Introduction

Over the past decade, covalent drug discovery has undergone a surge of interest driven by the approvals of targeted covalent drugs and the rise of technological advancements accelerating covalent drug development. Previous safety concerns about off-target reactivity and toxicity steered researchers away from covalent mechanisms1. However, covalent inhibitors such as ibrutinib and osimertinib have achieved impressive efficacy and safety in patients with durable engagement of their targets. Moreover, the discovery of KRASG12C covalent inhibitors reveals the potential of covalency to target traditionally undruggable proteins with shallow pockets2,3. Ibrutinib and most other targeted covalent inhibitors were developed by grafting an electrophile onto a known high-affinity ligand. Building on a disulfide-based tethering method of fragment discovery4, a newer electrophile-first approach starts with small fragments capable of forming covalent bonds with proteins to identify weak modifiers that are subsequently optimized. This fragment-based approach was used in the discovery of sotorasib and covalent inhibitors for other targets beyond KRAS5,6,7,8,9,10,11,12.

In covalent hit-to-lead cascades, mass spectrometry (MS) often plays an instrumental role by enabling direct detection of covalent protein-ligand adducts. MS does not require a competitive probe and is not limited to functional binding events. Thus, MS can be used in programs targeting poorly characterized proteins with no existing chemical probes and can be used to develop site-specific ligands that bind at sites that are non-functional, as is the case for finding novel binders to E3 ligases. In addition, compared to fluorescence or luminescence readouts, MS is less susceptible to interference and false positives13. MS can also determine binding stoichiometry and confirm sites of modification using enzymatic cleavage or top-down MS/MS14.

Intact protein MS (iMS) is frequently employed to screen electrophilic fragment libraries and characterize covalent modifiers in dose-response (DR) and time-course (TC) format12,15. iMS is commonly performed via liquid chromatography coupled to MS (LC-MS), though the practical utility of conventional LC-MS workflows has been limited by modest sample throughput. Pairing Time-of-Flight (ToF) MS with an Agilent RapidFire system improves acquisition time from minutes to seconds16, and additional analytical advancements have led to techniques such as infrared matrix-assisted laser desorption electrospray ionization (IR-MALDESI), with reported sample rates as high as 1.5–22 Hz17.

Despite the advantages of MS for covalent drug discovery, full characterization of the potency of covalent compounds involves characterizing both time- and dose-dependent behavior, which can be resource-intensive. For reversible drugs, equilibrium constants (KD) or IC50 values are often used to drive SAR. However, the time-dependent mechanism of irreversible modifiers makes IC50 values inadequate; an IC50 measurement taken at five minutes might be very different from one taken at two hours, and multiple researchers have written of the limitations of relying on IC50 alone to guide SAR18,19,20. A more rigorous assessment of irreversible modifiers is the ratio kinact/KI, a time- and concentration-independent inactivation rate constant that incorporates a covalent modifier’s reactivity towards the target (kinact) and factors influencing kinact, as well as its reversible affinity (KI)14,15,16.

Irreversible covalent binding occurs in two steps as described in Eq. (1), where P, L, P:L, and PL denote protein, ligand, noncovalent binary complex, and covalent binary complex, respectively.

$${{\rm{P}}}+{{\rm{L}}}{\rightleftharpoons }^{{K}_{{{\rm{I}}}}}{{\rm{P}}}:{{\rm{L}}}{\longrightarrow }^{{k}_{{{\rm{inact}}}}}{{\rm{PL}}}$$
(1)

kinact/KI allows for the relative ranking of covalent binding efficiencies, even with low reversible affinity21. It is also the preferred means of predicting in vitro and in vivo target occupancy and therapeutic effect20. Furthermore, the ratio has demonstrated greater accuracy in predicting drug-drug interactions and off-target effects compared to IC5022,23; guidance from the U.S. Food and Drug Administration and European Medicines Agency now advises investigating the inhibition mechanism and inactivation kinetics to evaluate the risk of drug-drug interactions24,25.

