucleotide-dependent changes at later timepoints. Thus, our data do show dynamic changes in the AP-2 pocket when we compare ErbB3-PsKD in a nucleotide-bound state versus unliganded (unlikely to be seen in cells). However, these effects do not appear to be specific for the pocket as proposed by Campbell et al.³⁷ Rather, the changes seem to reflect a larger and more general stabilization of the ErbB3-PsKD N-lobe upon nucleotide binding. Our results argue against specific allosteric communication between the ATP-binding site and the AP-2. Indeed, Figure 1C shows that the N-lobe regions listed above are among the most dynamic parts of unliganded ErbB3-PsKD (orange when projected onto the structure in Figure 1D), and correspondingly showed the largest reduction in HDX upon AMP-PNP/Mg²⁺ binding (Figure 2A).

Figure 2. Nucleotide-induced changes in ErbB3-PsKD dynamics are limited to the N-lobe.

A. HDX-MS delta exchange data at the 10 min time point for ErbB3-PsKD with- and without AMP-PNP/Mg²⁺ are mapped onto the crystal structure (PDB ID: 3LMG). Percent exchange for each amino acid was assigned as described in STAR Methods, and delta exchange upon AMP-PNP/Mg²⁺ binding is color coded on a scale from blue (10% reduction in exchange) to red (10% increase in exchange) as illustrated. Regions with no coverage (e.g. helix αD) are represented as white transparent ribbons rather than PyMol cartoon, and key structural elements are labeled as in Figure 1. Woods difference plots are shown for all time points in Figure S2. For AMP-PNP/Mg²⁺ data, n = 9 from three independent labeling campaigns; for unliganded data, n = 6 from two independent labeling campaigns.

B. – I. Uptake plots are shown for key peptides discussed in the text, comparing HDX-MS for unliganded ErbB3-PsKD (open points, dashed curve) or with bound AMP-PNP/Mg²⁺ (closed points, solid curve). Data are shown ±SD (where larger than point size) across 5 time points as percentage of maximum exchange. Amino acid sequences of each highlighted peptide are noted at top, as are the structural feature/motif that they include. Uptake plots for all peptides are shown in Figure S1. For AMP-PNP/Mg²⁺ data, n = 9 from three independent labeling campaigns; for unliganded data, n = 6 from two independent labeling campaigns.

See also Figure S3.

Not unexpectedly, the ErbB3-PsKD C-lobe was generally unaffected by AMP-PNP/Mg²⁺ binding (Figures 2A and S2). We were interested, however, in knowing whether nucleotide binding to ErbB3 might specifically affect structural dynamics around the positions of C-lobe residues reported to be mutated in cancer, with mutations including Q809R, S846I and E928G (Figure S3A).⁵⁶ These mutated sites all map to points of contact with the EGFR-TKD N-lobe in an ErbB3-PsKD/EGFR-TKD heterodimer crystal structure,⁵⁷ with Q809 and S846 in ErbB3 contacting EGFR’s juxtamembrane region and E928 contacting the EGFR-TKD N-lobe (Figure S3B). As shown in Figure S3, peptides in the region of Q809 and E928 showed very little (if any) change in HDX upon AMP-PNP/Mg²⁺ binding. However, nucleotide binding did slightly reduce backbone amide exchange in an ErbB3 peptide that contains S846 (Figure S3D), which lies between strands β6 and β7 and contacts the EGFR-TKD juxtamembrane latch in the ErbB3-PsKD/EGFR-TKD heterodimer crystal structure (Figure S3B).⁵⁷ Reduced conformational fluctuation in this region of ErbB3’s C-lobe upon nucleotide binding – against which the EGFR juxtamembrane ‘latch’ docks – could allow stabilization of the EGFR-TKD/ErbB3-PsKD kinase domain heterodimer as suggested by Littlefield et al.,⁵⁷ potentially promoting the ability of ErbB3 to allosterically activate EGFR (or ErbB2).

Modulating ErbB3 pseudokinase conformation from the ATP-binding site

A published screen identified several small molecule TKIs that bind ErbB3-PsKD⁵⁸. Among these, bosutinib (SKI-606) had the highest affinity (K_D ~ 0.8 nM, also stabilizing ErbB3-PsKD by 15°C), with both dasatinib and gefitinib binding with lower affinities (K_D values of ~ 18 nM and ~ 800 nM respectively).⁵⁸ Previous work has shown that the signaling functions of ErbB3 in heterodimers with ErbB2 or EGFR are maintained (or even enhanced) when the ErbB3 ATP-binding site is occupied by any of these ATP-competitive TKIs.^37,54 Consistent with this, a crystal structure (PDB ID: 6OP9)³⁷ indicated that bosutinib stabilizes the same (inactive-like) αC-’out’ conformation of ErbB3-PsKD seen when AMP-PNP/Mg²⁺ is bound^38,39 – despite bosutinib being considered a ‘type I’ inhibitor (at least for ABL) that typically stabilizes the active-like (‘DFG-in’, αC-’in’) conformation of kinase domains.⁵⁹ Anticipating that these static crystallographic views may conceal differences in dynamics, we used HDX-MS to compare effects of binding of bosutinib and AMP-PNP/Mg²⁺ to ErbB3-PsKD. Plotting differences in HDX for ErbB3-PsKD bound to bosutinib versus AMP-PNP/Mg²⁺ in a differential butterfly plot (Figure 3A) revealed only small differences across most of the PsKD, which were not statistically significant according to an empirical Bayes analysis⁶⁰ (see STAR Methods). The region encompassing β4 and β5 of the PsKD (adjacent to the inhibitor binding site) was a clear exception (Table S2), showing significantly more protection with bosutinib bound (see also Woods plots in Figure S5A). Interestingly, the pattern of protection from exchange was different for dasatinib and gefitinib (Figure 3B), which are both also type I TKIs, but bind ErbB3-PsKD with 20 to 1000-fold lower affinity than bosutinib. These two TKIs had very similar effects on ErbB3-PsKD HDX-MS to one another, but neither showed the increased protection in the β4/β5 region seen with bosutinib. In fact, both dasatinib and gefitinib significantly increased HDX in this region to a small degree, when compared with AMP-PNP/Mg²⁺ (Figure 3B: see also Woods plots in Figures S5B and S5C). A significant small increase in exchange was also seen in the activation loop with dasatinib and gefitinib (but not bosutinib). Despite these small differences, the effects of dasatinib and gefitinib binding on ErbB3-PsKD dynamics are very similar to those of AMP-PNP/Mg²⁺. This suggests that the inactive-like αC-’out’ conformation seen crystallographically^38,39 is retained when these inhibitors are bound, despite them being type I TKIs. Bosutinib, with broadly similar HDX effects (apart from β4/β5 stabilization), was seen bound to this conformation crystallographically.³⁷

Figure 3. TKIs induce different effects on ErbB3-PsKD conformational dynamics.

