Mass spectrometry methods and mathematical PK/PD model for decision tree-guided covalent drug development

Abstract

Covalent drug discovery efforts are growing rapidly but have major unaddressed limitations. These include high false positive rates during hit-to-lead identification; the inherent uncoupling of covalent drug concentration and effect [i.e., uncoupling of pharmacokinetics (PK) and pharmacodynamics (PD)]; and a lack of bioanalytical and modeling methods for determining PK and PD parameters. We present a covalent drug discovery workflow that addresses these limitations. Our bioanalytical methods are based upon a mass spectrometry (MS) assay that can measure the percentage of drug-target protein conjugation (% target engagement) in biological matrices. Further we develop an intact protein PK/PD model (iPK/PD) that outputs PK parameters (absorption and distribution) as well as PD parameters (mechanism of action, protein metabolic half-lives, dose, regimen, effect) based on time-dependent target engagement data. Notably, the iPK/PD model is applicable to any measurement (e.g., bottom-up MS and other drug binding studies) that yields % of target engaged. A Decision Tree is presented to guide researchers through the covalent drug development process. Our bioanalytical methods and the Decision Tree are applied to two approved drugs (ibrutinib and sotorasib); the most common plasma off-target, human serum albumin; three protein targets (KRAS, BTK, SOD1), and to a promising SOD1-targeting ALS drug candidates.

Fig. 1. Method development for intact protein MS.

Whole blood from a transgenic fALS mouse model (SOD1^G93A) fractionated using chloroform-ethanol without (a–e) and with (b–f) spiked S-XL6 (200 µM). S-XL6 is a kinetic stabilizer that cross-links the SOD1 dimer via two cysteines (Cys111_{subunit A} and Cys111_{subunit B}). Cross-linking forms two disulfide bonds that may be susceptible to cleavage by thiol-disulfide exchange with endogenous thiols (e.g., glutathione). To address this, we compared target engagement without (a, b, green) and with (c, d, purple; e, f, light blue) thiol alkylating (i.e., endogenous thiol blocking) agents iodoacetamide (IAA) and N-ethyl maleimide (NEM). The inset panels show the deconvoluted spectra. Ubiquitin peak eluted at 23.5 min. Alkylating agents did not improve cross-linking yield, indicating that a thiol blocking step was unnecessary in the final method. Mass spectra were obtained using reversed-phase liquid chromatography and quadrupole time-of-flight MS (Agilent 6560 LC-QToF-MS). % TE is defined as percentage of deconvoluted intensity of the drug-protein complex [drug-protein complex intensity / (drug-protein complex intensity + unbound protein intensity)]. Total ion current is represented as a.u. (arbitrary units) and mass is reported in kDa.

Fig. 2. An intact MS-based Decision Tree for covalent drug development.

Rectangles signify objectives and diamonds signify decisions. Blue arrows indicate advancement (go decisions), pink arrows indicate an unsuitable drug candidate, and black arrows indicate re-estimation. Once a covalent drug candidate reaches the green line, it can be advanced into safety pharmacology studies in higher animal species and Good Laboratory Practice (GLP) ADME studies. *Failures at these stages require input from traditional ADME assays. D# = Decision number, LD₅₀ = lethal dose 50, EC₅₀ = Half-maximal effective concentration, MoA = Mechanism of action, METE = Minimally effective target engagement.

Fig. 3. Intact protein LC-MS analysis applied to covalent drug screening.

Covalent drug candidates—disulfiram (green, column 1), cisplatin (light blue, column 2), ebselen (purple, column 3), and S-XL6 (pink, column 4) were chosen for screening as they target SOD1^G93A, a fALS variant of SOD1 protein involved in ALS. (First row) The proposed mechanism of binding and the predicted mass shift for each covalent drug using intact protein LC-MS analysis. (Second row) Control sample for target protein SOD1^G93A. Representative raw and deconvoluted spectra (inset) for each covalent drug. (Third row) Assessment of target engagement of covalent drug using purified SOD1^G93A. a Disulfiram shows the predicted molecular mass at 16006 Da ( + 148 Da shift). *However, it also forms a SOD1^G93A dimer via a non-native disulfide bond with an unknown mechanism, which can result in misfolding and aggregation^,. b Cisplatin shows the predicted molecular mass at 16311 Da ( + 452 Da i.e., 226 × 2). c Ebselen shows a molecular mass of 16133 Da ( + 274 Da shift), which confirms its proposed MoA. d S-XL6 cross-links SOD1^G93A to form a dimer at 31834 Da [15976 Da shift from (15858*2 + 136 – 18 = 118 Da)], that also confirms the proposed MoA. (Fourth row). Specifically, the mechanism of S-XL6 involves (step 1) reversible thiolate disulfide interchange between a protein cysteine and S-XL6, which due to Brönstead characteristics results in ring-opening to expose a terminal sulfenic acid, which (step 2) forms a second disulfide (and crosslink) with a second cysteine via condensation. Evaluation of in vitro target engagement in target tissue homogenate (i.e., transgenic ALS^G93A mouse brain homogenate). Both ebselen and S-XL6 show the predicted mass shifts at 16133 Da and 31834 Da, respectively. (Fifth row) In vivo target engagement (SC RoA for all candidates). Only S-XL6 in vivo dosing was feasible and resulted in successful detection of cross-linked dimer at 31834 Da in brain (1-hr post-dose via SC). A red “no” symbol indicates a “no-go” decision.