MS-enabled programs traditionally obtain kinact/KI values with a full iMS DRTC, a resource-intensive experiment involving a two-dimensional matrix varying time on one axis and dose on the other, leading to a sample size of n doses multiplied by n time points for a single compound. After measuring occupancy across a range of time points, the observed rate of inactivation (kobs) values estimated with Eq. (2) can be plotted as a function of inhibitor concentration and fit to determine kinact/KI with Eq. (3).

$$\%{{{\rm{Total\; Occupancy}}}}=100\left(1-{e}^{-{k}_{{{\mathrm{obs}}}}t}\right)$$
(2)
$${k}_{{{\mathrm{obs}}}}=\frac{{k}_{{{\mathrm{inact}}}}\left[{{\mathrm{L}}}\right]}{{K}_{{{\rm{I}}}}+\left[{{\mathrm{L}}}\right]}$$
(3)

Unfortunately, despite offering the most accurate determination of covalent potency, a full DRTC is not practical as a weekly SAR tool due to the protein consumption, instrument time, and complicated data analysis involved; a typical SAR campaign with around 35 compounds a week would require six 384-well plates for a full 8 × 8 DRTC because a single 384-well plate can accommodate only six compounds in this format.

Here, we report a practical alternative: the diagonal DRTC (dDRTC) method, which provides accurate measurement of kinact/KI and is sufficiently high-throughput for routine SAR for a hit-to lead optimization campaign. The number of samples is reduced from n x n samples for a full checkerboard down to just n samples in the diagonal slice, which is the same number of samples as a conventional IC50 experiment. Moreover, dDRTC data analysis can easily be automated by applying standard curve-fitting approaches commonly used during SAR campaigns. The method also provides superior coverage of dose-time space compared to DR or TC under the same assay conditions, and this expanded dynamic range offers valuable positive and negative SAR insights to medicinal chemists.

In this work, we acquire iMS data in both the traditional full DRTC and dDRTC format for twelve covalent fragments and lead-like modifiers of KRASG12C and previously unliganded E3 ligase complexes, SKP1-FBXW7 and SOCS2-EloB-EloC. In addition, we run dozens of simulations in KinTek Global Kinetic Explorer26 for a range of kinact and KI values to understand the optimal conditions where the dDRTC method delivers accurate and reliable kinact/KI values. Our results demonstrate reproducible, quantitative determination of kinact/KI values over more than three orders of magnitude while using eight-fold fewer resources, including both time and reagents.

Results

We sought to develop a streamlined method for obtaining kinact/KI that would be more amenable to routine SAR studies compared to a full DRTC. Our aim was to use a common DR curve format with the y-axis as our measurement (% occupancy quantified by iMS) and the x-axis with the units of the desired endpoint of kinact/KI (M−1s−1) and utilize only the diagonal slice of the checkerboard (Fig. 1). This approach is related to drug combination and drug synergy screens, where checkerboards are used to compare a dose range of one drug vs. another; the diagonal slice of a full checkerboard is the most informative sample of the checkerboard27,28.

Fig. 1: A schematic representation of the diagonal dose-response time-course (dDRTC) method to determine kinact/KI.
Fig. 1: A schematic representation of the diagonal dose-response time-course (dDRTC) method to determine kinact/KI.The alternative text for this image may have been generated using AI.
Full size image

For an n x n checkerboard with varying time on one axis and varying dose on another, only the diagonal sampling (in color) is measured. After testing samples in intact protein MS, an occupancy curve can be fit to determine OC50, the dose and time which yields 50% occupancy, which is then converted to kinact/KI. Y = % occupancy, and X = the inverse product of dose and time in units of kinact/KI (M−1s−1).

Derivation of the relationship between OC50 and k inact/K I

We sought to provide the theory to justify the dDRTC approach and derive a relationship between OC50 and kinact/KI both in units of (M−1s−1). Building on derivations by Copeland et al.29, we established a relationship between OC50 and kinact/KI beginning with the following pseudo first-order rate Eq. (4) under the assumption \([{{\mathrm{L}}}] > [{{{\rm{P}}}}]\), which generally holds true in the early stages of covalent drug discovery where protein concentrations are typically 1 µM or less:

$${[{{{\rm{PL}}}}]}_{t}={[{{{\rm{PL}}}}]}_{{eq}}\left(1-{e}^{-{k}_{{{\mathrm{obs}}}}t}\right)$$
(4)

[PL]t is the %PL (i.e., covalent occupancy) at time t and [PL]eq is 100% maximum occupancy. We converted Eq. (4) as follows:

$$\%{{{{\rm{PL}}}}}_{t}=100\left(1-{e}^{-{k}_{{{\mathrm{obs}}}}t}\right)$$
(5)

