LCK-targeting molecular glues overcome resistance to inhibitor-based therapy in T-cell acute lymphoblastic leukemia

Key Points

• •High-throughput screen identified LCK-targeted molecular glues that use a unique G-loop degron for CRBN engagement.
• •LCK MGDs overcome T-ALL resistance to inhibitors/inhibitor-based PROTACs, pointing to strategies to target drug-resistant kinases.

Visual Abstract

Abstract

Drug resistance is a major challenge in cancer therapy, especially in hematologic malignancies in which kinase inhibitors have transformed treatment yet are frequently undermined by drug resistance. Although targeted protein degradation (TPD) offers a mechanistically distinct mode of action compared with inhibition-based therapeutic therapies, the potential value of TPD in drug-resistant blood cancer remains unclear. Here, we report the discovery of cereblon (CRBN)–recruiting molecular glue degraders (MGDs) targeting lymphocyte-specific tyrosine kinase (LCK), an oncogenic kinase in T-cell acute lymphoblastic leukemia (T-ALL). By high-throughput screening and medicinal chemistry optimization, we developed a series of MGDs that induced CRBN-dependent degradation of LCK as well as potent cytotoxicity in T-ALL in vitro. Structure-activity relationship analysis and ternary complex modeling revealed a noncanonical degron at the LCK-CRBN interface involving the G-loop, whose mutation disrupts this interaction. Unlike inhibitors and inhibitor-based proteolysis-targeting chimeras, these MGDs engage LCK in regions distal to the ATP-binding site, and thus their activities in T-ALL are not affected by gatekeeper LCK mutations that drive resistance to inhibitor-based therapeutics. Taken together, our data highlight the potential of LCK-targeting MGDs as a strategy to overcome kinase inhibitor resistance in T-ALL, offering a framework for targeting kinase dependencies in drug-refractory hematologic malignancies more broadly.

Lenalidomide is the best-known example of a drug that works as a molecular glue degrader, a small molecule that induces the binding of 2 proteins that usually do not interact, thereby enhancing destruction of the protein of interest. Yoshimura et al generated and tested a novel molecular glue degrader that targets lymphocyte-specific protein tyrosine kinase (LCK), which is highly activated in 40% of T-cell acute lymphoblastic leukemias. They demonstrate antileukemic efficacy in preclinical models and the ability of this compound to overcome resistance to LCK inhibitors such as dasatinib, encouraging ongoing development toward clinical application.

Introduction

Targeted protein degradation (TPD) has garnered growing popularity in drug discovery thanks to several key advantages over protein inhibition, including a catalytic mechanism of action, greater efficacy, and the ability to target nonfunctional binding sites.^1,2 The 2 most explored chemical approaches in TPD are proteolysis-targeting chimeras (PROTACs) and molecular glue degraders (MGDs).³ Both modalities harness the endogenous cellular ubiquitin proteasome system by bringing a protein of interest (POI) into proximity to an E3 ligase to induce ubiquitination and subsequent degradation of the former by the proteasome.³ PROTACs can be rationally designed as bifunctional molecules, that is, chemically linking a small-molecule binder of a POI (eg, a known inhibitor of a therapeutic target) to an E3 ligase ligand.¹ Although the irreversible target degradation induced by PROTACs may be translated into greater efficacy than inhibitors, drug resistance is a significant potential challenge. For example, any mutation in the POI that compromises the binding of the inhibitor would, in theory, result in resistance to PROTACs designed around this inhibitor. For instance, in the case of leukemia, gatekeeper mutations in ABL (eg, T315I) that cause resistance to dasatinib would greatly reduce the activity of all dasatinib-based PROTACs.^{4, 5, 6, 7}

By contrast, MGDs induce conformational changes at the surface of the E3 ligase that are complemented by a POI to form the required ternary complex; this proximity leads to target protein ubiquitination and eventually proteasomal degradation.⁸ The key feature of an MGD is that it usually engages a POI through surface interaction, which can be distant from the binding pocket deep inside the protein that is required for inhibitor-target interaction. Because of this, we reason that MGDs may be less affected by mutations in the inhibitor binding pocket and should remain active in cells resistant to inhibitor or inhibitor-based PROTACs. However, there is a paucity of examples clearly demonstrating the potential of MGDs to overcome PROTAC resistance, partly because very few therapeutic targets have been evaluated with both types of protein degradation modalities.⁹