Fig. 4. Estimating effective concentration and dose to infer iPK/PD parameters for S-XL6.

S-XL6’s concentration-%TE relationship was established by titration into target tissue homogenate, a minimum effective dose was estimated and tested in vivo. a Titration of covalent drug in whole blood and target tissue homogenate followed by intact protein LC-MS analysis. S-XL6 was spiked at increasing concentrations in whole blood and brain homogenate (target tissue) prepared from fast-line B6SJL-Tg (SOD1*G93A)1Gur/J transgenic mouse model, incubated at 37°C for 1 h, extracted using chloroform-ethanol precipitation, and %TE was analyzed by LC-MS. Dose estimation of S-XL6 was based upon 37% METE. b In vivo administration of covalent binder S-XL6 to achieve effective target engagement in systemic (whole blood) and target organ (brain). Dosed ALS SOD1^G93A mice achieved METE at all predicted doses (10 mg/kg intravenously, 12.5 mg/kg and 30 mg/kg subcutaneously). In vitro and in vivo experiments were performed in singlet and duplicate, respectively. c Intravenous (IV) administration of covalent drug S-XL6 at 10 mg/kg and (d) oral (PO) administration of S-XL6 at 150 mg/kg in transgenic SOD1^G93A mice (n = 2 per RoA). Whole blood was collected at 1, 4, 8 24-, 48-, 72-, and 168-h post-dose (for IV), and up to 72-h for PO, and extracted and analyzed as described above. Area under the curves (AUCs) were estimated for IV and PO curves. Additionally, brain (target) tissue was collected 1-hr post IV dose and processed as described above. K_p[brain] was calculated using %TE values (%TE _brain divided by %TE _blood) was c.a. ~1. For c and d raw %TE data are shown as discrete data points, and the lines illustrate fits obtained to the two term iPK/PD model [$C\left(t\right)=A{e}^{-\alpha t}+B{e}^{-\beta t}$] derived in Supplementary Fig. 2 and described in Supplementary File. Data (c, d) are presented as Mean ± SD. RoA = route of administration, METE = Minimally effective target engagement. Source data are provided as a Source Data file.

Fig. 5. Application of intact protein LC-MS to oncology targets.

Covalent inhibitors sotorasib and ibrutinib targeting the undruggable KRAS^G12C and BTK protein, correspondingly, were chosen to test the scope of the methods presented here. Representative raw and deconvoluted spectra (inset) with drug-related mass shift for each of the protein targets are shown. Together (a) and (b) confirm the expected sotorasib-related mass shift ( + 560.6 Da) using a commercially sourced KRAS^G12C (green). In addition, the mechanism of action (Decision#1, D1) and METE (D2) were confirmed using an in-house purified (c), (d) GNP (blue) and (e), (f) GDP-bound (purple) versions of KRAS, confirming sotorasib selectively binds the inactive GDP-KRAS^G12C, thereby trapping the protein in its inactive conformation. (g), (h) Ibrutinib covalently binds to BTK ( + 438.6 Da) (pink) confirming the MoA and METE criteria. The phosphorylated proteoforms of BTK are delineated with a bracket. Notably, this analysis shows that unmodified BTK as well as BTK phospoproteoforms with 1, 2, and 3 phosphorylation sites fully bind ibrutinib. The unbound mass of BTK (average mass: 77512.5 Da) corresponds to an unknown +88 Da modified proteoform of BTK that cannot bind ibrutinib. Total ion current is represented as a.u. (arbitrary units).

Fig. 6. Plasma protein binding of covalent drug candidates using intact protein LC-MS.

Covalent drugs were incubated with human serum albumin (HSA), the most abundant plasma protein, to test albumin-binding and to provide a go/no go decisions in the covalent discovery workflow. a Raw and deconvoluted (inset) spectra for control HSA (black). b Disulfiram ( + 147 Da) (purple), (c) cisplatin ( + 225 Da) (pink), and (d) ebselen ( + 274 Da) (blue) showed drug-related mass shifts with HSA, which affects their METE in vivo (D5 = No Go). As expected, rationally designed covalent drugs (e) afatinib (green), and (f) ibrutinib (orange), showed minimal binding to HSA. Additionally, (g) sotorasib (dark red) and (h) S-XL6 (red) showed negligible binding to HSA, which with S-XL6 is enabled by the unique ability to reversibly bind off-target lone cysteine residues, while covalently cross-linking closely spaced pairs of cysteines such as SOD1’s. *denotes cysteinylation (average mass: 66564.28 Da), which blocks the binding of cysteine-targeting drugs disulfiram and ebselen, but does not block the binding of cisplatin, presumably due to reaction with additional nucleophilic amino acids. Total ion current is represented as a.u. (arbitrary units).