A Covalent Allosteric Molecular Glue Suppresses NRF2-Dependent Cancer Growth

A first-in-class NRF2 inhibitor functions by stabilizing the KEAP1–CUL3 complex to enhance NRF2 degradation, effectively inhibiting the growth of NRF2-dependent malignancies.

Abstract

The NRF2 transcription factor is constitutively active in cancer, in which it functions to maintain oxidative homeostasis and reprogram cellular metabolism. NRF2-active tumors exhibit NRF2 dependency and resistance to chemotherapy/radiotherapy (RT). In this study, we characterize VVD-065, a first-in-class NRF2 inhibitor that acts via an unprecedented allosteric molecular glue mechanism. In the absence of stress or mutation, NRF2 is rapidly degraded by the Kelch-like ECH-associated protein 1 (KEAP1)–cullin3 (CUL3) ubiquitin-ligase complex. VVD-065 specifically and covalently engages Cys¹⁵¹ on KEAP1, which in turn promotes KEAP1–CUL3 complex formation, leading to enhancement of NRF2 degradation. Previously reported Cys¹⁵¹-directed compounds decrease KEAP1–CUL3 interactions and stabilize NRF2, thus establishing KEAP1^C151 as a tunable regulator of the KEAP1–CUL3 complex and NRF2 stability. VVD-065 inhibited NRF2-dependent tumor growth and sensitized cancers to chemotherapy/RT, supporting an open phase I clinical trial (NCT05954312).

Significance:

NRF2 hyperactivation is frequently observed in various solid tumors, including lung, esophageal, and head and neck cancers, highlighting NRF2 as a potential therapeutic target. We report a first-in-class KEAP1-dependent allosteric molecular glue degrader of NRF2, which demonstrated robust monotherapy responses in NRF2-activated cancers and effectively sensitized chemo-refractory tumors to chemotherapy.

See related commentary by Hintzen and Burslem, p. 829

Introduction

The NRF2 transcription factor (encoded by the NFE2L2 gene) functions as a primary cellular defense against oxidative and electrophilic stress (1, 2). In the absence of cellular stress, NRF2 expression is restricted at the protein level by the Kelch-like ECH-associated protein 1 (KEAP1) and cullin 3 (CUL3) E3-ubiquitin ligase complex. KEAP1 functions as the substrate adapter protein on the CUL3 scaffold, directly binding and positioning NRF2 for ubiquitylation. Hundreds of cellular stressors, including reactive oxygen species and electrophilic xenobiotics, lipids, and metabolites, inactivate KEAP1–CUL3–mediated NRF2 degradation. Accumulated NRF2 protein then enters the nucleus to activate the transcription of ∼300 cytoprotective genes. Reactive cysteine residues in KEAP1 sense electrophilic stress, but precisely how this alters KEAP1 function to suppress NRF2 degradation is poorly understood (3, 4).

Cancer evolution favors NRF2 activation, in which it promotes redox homeostasis, metabolic reprogramming, immune suppression, and resistance to chemotherapy, radiotherapy (RT), and immune checkpoint inhibitors (5–7). Gain-of-function “hotspot” mutations in NFE2L2 and loss-of-function mutations in KEAP1 or CUL3 are common in lung and upper aerodigestive cancers, resulting in constitutive NRF2 transcriptional activity (1). Mutation-independent mechanisms of NRF2 activation are also common, including KEAP1 posttranslational modifications and altered protein–protein interactions (PPI) that sterically displace NRF2 from KEAP1 (1, 8). Characterization of these genetic aberrations and functional phenotyping of cells lacking NRF2 have provided compelling evidence in support of targeting NRF2 in cancer (9–11). Mouse models have also established that mutational activation of NRF2 functions in concert with hallmark oncogenes or tumor suppressors to enhance tumor initiation and progression (10, 12, 13). Conversely, genetic disruption of Nfe2l2 suppresses tumor progression in transgenic mouse models of lung, pancreatic, and colorectal cancers (8, 14), strongly supporting NRF2 inhibition as a potential cancer therapy. Because Nfe2l2-knockout (KO) animals are viable, fertile, and exhibit physiologic disorders only at older ages, NRF2-targeted therapies may be well tolerated (15–17). However, transcription factors such as NRF2 are challenging to drug, and currently no direct NRF2 antagonistic therapies exist.