Lymphocyte-specific tyrosine kinase (LCK) is constitutively activated in ∼40% of T-cell acute lymphoblastic leukemia (T-ALL).^{10, 11, 12, 13} Therapeutic targeting of LCK with small-molecule inhibitors such as dasatinib can be highly effective in inducing leukemia cell death both in the laboratory and in patients.^10,12,14,15 Dasatinib-based PROTACs induce rapid degradation of LCK in T-ALL, with cytotoxicity far exceeding that of dasatinib itself as well as greatly superior pharmacodynamic effects in vivo.^16,17 However, an LCK mutation has been identified (T316M or T316I) that abolishes dasatinib binding,^14,18 with expected resistance to a dasatinib-based PROTAC. We hypothesized that we could overcome resistance associated with kinase domain mutations by developing LCK-targeted MGDs. In fact, LCK contains the common β-hairpin G-loop motif, known as degron, present in the most frequently reported neosubstrates (G1 to S phase transition 1 [GSPT1], casein kinase 1α [CK1α], and IKAROS family zinc finger 1 [IKZF1]) of cereblon (CRBN) ligase, providing a rationale for this approach.

Herein, we report the discovery and optimization of a series of LCK-targeting MGDs that shows superior cytotoxicity in dasatinib-resistant T-ALL cells compared with dasatinib and dasatinib-based PROTAC.¹⁶ Importantly, our data point to the potential of switching degradation modalities from bivalent to monovalent as a means of overcoming PROTAC resistance.

Methods

Cell lines

An LCK-activated T-ALL cell line KOPT-K1 was kindly gifted by Takaomi Sanda at National University of Singapore. HiBiT-tagged LCK was introduced into KOPT-K1 using CRISPR–CRISPR-associated protein 9, as previously reported.^17,19 Dasatinib-resistant clones of KOPT-K1 with a heterozygous LCK T316I mutation was generated as previously reported.¹⁴ KOPT-K1 cells with LCK G415I knockin were generated via CRISPR–CRISPR-associated protein 9 editing (∼75% editing efficiency)²⁰; genotyping primers and guide RNAs are shown in supplemental Table 2, available on the Blood website. Another LCK-activated T-ALL line, HSB-2, as well as CCRF-CEM, Jurkat, Loucy, and SUP-T1 were purchased from the America Type Culture Collection; ALL-SIL and DND-41 were from Leibniz Institute DSMZ. T-ALL molecular subtypes were determined using RNA sequencing and/or whole-genome sequencing (St. Jude Children’s Research Hospital PROPEL data portal)²¹ based on the classification schema described by Polonen et al.²² All T-ALL cell lines were cultured in RPMI-1640 supplemented with 10% fetal bovine serum.

Compound screening

A total of 3424 compounds from the St. Jude Children’s Research Hospital proprietary molecular glue library were screened against KOPT-K1 expressing HiBiT-tagged LCK, as previously described.¹⁷ Compounds were prepared in 384-well plates at 10mM in 100 nL. Cells were added at 7500 cells per well in 25 μL using a VIAFLO 384 electronic pipette (Integra). After incubation of 4 or 24 hours at 37°C/5% CO₂, the levels of HiBiT-tagged LCK protein were quantified by the Nano-Glo HiBiT lytic detection system (catalog no. N3030; Promega) per the manufacturer’s instruction. Luminescence was measured using a Synergy H4 hybrid microplate reader (BioTek). Percentage LCK degradation was calculated by normalizing luminescence values to dimethyl sulfoxide (DMSO)–treated controls and subtracting background signal from parental KOPT-K1 cells.

LCK MGDs synthesis

Details for the synthesis of all LCK MGDs are provided in the supplemental Methods. Previously reported selective CK1α degraders SJ3149 and 0040,²³ as well as the dasatinib-based PROTAC SJ11646, were included in comparative assays.¹⁶

Ternary complex assay (LCK/CRBN-DDB1)

Assay buffer consisted of 25mM HEPES (N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid), pH 7.4, 100mM NaCl, 0.1% bovine serum albumin, and 0.05% Tween 20. Compounds were acoustically dispensed (ECHO 650, Labcyte) as 10-point dilution curves into 384-well OptiPlates (PerkinElmer, catalog no. 6007290) giving a final DMSO concentration of 0.1%. Each well received 5 μL of a 5× mixture of 100nM His-tagged CRBN–damage-specific DNA binding protein 1 (DDB1) and 75nM glutathione S-transferase (GST)–tagged LCK or LCK (G415N), followed by 1 hour incubation at room temperature. Subsequently, 10 μL each of 20 μg/mL nickel chelate acceptor and glutathione donor beads (PerkinElmer, catalog no. 6765300 and catalog no. 6760141) were added under subdued lighting. Plates were sealed, mixed (MixMate, Eppendorf) for 1 hour, and then luminescence was measured on an Envision plate reader (PerkinElmer). A 4-parameter dose-response model was used to fit the data using GraphPad Prism software (version 10.3.1). Protein production details are provided in the supplemental Methods.