With \({k}_{{{\mathrm{obs}}}}=\frac{{k}_{{{\mathrm{inact}}}}[{{\mathrm{L}}}]}{{K}_{{{\mathrm{I}}}}+[{{\mathrm{L}}}]}\) from Eq. 3,

$${\%{{\mathrm{PL}}}}_{t}=100\left(1-{e}^{-\frac{{k}_{{{\mathrm{inact}}}}\left[{{\mathrm{L}}}\right]}{{K}_{{{\mathrm{I}}}}+\left[{{\mathrm{L}}}\right]}t}\right)$$
(6)

When \([{{\mathrm{L}}}]\ll {K}_{{{\mathrm{I}}}}\),

$${\%{{\mathrm{PL}}}}_{t}=100\left(1-{e}^{-\frac{{k}_{{{\mathrm{inact}}}}}{{K}_{{{\mathrm{I}}}}}\left[{{\mathrm{L}}}\right]t}\right)$$
(7)
$${e}^{-\frac{{k}_{{{\mathrm{inact}}}}}{{K}_{{{\mathrm{I}}}}}\left[{{\mathrm{L}}}\right]t}=1-\frac{{\%{{\mathrm{PL}}}}_{t}}{100}$$
(8)
$$-\frac{{k}_{{{\mathrm{inact}}}}}{{K}_{{{\mathrm{I}}}}}\left[{{{\mathrm{L}}}}\right]t={{{\mathrm{ln}}}}\frac{100-{\%{{{\mathrm{PL}}}}}_{t}}{100}$$
(9)
$$\frac{{k}_{{{\mathrm{inact}}}}}{{K}_{{{\mathrm{I}}}}}=-\frac{{{{\mathrm{ln}}}}\frac{100-{\%{{{\mathrm{PL}}}}}_{t}}{100}}{\left[{{{\mathrm{L}}}}\right]t}$$
(10)

We defined the OC50 as an estimate of the dose and time value (M−1s−1) that yields 50% occupancy: 1/(\([{{{\rm{L}}}}]\times t\))%PL = 50%. Thus,

$$\frac{{k}_{{{\mathrm{inact}}}}}{{K}_{{{\mathrm{I}}}}}=\frac{{{\mathrm{ln}}}2}{[{{{\rm{L}}}}]t}={{\mathrm{ln}}} \; 2\cdot {{{{\rm{OC}}}}}_{50}$$
(11)

Consistent with our goal of applying a streamlined method to determine the efficiency of covalent bond formation (kinact/KI) for targets beyond enzymes, the relationship described in Eq. (11) does not require knowledge of substrate concentration or KM.

k inact/K I from diagonal sampling of the checkerboard

We designed an 8 × 8 checkerboard of time and dose conditions to assess ARS-85330,31, a cell-active, selective, covalent KRASG12C inhibitor, and prepared iMS samples according to the conditions across the diagonal slice (Series 1 in Table 1). To obtain OC50, we plotted the percentage of KRASG12C bound to ARS-853 against the inverse product of time and dose (Fig. 2A) and fit the data using a commonly applied four-parameter logistic regression, where Y = % occupancy and X = the inverse product of dose and time (in units of kinact/KI) as both dose and time were varied during the study.

Fig. 2: Comparison of dDRTC vs. gold-standard full DRTC checkerboard approach using KRASG12C occupancy by ARS-853 quantified via intact protein MS.
Fig. 2: Comparison of dDRTC vs. gold-standard full DRTC checkerboard approach using KRASG12C occupancy by ARS-853 quantified via intact protein MS.The alternative text for this image may have been generated using AI.
Full size image

Assay dose and time conditions prepared according to the Series 1 scheme. Error bars indicate SEM. A OC50 curve from the diagonal slice of the 8 × 8 checkerboard. Each data point is the mean of two technical replicates with 8 data points per experiment. Data were fit using a common curve-fitting approach of four-parameter logistic regression. B Full DRTC curves with global fitting. Each data point is the mean of two technical replicates with 64 data points per experiment. Source data are provided as a Source Data file.

Table 1 Dose and time schemes for dDRTC
Full size table

The average OC50 for ARS-853 in the dDRTC method for two independent experiments was 310 ± 10.3 M−1s−1 (mean ± SEM). Using Eq. (11) to convert the OC50 for ARS-853, we found that the kinact/KI of 215.0 ± 7.1 M−1s−1 from dDRTC was within 20% of the reported literature value of 250 ± 20 M−1s−130.