A. – C. HDX-MS analysis of ErbB3-PsKD in the presence of: A bosutinib (magenta); B dasatinib (maroon) or gefitinib (gold); or C compound 2 (teal). Differential butterfly plots⁵⁵ represent HDX differences (Δ% exchange) between TKI-bound PsKD and AMP-PNP/Mg²⁺-bound ErbB3-PsKD, plotted against median residue number (in ErbB3 sequence) for the peptide, as in Figure 1F. Less exchange (negative values) are seen when TKI protects from exchange compared with AMP-PNP/Mg²⁺, and more exchange (positive values) when TKI increases HDX compared with AMP-PNP/Mg²⁺. Peptides that are protected or deprotected to a statistically significant degree were determined using hdxstats⁶⁰ with the criteria described in STAR Methods. Significantly altered peptides are denoted below each plot with asterisks and are listed in Table S2. ErbB3-PsKD secondary structure is indicated above each plot. For bosutinib and compound 2 data, n = 9 from three independent labeling campaigns. For gefitinib and dasatinib data, n = 3 from one independent labeling campaign. Uptake plots for all peptides are shown for all TKIs in Figure S4, and Woods difference plots are shown for all TKIs at all time points in Figure S5.

D. – F. Uptake plots for peptides containing key structural motifs in ErbB3-PsKD, comparing dynamics when bound to AMP-PNP/Mg²⁺ (black), bosutinib (magenta), and compound 2 (teal) across 5 time points. Data are plotted as percentage of maximum exchange ±SD (where larger than point size). Amino acid sequences of each highlighted peptide and corresponding structural element are noted at top.

See also Figures S4 and S5

We next asked whether it might be possible to ‘switch’ ErbB3-PsKD to an active-like conformation by binding a small molecule. Although the type I TKIs studied above failed to achieve this, we suspected that compound 2 might.⁴⁴ Compound 2 emerged from previous efforts to identify inhibitors of ErbB2/ErbB3 heterodimer signaling,⁴⁴ and appears to bind all ErbB kinase-like domains but to be highly selective for the active-like (αC-’in’) conformation.^44,61 We showed previously that compound 2 does bind ErbB3-PsKD, although its ability to block ErbB2/ErbB3 heterodimer signaling appears to result from inhibition of ErbB2 tyrosine kinase activity. We used HDX-MS to ask whether compound 2 has different conformational effects on ErbB3-PsKD from those seen with bosutinib, dasatinib and gefitinib, which all paradoxically maintain ErbB3-PsKD in an αC-’out’ conformation. As shown in Figures 3C–F, compound 2 stood out in its effects on HDX of ErbB3-PsKD. By contrast with the other TKIs analyzed, compound 2 binding substantially increased exchange in both the αC/β4/β5 region of ErbB3-PsKD and the activation-loop (see also Woods plots in Figure S5D). Exchange plots are compared in Figure 3 (and Figure S4) for peptides that include elements of αC (Figure 3D), the αC-β4 loop (Figure 3E), and the DFG motif/activation-loop (Figure 3F) – in which the inactive-like conformation contains a small 3₁₀ helix that melts upon transition to the active-like state. We were not successful in determining a crystal structure of ErbB3-PsKD bound to compound 2, but our HDX-MS data are consistent with compound 2 stabilizing a more active-like ErbB3-PsKD conformation than seen with other TKIs.

Implications of ErbB3 conformational changes

Our results reveal that different TKIs can have quite distinct effects on ErbB3-PsKD conformation (Figure 4) and argue that compound 2 stabilizes an active-like, αC-’in’, conformation not seen with the other inhibitors tested. Importantly, neither bosutinib nor compound 2 caused any significant HDX changes in the ErbB3-PsKD helical C-lobe, and the only significant differences seen in this region with dasatinib or gefitinib were small. These TKIs therefore do not impact the region of ErbB3-PsKD responsible for allosteric activation (as ‘activator’) of the EGFR or ErbB2 TKDs in asymmetric ErbB TKD heterodimers.⁵⁷ Rather, all significant TKI-induced conformational alterations that we can infer from our HDX-MS studies are restricted to the N-lobe and activation-loop regions of ErbB3-PsKD. Intriguingly, Claus et al. have argued⁵⁴ that ErbB2/ErbB3 activation can also occur through a structurally distinct mode of association involving TKD N-lobe interactions similar to those seen in crystals of AMP-PNP/Mg²⁺-bound ErbB3-PsKD (Figure S6).^38,39 They reported that this N-lobe mediated heterodimerization mode of ErbB2 and ErbB3 kinase-like domains is promoted by binding of lapatinib to the ErbB2 TKD (stabilizing its αC-’out’ inactive-like conformation),⁵⁴ and that it requires both ErbB2-TKD and ErbB3-PsKD to be in αC-’out’ configurations. The same group subsequently identified an ErbB3-specific small molecule (AC3573) that binds to the ErbB3-PsKD ATP-binding and inhibits N-lobe mediated ErbB3-PsKD/ErbB2-TKD heterodimerization.⁶² This would be the expected result if AC3573 stabilizes an active-like (αC-’in’) conformation of ErbB3-PsKD (or destabilizes the αC-’out’ conformation). We suggest that compound 2 may have emerged in the screens of Novotny et al.⁴⁴ as a particularly potent hit as a result of its ability both to block ErbB2 kinase activity and to inhibit engagement of ErbB3-PsKD in N-lobe mediated interactions (by stabilizing ErbB3-PsKD in an αC-’in’ conformation). Although the full complement of N-lobe interactions in which ErbB3-PsKD is involved has not yet been elucidated, it seems likely that their sensitivity to disruption by small molecules could allow useful future approaches for therapeutic targeting of aberrant signaling by this pseudokinase.

Figure 4. Different TKIs have distinct effects on ErbB3-PsKD conformational dynamics.