CTG assay for T-ALL cell lines

CellTiter-Glo (CTG; catalog no. G9243; Promega;) was used as a proxy measure of cell number based on ATP content. T-ALL cell lines were plated at 120 000 cells per mL in 25 μL per well onto 384-well plates containing compounds (see aforementioned LCK-HiBiT assay). After 72 hours of incubation, an equal volume of CTG reagent was added, incubated briefly at room temperature, and luminescence was detected. Cell abundance was normalized to DMSO-treated controls. Compounds were tested in triplicate, and LC₅₀ (the concentrations lethal to 50% of cells) values were calculated by a 4-parameter dose-response model.

CyQUANT direct cell proliferation assay for T-ALL cells from PDX models

A patient-derived xenograft (PDX) LCK-activated T-ALL model (SPI1 subtype) was provided by David T. Teachey at Children’s Hospital of Philadelphia. Drug response of PDX T-ALL blasts was evaluated using a mesenchymal stem cell coculture system and determined by an imaging-based assay, as described previously.²⁴ Briefly, human telomerase reverse transcriptase–immortalized mesenchymal stem cells were seeded 1 day before the assay, followed by plating of the cells with serial dilutions of drug in AIM V serum-free medium (catalog no. 12055083; Gibco). After 96 hours of incubation at 37°C with 5% CO₂, leukemia cells were evaluated by high-content imaging. Drug-induced cell death was estimated by comparing treated wells with medium-only controls using CyQUANT direct cell proliferation assay green (catalog no. C35012; Invitrogen). The number of viable cells was based on the fluorescence from the cell-permeant CyQUANT dye that stains DNA, whereas a suppressor was also added to block staining in dead or damaged cells. Cell viability data were analyzed by Harmony high-content imaging and analysis software (version 4.9).

Statistical analysis

The association of LCK DC₅₀ (the concentration that can degrade half of the proteins) values between 4 hours and 24 hours was evaluated by Wilcoxon matched-pairs signed-rank test. The comparison of LC₅₀ values in KOPT-K1 between MGDs capable of degrading both LCK and CK1α and those degrading CK1α only was assessed by Mann-Whitney U test. P values were considered significant if <.05. All analyses were performed with GraphPad Prism (version 10.1.2).

Results

High-throughput screening of molecular glue library identifies LCK degraders

To identify MGDs that target LCK, we designed a screening strategy in which the St Jude proprietary molecular glue library was assayed in a HiBiT-tagged system and evaluated for degradation of LCK (Figure 1A-B).²⁵

Figure 1.

High-throughput screening to identify LCK-targeting MGDs. (A) Schematic workflow of high-throughput screening of LCK-targeting MGDs. (B) Visualization of the chemical scaffold diversity of the molecular glue library. In this t-SNE plot, each dot represents a compound and the proximity between points indicates their chemical structure similarities. (C) LCK degradation induced by each MGD at early (4 hours) and late (24 hours) time points. Each dot represents the mean percentage of LCK remaining, calculated from duplicate measurements for each compound. Five compounds (orange) met the threshold of ≥30% and ≥75% LCK-HiBiT reduction at 4 hours and 24 hours, respectively. t-SNE, t-distributed stochastic neighbor embedding. Panel A created with BioRender.com. Yoshimura S. (2026) https://biorender.com/0pu10n3.

Using CRISPR editing, we first engineered a T-ALL cell line model (KOPT-K1) in which the constitutively activated LCK is fused with a HiBiT tag,¹⁷ enabling quantification of LCK degradation using luminescence. Then, the St Jude proprietary molecular glue library,²³ consisting of 3424 compounds designed for potential CRBN engagement and comprising of chemically diverse chemotypes, was screened in the T-ALL cell line using the LCK-HiBiT assay at a single concentration (40μM; Figure 1A). For each compound, the LCK-HiBiT protein level in treated cells was measured and normalized to that of nontreated cells. Because LCK is essential for T-ALL survival, degraders are expected to induce apoptosis.¹⁶ To account for the effect of cell death on LCK-HiBiT protein level, we performed the assay after both 4 and 24 hours of drug exposure. Of 3424 compounds screened, there was a greater degree of reduction in LCK protein at 24 hours than at 4 hours (P < .0001), and more compounds degraded LCK to the same degree at 24 hours than at 4 hours. Five hits (SJ1473, 1481, 1581, 1652, and 2917) induced ≥30% and ≥75% of LCK-HiBiT degradation at 4 and 24 hours, respectively, and were selected for further characterization (Figure 1C). Notably, there was a trend for positive correlation between the degree of LCK degradation and the decrease in cell metabolic activity (a proxy for cytotoxicity) measured by CTG assay, although this did not reach statistical significance (Spearman ρ = 0.90; P = .08; supplemental Figure 1).