The KEAP1 protein includes 27 cysteines, several of which act as sensors to detect oxidative stress as well as various electrophilic metabolites and xenobiotics (18). Electrophilic attack of these cysteines inactivates KEAP1-mediated NRF2 degradation (19, 20). Cys¹⁵¹ is the most extensively characterized sensor cysteine in KEAP1 and has been the subject of extensive drug discovery efforts (21, 22). FDA-approved KEAP1 inhibitors, such as dimethyl fumarate for multiple sclerosis and omaveloxolone for Friedreich’s ataxia, inhibit KEAP1 function through covalent modification of Cys¹⁵¹ (23, 24). Despite the availability of these approved KEAP1 inhibitors, there remains a strong rationale for developing KEAP1 inhibitors with differentiated profiles for autoimmune disease indications. In search of additional structurally distinct classes of KEAP1 inhibitors, we initiated a chemoproteomics screening–based drug discovery campaign aiming to identify KEAP1_Cys¹⁵¹ ligands that would activate NRF2 for treating autoimmune disorders. Unexpectedly, during the optimization of our early leads, we discovered selective KEAP1^C151-reactive molecules that allosterically activate, rather than inhibit, KEAP1. These molecules operate through a novel molecular glue mechanism, defined by their ability to stabilize an existing PPI (25), specifically enhancing the KEAP1–CUL3 interaction and thereby accelerating NRF2 degradation. Herein, we report the identification, optimization, mechanism of action, and anticancer phenotypes for highly selective, potent, and in vivo efficacious NRF2 inhibitors.