HDX-MS delta exchange data at the 10 min time point for the ErbB3 PsKD bound to A bosutinib or B compound 2 (referenced to the AMP-PNP/Mg²⁺-bound state). Differences are mapped onto the ErbB3-PsKD crystal structure (PDB ID: 3LMG) that has αC ‘out’. The difference in percent exchange for each amino acid was determined as described in STAR Methods, and is color coded from blue (20% reduction in exchange) to red (20% increase in exchange) as noted in the scale. Regions with no coverage (e.g. helix αD) are represented as white transparent ribbons rather than PyMol cartoon, and key structural elements are labeled as in Figure 1. Woods difference plots are shown for all time points in Figure S5.

See also Figure S6.

Eph receptor pseudokinase domains

The range of different TKI effects on ErbB3-PsKD conformation described above motivated us to turn to two pseudokinase members of the Eph (for erythropoietin-producing hepatoma) family of RTKs, EphA10 and EphB6.^42,63 These two RTK pseudokinases belong to a family of 14 RTKs in humans that otherwise all have active TKDs, and are collectively involved in regulating a host of biological functions.^64,65 Eph receptors are unique among RTK families in containing both a TKD (or PsKD) followed by a C-terminal SAM (for sterile alpha motif) domain⁶⁶ in their intracellular regions.²⁰ The PsKD/TKD and SAM domains influence each other structurally.⁶⁷ Liang et al.⁴¹ showed that both the EphA10 and EphB6 PsKDs bind nucleotides with K_D values in the 100 μM range (measured in the absence of Mg²⁺). We confirmed using DSF that the presence of nucleotide alone stabilizes the EphA10 and EphB6 intracellular regions to a greater extent than when ATP/Mg²⁺ is added (Figure S7A).

We subjected the soluble intracellular regions of EphA10 and EphB6 bound to AMP-PNP (with no added Mg²⁺) to HDX-MS, and again used PyHDX⁵¹ to infer residue-level protection values and free energies (ΔG_open), as plotted in Figure 5. The HDX-MS profiles of both PsKD regions resemble those of EGFR-TKD and ErbB3-PsKD (Figure 1C) as well as that published for the conventional TKD seen in EphA2.⁶⁷ For both EphA10 and EphB6, the linker between the PsKD and the C-terminal SAM domain showed very rapid exchange (ΔG_open close to zero: Figure 5A,B), as previously described for with EphA2.⁶⁷ Interestingly, however, the SAM domains of EphA10 and EphB6 appeared to differ from one another. The EphA10 SAM domain (Figure 5A) generally showed ΔG_open values closer to zero than the EphB6 SAM domain (Figure 5B) – indicating less protection from exchange – with EphB6 more closely resembling what was previously reported for the SAM domain of EphA2.⁶⁷

Figure 5. HDX-MS data for EphA10 and EphB6 intracellular regions.

Free energies of exchange (ΔG_open) estimated from HDX-MS exchange data using PyHDX⁵¹ for AMP-PNP-bound intracellular regions from: A EphA10, and B EphB6 in the absence of Mg²⁺ (see text). Data are represented as dark red points (EphA10) or gold points (EphB6) connected with solid lines bounded by estimated SD (dashed lines). Numbers on the x axis correspond to residue number. PsKD and SAM domain boundaries, as well as secondary structure elements are indicated above the plots, and locations of the glycine-rich loop (G-loop), αC helix, HRD motif, and activation loop are also noted. Peptide exchange data for the EphA10 and EphB6 intracellular regions are also shown in Figures S8 and S10. For both datasets, n = 6 from two independent labeling campaigns.

See also Figure S7.

The EphA10 and EphB6 PsKDs were both previously shown to bind strongly to the type I TKI dasatinib and the type II TKI AMG-Tie 2–1.⁴¹ We were interested in identifying other potential modulators of these PsKDs, and used DSF to screen for compounds in the Published Kinase Inhibitor Set (PKIS)⁶⁸ that thermally stabilize the EphA10 or EphB6 PsKDs in the context of the purified recombinant receptor intracellular regions. As detailed in Figure S7B and Table S3, we observed stabilization of the EphA10 intracellular region by more than 2°C (our threshold) by 32/361 (~8%) of the PKIS compounds and of the EphB6 intracellular region by 135/361 (~37%) compounds. The type I TKI dasatinib stood out as the only inhibitor that stabilized both the EphA10 and EphB6 intracellular regions by more than 10°C (Table S3). Foretinib also stood out by substantially stabilizing both proteins (EphA10 by 5.7°C and EphB6 by 15.3°C). Foretinib is a type II TKI with an additional ring that occupies the allosteric pocket in αC-’out’ kinases (Figure S7C).⁶⁹ The relative permissiveness of EphB6 to small molecule binding in this screen likely reflects its unusually short activation-loop (~14 residues shorter than those of other Eph RTKs), resulting in a lack of autoinhibitory interactions in this region and a relatively ‘unprotected’ ATP-binding site as suggested by our previous computational analysis.²¹

Following a similar logic to the ErbB3-PsKD studies outlined above, we used HDX-MS to ask whether binding of these type I (dasatinib) and type II (foretinib) TKIs to the EphA10 and EphB6 intracellular regions has different structural and dynamic consequences compared with AMP-PNP binding. The butterfly plots shown in Figures 6A and 6B show significant⁶⁰ changes distributed across the intracellular region. Binding of either dasatinib or foretinib increased HDX across the glycine-rich (G-rich) loop (Figure 6C) and αC/β4/β5 elements (Figure 6D) of the EphB6 PsKD – with little difference between these two TKIs. This is reminiscent of what we observed upon binding of dasatinib or gefitinib to ErbB3-PsKD (Figure 3B). Unexpectedly, there were small increases in exchange extending across the entire PsKD in EphA10 (Figure 6A), including the C-lobe – for which Δexchange was 10–20% higher when either dasatinib or foretinib were bound than when AMP-PNP was bound (notably between αG and αI). For EphB6, dasatinib but not foretinib increased exchange in the region of the PsKD surrounding the degenerate catalytic HRD motif (HRS in EphB6) compared with when AMP-PNP was bound (Figures 6B,E). Interestingly, binding of foretinib but not dasatinib also slightly increased exchange in the shortened EphB6 activation-loop region (Figure 6B), close to the motif that is proposed to bind SH3 domains in EphB6 signaling.⁴² With the exception of some small changes in αG, EphB6 PsKD dynamics between αEF and αI appeared similar regardless of whether the ATP-binding site was occupied by dasatinib, foretinib, or AMP-PNP (Figure 6B) – and less altered than for EphA10 (Figure 6A).

Figure 6. TKI-induced conformational modulation of the EphA10 and EphB6 intracellular regions.