Next, these 5 compounds were tested in a dose-response format using the LCK-HiBiT assay at serial concentrations ranging from 2nM to 40μM of drug exposure. Compound 1 (SJ1581) exhibited the greatest potency (Figure 2A-B), whereas the other 4 induced only modest degradation. We also directly measured each compound’s ability to promote the formation of the ternary complex with CRBN and LCK, using an in vitro AlphaScreen assay,¹⁶ because the induced proximity of the 2 is required for ubiquitination and subsequent proteasomal degradation. The AlphaScreen signal was measured for each molecular glue at varying concentrations. As a positive control, a previously described dasatinib-based LCK PROTAC (SJ11646) was included,¹⁶ giving the characteristic bell-shaped curve (Figure 2C). Of the 5 compounds, compound 1 was most effective at forming a ternary complex with a half-maximum concentration (EC₅₀) of 123.1nM (Table 1), in line with its potency in degrading LCK quantified by the HiBiT assay.

Figure 2.

Compound 1 is validated as a hit MGD identified through compound screening. (A) Dose-dependent LCK degradation by the 5 compounds identified from the screening, measured using HiBiT assay after 24 hours of treatment. Compound 1 exhibits the greatest potency among them. Data represent mean ± standard deviation (SD) from 3 biological replicates. (B) Structures of the 5 compounds identified from the screening. (C) MGD-induced formation of a ternary complex between GST-LCK and His-CRBN-DDB1 proteins measured using an in vitro AlphaScreen assay. Data are shown as mean ± SD from 3 technical replicates. (D) Representative immunoblots showing LCK degradation in KOPT-K1_LCK_HiBiT cells treated with serial concentrations of compound 1 for 24 hours. Quantified LCK degradation normalized to the DMSO control from 3 biological replicates are also shown. (E) Dose-dependent LCK degradation by compound 1 after 24-hour treatment demonstrated by HiBiT assay, which was attenuated by cotreatment with lenalidomide 10μM. Data represent mean ± SD from 3 biological replicates. GAPDH, glyceraldehyde-3-phosphate dehydrogenase.

Table 1.

Properties of LCK MGDs (compounds 1-8)

Cl, chloro; CO, carbonyl (C=O); Et, ethyl; Me, methyl; Pr, propyl.

^∗AlphaScreen ternary complex formation half-maximum concentration reported as the mean with 95% confidence interval for 3 technical replicates.

^†Degradation measured at the indicated time points using HiBiT-LCK, reported as the mean ± standard deviation for 3 biological replicates.

LCK degradation by compound 1 was also confirmed using immunoblotting (Figure 2D; supplemental Figure 2). Pretreatment with excess lenalidomide abolished CRBN-LCK ternary complex formation (supplemental Figure 3) and reversed the compound 1–induced LCK degradation in T-ALL cells (DC₅₀ >40μM), confirming these effects as CRBN dependent (Figure 2E).

These results qualified high-throughput screening hit compound 1 as a potent and CRBN-dependent LCK degrader for further optimization.

Medicinal chemistry optimization to improve potency and solubility of LCK MGDs

To understand the structure-activity relationship (SAR) for LCK activity, we first generated a series of analogs of compound 1 probing key structural features, that is, the halide at the 3-position of the indole (R₁), the N-H group of the indole and amide (R₂ and R₃, respectively), and the carbonyl (X) linking the CRBN-binding moiety to the indole (Table 1). Using the AlphaScreen assay, we observed a loss in the ability to form a ternary complex when removing the halide (compound 3, EC₅₀ = 234.7nM vs compound 1, EC₅₀ = 123.1nM), which also led to reduced degradation potency as determined using the LCK-HiBiT assay (DC₅₀ of 268.4nM at 24 hours treatment). Replacing the chloro at R₁ with a methyl substitution (compound 2) slightly improved its ability to form the ternary complex (EC₅₀ = 109.9nM), whereas incorporating an ethyl substitution provided further improvement (compound 7; EC₅₀ = 97.4nM) and greatly increased LCK-HiBiT degradation activity, providing the first analog with activity at 4 hours of drug treatment (DC₅₀ = 20.4nM). These improvements diminished when incorporating a larger propyl group at R₁ (compound 8), suggesting there is a limit to the size tolerated at that position.

