Reactivities of the Front Pocket Ncap Cysteines in Human Kinases

Abstract

The front pocket (FP) Ncap cysteine is the most popular site of covalent modification in kinases. A long-standing hypothesis associates the Ncap position with cysteine hyper-reactivity; however, traditional computational predictions suggest the FP Ncap cysteines as predominantly unreactive. Here we applied the state-of-the-art continuous constant pH molecular dynamics (CpHMD) to test the Ncap hypothesis. Simulations found that the Ncap cysteines of BTK/BMX/TEC/ITK/TXK, JAK3, and MKK7 are reactive to varying degrees; however, those of BLK and EGFR/ERBB2/ERBB4 possessing a Ncap+3 aspartate are unreactive. Analysis suggested that hydrogen bonding as well as electrostatic interactions drive the reactivity, and their absence renders a Ncap cysteine unreactive. To further test the Ncap hypothesis, we examined the FP Ncap+2 cysteines in JNK1/JNK2/JNK3 and CASK. Our work offers a systematic understanding of the cysteine structure-reactivity relationship and illustrates the use of CpHMD to differentiate cysteines towards the design of targeted covalent inhibitors with reduced chemical reactivities.

Graphical Abstract

INTRODUCTION

Protein kinases are enzymes that catalyze protein phosphorylation reactions, and represent one of the largest gene families with over 500 genes discovered so far.¹ Dysregulation of kinase functions plays significant roles in cancer, immunology, and a host of other diseases.¹ In the past decade, the highest number of approvals by the US Food and Drug Administration (FDA) have gone to inhibitors targeting kinases as compared to other protein families.² While traditional kinase inhibitors are small molecules that bind the ATP-binding site via reversible interactions,¹ targeted covalent kinase inhibitors (TCKIs) are emerging at a rapid pace^3,4 due to the advantages such as prolonged duration of action, enhanced potency, and increased selectivity.^5,6 Starting from a low affinity reversible inhibitor, a TCKI is typically developed by structure-guided incorporation of an electrophilic warhead that covalently links with a nearby nucleophilic side chain.⁵

Owing to the high nucleophilicity of thiolate and the non-catalytic role in kinases, cysteine is the most targeted amino acid in covalent drug discoveries.^1,5,21 According to a survey from 2018,³ over half of the TCKIs are directed at a front pocket (FP) Cys located in the linker as the N-terminal cap (Ncap) of the αD helix (Fig. 1a, 1b). Sequence alignment (by us and others^5,22) showed that 11 kinases in the human kinome possess such a FP Ncap Cys, including BTK/BMX/TEC/ITK/TXK from the Tec family, BLK from the Src family, JAK3 from the JakA family, EGFR/ERBB2/ERBB4 from the EGFR family, and MKK7 from the STE7 family (Fig. 1), all of which are therapeutic targets. BTK/BMX/TEC/ITK/TXK are key components of T-cell receptor signaling and contribute to acting reorganization and cell polarization.^8,23-25 BLK is an oncogene and a potential therapy target in cutaneous T-cell lymphoma.²⁶ JAK3 drives signaling through cytokine receptors, and is associated with various immune-mediated diseases.^27,28 EGFR (or ERBB1 or HER1), ERBB2 (or HER2), and ERBB4 (or HER4) play essential roles in regulating cell proliferation, survival, differentiation and migration, and understanding their roles has fueled the development of targeted therapies for cancer in the past decade.^29-32 MKK7 is associated with multiple myeloma and various neurological diseases.^15,16,33

Figure 1. Human kinases with a FP Ncap or Ncap+2 Cys.

(a) The 15 kinases possessing a FP Ncap or Ncap+2 Cys displayed in the human kinome tree as orange or yellow circles, respectively. The kinome tree was generated with KinMap.⁷ (b) Structure of the BMX kinase domain showing a FP Ncap Cys (orange sphere, PDB ID: 3sxs⁸). (c) Names of the kinases possessing a FP Ncap (orange) or Ncap+2 Cys (yellow), the families and groups they belong to, and the nearby amino acid sequences. The nearby residues making a significant contribution to its pK_a shift are colored blue, magenta, green, or brown (same as in Fig. 2). For JAK3, the second domain which has a crystal structure is studied here.

The FP Ncap Cys in all human kinases have been covalently modified. Most notably, the FDA-approved drug ibrutinib covalently modifies the Ncap Cys in BTK/BMX/TEC/ITK, BLK, JAK3, and EGFR/ERBB4 (Table 1).^22,34,35 The same Cys in EGFR is also targeted by newer covalent drugs such as afatinib, neratinib, acalabrutinib, spebrutinib, mavelertinib, and osimertinib. The FP Ncap Cys in BTK was found to react with a range of chloroacetamide and acrylamide electrophilic fragments.³⁶ The popularity of the FP Ncap Cys in TCKI discovery can be explained by its accessibility (near the ATP binding site) and expected reactivity (see below). A Ncap Asp has been known to stabilize a helix, which was attributed to the favorable interaction between the negative aspartate and the positive helix dipole.^37,38 Similarly, a Ncap Cys was found to be helix stabilizing, which was attributed to the thiolate-helix dipole interaction.^39,40 It was argued that Cys is the least frequent Ncap residue, because it is in the thiolate form prone to chemical modification.³⁹ Indeed, the Ncap Cys in myoglobin has a measured pK_a of 6.5, which is 1.9 or 1.6 units lower than the (mutant) Cys at N2 or N1 position, and 2 units lower than the pK_a of 8.6 for the model Cys in a peptide.⁴⁰ Thus, the consensus is that a Ncap Cys is in the hyper-reactive thiolate form at physiological pH, which we will refer to as the Ncap hypothesis.