A – B. HDX-MS data for the intracellular regions of A EphA10 and B EphB6 when the PsKD is occupied by dasatinib (orange: top) or foretinib (purple: bottom), revealing that subdomains are differentially influenced by these TKIs. Differential butterfly plots represent differences in exchange between the TKI-bound protein and AMP-PNP-bound protein. Peptides protected or deprotected to a statistically significant degree were determined using hdxstats⁶⁰ with the criteria for significance described in STAR Methods, and are denoted below each plot with asterisks and listed in Table S2. Domain boundaries and secondary structure elements for each protein are indicated at the top. For both datasets, n = 6 from two independent labeling campaigns. All uptake plots used to generate these representations are shown in Figure S8 (EphA10) and Figure S10 (EphB6), and Woods difference plots are presented for all time points in Figure S9A (EphA10/dasatinib), Figure S9B (EphA10/foretinib), Figure S11A (EphB6/dasatinib), and Figure S11B (EphB6/foretinib).

C – E. Uptake plots of EphB6 peptides corresponding to key elements within the PsKD, comparing HDX-MS data when bound to AMP-PNP (black), dasatinib (orange), and foretinib (purple) across five different time points – with HDX represented as percentage of maximum exchange ±SD (n = 6 from two independent labeling campaigns). The amino acid sequence and element/motif for each highlighted peptide is noted.

F – G. Corresponding uptake plots for two EphB6 peptides in the SAM domain. Peptides show significantly increased HDX when the PsKD is occupied by foretinib rather than dasatinib or AMP-PNP, with significant denoted in B.

See also Figures S8 to S11.

TKI effects on SAM domain dynamics

Intriguingly, the most prominent differences between the type I (dasatinib) and type II (foretinib) TKIs were seen in the SAM domain of EphB6 (Figure 6B). Whereas binding of dasatinib and AMP-PNP to the PsKD had essentially the same effects on the EphB6 SAM domain, foretinib binding substantially increased exchange (Figures 6B,F,G). By contrast, HDX-MS results for EphA10 were very similar for dasatinib and foretinib: both showing slightly more exchange than when AMP-PNP was bound (Figure 6A). Thus, the dynamics of an adjacent domain C-terminal to the PsKDs of these Eph family members are unexpectedly influenced by the nature of the ligand bound to the ATP-binding site of the PsKD. Since the Eph family SAM domains play important roles in signaling through SAM-SAM interactions and interactions with other entities,⁶⁴ these results suggest that ligands that bind the EphA10 and EphB6 ATP-binding sites may be able to modulate signaling by these receptors.

How might ligands for the PsKD ATP-binding site influence SAM domain dynamics and interactions? It is clear from crystal structures⁶⁷ and small angle X-ray scattering (SAXS) studies^41,67 that the SAM domain and TKD (or PsKD) of Eph RTKs do not have a fixed relationship. The purified intracellular region is monomeric (see STAR Methods)^41,67. SAXS studies of the EphA10 and EphB6 intracellular regions have suggested that they adopt multiple conformations in solution,⁴¹ being extended on average as in the AlphaFold⁷⁰ models shown in Figure 7. Cross-linking experiments have suggested frequent interdomain contact, however,⁴¹ including interactions that involve the αH/αI region of the PsKD highlighted by our HDX-MS studies as being modulated by TKI binding in EphA10. Similarly, Lechtenberg et al.⁶⁷ visualized multiple TKD/SAM arrangements in crystal structures of the EphA2 intracellular region – with the SAM domain abutting the αH/αI loop in the C-lobe of the TKD in the most intimate contacts. They also suggested that the positively-charged C-terminal part of the SAM domain may interact with an acidic region between helices αF and αG,⁶⁷ where TKI binding to EphA10 also causes altered HDX. Moreover, Lechtenberg et al. showed that phosphomimetic mutations in the EphA2 TKD/SAM domain linker lead to increased exchange in both the TKD C-lobe and the SAM domain.⁶⁷ Our data suggest possible allosteric coupling between the PsKD ATP-binding site and the SAM domain of EphB6, as foretinib produced coordinated deprotection at both the αC/β4/β5 region of the PsKD N-lobe and the SAM domain relative to AMP-PNP (Figure 7B). Because the ATP-binding site is conformationally connected to the kinase-domain surfaces that may contact the SAM domain, a ligand that reshapes the ATP-binding site could alter these docking interactions and loosen association of the SAM domain with the PsKD – thereby increasing its flexibility and solvent exposure. Thermodynamic linkage between these two domains is also suggested by the fact that inclusion of the SAM domain was required for us to be able to stably express and purify proteins containing the EphA10 and EphB6 PsKDs.

Figure 7. Differential conformational modulation in the RTK pseudokinase EphB6.

AlphaFold2⁷⁰-based structural representations of intracellular regions from A EphA10 and B EphB6 with delta uptake data at the 10 min time point superimposed for dasatinib minus AMP-PNP (left) and foretinib minus AMP-PNP (middle). The change in percent exchange for each amino acid was determined using the DynamX software package (Waters) as described in STAR Methods, and is color coded from blue (20% reduction in exchange) to red (20% increase in exchange) as depicted in the legend. Regions with no coverage are colored grey. Regions with no coverage are represented as white transparent ribbons rather than PyMol cartoon, and key structural regions are labeled. Woods difference plots are presented for all time points in Figure S9A (EphA10/dasatinib), Figure S9B (EphA10/foretinib), Figure S11A (EphB6/dasatinib), and Figure S11B (EphB6/foretinib). The structures were predicted for monomeric intracellular regions using AlphaFold2, with statistics shown in Figure S6D.⁷⁰ The rightmost representations of the structures are colored by residue according to pLDDT bins as depicted in the legend.

See also Figures S10 and S11.

CONCLUSIONS

Our findings provide biophysical evidence to support the general principle of pseudokinase conformational modulation and lend credence to the idea that small molecules that target the ATP-binding site of PsKDs can alter their conformational dynamics in ways that might be exploitable for modulating cell signaling.