Next, we explored the impact of the indole N-H and amide functional group. As shown in Table 1, methylating R₂ or R₃ gave compounds (4 and 5, respectively) with loss of activity across assays, highlighting the importance of hydrogen bond donors in the formation of the ternary complex. Removing the carbonyl of the amide functional group and losing this hydrogen bond acceptor resulted in inactive compound 6. The SAR pointed to an essential network of H-bond donors and acceptors for ternary complex formation.

Based on these results, we selected compound 7 for further structural modifications with the goal of systematically exploring the impact of indole substitutions on ternary complex formation and LCK degradation activity, using DC₅₀ values at 4 hours as the end point to prioritize the most potent degraders (Table 2).

Table 2.

Properties of LCK MGDs (compounds 9-15)

Me, methyl; OMe, methoxy; Sol., solubility.

^∗AlphaScreen ternary complex formation half-maximum concentration reported as the mean with 95% confidence interval for 3 technical replicates.

^†Degradation measured using HiBiT-LCK reported as the mean ± standard deviation for 3 biological replicates.

^‡Kinetic aqueous solubility measured at pH 7.4 and reported as the mean ± standard deviation for 3 replicates.

Firstly, focusing on bromo (Br) substitutions, we found that the 5-Br (compound 9) and 7-Br (compound 11) analogs were favored for inducing proximity in the AlphaScreen assay (EC₅₀ = 84.7nM and 62.1nM, respectively) over the 6-Br substitution (compound 10; EC₅₀ = 128.6nM). In agreement with the biochemical data, compounds 9 and 11 were highly effective in LCK degradation (DC₅₀ = 22.1nM and 68.6nM, respectively). Incorporating a methoxy group at these positions provided compounds 12, 13, and 14 with comparable potencies in LCK:CRBN ternary complex assay and LCK degradation (DC₅₀ = 33.2nM, 23.1nM, and 15.9nM, respectively). This representative set of compounds suggested that indole substitutions were tolerated, and that they could be exploited to improve on the absorption, distribution, metabolism, and excretion (ADME) properties later in the optimization process. To mechanistically validate the activity of these molecules, compound 15 was prepared, featuring a methyl substitution on the glutarimide nitrogen, which blocks key interactions in the thalidomide binding pocket of CRBN. Compound 15 completely lost its ability to form a ternary complex, which correlated with loss of activity for LCK degradation (Table 2). These results support a CRBN-dependent degradation mechanism.

Next, we focused on physicochemical physical properties and profiled compounds in Table 2 against our aqueous solubility assay. All compounds displayed poor to modest aqueous solubility (0.44μM-6.77μM) and to overcome this liability, we explored adding polar substitutions at the 7-position of the indole ring. The addition of an amide group in compounds 16 and 18 yielded glues with strong LCK degradation (Table 3), whereas only compound 16 gave improved solubility (22.6μM) compared with 17 (3.7μM). Incorporating a primary amine (17) or alcohol (19) significantly improved solubility (67.9μM and 27.0μM, respectively) while retaining LCK degradation activity (DC₅₀ = 103.4nM and 32.4nM, respectively). Finally, we also validated LCK degradation and cytotoxicity of these MGDs in an additional T-ALL cell line HSB2 and PDX model of an LCK-activated T-ALL (supplemental Figures 4 and 5). Encouraged by the profiles of compounds 17 and 19, we pursued further characterization of these chemical tools, probing their binding modes via molecular modeling, neosubstrate selectivity, and activity in dasatinib-resistant cell lines.

Table 3.

Properties of LCK MGDs (compounds 16-19)

Me, methyl.

^∗AlphaScreen ternary complex formation half-maximum concentration reported as the mean with 95% confidence interval for 3 technical replicates.

^†Degradation measured using HiBiT-LCK reported as the mean ± standard deviation for 3 biological replicates.

^‡Metabolic activity measured using CTG assay was used as a proxy for drug cytotoxicity and was reported as the mean ± standard deviation for 3 biological replicates.

^§Kinetic aqueous solubility measured at pH 7.4 and reported as the mean ± standard deviation for 3 replicates.

Modeling of CRBN-MGD-LCK interaction rationalizes SAR

Generally, immunomodulatory drug–based MGDs are thought to degrade target protein through a glycine-containing β-hairpin loop recognition motif, known as the “degron.” However, this view was recently challenged in work by Petzold et al,²⁶ which described non–β-hairpin glycine–containing degrons recognized by CRBN to form the required ternary complex. In particular, the authors reported an MGD-mediated interaction of CRBN with inactive LCK through its helical G-loop motif, specifically at G415.