Results

Identification of Covalent Ligands That Activate KEAP1

We utilized an industrialized targeted mass spectrometry (MS)–based chemoproteomic platform to screen a custom library of several thousand electrophilic small molecules for reactivity with several hundred cysteine residues, including KEAP1_Cys¹⁵¹ in native cells or cell lysates. This method measures changes in the covalent binding of cysteine residues to a biotinylated iodoacetamide probe following pretreatment with candidate covalent ligands (26, 27). A primary screen of our covalent fragment library (>1,000 compounds at 500 μmol/L), followed by a secondary screen of >500 compounds at 50 μmol/L, identified VVD-944 as a promising starting point for subsequent exploration (Supplementary Fig. S1A–S1H). Molecules showing potent and selective KEAP1_Cys¹⁵¹ engagement in the chemoproteomics screen were assessed for functional impact on NRF2 using an engineered transcriptional reporter assay (HEK293–ARE-Luc), in which luciferase is under the control of the antioxidant response element (ARE), a canonical binding site for NRF2. Two well-characterized KEAP1 inhibitors, PSTC and bardoxolone methyl, served as reference compounds which are known to increase NRF2 levels through covalent engagement to KEAP1_Cys¹⁵¹ (Fig. 1A; Supplementary Fig. S2A–S2C). Among the many compounds that engaged KEAP1_Cys¹⁵¹, we selected VVD-325 (VVD-944 is the racemate of VVD-325), a simple morpholine acrylamide fragment that displayed submicromolar half-maximal target engagement potency (TE₅₀) as an attractive and ligand-efficient starting point (Fig. 1A; Supplementary Fig. S2A). Further chemical exploration of this chemotype led to VVD-330, showing a KEAP1_Cys¹⁵¹ TE₅₀ equivalent to PSTC and bardoxolone, but, surprisingly, lacking cellular efficacy in the HEK293–ARE-Luc assay (Fig. 1A; Supplementary Fig. S2C). To complement the HEK293 model, which has low basal NRF2 activity, we tested VVD-330 and its related analogues in KYSE70 cells, an esophageal squamous cell carcinoma (ESCC) cell line expressing a constitutively active NRF2^W24C mutation. Remarkably, VVD-330 reduced ARE-Luc activity in this NRF2-activated setting by 52% (Fig. 1B). The terminal phenyl group provided a useful handle for modulating the extent of NRF2 inhibition (I_max), which behaved independently of engagement potency (Supplementary Fig. S2D). For example, I_max of compounds VVD-446, VVD-860, and VVD-065 ranged from 45% to 94% despite having very similar TE₅₀ values (Fig. 1A and B). VVD-065 emerged as an attractive lead compound from this campaign. It potently engaged KEAP1_Cys¹⁵¹ in KYSE70 cells (TE₅₀ = 0.009 μmol/L; Fig. 1C), leading to a robust decrease in NRF2 (Fig. 1D), inhibition of NRF2 transcriptional activity (Fig. 1B), and reduced expression of NRF2 target genes (Fig. 1E). VVD-064, the enantiomer of VVD-065 (TE₅₀ = 0.18 μmol/L; Supplementary Fig. S2E), was functionally inactive (Fig. 1D; Supplementary Fig. S2F–S2I). Chemoproteomic analysis by cysteine-competitive profiling as well as protein enrichment using a VVD-065–derived affinity probe confirmed exquisite binding selectivity of VVD-065 (Fig. 2A and B; Supplementary Fig. S3A–S3G; Supplementary Table S1). Furthermore, global proteomic profiling of cells treated for 24 hours with VVD-065 was conducted to determine the overall impact of VVD-065 on protein networks (Fig. 2C and D; Supplementary Fig. S4A and S4B; Supplementary Table S2). Protein abundance changes with VVD-065 cellular treatment were predominantly restricted to canonical NRF2 target proteins, and these changes were not observed in KYSE70 cells in which KEAP1 or NFE2L2 had been genetically ablated. Given the minimal impact on noncanonical KEAP1 substrates (e.g., PGAM5 and IKBKB), it is evident that KEAP1 is the primary E3 ligase for NRF2 protein stability, although the regulation of these noncanonical substrates is likely cell type–specific or context-specific, potentially involving other E3 ligases (Supplementary Table S2). VVD-065’s impact on NRF2 expression depended on CUL3, KEAP1_Cys¹⁵¹ (Supplementary Fig. S5A–S5C), and proteasome activity (Supplementary Fig. S5D) and was minimally affected by redox stress (Supplementary Fig. S5E and S5F).

Figure 1.

KEAP1_Cys¹⁵¹ liganding by VVD-065 induces NRF2 degradation and inhibits NRF2 target gene expression. A, Progression of KEAP1_Cys¹⁵¹ ligands from the initial hit VVD-325 to more advanced VVD-860 and VVD-065. Bardoxolone methyl and PSTC served as reference compounds. (*, Bardoxolone methyl is a covalent reversible compound, and therefore the value is not directly comparable with the other values.) B, ARE-luciferase reporter activity in NRF2^W24C KYSE70 cells treated for 18 hours with PSTC, bardoxolone methyl, and VVD KEAP1_Cys¹⁵¹ ligands. C, Chemoproteomic determination of KEAP1_Cys¹⁵¹ engagement by VVD-065 in KYSE70 lysates. D, Simple Western analysis of NRF2 and actin in KYSE70 cells treated for 18 hours with VVD-064 and VVD-065 at indicated concentrations. E, qPCR-based expression analysis of NRF2 target genes in KYSE70 cells treated for 18 hours with VVD-065 at indicated concentrations. N.D. = Not Determined (B, Created with BioRender, BioRender.com.)

Figure 2.