Over 100 small molecule kinase inhibitors have now been approved by the U.S. Food and Drug Administration,⁵ among which just one targets a pseudokinase – deucravacitinib, which binds to the TYK2 PsKD. The success of deucravacitinib may herald a new era of pseudokinases as bona fide drug targets, demonstrating that small molecules can stabilize PsKD interactions that counter dysregulated signaling. We show here that PsKDs of RTKs that have retained the ability to bind ATP can be differentially modulated with different TKIs – in ways that have distinct conformational and dynamic effects on signaling-relevant parts of the molecule, including the N-lobe, the EphB6 SH3 domain-binding site, and the SAM domain. Our previous work with ROR1 – which we showed can be conformationally modulated using ponatinib²¹ – implies that these properties also extend across the RTK PsKDs that do not bind ATP. Work with KSR³³, STRADα³⁴, and JAK2⁷¹ suggests a similar story in each of these cases, and deucravacitinib establishes the potential for this approach in the clinic as mentioned above.⁷² These examples argue that there is nothing ‘undruggable’ about many pseudokinases – a conclusion that is further underlined by the data presented here for ErbB3, EphA10 and EphB6. Rather, the challenge is that we do not yet fully understand how most PsKDs signal – beyond the broad notion that they function as allosterically regulated scaffolds.^7,12,73,74 A structural understanding of how different PsKD-binding small molecules influence PsKD signaling should provide an experimental route to teasing out mechanisms of pseudokinase signaling. This idea extends to the growing set of pseudoenzymes that are similarly thought to undergo functionally important conformational changes that reflect transitions between active and inactive forms of their conventional enzyme counterparts.^17,75

The potential value of small molecules that target particular conformations of pseudokinases (or other pseudoenzymes) is further emphasized by recent reports that inhibitors of conventional kinases frequently induce the degradation of their targets by stabilizing particular conformations.⁷⁶ This phenomenon was also reported for off-target binding of irreversible TKIs to pseudokinases, with Tribbles 2 being rapidly degraded in response to the pan-ErbB inhibitor afatinib, for example.³⁵ These findings further increase the likely value of identifying small molecules that will bind pseudokinases as potential leads for new therapeutics in the many disease settings in which pseudokinase domains play important roles.

RESOURCE AVAILABILITY

Lead contact

Requests for further information or reagents may be directed to the lead contact, Mark A. Lemmon (mark.lemmon@yale.edu).

Materials Availability

All unique and stable reagents generated in this study are available upon request from the lead contact (mark.lemmon@yale.edu).

Data and Code Availability

• All data reported in this paper will be shared by the lead contact upon request.

• PDB entry 3LMG³⁹ was used to generate Figures 1A, 1D, 2A, 4A, 4B, S4.

• PDB entry 2GS7⁴⁷ was used to generate Figure 1B.

• PDB entry 5JEB⁴⁵ was used to generate Figure 1B.

• PDB entry 4RIW⁵⁷ was used to generate Figure S3B.

• AlphaFold2⁷⁰ was used to generate structural predictions in Figure 7.

• This paper does not report original code.

• HDX-MS mass spectrometry data have been deposited with ProteomeXchange via the PRIDE⁷⁷ database partner repository with the dataset identifier PXD071556 and are publicly available as of the date of publication.

• Any additional information required to analyze or reanalyze the data reported in this paper is available from the lead contact upon request.

STAR★Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell lines and culture conditions

Spodoptera frugiperda Sf9 cells (Expression Systems), originally established from immature ovaries of female S. frugiperda pupae, were propagated at 27°C with constant shaking at 120 rpm in either serum-free ESF 921 Insect Cell Culture Medium (Expression Systems) containing 50 U/ml penicillin/streptomycin or Insect-XPRESS medium (Lonza), and were used for production of all proteins.

METHODS DETAILS

DNA for recombinant protein expression

For producing purified ErbB3 protein, a previously described plasmid³⁹ based on pFastBac1 that contains the human ErbB3 tyrosine kinase-like domain/PsKD (amino acid residues 684–1020 using numbering of UniProt entry: P21860), with an N-terminal hexahistidine tag was used. For producing purified EphA10 and EphB6 proteins containing the intracellular PsKD and SAM domains (PsKD-SAM), PCR fragments corresponding to human EphA10 amino acids 624–1008 (UniProt entry: Q5JZY3) and human EphB6 amino acids 653–1021 (UniProt entry: O15197) were amplified from a Mammalian Gene Collection (MGC) cDNA template and inserted into the vector pFastbac-6HZB (Addgene #218680) using ligation-independent cloning as previously described.⁷⁹ The final plasmids contain open reading frames that encode either the EphA10 or the EphB6 intracellular regions along with a series of N-terminal tags consisting of a hexahistidine motif, followed by a protein Z domain from the IgG-binding part of Staphylococcal Protein A (to improve expression), followed by a TEV protease cleavage sequence.

Protein expression and purification

Baculovirus-based expression in insect cells was used for protein production.⁸⁰ Cells were grown to a density of 2 × 10⁶ cells/ml and inoculated with 15–25 ml of freshly prepared P2 viral stock per liter culture and incubated at 27°C for three days while shaking at 90 rpm. Sf9 cells expressing recombinant protein were lysed by sonication in Buffer A [50 mM Tris-HCl (pH 8.0), 300 mM NaCl, 10 mM 2-mercaptoethanol, 20 mM imidazole, 1 mM PMSF, protease inhibitor cocktail (Roche)]. Lysate was then clarified by centrifugation for 30 min at 16,000 rpm at 4°C and loaded onto pre-washed Ni-NTA beads (1 ml per liter of culture) for 1 h at 4°C. Beads were subsequently washed with 50 column volumes of Buffer A, and protein eluted with increasing concentrations of imidazole in Buffer B (Buffer A + 400 mM imidazole). ErbB3 PsKD protein was next loaded onto a Fractogel TMAE (trimethylaminoethyl) anion-exchange column (EMD) equilibrated with Buffer AEX [50 mM Tris-HCl (pH 8.0), 50 mM NaCl, 2 mM DTT (dithiothreitol)], and eluted with a NaCl gradient up to 1 M. EphA10 and EphB6 intracellular region proteins were instead loaded after the NTA-step onto a SP-Sepharose cation exchange column (Cytiva), washed with buffer containing 250 mM NaCl and eluted with increasing NaCl, followed by overnight cleavage of N-terminal tags with TEV protease. Proteins were finally spin-concentrated and subjected to size-exclusion chromatography using a Superdex 200 16/600 column (GE Healthcare) equilibrated in 20 mM HEPES (pH 7.5), 150 mM NaCl, 100 μM TCEP [tris-(2-carboxyethyl)phosphine], and 5% (w/v) glycerol. All proteins eluted at the position expected for monomers in size exclusion chromatography. Purified proteins were characterized during purification using standard quality-control methods (SDS-PAGE and mass spectrometry), and we did not observe evidence of significant post-translational modification. Any modified peptides species would represent a very low-abundance population and were not included in our HDX analysis.