We also ran the full 8 × 8 DRTC with ARS-853 for comparison and applied a global fit according to Eqs. (2) and (3) (Fig. 2B). Global fitting ensured consistent, robust, and accurate parameter estimates by simultaneously fitting the model to all curves within the group. The average kinact/KI over two independent experiments was 274.2 ± 0.8 M−1s−1, also within 20% of the reported value.

General correlation of k inact/K I from dDRTC vs. full DRTC

Beyond the ARS-853 example, we sought to further test the agreement between a full checkerboard approach and dDRTC by comparing their respective kinact/KI values. We selected four additional covalent inhibitors of KRASG12C as well as seven covalent modifiers of two E3 ligase complexes, SKP1-FBXW7 and SOCS2-EloB-EloC (Supplementary Table 1). Potencies spanned 1 to 2000 M−1s−1, covering early hit optimization into lead generation potencies and compound sizes7. Dose-time schemes evolve with compound potency, and we have provided general guidance for condition selection in Supplementary Table 2.

The linear regression analysis for kinact/KI determined by the full DRTC vs. dDRTC method demonstrated a near 1:1 and statistically significant relationship as well as strong correlation (y = 0.8141x, p = 1.42974 × 10−11, R2 = 0.9817), indicating a high degree of alignment between methods (Fig. 3). In addition, the median percent difference (n = 12) between the gold-standard approach and dDRTC was 19% (Supplementary Tables 3 and 4), showing the dDRTC approach can accurately reflect kinact/KI while reducing resource (reagents and time) consumption and increasing throughput to support iterative hit-to-lead optimization campaigns.

Fig. 3: Linear regression analysis for kinact/KI determined from full DRTC vs. diagonal DRTC (n = 12 compounds, p = 1.42974E-11, R2 = 0.9817).
Fig. 3: Linear regression analysis for kinact/KI determined from full DRTC vs. diagonal DRTC (n = 12 compounds, p = 1.42974E-11, R2 = 0.9817).The alternative text for this image may have been generated using AI.
Full size image

Each data point for both dDRTC and full DRTC is the mean of two technical replicates. Error bars indicate SEM, and gray shading indicates the 95% confidence interval. Source data are provided as a Source Data file.

Simulations suggest the dDRTC method is ideal for early-stage programs

To understand the applicable range where dDRTC is best suited, we simulated datasets across a broad range of kinact (0.00005 − 1 s⁻¹) and KI (1 − 1000 μM) values (see Methods/SI for details). These simulations were agnostic to the molecular weight of the compounds and the target being modified. Consistent with the assumption KI \(\gg\) [L] used in deriving the relationship between dDRTC OC50 and true kinact/KI (Eq. (11)), dDRTC accurately quantified kinact/KI values for covalent modifiers with low reversible affinities. The simulations did not predict significant deviation ( ≥ two-fold difference) for modifiers with KI ≥ 50 μM within the wide potency range of 1 to 10,000 M⁻¹s⁻¹, aligning with a typical kinetic profile for early covalent leads or compounds that are optimized for kinact32, such as sotorasib, which has a reported KI of > 50 μM (Fig. 4)3,33. Improving KI to less than 50 μM led to greater deviation between true kinact/KI and dDRTC-derived values, particularly at low kinact/KI, thus defining the utility of the dDRTC method as a highly efficient early lead discovery tool, or perhaps later in kinact driven programs.

Fig. 4: Accurate quantification of the covalent potencies (kinact/KI) by dDRTC.
Fig. 4: Accurate quantification of the covalent potencies (kinact/KI) by dDRTC.The alternative text for this image may have been generated using AI.
Full size image

Simulation predicts no deviation between dDRTC OC50-derived kinact/KI values \(\left({{\mathrm{ln}}} \; 2\cdot {{{{\rm{OC}}}}}_{50}\right)\) plotted on the y-axis and true kinact/KI (x-axis) for the covalent modifiers with reversible affinities (KI) ≥ 50 μM (filled circles). The solid line represents a perfect positive correlation where \({{\mathrm{ln}}} \; 2\cdot {{{{\rm{OC}}}}}_{50}\) = kinact/KI, and the dotted lines indicate two-fold deviation margins. Optimized leads with KI < 50 μM (empty circles) exhibit increasing deviation with improving reversible affinities.