VVD-065 is a highly selective ligand of KEAP1_Cys¹⁵¹. A, Cysteine-directed selectivity analysis of VVD-065 in MDA-MB-468 cells (>24,000 sites were surveyed). B, Left, chemical structure of the alkyne probe; right, protein-directed selectivity analysis of VVD-065 in KYSE70 cells following click-chemistry and pull-down of proteins with VVD-369. C, MS analysis of KYSE70 parental cells treated with 100 nmol/L VVD-065 for 24 hours. D, NRF2 expression change in VVD-065–treated KYSE70 parental, KEAP1 KO, and NFE2L2 KO cells as measured by MS.

Removal of the acrylamide olefin in VVD-065 resulted in loss of activity in the ARE-Luc assay (Supplementary Fig. S6A and S6B), and time/dose–response evaluation of VVD-065 engagement rates showed no signs of rate saturation at the highest dose tested of 25 μmol/L, supporting relatively weak reversible affinity (Ki > 25 μmol/L) and a Kinact/Ki value of 3,497 M⁻¹ s⁻¹ (Supplementary Fig. S6C and S6D). Even in the absence of high reversible binding affinity, low reactivity [glutathione (GSH) Kobs/(i): 0.068 M⁻¹ s⁻¹] and shape complementarity probably contributed to such high degree of selectivity, as has been previously observed for fragment-derived covalent drugs (27, 28).

Structural and Mechanistic Evaluation of VVD-065 Activity

To gain mechanistic insights into how VVD-065 affects KEAP1 function, we determined the crystal structure of the KEAP1 BTB domain in complex with this compound at 1.87 Å resolution [Protein Data Bank (PDB) ID: 9DU7; Fig. 3A; Supplementary Fig. S6E–S6G]. As expected, the terminal carbon of the acrylamide of VVD-065 covalently binds Cys¹⁵¹. A hydrogen bond is formed between the side chain of R135 and the carbonyl of the acrylamide, and the morpholine ring projects into the solvent, acting as a linker to spatially orient the reactive acrylamide relative to the binding cleft occupied by the biaryl group. The chlorophenyl moiety sits in a hydrophobic, cleft and the triazine forms π-stacking interactions with H129 and H154. The terminal amine forms a water-mediated hydrogen bond with the backbone amide of H129 (Fig. 3A). Given the distance between the VVD-065 binding site and the KEAP1–NRF2 interface, it was not immediately clear how VVD-065 promotes NRF2 degradation (Fig. 3B; Supplementary Fig. S6F). It has been suggested that the binding of KEAP1_Cys¹⁵¹ inhibitory ligands in this groove decreases the KEAP1–CUL3 interaction through steric displacement of the N-terminal peptide of CUL3 (29); however, this explanation is implausible because occupancy of the same binding site by VVD-065 promotes KEAP1 activity. Comparison of the KEAP1 BTB domain structure in the VVD-065 bound to unliganded or bardoxolone-bound state revealed major conformational changes induced by VVD-065 binding. The side chain of Cys¹⁵¹ is in the p-rotamer (χ angle 55 degrees) when bound to VVD-065, contrasting to other published structures of both APO- (Supplementary Fig. S6H) and bardoxolone-bound (Fig. 3C) KEAP1 BTB domain in which this residue is observed in the m-rotamer (χ angle −65 degrees). To accommodate the rotation of Cys¹⁵¹, M147 (β-strand 5) and F52 (β-strand 1) are pushed down and away from the BTB core. These movements change the hydrophobic packing of the BTB domain and enable helix 6 to pull in toward Cys¹⁵¹. The end of helix 6, defined by the α-carbon of Q177 in the truncated crystallization construct, moves by 2.9 Å. Importantly, the conformation of the KEAP1 BTB domain bound to VVD-065 can be nearly superimposed with the structure of the KEAP1 BTB domain when complexed with CUL3 (Fig. 3D). Both the shifted position of helix 6 and the p-rotamer of Cys¹⁵¹ are consistent between these two structures. The similarity between the KEAP1–VVD-065 and KEAP1–CUL3 complexes suggests that VVD-065 may increase KEAP1 activity by stabilizing a KEAP1 conformation that favors CUL3 binding.

Figure 3.