Table 1.

Summary of kinases with a FP Ncap or Ncap+2 Cys and the covalent inhibitors^a directed at the location

Kinase	Family	Group	PDB	Residue	PROPKAb	TCKI
FP Ncap Cys
BTK	Tec	TK	3pj3	C481	9.6	ibrutinib, acalabrutinib, zanubrutinib, tirabrutinib,9 orelabrutinib10
BMX	Tec	TK	3sxs	C496	9.9	ibrutinib
TEC	Tec	TK	4hcta	C449	9.2	ibrutinib, ritlecitinibc
ITK	Tec	TK	3miy	C442	10.7	ibrutinib
TXK	Tec	TK	3t9ta	C350	10.4	acalabrutinib
BLK	Src	TK	4mxya	C319	10.6	ibrutinib, acalabrutinib
JAK3	JakA	TK	5lwm	C909	10.3	ritlecitinib11
EGFR	EGFR	TK	4zjv	C797	11.4	afatinib, osimertinib, dacomitinib, neratinib, olmutinib,12 pyrotinib,13 mobocertinib,14 ibrutinib, acalabrutinib
ERBB2	EGFR	TK	3pp0	C805	10.8	acalabrutinib
ERBB4	EGFR	TK	2r4b	C803	11.0	ibrutinib
MKK7	STE7	STE	6qfr	C218	9.1	investigational inhibitors15,16
FP Ncap+2 Cys
JNK1	MAPK	CMGC	2xrw	C116	10.0	investigational inhibitors17
JNK2	MAPK	CMGC	3e7o	C116	10.2	investigational inhibitors17
JNK3	MAPK	CMGC	6emh	C154	10.4	investigational inhibitors18
CASK	CASK	CAMK	3mfr	C100	13.0	unkown

^aInhibitors without references are from the FDA approved ones from the FDA Orange Book (https://www.accessdata.fda.gov/scripts/cder/ob/index.cfm). Those with references are in the last stage of clinical trial; some of them have gained regulatory approval in foreign countries.

^bPROPKA calculations were performed with PROPKA3.¹⁹

^cTemplate structure used for homology modeling with SWISS-MODEL.²⁰

Kinetics experiments demonstrated that the observed (apparent) rate constants of thiol reactions with electrophiles are inversely related to the thiol pK_a’s.^41-44 Thus, it is generally accepted that cysteines with low pK_a’s are more prone to covalent modification due to the increased availability of the nucleophilic thiolate form, and the Cys pK_a can be used as a proxy for reactivity in TCKI design.^1,5,21,39 Curiously, however, according to the popular empirical pK_a prediction program PROPKA,¹⁹ the Ncap Cys in the aforementioned 11 kinases have pK_a’s of 9.1–11.4 (Table 1), suggesting that they are mainly in the thiol form and consequently unreactive or have low reactivities towards electrophiles. Note, despite the fact that the observed thiol nucleophilicity increases with lower pK_a, the intrinsic nucleophilicity may decrease or stay constant depending on the specific molecule.^41,44

Accurate prediction of Cys pK_a’s is challenging using traditional computational methods. For a dataset of 18 proteins, Rowley and coworker showed that the root-mean-square errors (RMSE’s) from the experimental pK_a’s are 3.4–4.7 using the static structure based Poisson-Boltzmann (PB) and empirical PROPKA calculations.⁴⁵ They obtained a RMSE of 2.4–3.2 using the molecular dynamics (MD) based thermodynamic integration (TI) method, which is several orders of magnitude slower than the PB or PROKA calculations. This level of accuracy is however on par with that (RMSE of 2.7) from the null model which assumes the solution (or model) pK_a of 8.6⁴⁶ for all cysteines.

Recently, we developed a GPU-accelerated continuous constant pH MD (CpHMD) method in Amber package⁴⁷ for accurate and rapid prediction of protein pK_a’s based on independent pH⁴⁸ or pH replica-exchange titration⁴⁹ simulations in the GB-Neck2 implicit solvent.⁵⁰ Using replica-exchange CpHMD, which significantly accelerates the pK_a convergence, we scanned the kinome structure database and found that the catalytic (roof) Lys in a dozen of human kinases can become reactive in the rare DFG-out/αC-out inactive conformation.⁵¹ Most recently, we implemented an asynchronous algorithm to allow replica-exchange simulations on a single or any number of GPU cards.⁵² Benchmark study based on 24 proteins, including those with large pK_a downshifts examined by Rowley and coworker⁴⁵ and additional ones with large pK_a upshifts relative to the model value, gave a RMSE of 1.2, which is more than two units lower than the PB or empirical calculations. The accuracy of predicting thiolates at physiological pH is about 81% with the replica-exchange CpHMD, in contrast to the accuracy below 50% with the PB or empirical methods.⁵³

Motivated by the importance of the Ncap Cys in TCKI design and intrigued by the contradiction between the Ncap hypothesis and the high pK_a’s predicted by the empirical method, here we applied the asynchronous replica-exchange CpHMD to determine the pK_a’s of all 11 human kinases that possess a FP Ncap Cys. We found that most Ncap Cys have pK_a’s in the range of 7.7–8.5, thus providing nucleophilic thiolates for the direct thiol-Michael addition with electrophiles at physiological pH; however, surprisingly, the Ncap Cys in EGFR/ERBB2/ERBB4 and BLK remain in the thiol form up to pH 10.5. Analysis of the pH-dependent conformational environment surrounding the Ncap Cys provides insights into the varied reactivities and support to the general base assisted mechanism for the Michael addition with the Ncap Cys in EGFR/ERBB2/ERBB4 and BLK. To further explore the Ncap hypothesis, we also determined the pK_a’s of the Ncap+2 Cys in JNK1/JNK2/JNK3 and CASK. Additionally, our study identified reactive Cys and Lys locations in all 15 kinases, offering new opportunities for TCKI discovery. Together, our findings offer a systematic understanding of the Cys structure-reactivity relationship and demonstrate the utility of CpHMD simulations in aiding the rational design of Cys- and Lys-targeted covalent kinase inhibitors.