Discussion

We present a practical, resource-efficient method for estimating kinact/KI over a broad dose-time space. We applied common curve fitting techniques to the information-rich diagonal datapoints of a full DRTC checkerboard and approximated kinact/KI from OC50 initially with the cell-active compound ARS-853. We then extended the study to a dozen covalent modifiers spanning both fragment and beyond-rule-of-three lead-like space, for KRASG12C and two E3 ligase complexes, SKP1-FBXW7 and SOCS2-EloB-EloC, for which no approved drugs exist. We analyzed the kinact/KI across more than three orders of magnitude via iMS and validated the results against the full DRTC checkerboard method, observing a strong correlation between the two methods. Kinetic simulations further highlighted the strengths of dDRTC as an early-stage tool that enables teams to obtain kinact/KI values earlier and more frequently in a drug discovery program, overcoming the resource barriers of the traditional approach.

Outside of the optimal dDRTC kinact and KI boundaries indicated by the simulations, other methods, such as a bimolecular reaction model34 may be more appropriate for highly potent modifiers with kinact/KI > 10,000 M−1s−1 and full DRTC or SPR analysis for leads with KI < 50 μM; Compound 9 was the only compound that deviated from the full DRTC ~two-fold, and it had the lowest KI (10 μM, Supplementary Table 5), consistent with the simulation results. In general, reaching these potencies indicates significant progress for a project team and confirmation that a protein can be targeted with covalent modifiers. At this stage, the dDRTC assay has enabled the team with suitable chemical matter for probe-competition assays, which can be used to further drive SAR and determine kinact/KI relative to the known kinact/KI of the probe (Supplementary Fig. 1)35. In addition, at values of kinact/KI > ~ 400 M−1s−1, cellular activity is expected, and this becomes more important in driving further optimization7.

Though MS is particularly well-suited for assessing covalent potency, other methods have been used to derive kinact/KI from biochemical data, including time-dependent endpoint probe competition assays and pre-incubation assays such as EPIC-FIT35,36. However, probes may not be available for new targets, and not all covalent modifiers are amenable to functional IC50 shift assays that require the covalent modifier to be an inhibitor. Indeed, some covalent modifiers may not alter protein function, such as covalent ligands designed to induce proximity between two proteins, e.g., the ligands shown earlier that modify E3 ligase complexes (note: though kinact may not be the most accurate term in this case as we are not inactivating the E3 ligases, for the purposes of this manuscript, we have chosen to maintain terminology commonly used in the field of covalent drug discovery). For a broader understanding of covalent occupancy’s impact on function, which can differ for each target, a very helpful concept known as target vulnerability has been described22. To further aid the reader in positioning dDRTC as a tool in their discovery toolbox, Table 2 outlines the pros and cons of kinetic IC50-based assays, iMS, and SPR, each appropriate for different experimental contexts and research goals.

Table 2 Summary of pros and cons of kinetic assays for the quantification of the rate of covalent modification, kinact/KI
Full size table

iMS DR and TC assays can guide SAR in the early stages of a program, but both assays have shortcomings that can be addressed with the dDRTC method. For an iMS assay with sensitivity limits above 1 µM of protein, the effective assay range for a DR curve is the solubility limit of a compound (often 400 µM or less based on experimental measurement and in-house solubility prediction algorithms), down to a dose that maintains pseudo-first-order kinetics, i.e., [L] > [P] to allow for 100% modification. This could be addressed by digesting the protein into peptides after the reaction or concentrating the protein prior to MS, both of which could lower the starting concentration of protein required, but are beyond the scope of this manuscript. To extend the range of SAR using the DR approach, multiple incubation times are needed, and in the case of a fixed-dose TC approach, researchers are limited by practical, user-friendly considerations such as standardized time points. Thus, by modulating time and dose simultaneously, dDRTC provides SAR across a wider range of potencies under the same assay conditions (see Supplementary Fig. 2, Supplementary Fig. 3, and Supplementary Table 6) and eliminates the need to run multiple curves over a range of time points and/or doses.

Another attractive feature of dDRTC is that kinact/KI is estimated directly from the OC50 curve, which provides a simple, visual representation similar to more commonly used IC50 values. Previously reported approaches rely on correlating a value such as IC50 to kinact/KI and then using regression to derive kinact/KI37,38,39. As demonstrated in those studies, it would be necessary to run a full DRTC checkerboard for many compounds to establish the initial calibration curve, which may not be possible in the early stages of a program due to a lack of compounds across a wide range of potencies. The dDRTC method eliminates the need for this step.