Differential scanning fluorimetry (DSF)

For thermal shift assays to study ATP binding, proteins were diluted in 20 mM HEPES (pH 7.5), 150 mM NaCl, and 100 μM TCEP to a final protein concentration of 5 μM. Nucleotide (5 mM) and/or MgCl₂ (10 mM) was added. For studies of inhibitor binding, protein was diluted in the same buffer to a final concentration of 10 μM drug (containing a final concentration of 5% DMSO). SYPRO Orange (ThermoFisher Scientific) was then added (diluting 2500-fold into the sample), and 25 μl aliquots of the reaction mixtures were transferred to Concord 96-well polycarbonate PCR plates (Bio-Rad) with three technical replicates per condition. A Bio-Rad CFX96 Touch Real Time PCR machine was then used to monitor fluorescence using the FRET filter setting (exciting at ~450–490 nm with broad detection from ~530–650 nm), while the temperature was raised by 1°C per minute from 25°C to 95°C, with fluorescence measured at each increment. Fluorescence values were then plotted as a function of temperature after normalizing to the maximum fluorescence signal. Melting temperatures (T_M) were determined as the temperature at half-maximum fluorescence. Plots were generated using GraphPad Prism and represent the mean from three technical replicates.

For small molecule screening using thermal shift assays, a 2 μM solution of the relevant purified PsKD in assay buffer (20 mM HEPES pH 7.4, 150 mM NaCl, 0.5 mM TCEP, 5% glycerol) was mixed 1:1000 with SYPRO Orange (Millipore Sigma). Test compounds were added to a final concentration of 10 μM (containing a final 2% DMSO). 20 μl of each sample was then placed in a white 96-well plate (Biozym #710874), and the temperature was increased stepwise from 25°C to 95°C using a 3°C/min ramp rate. Fluorescence was monitored every 1°C increment using a Mx3005P real-time PCR instrument (Stratagene) with excitation and emission filters set to 465 and 590 nm, respectively. Data were analyzed using MxPro software. Compound screening was performed as a single shot experiment. Plots were generated using GraphPad Prism and represent the mean from technical replicates.

Small molecule screening for pseudokinase binders

Small molecule screening was performed using the DSF assay described above. Our screening collection consisted of the published kinase inhibitor set (PKIS)⁶⁸ library (#L1200 Selleckchem), which includes a total of 361 compounds. The PKIS was stored protected from light at −80°C in 500 μM stock concentrations solubilized in DMSO. Compounds were transferred for screening using a multi-channel pipette and mixed by pipetting up and down. Thermal shifts were determined by subtracting the temperature at which 50% of the protein was unfolded (T_M) from the T_M value of a compound-free (DMSO only) control condition. Resupplies of top hits dasatinib and foretinib were purchased from the independent vendor Selleckchem for follow up HDX-MS studies.

Chemical synthesis of compound 2

Molecular sieves (4 Å) were dried at 160°C under 0.25 mTorr vacuum prior to use. Unless otherwise mentioned, solvents were purified as follows. Methanol and 1,2-dimethoxyethane were purchased from Sigma Aldrich and incubated over 4 Å molecular sieves for 48 h prior to use. CH₂Cl₂ was dried over CaH₂ and freshly distilled prior to use. NMR spectra were obtained using a 400 MHz Bruker NMR spectrometer and referenced using the residual ¹H peak from the deuterated solvent. An Agilent 6550A (QTOF) instrument was used for HRMS (ESI) analysis with polyethylene glycol (PEG-400–600) as a reference. Column chromatography was performed using Silicycle 60 Å, 35–75 μm silica gel. Pre-coated 0.25 mm thick silica gel 60 F254 plates were used for analytical TLC and visualized using UV light, iodine, potassium permanganate stain, p-anisaldehyde stain or phosphomolybdic acid in EtOH stain. Compound 2 was synthesized based on the reported protocol by Novotny and colleagues, where it is referred to as compound 2⁴⁴. The purified product was obtained as an off-white powder in about 53% overall yield.

Analytical data for 3-(furan-2-yl)-N-(5-(furan-2-yl)-2-methoxyphenyl)-1H-pyrazolo[3,4-d]pyrimidin-4-amine: ¹H NMR (400 MHz, DMSO-d₆) δ 4.01 (s, 3H), 6.59 (dd, 1H, J = 4.0, 2.0 Hz), 6.80 (dd, 1H, J = 4.0, 2.0 Hz), 6.82 (dd, 1H, J = 4.0, 2.0 Hz), 7.08 (d, 1H, J = 2.0 Hz), 7.20 (d, 1H, J = 8.0 Hz), 7.44 (dt, 1H, J = 12.0, 4.0 Hz), 7.73 (s, 1H,), 8.01 (d, 1H, J = 4.0 Hz), 8.56 (s, 1H), 9.24 (t, 1H, J = 4.0 Hz), 9.62 (s, 1H), 14.03 (s, broad, 1H); ¹³C NMR (DMSO-d₆, 100 MHz) δ 56.5, 97.5, 104.4, 108.0, 111.2, 111.9, 112.6, 116.0, 118.8, 123.2, 128.3, 142.2, 143.4, 147.5, 148.1, 153.3, 154.4, 155.6, 156.4. HRMS: ES⁺ (C₂₀H₁₆N₅O₃): Calc. [M + H]⁺: 374.1253, Found [M + H]⁺: 374.1251.

Hydrogen-Deuterium Exchange Mass Spectrometry (HDX-MS)

Methods for HDX-MS data collection and analysis are described previously.^21,36 Starting with purified protein at 0.4 mg/ml in H₂O-based buffer (20 mM HEPES pH 7.4, 150 mM NaCl, 100 μM TCEP, 5% glycerol), the relevant TKI was added from a freshly prepared 1 mM TKI stock solution in DMSO to a final inhibitor concentration of 50 μM (and final DMSO concentration of 5% v/v). Alternatively, AMP-PNP was added from a freshly prepared 10 mM stock to a final concentration of 500 μM in H₂O-based buffer (plus final [MgCl₂] of 1 mM when included) for 30 min. The resulting mixtures were then diluted 20-fold with D₂O-containing buffer that contained 20 mM HEPES pD 7.4, 200 mM NaCl, in D₂O for H/D exchange, or the deuterium-free equivalent for undeuterated standards. This scheme ensured a final D₂O concentration of 95% in all samples for amide proton exchange. To maintain TKI or nucleotide concentrations during exchange, a 20X stock of the relevant TKI (in DMSO) or AMP-PNP +/− MgCl₂ (in H₂O-based buffer) was also diluted 20-fold into the D₂O-based labeling buffer before use – so that the v/v dilution of ligand and the final D₂O concentration was consistent across all labeling conditions. Labeling was performed at 25°C for a range of times (10 s, 1 min, 10 min, 1 h, or 2 h), and the reaction was quenched by adding cold (4°C) 200 mM citric acid buffer at pH 2.4. Fully deuterated controls were prepared by incubating proteins in the presence of 7 M urea-D₂O (20 mM HEPES, 200 mM NaCl, 7M Urea, pD = 7.4, 25°C) for 10 min prior to quenching, to allow complete exchange. All deuterium labeling time points and controls were repeated three times (technical replicates) for each biological replicate.