RESULTS AND DISCUSSION

CpHMD titration showed varied reactivities for the FP Ncap Cys.

We performed the GPU-accelerated replica-exchange CpHMD titration simulations^48,49,52 to determine the Cys and Lys pK_a values for the aforementioned 11 human kinases possessing a FP Ncap Cys, which is located in the linker as the N-terminal cap of the αD helix (position 52 of the ATP binding site sequence according to KLIFS database⁵⁴). Except for TEC, TXK, and BLK, for which homology models were built using SWISS-MODEL,²⁰ simulations started from the X-ray crystal structures with inhibitors, ions, and crystal water molecules removed. A pH range 6.0-10.5 was used, and each replica lasted 30–50 ns until convergence (for more details see Methods). Considering the inverse relationship between the thiol reactivity and pK_a^41,55 and to facilitate discussion, we group the Cys (or later Lys) in four categories (Table 2): unreactive (pK_a > 9.4), somewhat reactive (pK_a 8.4–9.4), reactive (pK_a 7.4–8.4), and hyper-reactive pK_a < 7.4. Accordingly, the fraction of thiolate (or neutral lysine) state at physiological pH is < 1% (unreactive), 1–10% (somewhat reactive), 10–50% (reactive), and > 50% (hyper-reactive).

Table 2.

Reactivity definitions used in this work

Cys/Lys pKa	Deprot. fraction	Reactivity
< 7.4	> 50%	hyper reactive
7.4–8.4	10–50%	reactive
8.5–9.4	1–10%	somewhat reactive
> 9.4	< 1%	unreactive

Our simulations gave a wide pK_a range from 7.6 to above 10.5 for the FP Ncap Cys, suggesting that its reactivity varies despite the same Ncap location. Among all, the Ncap Cys in BTK/BMX of the Tec family and JAK3 of the JakA family are nearly hyper-reactive, with the calculated pK_a’s around 7.7 (Fig. 2 and Table S1). The Ncap Cys in TEC/TXK/ITK of the Tec family are also reactive, with the calculated pK_a’s around 8.4 (Fig. 2 and Table S1), close to the value (8.6) for the model alanine penta-peptide (AACAA) in solution.⁴⁶ The Ncap Cys in MKK7 is somewhat reactive with a calculated pK_a of 9.1 (Fig. 2 and Table S1). Interestingly, the Ncap Cys in EGFR/ERBB2/ERBB4 of the EGFR family and BLK of the Src family are predicted as unreactive with pK_a values near or above 10.5 (Fig. 2 and Table S1), consistent with our previous work.⁵¹ Noticeably, our calculated pK_a’s of 7.7 and 8.4 for the FP Ncap Cys in BTK and ITK are in agreement with the respective values of 7.7 and 8.5 noted by Zapf and coworkers in the covalent inhibition study of ITK although the source of these pK_a’s was not mentioned.⁵⁶ However, curiously, our simulation showed that EGFR’s Ncap Cys remained protonated in the pH range 7.5–10.5, in disagreement with the pK_a of 5.5 estimated based on the Bromobimane fluorescence labeling experiment.⁵⁷ We will come back to the discussion of this discrepancy.

Figure 2. Predicted pK_a’s and reactivities for the FP Ncap and Ncap+2 Cys in 15 kinases.

The reactivity definitions are given in Table 2. The exact pK_a values for the Ncap Cys in EGFR/ERBB2/ERBB4 and BLK are not given, as the thiol form was dominant even at the highest simulation pH 10.5. The kinase names are shown in different colors based on the environment of the FP Ncap Cys: blue (Ncap+3 Asn), magenta (Ncap+3/2 Asp/Glu and Ncap+2/3 Ser/Arg/Lys), and green (Ncap+3 Asp/Glu). The kinases with a FP Ncap+2 Cys are shown in brown.

Sequence alignment of the 11 kinases showed that the FP Ncap Cys is often followed by an Asn or Asp residue at Ncap+3 position (Fig. 1 b and c). A previous experimental study suggested that the Ncap+3 residue may form hydrogen bond (h-bond) or electrostatic interaction with the Ncap residue.³⁹ Our previous computational work^51,53 demonstrated that the pK_a of a Cys can be significantly shifted by h-bonding with proximal residues. Experimental work on various proteins also suggested the role of local h-bonding and electrostatics in stabilizing Cys thiolates.^40,44,58 Thus, to understand the varied reactivities of the FP Ncap Cys, we binned the 11 kinases into three groups according to the nearby residues that provide stabilization or destabilization of the Ncap thiolate: Asn at Ncap+3 (BTK/BMX/TEC/TXK); Asp/Glu at Ncap+3/2 and Ser/Arg/Lys at Ncap+2/3 positions (ITK, JAK3, and MKK7); and Asp/Glu at Ncap+3 without other nearby stabilizing h-bond or electrostatic partners (EGFR/ERBB2/ERBB4 and BLK). Below we present analysis of the pH-dependent h-bond and electrostatic interactions to elucidate the FP Ncap Cys reactivities.

BTK/BMX/TEC/TXK: FP Ncap thiolate stabilized by the Ncap+3 Asn and other h-bond donors.