To validate the importance of the G415-containing degron, we generated a mutant LCK protein (G415N), with which we screened a panel of LCK MGDs for their ability to form a ternary complex with CRBN. None of the LCK MGDs tested (16-19) induced the formation of the ternary complex between CRBN and mutant LCK (Figure 3A). By contrast, the mutant LCK can still interact with CRBN in the presence of LCK PROTAC SJ11646 because this interaction does not involve the G415 degron. This was further supported by immunoblotting, which showed that LCK degradation by MGDs was impaired in KOPT-K1 with LCK G415N (Figure 3B). It should be noted that mutant cells still exhibited some degree of LCK degradation, which may be because of incomplete editing at the G415 position (∼75% based on editR estimate).²⁰

Figure 3.

Modeling of CRBN-MGD-LCK interaction rationalizes SAR. (A) In vitro AlphaScreen assay demonstrating ternary complex formation between His-CRBN-DDB1 and GST-LCK (G415N) in the presence of MGD. (B) Representative immunoblots showing attenuated LCK degradation KOPT-K1 with LCK G415N mutant treated with compound 17 at 1000nM for 24 hours, compared with cells expressing WT LCK. Quantified LCK degradation normalized to the DMSO control from 3 biological replicates are also shown. (C-E) Model of the ternary complex, CRBN-1-LCK, with the helical G-loop motif G415 of LCK. (C) Wide view of the modeled ternary complex; β-hairpin loop (top) and G-loop (middle) are highlighted in red. (D) Close-up view of the interactions between CRBN, compound 1, and LCK with G415 shown in black with C-α as sphere. (E) Binding pocket of LCK corresponding to the 3-position in the indole ring (X = Cl). The pocket shown (black) is a representative size from the volume calculations. (F) Ternary complex EC₅₀ values for 8 different aliphatic R groups (R = H, Me, Et, Pr, iPr, cycloPr, Ph, and Bn). The relative volume of the Et group most closely matched the volume calculated from the structure and was associated with the most robust ternary complex formation. The red arrowhead indicates the average modeled pocket volume (23.2 ų). The red dashed line represents a quadratic trendline. Bn, benzyl; cycloPr, cyclopropyl; Et, ethyl; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; iPr, isopropyl; Me, methyl; Ph, phenyl; Pr, propyl; WT, wild-type.

To provide structural insight into the observed activities of different LCK MGDs in the absence of a cocrystal structure, we developed models of the ternary complex for various compounds in Table 1., Table 2., Table 3.. Figure 3C shows a model of CRBN-1-LCK. We prioritized models in which the 3 backbone carbonyls (G-3, G-2, and G-1) on the helical G-loop motif on LCK form hydrogen bonds (H-bonds) to N351, H357, and W400 in CRBN, respectively (Figure 3D), to understand the SARs in Table 1., Table 2., Table 3.. First, our models contained H-bonds between the 2 N-H groups (R₂ and R₃ in Table 1) on the MGD to E377 on CRBN, matching the SAR of N-H over N-methyl preferred at that position (Table 1). Similarly, we observed an H-bond between the carbonyl (X in Table 1) and H353 in CRBN, that has been shown to be involved in other quaternary complexes.²³ In our models, substitutions at the 3-position of the indole point in the direction of LCK and engage a small, hydrophobic pocket. Based on an ensemble of 250 models of CRBN-1-LCK, the volume of the pocket engaged by substitutions at this position was on average 23.2 ų (Figure 3E). The pocket size agrees well with the SAR observed in Table 1 in which the ethyl group in compound 7 (with an approximate volume of 23 ų) outperforms all other analogous compounds (larger and smaller), as shown in Figure 3F for 8 aliphatic groups (supplemental Table 3).

Although our model explains this SAR, to accommodate binding at the 4 to 7 positions of the MGD indole ring, the flexible LCK activation loop, known to play a role in inactive LCK, requires a conformational change (Table 2). This suggests that substitutions at the 7-position are directed toward a more solvent exposed region (also toward CRBN) may be preferable because these may not need to pay the energetic cost to reorganize the activation loop. Our model suggests 2 possible interactions for the substitutions at the 7-position in Table 3: the backbone carbonyl in A386 in the LCK activation loop within ∼3 Å, and E377 in CRBN at ∼4.5 Å; both distances are consistent with a ∼2 bond-length group, such as those in compounds 17 and 19.