Quenched samples were flash-frozen in liquid N₂ and stored at −80°C until mass spectrometry analysis. Quenched samples were quickly thawed and immediately injected onto a Waters HDX nanoAcquity UPLC (Waters, Milford, MA) with in-line digestion using an Enzymate BEH pepsin column (Waters, Milford, MA). Peptic fragments were trapped on an either an Acquity UPLC CSH C18 (ErbB3 and EphB6) or an Acquity UPLC BEH C18 (EphA10) peptide trap and separated either using an Acquity UPLC CSH C18 (ErbB3 and EphB6) or an Acquity UPLC BEH C18 (EphA10) column. A 7 min, 5%−35% linear acetonitrile (0.05% trifluoracetic acid) gradient was used to elute peptides. The mass-to-charge ratio (m/z) of peptic peptides was acquired using a Waters Synapt G2-Si mass spectrometer operated in ion mobility mode (HD-MS^e). Leucine enkephalin was used for continuous mass and ion mobility correction. The scan times were 0.4 s, 0.2 s, and 0.3 s for ErbB3, EphA10, and EphB6, respectively. Other instrument parameters were as follows: capillary voltage, 3 kV; source temperature, 90°C (ErbB3), 85°C (EphA10), and 80°C (EphB6); desolvation temperature, 150°C; IMS gas flow, 90 ml min⁻¹; scan range, 250–1400 m/z. Peptide identification and protein sequence coverage maps were obtained from the undeuterated controls. Peptides were identified using the ProteinLynx Global SERVER 3.0.3 (PLGS) from Waters. Data and statistical analyses were carried out as described previously.⁵⁵ Briefly, the deuterium uptake by the identified peptic fragments through increasing deuteration time and for the fully deuterated control was determined using DynamX 3.0 (Waters). The normalized percentage of deuterium uptake at incubation time t (%D_t) at an incubation time t for a given peptide was calculated as follows:

$$\%D_t=100*\frac{m_t-m_0}{m_f-m_0}$$

With m_t being the centroid mass at incubation time t, m₀ the centroid mass of the undeuterated control and m_f the centroid mass of the maximally deuterated condition. For studies of changes in pseudokinase dynamics upon ligand binding, the percent deuteration difference plots Δ%D_exchange (drug bound – AMP-PNP bound) were generated by subtracting the corresponding percent deuteration at incubation time t calculated for the nucleotide-bound condition (AMP-PNP) from that calculated with drug thus:

$$\Delta \%D_{\mathit{\text{exchange}}}=\%D_{x,t}(\mathit{\text{drug}})-\%D_{x,t}(\mathit{\text{AMPPNP}})$$

where x is each peptide, and t is the D₂O incubation time. HDX-MS data collection followed consensus guidelines.⁸¹

Raw HD-MS^e mass spectra of undeuterated controls were used for peptide identification using ProteinLynx Global Server 3.0.3 (PLGS) from Waters. HD-MS^e data from all controls and labeling conditions were then processed using DynamX to identify the centroid mass of each peptic peptide. Following standard HDX-MS guidelines,⁸² peptides were filtered with DynamX using minimum precursor ion intensity = 2000, minimum products per amino acid = 0.2–0.4, minimum score = 6.62, minimum consecutive products = 1, minimum sequence length = 5, maximum sequence length = 20, and maximum MH+ error = 10 ppm. All raw spectra for each peptide, labeling condition, drug condition, and charge state were then manually assessed for quality and for accurate peak assignment, at which point poor quality or incorrectly assigned peaks were unassigned. Average mass shifts of centroids and their standard deviations were then used to calculate percent uptake for each time point relative to a fully deuterated standard.

QUANTIFICATION AND STATISTICAL ANALYSIS

Analysis of HDX dynamics

Peptic peptides from HDX-MS experiments were identified and analyzed as described above. Two complementary types of analysis were then performed: (i) residue-level protection-factor/free-energy (ΔG) determination for cross-protein and cross-state comparisons, and (ii) functional statistical analysis of differential uptake kinetics.

Residue-level protection factors and free energies of exchange (PyHDX).

To enable comparison of conformational dynamics across different protein sequences and ligand states, residue-level protection factors and the corresponding free energies of exchange (ΔG_open) were determined using PyHDX (v0.4.3)⁵¹. Because ΔG/protection-factor analysis normalizes for the sequence-dependent intrinsic exchange rate (k_int), it can provide a sequence-independent estimate of local backbone stability that is suitable for comparing HDX-MS results across proteins. DynamX cluster output was processed with peptide-level filtering, and the fully deuterated control obtained for each dataset was used to convert uptake to relative fractional uptake. Intrinsic exchange rates were then calculated for the experimental conditions⁸³ (25°C, pH 7.0 [read uncorrected from the pD = 7.4 reading per the Glasoe–Long relationship⁸⁴]). Exchange rates were first estimated by weighted-averaging fits, and residue-level ΔG values were then obtained by global fitting, yielding per-residue ΔG values, protection factors, and covariance-derived standard errors. Residues were considered reliably determined and retained for downstream analysis if their fitted standard error was < 2 kcal/mol and − ΔG ≤ 0; prolines and residues lacking coverage were excluded. ΔG_open values are reported throughout in the exchange convention (with more negative values indicating less protection).

Cross-protein comparison and RMSD analysis.

To compare HDX-MS results for EGFR-TKD and ErbB3-PsKD (Figure 1C), the two sequences were aligned by global pairwise alignment (Biopython, BLOSUM62), and conserved structural landmarks (the DFG motif, gatekeeper, and HRD/HRN positions) were used to confirm the register. Estimated ΔG values were then compared at aligned positions present in both proteins (n = 185). The root-mean-square difference (RMSD) and Pearson correlation were computed on log₁₀(protection factor) (equivalently ΔG), since the protection factors span several orders of magnitude.