Comparing the throughput of dDRTC and full DRTC, the dDRTC method significantly reduces protein consumption and instrument runtime and is amenable to weekly SAR, i.e., less than 100 compounds per week. An 8-point dDRTC can provide kinact/KI for 48 compounds per 384-well plate in contrast to six compounds in an 8 × 8 full DRTC. Thus, for a typical weekly SAR assay with 35 compounds, a single 384-well plate has more than enough capacity for the full set of compounds in an 8-point dDRTC format, which would require 1960 fewer samples, saving approximately 1.2 mg of protein (using KRAS as an example), compared to an 8 × 8 full DRTC. This is particularly advantageous for targets that are difficult to express. Although the full checkerboard is not practical for routine compound profiling during hit-to-lead optimization, because it is the gold standard and can discriminate kinact from KI, we recommend teams spot-check values using full DRTC when there is a ten-fold increase in potency to assess whether the lead series is still within the suggested dDRTC kinact and KI boundaries.

The dDRTC method can deliver kinact/KI values for covalent modifiers over at least three orders of magnitude of potency with an eight-fold increase in throughput by utilizing the diagonal slice of an 8 × 8 checkerboard, all with a more simplified data analysis. It can also be applied to proteins for which no previous binders or functional assays exist. dDRTC enables kinact/KI determination earlier and more frequently in a drug discovery campaign where these values can be used as a comparative metric for covalent fragments and leads through methods such as Rate Enhancement Factor analysis21. Our simulations extend the characterization of the dDRTC method and provide guidelines for implementation according to the stage of a program. We routinely use dDRTC at Frontier Medicines to profile our compounds, and we believe this method will help others in covalent fragment drug discovery accurately define inactivation kinetics on a scale amenable for SAR studies.

Methods

The compounds in this study were selected from our internal library as well as published compounds. While Compound 3 (ARS-853) was ordered directly from Selleck Chemicals’ product catalog, the remaining compounds were synthesized and supplied from WuXi AppTec all at > 90% purity. Compounds were QC’ed internally, resolubilized in DMSO, and diluted to 1 mM and 10 mM working concentrations.

Sample Preparation for MS

KRASG12C inactive-state protein (produced at Viva Biotech) in stock solution consisting of 50 mM HEPES, 2 mM MgCl2, 100 mM NaCl, 100 µM GDP, and with a resulting pH of 7.4 was diluted to a 0.8 µM protein concentration master mix solution consisting of 25 mM HEPES (Fisher Scientific, NC0470071), 2 mM MgCl2 (Fisher Scientific, 50-983-241), 10 mM NaCl (Fisher Scientific, 50-983-260), 100 µM GDP (Sigma Aldrich, G7127-1G), and with a resulting pH of 7.4.

SKP1-FBWX7 protein complex (produced internally) in stock solution consisting of 50 mM HEPES, 200 mM NaCl, 2 mM TCEP, and with a resulting pH of 6.8, was diluted to a 0.5 µM protein concentration master mix solution in Dulbecco’s Phosphate-Buffered Salt Solution (Fisher Scientific, MT21031CM). The protein complex was shown to have a purity of at least 90% by SDS-Page and HPLC-SEC.

SOCS2-EloB-EloC protein complex (produced internally) in stock solution consisting of 50 mM HEPES, 250 mM NaCl, 1 mM TCEP, and with a resulting pH of 7.5, was diluted to a 0.5 µM protein concentration master mix solution in Dulbecco’s Phosphate-Buffered Salt Solution (Fisher Scientific, MT21031CM). The protein complex was shown to have a purity of at least 90% by SDS-Page and HPLC-SEC.

Compounds were dispensed with an Echo 650 liquid handler into 384-well plates (Fisher Scientific, 07-000-890) at assay concentrations with 3% final DMSO (Fisher Scientific, BP231-100) at concentrations according to Table 1. Protein master mix was added to assay plates at room temperature (22 °C incubator) according to assay time points, and samples were quenched with 4% formic acid (Fisher Scientific, A117-50) solution following incubation. We determined compound solubilities experimentally and generally observed reproducible results and reliable solubility for dose conditions below 400 µM.