LCK MGDs overcome leukemia resistance to LCK inhibitor and inhibitor-based PROTAC

LCK is a selective therapeutic target in T-ALL and inhibitors such as dasatinib are highly effective.^12,15 However, leukemia can develop drug resistance by acquiring mutations in the LCK gene, especially at the gatekeeper position T316.^14,18 The mutant LCK protein is expected to lose its ability to bind dasatinib and this should also extend to dasatinib-based PROTACs. To test this, we established isogenic clones of KOPT-K1 cells with either mutant or wild-type LCK. As shown in Figure 4A, T316I mutation led to a 4.8- and 10.0-fold increase in the area under the curve of dasatinib and SJ11646, respectively, compared with cells with wild-type LCK. By contrast, compounds 16 through 19 retained strong cytotoxicity even in KOPT-K1 cells harboring LCK mutation (LC₅₀: 27.9nM-85.96nM; Figure 4B). Consistent with this, LCK degradation by the PROTAC was greatly compromised in the mutant cell line relative to the parental control, whereas MGDs induced a comparable level of LCK degradation in KOPT-K1 cells regardless of LCK genotype (Figure 4C). We reason that this can be attributed to the fact that MGDs engage LCK in a region distal to the ATP-binding pocket used by dasatinib and dasatinib-derived PROTACs. (Figure 4D). Therefore, LCK MGDs can overcome small-molecule inhibitor resistance, which is a liability to treatment with an inhibitor-based PROTAC.

Figure 4.

LCK MGDs overcome leukemia resistance to LCK inhibitors and PROTAC. (A-B) Dose-dependent cytotoxicity (measured as metabolic activity based on ATP content) in parental and dasatinib-resistant KOPT-K1 harboring a heterozygous LCK gatekeeper mutation T316I by dasatinib or a dasatinib-PROTAC SJ11646 (A), or LCK MGDs (B), measured by CTG assay. The activity was retained by LCK MGDs but greatly shifted by dasatinib or the dasatinib-PROTAC in LCK T316I cells compared with LCK WT cells. Cells were treated with compounds for 72 hours. Data represent mean ± SD from 3 biological replicates. (C) Representative immunoblots showing LCK protein levels in KOPT-K1 with LCK WT and with LCK T316I after treatment with compound 17 at 1000nM or SJ11646 at 10nM for 24 hours. Quantification of LCK degradation from 5 biological replicates normalized to DMSO-treated controls, in parental and dasatinib-resistant KOPT-K1 cells are shown below. (D) LCK structural model showing that binding locations to dasatinib (cyan) and to LCK MGDs (eg, compound 1, green) are different. The C-α of G415 is highlighted as a dark green sphere. GAPDH, glyceraldehyde-3-phosphate dehydrogenase; WT, wild-type.

Effects of LCK MGD on common immunomodulatory drug neosubstrates

To explore the selectivity of our LCK MGDs, we used HEK293 cell line models in which HiBiT was tagged with either of 2 common neosubstrates, namely GSPT1 and CK1α.²³ Degradation of GSPT1 was not observed for compounds 16 through 19 (supplemental Table 1), whereas CK1α was degraded with a DC₅₀ ranging from 10nM to 40nM (supplemental Table 1). Immunoblot analyses also confirmed substantial CK1α degradation in KOPT-K1 cells treated with MGDs 17 and 19 (Figure 5A). As shown in Figure 5B, there is indeed significant structural similarity between LCK and CK1α. On the basis of DC₅₀ for either LCK or CK1α, our series of MGDs clustered into 2 groups (LCK DC₅₀ of ≥100nM or <100nM; Figure 5C). Importantly, compounds that degraded both LCK and CK1α (DC₅₀ of <100nM for both targets; n = 14) exhibited significantly greater activity against KOPT-K1 cells compared with those preferentially degrading CK1α (n = 5; Figure 5D).

Figure 5.

MGDs also show activity against common immunomodulatory drug neosubstrates. (A) Representative immunoblots showing CK1α degradation in KOPT-K1 treated with compounds 17 and 19, or a selective CK1α degrader, SJ0040 (1000nM, 24 hours). (B) Structural overlay illustrating similarity between the active form of LCK (pink) and CK1α (cyan). (C) Degradation activity against LCK and CK1α by MGDs (compounds 1-19, excluding compounds 6 and 15) and CK1α-selective degraders (SJ0040 and SJ3149), as determined by DC₅₀ values at 24 hours. The red dashed line indicates LCK DC₅₀ threshold of 100nM. (D) Comparison of cytotoxicity (measured as metabolic activity based on ATP content) in KOPT-K1 between MGDs capable of degrading both LCK and CK1α (DC₅₀ of <100nM for both; n = 14) and those degrading CK1α (DC₅₀ of <100nM for CK1α only; n = 5). Statistical significance was assessed by Mann-Whitney U test. GAPDH, glyceraldehyde-3-phosphate dehydrogenase.