BTK/BMX/TEC/TXK of the Tec family contain an Asn residue at the Ncap+3 position relative to the FP Ncap Cys. The calculated pK_a’s of the Ncap Cys in BTK (C481) and BMX (C496) are 7.7 and 7.8, respectively, indicating that they are nearly hyper-reactive according to our definition (Table 2). Trajectory analysis shows that when deprotonated, Cys481 in BTK accepts h-bonds from both the side chain and backbone amide of Asn484 at Ncap+3 as well as the hydroxyl group and backbone amide of Thr410 (Fig. 3a). The pH-dependent occupancy of the h-bond formation and the pH-dependent deprotonation of Cys481 are perfectly correlated, suggesting that the h-bond interactions contribute to the pK_a downshift of Cys481 (Fig. 3b), consistent with our previous findings that thiolates tend to be stabilized by nearby h-bond donors.^51,53 Similarly, the thiolate form of Cys496 in BMX is stabilized by h-bonding with the sidechain and backbone of Asn499 at Ncap+3 as well as the salt-bridge interaction with Arg540 (Fig. S1).

Figure 3. The FP Ncap thiolate is stabilized by hydrogen bonding with Ncap+3 Asn and other residues in BTK.

(a) A zoomed-in view of the structural environment of Cys481 in BTK with the Asn484–Cys481 and Thr410–Cys481 h-bonds shown. A snapshot from simulation at pH 9.0 is used. (b) Deprotonated fraction of C481 at different pH. Curve represents the best fit to the Henderson-Hasselbalch equation. (c) Occupancy of the h-bond formation between Cys481 thiolate and Asn484 (sidechain and backbone) or Thr410 (sidechain and backbone) at different pH. A h-bond is defined by the heavy-atom donor-acceptor distance of 3.5 Å and the donor-hydrogen-acceptor angle of 150°.

The calculated pK_a’s of the FP Ncap Cys in TEC (Cys449) and TXK (Cys350) are about 8.5, suggesting that they are somewhat reactive at physiological pH. In TEC, Cys449 thiolate accepts h-bonds from Asn452 at Ncap+3 and Ser378, as well as a salt bridge with Arg493 positioned at Ncap+44. Cys350 in TXK forms h-bonds with Asn353 at Ncap+3 and salt-bridge with Arg394 at Ncap+44. It is interesting to notice that the Ncap+44 Arg which interacts with the Ncap thiolate is the second Arg of the highly conserved sequence HRDLAARN on the catalytic loop.⁵⁹ It is also noteworthy that the Ncap Cys in TEC and TXK does not form h-bond or salt-bridge with Ncap-2 or Ncap-3 residues such as Asn, Ser, Glu, and Arg (Fig. S1).

ITK, JAK3, and MKK7: FP Ncap thiolate destabilized by Ncap+3/2 Asp/Glu but stabilized by Ncap+2/3 Ser/Arg/Lys or Ncap+1 Lys/Thr.

Several of the Ncap Cys are in close proximity to an anionic side chain (Asp or Glu), which would increase the Cys pK_a due to electrostatic repulsion. However, h-bonding and salt-bridge formation with the nearby residues can stabilize the thiolate form and thereby decreasing the Ncap Cys pK_a. These opposing effects are experienced by the Ncap Cys in ITK (Cys442), JAK3 (Cys909), and MKK7 (Cys218), resulting in the calculated pK_a’s of 8.4, 7.8, and 9.1, respectively.

Simulations of ITK showed that the deprotonated Cys442 can accept h-bonds from the side chain hydroxyl and backbone amide of Ser444 at Ncap+2 and the backbone amide of Asp445 at Ncap+3, as well as form a salt-bridge with Arg486 positioned at Ncap+44 (Fig. 4a). The excellent correlation between the increasing occupancies of h-bond (or salt bridge) formation and the increasing degree of Cys deprotonation suggests that these interactions stabilize the thiolate form, which compensates for the electrostatic repulsion from the nearby carboxylate group of Asp445 (Fig. 4b and c). As a result, the calculated pK_a of Cys442 is 8.4, indicating that it is somewhat reactive towards electrophiles.

Figure 4. Stabilizing interactions for the FP Ncap Cys thiolate in ITK.

(a) Zoomed-in view of the structural environment of Cys442 in ITK taken from the simulation at pH 9.5. (b) pH-dependent deprotonated fraction of the FP Cys442. Curve represents the best fit to the Henderson-Hasselbalch equation. (c) pH-dependent occupancies of the h-bond formation between Cys442 thiolate and the sidechain and backbone of Ser444 (purple), backbone of Asp445 (yellow), and the side chain of Arg486 (light pink).

The compensating interactions are somewhat different for JAK3 and MKK7, in which an attractive electrostatic interaction with a nearby cationic sidechain is also a major contributor. In JAK3, the Ncap Cys909 thiolate is stabilized by salt-bridge interactions with Arg953 at Ncap+44 and h-bonding with Tyr904 at Ncap-5 below pH 8 (Fig. S2). As a result, despite the repulsion with Asp912 at Ncap+3, the pK_a of Cys909 is downshifted to 7.8, lower than the corresponding Ncap Cys in ITK.

In MKK7, the sidechain-to-sidechain salt-bridge and sidechain-to-backbone h-bond interactions with Lys221 at Ncap+3 as well as the h-bond formation with Ser144 are the stabilizing forces for the thiolate state of the Ncap Cys218 (Fig. S2); these forces compensate for the destabilization due to Glu220 at Ncap+2 position. Overall, the number of stabilizing h-bond/salt bridge interactions with the Ncap Cys in MKK7 is less than those in ITK or JAK3, which may explain why the Ncap Cys pK_a in MKK7 (9.1) is the highest among the three kinases.

EGFR/ERBB2/ERBB4 and BLK: FP Ncap thiolate destabilized by Ncap+3 Asp/Glu and desolvation without compensating interactions.