Differential uptake-kinetics analysis (hdxstats).

Statistically significant differences in deuterium-uptake kinetics between differently liganded states of the same protein were determined using hdxstats, an empirical Bayes functional method for HDX-MS analysis.⁶⁰ Uptake versus time profiles (timepoints 10 s, 60 s, 600 s, 3600 s, and 7200 s; fully deuterated controls excluded from the kinetic fit) were fit with the full Weibull kinetic model (value ~ a × (1 - exp(−b × t^p)) + d, with a, b, d, and p free), and a model in which both states share a common curve was compared to one in which they differ using a moderated F-test with empirical Bayes shrinkage of the variance and Benjamini-Hochberg correction for multiple testing. A peptide-charge feature was considered significantly different if it satisfied two criteria: a Benjamini-Hochberg-adjusted false discovery rate < 0.05 AND a peak per-timepoint difference in deuterium uptake (the per-timepoint Δ uptake of maximum absolute magnitude across the labeling time course) that exceeded an effect-size threshold of ~3 × √2 × the median within-condition replicate standard deviation for that comparison. Each charge state was analyzed as a separate feature. Each ligand-bound condition was compared to its corresponding AMP-PNP-bound reference. The complete list of significant peptide-charge features, with uptake differences, false discovery rates, and per-comparison effect-size thresholds, is provided in Table S2.

Structural representations and data summary.

Structural representations of HDX-MS and ΔG/ΔΔG datasets were generated by mapping per-residue values onto structures as described previously²¹ or using custom PyMOL scripts to color B-factor fields by ΔΔG or pLDDT. A summary of HDX-MS conditions, replicates, and uptake data can be found in Table S1

Supplementary Material

1

Document S1. Figures S1-S11 and supplemental references.

2

Table S1. HDX-MS summary and uptake Data, related to Figures 1–7. Experimental details are listed for all HDX-MS experiments performed in this study, and the uptake data for each peptide are reported.

3

Table S2. Lists of all peptides showing statistically significant exchange differences in each HDX-MS experiments performed in this study, related to Figures 1–7.

4

Table S3. Screen of EphA10 and EphB6 PsKD binders, related to Figure 6. A library of ~350 small molecules that comprise the published kinase inhibitor set is listed, and the respective effects of each compound on the differential scanning fluorometry-based T_M values of EphA10 and EphB6 (relative to the unliganded T_M values) are reported.

KEY RESOURCE TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
MAX Efficiency DH5alpha Competent Cells	Invitrogen	Cat#: 18258012
MAX Efficiency DH10Bac Competent Cells	Invitrogen	Cat#: 10361012
Chemicals, peptides, and recombinant proteins
Adenylyl-imidodiphosphate (AMP-PNP)	Fisher Scientific	Cat#: NC1050153
Bosutinib (SKI-606), >99% (HPLC)	Selleckchem	Cat#: S1014
Cellfectin II Reagent	Thermo Fisher Scientific	Cat#: 1036200
cOmplete, mini, EDTA-free Protease Inhibitor Cocktail	Sigma-Aldrich/Roche	Cat#: 11836170001
Compound 2	Novotny et al.44	N/A
Dasatinib (BMS-3544825)	Selleckchem	Cat#: S1021
Deuterium Oxide (99.9%) Low paramagnetic	Cambridge Isotope Labs	Cat#: DLM-11-100
FDA-approved Drug Library	Selleck Chemicals	Cat#: L1300
Foretinib (GSK1363089)	Selleckchem	Cat#: S1111
Fractogel EMD TMAE	Millipore Sigma	Cat#: 116881
Fractogel EMD SO3-	Millipore Sigma	Cat#: 116882
Gefitinib (ZD1839)	Selleckchem	Cat#: S1025
Kinase Inhibitor Library	Selleck Chemicals	Cat#: L1200
Ni-NTA Agarose Beads	Qiagen	Cat#: 30210
Penicillin-Streptomycin (10,000 U/ml)	Thermo Fisher Scientific	Cat#: 15140122
SYPRO Orange Protein Gel Stain	ThermoFisher Scientific	Cat#: S6650
SYPRO Orange Protein Gel Stain	Millipore Sigma	Cat#: S5692
Recombinant ErbB3TKD	This study	N/A
Recombinant EphA10TKD-SAM	This study	N/A
Recombinant EphB6TKD-SAM	This study	N/A
Deposited data
HDX-MS data for ErbB3TKD	This study	PRIDE: PXD071556
HDX-MS data for EphA10TKD-SAM	This study	PRIDE: PXD071556
HDX-MS data for EphB6TKD-SAM	This study	PRIDE: PXD071556
Experimental models: Cell lines
Insect: S. frugiperda Sf9 cells	Expression Systems	Cat#: 94-001F
Oligonucleotides
EphA10F: TACTTCCAATCCCAGAGCTGTGGGGACCTGCTGC	This study	N/A
EphA10R: TATCCACCTTTACTGTCACACCTGCACCCCCTGGCCCTGC	This study	N/A
EphB6F: TACTTCCAATCCGACCCCTGTCAGGCCATCCGAGAAC	This study	N/A
EphB6R: TATCCACCTTTACTGTCAGACCTCCACTGAGCCCTGCTGCCTC	This study	N/A
Recombinant DNA
pFastBac1-6xHis-ErbB3TKD	Shi et al.39	N/A
pFastBac1-6HZD-TEV-EphA10TKD-SAM	This study	N/A
pFastBac1-6HZD-TEV-EphB6TKD-SAM	This study	N/A
Software and algorithms
Adobe Photoshop and Illustrator	Adobe Inc.	https://www.adobe.com
CFX Maestro v2.2	BioRad	N/A
GraphPad Prism 10.6.1	GraphPad Software, La Jolla California USA	https://www.graphpad.com
PyMol 3.1.0	Schrödinger Inc	https://www.pymol.org
ProteinLynx Global Server (PLGS) v3.03	Waters Corporation	N/A
DynamX v3.0	Waters Corporation	N/A
RStudio	RStudio, Inc.	N/A
PyHDX (v0.4.3)	Smit et al.51	N/A
hdxstats	Crook et al.60	N/A

HIGHLIGHTS.

• RTK pseudokinase domains are broadly targetable with protein kinase inhibitors.

• Pseudokinase domains sample both active-like and inactive-like conformation states.

• Modulation of the ErbB3/HER3 pseudokinase domain is limited to its N-lobe.

• Inhibitor modulation of Eph receptor pseudokinases can propagate beyond the N-lobe.