VVD-065 stabilizes KEAP1–CUL3 complex formation. A, Binding interactions of VVD-065. VVD-065 binds in a hydrophobic cleft and forms a hydrogen bond with the side chain of R135 and a water-mediated interaction with the backbone amide of His129. B, Model of the KEAP1–CUL3 complex highlighting the distance between the KEAP1–NRF2 interface (yellow), the KEAP1–CUL3 interface, and the VVD-065 binding site (red). KEAP1 (light and dark blue) forms a dimer through the BTB domain, and each copy independently binds to CUL3 (gray). VVD-065 binding site in the KEAP1 monomer. C, Superposition of bardoxolone-bound KEAP1 (orange, 4CXT) with VVD-065–bound KEAP1 (blue), highlighting the concerted movement of residues C151, M147, F174, and F52. D, Superposition of APO–KEAP1 (dark red, 4CXI), KEAP1–CUL3 complex (purple and gray, 5NLB), KEAP1–bardoxolone (orange, 4CXT), and KEAP1–VVD-065 (blue). E, Effects of bardoxolone methyl and VVD-065 on the KEAP1–CUL3 interaction measured by HTRF assay. F, MS analysis of KEAP1-interacting proteins in the presence and absence of VVD-065 using a miniTurbo-based proximity labeling assay. G, Viability levels of KYSE70 cells treated with VVD-065 under either the adherent (2D monolayer assay) or nonadherent (3D sphere assay) culture condition. H, Correlation analysis of NRF2 degradation and growth inhibition in cells treated with VVD-065. NRF2 expression was measured using Simple Western. Growth inhibition at a concentration of 0.2 μmol/L was plotted for all cell lines, except for NCI-H23 and KYSE180, for which a concentration of 0.3 μmol/L was used. Mutation status of each cell line is indicated by different colors. Cell lines showing >50% growth inhibition with genetic depletion of NRF2 (Supplementary Fig. S10C) are indicated by filled square. Cell lines with <50% growth inhibition in Supplementary Fig. S10C are indicated by filled circles.

To test this hypothesis biochemically, we used homogeneous time-resolved fluorescence (HTRF) assays to quantify PPIs between KEAP1 and either recombinant CUL3 or NRF2. The KEAP1 activator VVD-065 increased KEAP1–CUL3 binding, whereas the inhibitor bardoxolone decreased this interaction; neither compound affected KEAP1–NRF2 interactions (Fig. 3E; Supplementary Fig. S7A and S7B). Among all the BTB proteins that bind to CUL3, KEAP1 exhibits the weakest binding affinity (30). Consistent with this, kinetic analysis by surface plasmon resonance (SPR) showed weak binding between KEAP1 and CUL3 (K_D = 783 nmol/L). VVD-065 reduced the dissociation rate of the KEAP1–CUL3 complex from 0.01 s⁻¹ to 4.3 × 10⁻⁴ s⁻¹, increasing its binding affinity by more than 10-fold (K_D = 65 nmol/L; Supplementary Fig. S7C).

Consistent with KEAP1–CUL3 binding affinity modulation dictating our molecule’s impacts on NRF2 levels in cells, we found that across a series of molecules, the maximal impact on the KEAP1–CUL3 interaction (E_max) in the biochemical HTRF assay correlated precisely with the I_max observed in the cellular KYSE70–ARE-Luc assay (Supplementary Fig. S7D). VVD-065 promoted KEAP1–CUL3 ternary complex formation in a live-cell HTRF assay (Supplementary Fig. S7E) and enhanced NRF2 polyubiquitination in HEK293 cells (Supplementary Fig. S7F and S7G). KEAP1 turnover was not affected (Supplementary Fig. S8A). To more broadly assess the ability of VVD-065 to alter KEAP1 PPIs, KEAP1-proximal proteins were evaluated in the presence and absence of a ligand using miniTurbo-based biotin proximity labeling. VVD-065 treatment, but not the inactive enantiomer VVD-064, strongly increased CUL3 within the KEAP1 biotinylation sphere (Supplementary Fig. S8B and S8C), without affecting CUL3 levels (Supplementary Fig. S8D). NRF2 was detected as a KEAP1-proximal protein in dimethyl sulfoxide (DMSO)–treated cells, but not VVD-065–treated cells, likely reflecting the overall loss of NRF2 in VVD-065–treated cell lysates. The impact was specific to CUL3 as the KEAP1-associated proteins p62/SQSTM1, MCM3, PGAM5, TSC22D4, and SLK (31) were unaffected by VVD-065 treatment (Supplementary Fig. S8B). MS analysis further revealed strong increases in KEAP1-proximal CUL3 and additional proteins known to associate with the cullin/RBX1 E3 ligase systems such as CAND2 (32) in response to VVD-065 treatment (Fig. 3F).