The Ncap Cys is buried in EGFR/ERBB2/ERBB4 (of the EGFR family) and BLK (of the Src family) to a similar extent as in the aforementioned kinases; however, there is no nearby h-bond donor or attractive electrostatic interaction partner. Furthermore, the Ncap+3 position is occupied by an anionic Asp (or Glu in ERBB4) which can potentially destabilize the thiolate form of the Ncap Cys. Thus, it is not surprising that the calculated pK_a’s are above 10.5 (the highest simulation pH), suggesting that the Ncap Cys in these kinases are unreactive towards electrophiles.

Curiously, the Ncap Cys in all four kinases have been targeted by the FDA-approved covalent drugs. Most notably, Cys797 in EGFR was the target of the first TCKI,⁵ and it was subsequently targeted by numerous investigational and FDA-approved inhibitors including ibrutinib, afatinib, and neratinib (Table 1). Analysis of the X-ray structures of EGFR (e.g., PDB ID 4zjv) reveals that the Cys797 thiol may donate a h-bond to the carboxylate group of the Ncap+3 Asp800 (Fig. 5). Similarly, the X-ray structures of ERBB2/ERBB4 show that the Ncap thiol sulfur can be within 3.5 Å to the nearest carboxylate oxygen of the Ncap+3 Asp (Table S3).

Figure 5. The FP Ncap Cys thiol may donate a h-bond to the Ncap+3 aspartate in EGFR.

Zoomed-in view of the structural environment of the Ncap Cys797 in EGFR from the apo X-ray crystal structure (PDB: 4zjv³²). The distance between the thiol sulfur and the nearest carboxyl oxygen of Asp800 is 3.3 Å. The potential h-bond is indicated as a dashed line.

The high (calculated) pK_a of the Ncap and the potential formation of a thiol-aspartate h-bond in the holo structure together suggest that the Michael addition reactions with EGFR/ERBB2/ERBB4 and BLK may proceed via a base-assisted mechanism, in which the Ncap+3 Asp serves as a general base to deprotonate the Ncap Cys in the first stage of the reaction. Indeed, a recent combined quantum mechanical (QM)/MM simulation of EGFR⁶⁰ demonstrated that in the proximity of a covalent inhibitor carrying an acrylamide warhead, the proton on the Ncap Cys797 thiol can be transferred to the carboxylate of the Ncap+3 Asp, resulting in a stable thiolate/Asp-COOH pair, from which the thiolate then attacks the Cβ of acrylamide generating a carbanion before being reprotonated by the Ncap+3 Asp-COOH to result in the thiol adduct.⁶⁰ Our result is consistent with this base-assisted mechanism, although the proton transfer energetics may be underestimated in the QM/MM study⁶⁰ based on the SCC-DFTB method.⁶¹ We should also note that in our apo simulations of EGFR/ERBB2/ERBB4 and BLK, the Ncap thiol rarely formed a h-bond with the Ncap+3 Asp/Glu, supporting the hypothesis that a proton transfer may only occur in the presence of inhibitor due to the restricted motion of the Cys and Asp sidechains. Another possibility for the covalent modification of an unreactive Cys is to make use of a base group adjacent to the Michael acceptor warhead. The structure-activity data of EGFR inhibitors suggested that a basic amine, e.g., dialkylamino or pyrrolidine group can abstract a proton from the Cys thiol and thereby catalyzing the reaction.^30,62,63

Other reactive Cys and Lys locations.

Since in the CpHMD simulations all Cys and Lys were allowed to titrate simultaneously, we were able to determine the pK_a values. Table 3 lists the locations of all reactive Cys and Lys predicted by our simulations. Interestingly, a Cys on β4 sheet (Ncap-17/18) exists in 10 kinases and is found reactive or hyper-reactive in BMX/TEC/TXK of the Tec family and ERBB2/EGFR of the EGFR family. The positions of other reactive Cys are proximal to the ATP binding site and have already been targeted by clinical and investigational TCKIs. These locations include the P loop (Cys147 of MKK7),⁶⁴ DFG-1 (Cys276 of MKK7),⁶⁵ and the A loop (Cys891 of ERBB4).⁶⁶ The local conformational environment of the Ncap Cys is not affected by the conformation of the DFG-motif (in or out) or αC helix (in or out). As such, the calculated pK_a is insensitive to the variation of the kinase conformation. For example, using five X-ray structures (in different DFG/αC conformations according to KLIFS⁵⁴), the calculated pK_a of the Ncap Cys481 in BTK is 7.8–8.4 (7.8 with PDB 3pj3, 8.3 with PDB 6o8i, 8.4 with PDB 3piy, 8.3 with PDB 5p9j, and 7.9 with PDB 3oct). In contrast, although the roof catalytic Lys has a pK_a above 10.4 in the DFG-in or αC-in states, the pK_a value can be significantly downshifted when the kinase adopts the DFG-out/αC-out conformation⁵¹). Our simulations predicted that the roof Lys in BTK (Lys430, PDB 3pj3),²⁵ BMX (Lys445, PDB 3sxs),⁸ and JAK3 (Lys855, PDB 5lwm)⁶⁷ are reactive. Interestingly, although the X-ray structures for these kinases show a DFG out state, αC-Glu in BTK and JAK3 is within a salt-bridge distance to the roof Lys. During the simulations, however, the Glu-Lys salt-bridge disrupted, resulting in the sampling of the αC-Glu out conformations. The facile transition between the αC-Glu in and out positions is consistent with our previous explicit-solvent simulations of the c-Src kinase.⁶⁸ The Lys at the DFG-3 position has not been targeted before but it is also in the ATP binding site, and the simulations predicted that the DFG-3 Lys in ERBB2 and ERBB4 are reactive, thus providing a new opportunity for covalent targeting. We note that the P-loop and DFG-1 Cys as well as the roof Lys have been previously targeted in other kinases.³

Table 3.