Discussion

Drug resistance is a formidable challenge in oncology. Although targeted therapeutics offer improved therapeutic indices, they are often undermined by rapid emergence of resistance, often through mutations in the drug target.^4,27 Degradation-based agents, such as PROTACs and MGDs, have superior pharmacodynamic effects thanks to catalyst-like behavior,^2,8,28 but the patterns of drug resistance in this context remain unclear. In this study, we report on the identification and optimization of CRBN-recruiting MGDs that degrade LCK, an essential oncogenic kinase in T-ALL. Importantly, we provide a direct comparison of drug resistance for inhibitor, PROTAC, and MGD for the same target. The T316I mutation in LCK abrogates dasatinib binding and confers resistance both to the inhibitor itself and to dasatinib-based PROTACs. By contrast, our lead MGDs engage LCK through a distinct surface interaction at the helical G-loop degron and maintain activity in T316I-expressing cells. This confirms that MGDs can bypass resistance mechanisms that limit the efficacy of both small-molecule inhibitors and PROTACs based on the same scaffold.

Our data add important evidence for degron space targetable by MGDs beyond the canonical β-hairpin G-loop motif, as proposed by Petzold et al.²⁶ Despite their similarity in sequence and structure, the region in LCK corresponding to the β-hairpin degron of CK1α adopts a different inverted-loop conformation that is predicted to sterically prevent ternary complex formation (Figure 5B). By contrast, our modeling analyses described a helical G-loop that likely mediates the LCK-MGD-CRBN interaction, as confirmed by the G415N mutation studies (Figure 3A). These results will motivate further exploration of “noncanonical” degron motifs (such as a helical G-loop) that are targetable by MGDs, leading to the expansion of neosubstrate space for TPD. Given the importance of G-loop engagement in determining specificity, refinement of MGDs to fit this interface may enable selective recognition of different neosubstrates.²⁹

There are significant structural similarities between LCK and CK1α (Figure 5B), and therefore it was not entirely surprising that our lead MGDs can also potently degrade CK1α, a therapeutic target in hematologic malignancies in its own right (dependency data from DepMap in supplemental Figure 6).^{30, 31, 32, 33} We reason that there may be several potential advantages for a dual degrader over selectively targeting each one individually. First, CK1α dependency is more common than the dependency on LCK in ALL,³³ therefore compounds targeting both would be more broadly active than LCK-specific agents (Figure 5D). Second, in ALL cases for which both LCK and CK1α are essential, they represent 2 independent therapeutic vulnerabilities, and simultaneously targeting nonoverlapping pathways is advantageous because acquiring resistance mutations in 2 unrelated targets is much more difficult than developing resistance for 1 target.^34,35 Future optimization may focus on LCK selectivity to mitigate toxicity while maintaining antileukemic potency.

There are a few limitations of this study. For example, our primary MGD screen and subsequent characterization focused on a small number of LCK-activated T-ALL models. Previous data from us, and others, suggest that TAL1 and SPI1 T-ALLs are likely to have activated LCK and thus responsive to LCK-targeted therapeutics,^10,12 including degraders. Consistent with this, compound 17 demonstrated cytotoxicity similar to that of dasatinib and the PROTAC across a panel of T-ALL cell lines from 6 subtypes, supporting a shared mechanism of LCK-mediated effects (supplemental Figure 7). Notably, unlike dasatinib, LCK MGDs engage the kinase in a region away from the ATP-binding pocket and therefore may not bind other targets such as ABL the same way that dasatinib does. Accordingly, compound 17 showed minimal activity in an ABL-fusion–positive cell line ALL-SIL, which was highly sensitive to dasatinib, further validating the design of our LCK MGDs (supplemental Figure 7). It would be important to identify exactly which kinases can form a ternary complex with CRBN and our MGDs, which would then inform the clinical utility of this series of agents. Nevertheless, these findings will require validation in primary patient samples. Finally, this study focused on MGD identification and much remains unknown in terms of efficacy in vivo, or toxicity of these molecules. Although MGDs generally have better pharmacokinetic properties than PROTACs, more work is needed to optimize our lead compounds and also to characterize their metabolism, disposition, and pharmacodynamic effects in vivo. Given the observed CK1α degradation and its role in normal hematopoiesis,^{36, 37, 38} there may be risk of hematopoietic toxicity with these MGDs, which can be mitigated with targeted drug delivery (conjugation with tumor-recognizing antibodies).

In summary, we demonstrate that LCK-targeting MGDs represent a viable strategy to overcome clinically relevant drug resistance in T-ALL. This work highlights the potential of degrader-based therapeutics to complement or replace conventional kinase inhibitors, especially in settings in which mutation-driven resistance is common.

Conflict-of-interest disclosure: J.J.Y. receives funding from Takeda Pharmaceutical and AstraZeneca, outside the submitted work. The remaining authors declare no competing financial interests.

The current affiliation for Z.R. is Centre for Protein Degradation, The Institute of Cancer Research, Sutton, United Kingdom.