Overall, these structural and biochemical studies and cellular NRF2 degradation data indicate that VVD-065 binding to KEAP1_Cys¹⁵¹ leads to allosteric changes that increase KEAP1–CUL3 affinity and the abundance of ubiquitination-competent KEAP1–CUL3 complexes in cells.

VVD-065 Activity Requires the KEAP1–NRF2 Interaction

Cancer-derived mutations in NRF2 localize to either the ²⁹DLG³¹ or ⁷⁹ETGE⁸² motifs, which are required for binding each monomer of the KEAP1 homodimer. The ⁷⁹ETGE⁸² motif binds KEAP1 with ∼100-fold greater affinity than the ²⁹DLG³¹ motif, supporting a model wherein ²⁹DLG³¹-mutant NRF2 retains KEAP1 binding via the ⁷⁹ETGE⁸², whereas mutations in the high-affinity ⁷⁹ETGE⁸² motif result in a more profound loss of KEAP1 binding (33). We explored the impact of VVD-065 on a set of cell lines with NFE2L2 mutations that either preserve or eliminate KEAP1–NRF2 binding: (i) NRF2^W24C (KYSE70) and NRF2^G31A (NCI-H2228) that maintain KEAP1 interaction, (ii) NRF2^D77V (EBC1) that moderately compromises the association with KEAP1, and (iii) NRF2^E79K (LK2) that dramatically reduces binding to KEAP1 (33). Similar to what we observed in KYSE70 cells (Fig. 1B), VVD-065 induced a dramatic reduction of NRF2 transcriptional activity in NCI-H2228 cells (Supplementary Fig. S9A). EBC1 cells showed a partial decrease in NRF2 transcriptional activity, whereas no impact was observed in LK2 cells, consistent with the relative KEAP1–NRF2 affinity of the two mutants expressed in these cell models. As a tumor suppressor, KEAP1 mutations are not localized in hotspots and their impact on NRF2 binding is variable and complex (34). KEAP1 “anchor” mutants maintain NRF2 binding, whereas “nonanchor” mutants do not interact with NRF2 (34). We assessed the impact of VVD-065 on NRF2 transcriptional activity in a set of anchor and nonanchor mutants in an overexpression system. The activity of VVD-065 was significantly more pronounced in anchor mutants compared with nonanchor variants (Supplementary Fig. S9B). We expanded this analysis to an endogenous setting in A549 cells that harbor a KEAP1^G333C mutation, which completely disrupts the KEAP1–NRF2 interaction (34). VVD-065 failed to induce NRF2 degradation in these cells, whereas CRISPR editing to convert the mutant KEAP1 allele to wild-type (WT) restored VVD-065 suppression of NRF2 (Supplementary Fig. S9C). As expected, compound activity was seen in WT cell lines such as NCI-H520 and PC9 (Supplementary Fig. S9D). We also confirmed that VVD-065’s functionality is independent of the basal NRF2 activity level, as assessed by the NRF2 gene signature (Supplementary Fig. S9E–S9G; ref. 6). In summary, and consistent with our proposed mechanism of action, VVD-065–mediated NRF2 reduction requires the presence of KEAP1 and a residual ability of KEAP1 to interact with NRF2. These criteria are met in WT as well as a subset of KEAP1- and NFE2L2-mutant settings. It is to be noted that only such “mechanistic sensitive” settings, where VVD-065 degrades NRF2, are relevant for exploring the pharmacologic effects of VVD-065.