Summary of the predicted reactive Cys and Lys in the 15 kinases discussed in this work^a

Kinase	Family	Cys	Location	Reactivity	Major contributors	Lys	Location	Reactivity
FP Ncap Cys
BTK	Tec	C481	front pocket	reactive	N(+3), T(−71)	K430	roof	reactive
BMX	Tec	C479	β4 sheet	reactive		K445	roof	somewhat
		C496	front pocket	reactive	N(+3), R(+44)
		C549	β8 sheet	reactive
TEC	Tec	C406	αC helix	reactive
		C432	β4 sheet	reactive
		C449	front pocket	reactive	S(−71), N(+3), R(+44)
TXK	Tec	C333	β4 sheet	hyper
		C350	front pocket	somewhat	N(+3), R(+44)
		C402	β8 sheet	somewhat
ITK	Tec	C442	front pocket	reactive	S(+2), D(+3), R(+44)
JAK3	JakA	C909	front pocket	reactive	Y(−5), R(+2), R(+44)	K855	roof	hyper
		C1028	αG helix	reactive
MKK7	STE7	C147	P loop	hyper
		C218	front pocket	reactive	S(−74), K(+3)
		C276	DFG-1	reactive
		C341	loop (αF/αG)	hyper
EGFR	EGFR	C775	loop (αC/β4)	reactive
		C781	β4 sheet	reactive
ERBB2	EGFR	C789	β4 sheet	hyper		K860	DFG-3	reactive
ERBB4	EGFR	C891	A loop	reactive		K858	DFG-3	reactive
BLK	Src	C460	loop (αG/αH)	somewhat
FP Ncap+2 Cys
JNK1	MAPK	C116	front pocket	reactive	Q(+1), S(+39)
		C245		reactive	αH helix
JNK2	MAPK	C116	front pocket	reactive	Q(+1), S(+39)	K153	β7 sheet	somewhat
JNK3	MAPK	C154	front pocket	hyper	S(−82), Q(+1), S(+39)
		C283		reactive	loop (αG/αH)
CASK	CASK	C100	front pocket	somewhat

^aThe reactivity levels are defined in Table 2. Residues or locations that have been previously targeted by covalent inhibitors are labeled in bold font. Residues that make significant contributions to the pK_a shift of the Ncap Cys are listed with the sequence positions (relative to Ncap) given in the parenthesis. The calculated pK_a’s of all listed residues are given in Table S1.

Comparison to other pK_a calculations.

The CpHMD predicted pK_a’s are in stark contrast to the structure-based calculations based on the empirical PROPKA approach and on the solutions to the PB equation. Most Ncap Cys pK_a’s predicted by the PB calculation (based on the H++ server) are between 9.0 and 11.4, similar to those from PROPKA (Table S2). Most notably, while CpHMD simulations gave the pK_a’s of about 7.7–7.8 for the Ncap Cys in BTK, BMX, and JAK3 (the lowest among the 11 kinases, Table S1), the PROPKA calculated pK_a’s are 9.6, 9.9, 10.3, and 9.0, and those from PB are 9.6, 9.9, 8.8, and 9.0, respectively (Table S2). The large discrepancy, which has been noted in our previous benchmark studies,^49,51 can be attributed to the h-bonds and electrostatic interactions that are absent in the X-ray structures but emerge from the pH-dependent conformational dynamics captured by CpHMD and not by structure-based calculations.⁶⁹ For example, in BTK, the sidechain hydroxyl and backbone amide of Thr410 donate h-bonds to stabilize the Ncap Cys481 in a pH-dependent manner (Fig. 3); however, these interactions are absent in the X-ray structure and therefore not accounted for by the PROPKA or PB calculations. Similarly, in ITK, the stabilizing interactions with Ser444 and Arg486 (Fig. 4) are absent in the X-ray structure and therefore the pK_a’s from the PROKA and PB calculations are over 10.4, two units higher than the value of 8.4 from the CpHMD titration.

The TI simulations and alternative constant pH methods based on the hybrid-solvent Monte-Carlo (MC)/Md and non-equilibrium MD/MC schemes of Rowley and coworker⁷⁰ gave the pK_a’s of 10.4–13.0 for the Ncap Cys in BTK/BMX/ITK, JAK3, and 11.1–13.5 for EGFR (Table S2). Except for EGFR, these results are in disagreement with our CpHMD simulations which suggested that all four Ncap Cys are reactive with the pK_a’s of 7.7–8.4. With regards to the Ncap Cys797 in EGFR, Rowley and coworker calculated its pK_a with either protonated or deprotonated Asp800. Interestingly, the calculated pK_a was above 10 in both cases.⁷⁰

The FP Ncap+2 Cys in JNKs and CASK.

JNK1/JNK2/JNK3 of the MAPK family and CASK of the CASK family possess a Cys similar in position as the Ncap Cys discussed so far. This Ncap+2 Cys, Cys116 in JNK1/JNK2, Cys154 in JNK3, or Cys100 in CASK is the second N-terminal residue in the αD helix, and has an Ile in the Ncap+3 position instead of Asn/Asp/Glu. The calculated pK_a’s for Cys116 in JNK1/JNK2 and Cys154 in JNK3 are 7.5, 8.0, and 6.3 respectively, while that for Cys100 in CASK is 9.0 (Table S2). Analysis showed that in JNK1 and JNK2, Cys116 thiolate can accept h-bond formation with Gln117 and Ser155. In JNK3, Cys154 thiolate is additionally stabilized by the h-bond formation with Ser72, which may explain the lower pK_a (hyper-reactivity) of Cys154 compared to Cys116 in JNK1/JNK2. As to CASK, the deprotonation of Cys100 is mainly correlated with the salt-bridge formation with Arg302 and to a lesser extent with the h-bond formation with Asn299. The number of h-bonds formed by Cys100 in CASK is lower than that by Cys116 in JNK1/JNK2 or Cys154 in JNK3. As a result, the pK_a upshift due to desolvation is not offset, leading to a higher pK_a value of Cys100 as compared to Cys116 or Cys154 in JNKs.