Measuring Intact Protein Modification by RapidFire-ToF MS

Samples were analyzed by ESI+ mass spectrometry on an Agilent RapidFire high-throughput MS system connected to an Agilent 6230 Time-of-Flight mass spectrometer. Samples were injected onto a RapidFire™ C8 cartridge for KRASG12C and a C4 cartridge for SKP1-FBWX7 and SOCS2-EloB-EloC complexes. Samples were eluted into the ToF with the RapidFire settings as follows: pump 1 = 0.4 mL/min, pump 2 = 0.6 mL/min, pump 3 = 0.4 mL/min, aspiration = 400 ms, load = 7000 ms, extra wash = 5000 ms, elute = 10000 ms, re-equilibrate = 700 ms. Furthermore, the source settings were as follows: drying gas temp = 365 °C, drying gas = 12 L/min, VCap = 5550 V, fragmentor = 175 V, nozzle voltage = 2000 V, skimmer = 75 V, nebulizer = 60 psi, capillary = 0.051 µA, sheath gas flow = 12 L/min, sheath gas temp = 400 °C. The instrument was tuned in 3200 m/z mode, 2 GHz, with extended dynamic range. Throughout this study, 832 unique samples were acquired in technical duplicate for a total of 1664 samples.

MS Data analysis

Raw spectra were deconvoluted in Agilent MassHunter Bioconfirm Version 10 software using the Maximum Entropy deconvolution algorithm and settings from Table 3.

Table 3 Deconvolution settings for each protein or protein complex used in this study
Full size table

Deconvoluted spectra were analyzed for identification of parent and adduct peaks corresponding to specific compound masses and well locations using an in-house proprietary Python-based (version 3.8) analysis package. Exported results were transposed into GraphPad Prism version 10.1.2 for curve-fit generations based on the equations outlined above. The linear regression analysis was also performed in Prism using the Simple Linear Regression built-in analysis.

Kinetic data simulations

Data simulation was performed using KinTek Global Kinetic Explorer version 11.0.1 (KinTek Corp)26,40. Full DRTC progress curves were simulated for the two-step irreversible kinetic model:

$${{{\rm{P}}}}+{{{\rm{L}}}}\rightleftharpoons {{{\rm{P}}}}\times {{{\rm{L}}}}({k}_{{{{\rm{on}}}}},{k}_{{{{\rm{off}}}}})$$
$${{{\rm{P}}}}\times {{{\rm{L}}}}\to {{{\rm{P}}}}-{{{\rm{L}}}}({k}_{{{{\rm{inact}}}}})$$

where P = protein target, L = covalent modifier, P × L = noncovalent binary complex, P-L = covalent binary complex. The reversible binding step was treated as a rapid equilibrium, and the rate constant kon was fixed at the diffusion limit (108 M−1s−1). Datasets were simulated for each individual combination of KI and kinact/KI from Supplementary Table 7.

The observable signal (%Modification) was defined as a fraction of the covalent binary complex P-L as follows:

$$\%{{{\rm{Modification}}}}=100\%\times [{{{\rm{P}}}}-{{{\rm{L}}}}]/([{{{\rm{P}}}}]+[{{{\rm{P}}}}\times {{{\rm{L}}}}]+[{{{\rm{P}}}}-{{{\rm{L}}}}))$$
(12)

The %Modification data points were simulated for the [P] = 0.8 μM and [L] according to Series 1: 1.65, 1.65, 3.8, 5.8, 10.9, 25, 57, 85 μM, for 10,800 s, and 15 s interval between points (Table 1). In addition, an analogous dataset was generated for the concentrations according to Series 2: 1.56, 3.13, 6.25, 12.5, 25, 50, 100, 200 μM, for 72,000 s, and 60 s interval between points (Table 1). From each full DRTC dataset, the diagonal slices corresponding to the time-concentration combinations in Table 1 were extracted.

For each slice where the simulated signal reached at least 80% modification, %Modification was plotted versus [L]−1Time−1 and fitted to the four-parameter logistic nonlinear regression corresponding to the equation: %Modification = 100 + 100 / (1 + (([L]*t)−1 / OC50)^Hill), in GraphPad Prism version 10.1.2. Resulting OC50 values were converted to estimates of kinact/KI according to the equation: kinact/KI = ln2×OC50. The results were summarized in Supplementary Table 7.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.