CONCLUDING DISCUSSION

Ncap Cys has long been hypothesized as hyper-reactive due to the thiolate-helix dipole interactions.^39,40 In support of this hypothesis, the Ncap Cys in all human kinases have been covalently targeted. Our data demonstrated that the Ncap position alone may not stabilize the thiolate form making the Cys reactive at physiological pH. CpHMD simulations confirmed the reactivities of the Ncap Cys in BTK/BMX/TEC/TXK/ITK and JAK3, with the calculated pK_a’s ranging from 7.7 to 8.5. The calculated pK_a’s of BTK and ITK are very similar to those noted in an experimental study of ITK⁵⁶ (personal communication with the Pfizer team that performed the study).⁷¹ In comparison, the calculated pK_a is 9.1 for the Ncap Cys in MKK7, suggesting a lower reactivity compared to the aforementioned Tec family kinases and JAK3.

Surprisingly, the calculated pK_a’s are above 10.4 for the Ncap Cys in EGFR/ERBB2/ERBB4 and BLK, suggesting that the thiol form is predominant at physiological pH despite the Ncap position. Curiously, the calculated pK_a for EGFR is 5 units higher than that from the pH-dependent bromobimane-thiol reaction rate.⁵⁷ This discrepancy cannot be explained by the lack of polarization of additive force field based CpHMD simulations, which may lead to a small underestimation of the helix dipole⁷² and consequently the pK_a downshift due to the thiolate-dipole interaction. Base-catalyzed mechanism is common for thiol-Michael addition reactions in the absence of a deprotonated thiol.^73,74 Considering that EGFR and ERBB2/ERBB4 have a Ncap+3 Asp and the carboxyl group is within the h-bond distance to the Ncap thiol in some crystal structures, we speculate that the Ncap+3 Asp may act as a general base. Similarly, a QM/MM study suggested that in the proximity of an inhibitor, a Cys… Asp h-bonded pair may form, from which the thiolate then attacks the acrylamide warhead on the inhibitor.⁶⁰ Therefore, the Cys797 pK_a determined through the bromobimane-thiol reaction⁵⁷ might reflect the Cys797… Asp800 pair in the presence of bromobimane. However, we note that the proton transfer energy in the aforementioned QM/MM study⁶⁰ may be underestimated by the SCC-DFTB method.⁶¹

Other base-assisted mechanisms may also explain the thiol-Michael addition reactions. Several experimental studies of the structure-activity relationships of EGFR inhibitors suggested that an amine group proximal to the electrophilic warhead may act as a general base to abstract the proton from the Ncap thiol yielding the nucleophilic thiolate.^30,62,63 Additionally, a nearby water may in principle serve as a general base. Finally, in the absence of a general base, a direct addition mechanism is conceivable, whereby in the first step the carbonyl oxygen abstracts the proton from the Ncap thiol, as proposed by a most recent QM/MM study of BTK modification by ibrutinib assuming the Ncap Cys is protonated.⁷⁵

A corollary to the Ncap thiolate hypothesis is that a Cys placed further down the helical sequence, e.g., at N2 or N3 position, should have a higher pK_a. The CpHMD simulations of JNK1/JNK2/JNK3 and CASK showed that this needs not be the case. The calculated pK_a’s of the Ncap+2 Cys in JNKs range from 6.3 to 8.0, consistent with the fact that they have all been covalently modified.^17,18 The calculated pK_a of Ncap+2 Cys in CASK indicates it is somewhat reactive; no covalent inhibitor targeting this position has been reported yet.

Our analysis showed that solvent exposure of the Ncap or Ncap+2 Cys is very low, corroborating our previous finding that all reactive cysteines are largely buried⁵³ and consistent with an early bioinformatics survey which suggests that cysteine is the least exposed residue in proteins.⁷⁶ However, the h-bond and electrostatic environment of the Ncap and Ncap+2 Cys can be very different, which gives rise to the varied Cys pK_a’s. Specifically, interaction with the nearby h-bond donors is a major determinant for the pK_a downshift, while solvent exclusion is a major determinant of the pK_a upshift, consistent with our previous findings in the context of other proteins.^51,53 Notably, the presence of a h-bond donor at the Ncap+3 position (e.g., Asn) provides stabilization to the thiolate form in BTK/BMX/TXK. By contrast, the presence of an anionic sidechain (e.g. Asp or Glu) at the Ncap+3 position destabilizes the thiolate form. Although ITK, JAK3, BLK, and EGFR/ERBB2/ERBB4 all have an Ncap+3 Asp next to the Ncap Cys, only ITK and JAK3 have favorable h-bond interactions that stabilize the thiolate form. As a result, the Ncap Cys in ITK and JAK3 are reactive, whereas those in BLK and EGFR/ERBB2/ERBB4 are unreactive.

A unique capability of CpHMD simulations is to capture the pH-dependent formation of h-bond and electrostatic interactions that are either nascent or completely absent in the static (X-ray) structure but can nonetheless significantly impact the protonation state. Such interactions with the Ncap or Ncap+2 Cys cannot be accounted for by traditional structure-based PB or empirical calculations; the predicted high pK_a’s are dominated by the large contribution from the desolvation penalty.

The caveats of the current CpHMD approach include the neglect of polarization, which may lead to a small underestimation of the Cys pK_a downshift due to the underestimation of helix dipole as well as the neglect of a possible proton transfer between Cys and a nearby general base, which can be located on the protein (e.g. Ncap+3 Asp) or inhibitor. CpHMD simulations with the holo structures and QM/MM studies will be conducted in the future to test the hypothesis that a base-assisted mechanism underlies the Michael addition reactions with the Ncap Cys in EGFR/ERBB2/ERBB4 and BLK. Taken together, our work offers a systematic understanding of the cysteine structure-reactivity relationship and demonstrates the utility of CpHMD simulations in differentiating cysteine sites to assist in the design of targeted covalent inhibitors with reduced chemical reactivities. These capabilities would also allow the CpHMD tool to complement the chemical proteomic platform^36,77 in the discovery of new druggable sites in the proteome.

Experimental

Structure Preparation.

Except for TEC, TXK, and BLK, the initial structures were taken from the Protein Data Bank (PDB): BTK (3pj3),²⁵ BMX (3sxs),⁸ ITK (3miy),²⁴ JAK3 (5lwm),⁶⁷ EGFR (4zjv),³² ERBB2 (3pp0),³¹ ERBB4 (2r4b),³⁰ JNK1 (2xrw),⁷⁸ JNK2 (3e7o),⁷⁹ JNK3 (6emh),⁸⁰ and CASK (3mfr).⁸¹ The structures of TEC, TXK, and BLK were built using SWISS-MODEL.²⁰ For each structure, the open source tool pdbfixer from the OpenMM package⁸² was used to remove non-protein heavy atoms and to add missing residues, atoms, acetylated N terminus and amidated C terminus, disulfide bonds (if present), and hydrogen atoms. Lys and Cys had one dummy hydrogen, while His had two dummy hydrogens.

Simulation Protocol.

The generalized Born (GB) based CpHMD simulations^48,49 were performed using the pmemd engine of AMBER18 (AMBER 2018).⁴⁷ The pH replica-exchange titration simulations were performed on the GPUs using our asynchronous replica-exchange implementation.⁵² The proteins were represented by the ff14sb protein force field,⁸³ and solvent represented by the GBNeck2 (igb=8) implicit solvent model⁵⁰ with mbondi3 intrinsic Born radii and 0.15 M ionic strength. Modifications to the intrinsic Born radii were made for His and Cys sidechains, as discussed in our previous work.^48,49,51 Prior to the CpHMD runs, energy minimization was performed in the GB solvent using 500 steps of steepest descent and 500 steps of the conjugate gradient method with a harmonic force constant of 50 kcal/mol/Ų applied to the heavy atoms. Following minimization, four stages of equilibration were performed at 300 K using 2000 steps for each stage. A harmonic force constant applied to the heavy atoms was gradually decreased from 5.0, 2.0, 1.0, to 0 kcal/mol/Ų.

One set of pH replica-exchange CpHMD simulations was carried out for each protein. For BTK, ITK, and ERBB4, 8 independent pH titrations were also performed at pH conditions from 6.0 to 9.5, with a 0.5 unit interval. Each pH condition was run for 10 ns. These simulations gave very similar results as the replica-exchange runs. The pH conditions in the replica-exchange simulations were from 7.0 to 10.5, except BTK which had a pH range of 6.0–9.0 and MKK7 which had a pH range of 7.0–11.0. The pH conditions were placed at every 0.5 unit, except for MKK7 which had an extra pH condition of 9.25. Bonds involving hydrogen atoms were constrained with the SHAKE algorithm allowing a 2-fs timestep. Exchanges between adjacent pH replicas were attempted every 1000 steps (2 ps) according to the Metropolis criterion.⁸⁴ Each replica lasted 50 ns (aggregate time of 400 ns) for BTK, BMX, TXK, ITK, JAK3, MKK7, and 30 ns (aggregate time of 240 ns) for the other kinases (these proteins had faster convergence). Each replica underwent Langevin dynamics at 300 K with an effectively infinite cutoff (999 Å) for nonbonded interactions. All side chains of His, Cys, and Lys were allowed to titrate, with the experimental model alanine penta-peptide pK_a’s of 6.5, 8.5, and 10.4, respectively.⁴⁶ If not stated, all settings were identical to our previous work.^48,51,53

pK_a calculations.

The unprotonated fraction (S) of a titratable residue at different pH values was calculated using the definitions of the protonated (λ < 0.2) and deprotonated (λ > 0.8) states, as in our previous work.^48,49,53 The pK_a was calculated by fitting the S values to the generalized Henderson Hasselbalch (HH) equation, S = 1/(1 + 10^n(pK_a−pH)), where n is the Hill coefficient.

The bootstrap method was used to estimate the errors of the calculated pK_a’s.⁴⁹ The S values based on the 10-ns simulations at different pH were combined to generate mⁿ sets of titration data, where m is the number of the 10 ns simulations per pH and n is the number of pH conditions. For each set of titration data (S vs pH), a pK_a value was calculated. An average and standard deviation of all the calculated pK_a’s are reported (Fig. 2 and Table S1).

ABBREVIATIONS

CpHMD

continuous constant pH molecular dynamics

FDA

U.S. Food and Drug Administration

FP

front pocket

GB

Generalized Born

h-bond

hydrogen bond

HH

Henderson-Hasselbalch

MD

molecular dynamics

MC

Monte-Carlo

MM

molecular mechanics

Ncap

N-terminal cap

PB

Poisson-Boltzmann

RMSE

root-mean-square error

TCKI

targeted covalent kinase inhibitor

TI

thermodynamic integration

QM

quantum mechanics