VVD-065 Exhibits Cancer Cell Growth Inhibitory Effects

Genetic inactivation of NFE2L2 impedes the growth of NRF2-active human cancer cell lines, and this NRF2 dependency is more profound under nonadherent culture conditions (9, 11). Consistently, VVD-065 inhibited the growth of KYSE70 cells under nonadherent, but not adherent, culture conditions (Fig. 3G). VVD-064, the enantiomer of VVD-065, had no effect on KYSE70 cell proliferation (Supplementary Fig. S10A). Across VVD-065 and a set of related analogues, we observed a strong correlation between ARE-Luc I_max and antiproliferative effects, confirming that the observed antiproliferative activity is driven by NRF2 depletion (Supplementary Fig. S10B). A set of cell lines (n = 14) was chosen for a more comprehensive analysis of VVD-065’s impact on proliferation. These cell lines showed varying levels of NRF2 dependency under 2D culture and 3D spheroid growth conditions (Supplementary Fig. S10C). A subset of these NRF2-dependent cell lines demonstrated sensitivity to VVD-065 (Supplementary Fig. S10D). Mutation of KEAP1_Cys¹⁵¹ to serine in VVD-065–sensitive HCC95 cells abrogated the antiproliferative effect of VVD-065, and there was no inhibition of global protein synthesis, confirming on-target activity of the compound (Supplementary Fig. S10E–S10H). NRF2 expression analysis revealed that all four responsive lines (KYSE70, NCI-H1793, HCC95, and LN-18) are mechanistically sensitive (show loss of NRF2 protein expression) to VVD-065, whereas NRF2-dependent but VVD-065–insensitive cell lines (NCI-H2122, NCI-H2023, A549, and NCI-H1792) showed negligible changes in NRF2 protein levels (<10% reduction; Fig. 3H). Intriguingly, these VVD-065–insensitive cell lines all harbor KEAP1 mutations (Supplementary Fig. S10C), raising the possibility that many KEAP1 mutants do not bind NRF2 with sufficient residual affinity to support VVD-065–mediated NRF2 degradation (34).

VVD-065 Exhibits Antitumor Effects In Vivo

Following oral dosing of VVD-065 in immunocompromised mice bearing KYSE70 tumor xenografts, plasma exposures increased dose-dependently, leading to robust intratumoral KEAP1_Cys¹⁵¹ engagement (Fig. 4A) and decreases in the expression of canonical NRF2 target genes (Fig. 4B). Robust NRF2 degradation was confirmed at the lowest dose level of 5 mg/kg (Fig. 4C). A time-course study in KYSE70 tumors revealed that covalent engagement of KEAP1_Cys¹⁵¹ by VVD-065 reaches maximal levels shortly after a single 5 mg/kg dose (Fig. 4D) and steadily declines over 3 days, at a rate consistent with a combination of compound clearance and KEAP1 protein half-life (11 hours). Repeat daily dosing of VVD-065 for 7 days resulted in higher levels of maximal engagement when compared with single administration consistent with incomplete protein turnover at 24 hours after a single dose. Reduced expression of NRF2 targets both at the transcript and protein levels was observed following repeat dosing of VVD-065 (Fig. 4D; Supplementary Fig. S11A–S11C).

Figure 4.

VVD-065 exhibits antitumorigenic effects in vivo. A–C, Pharmacokinetic (PK) and PD properties of VVD-065 in a KYSE70 xenograft model. KYSE70 tumor-bearing animals were orally dosed at 5, 25, and 50 mg/kg once (A and B) or for 7 days (C). A, Plasma was collected 30 minutes after the dose for PK analysis, and tumors were collected 24 hours after the dose to measure covalent binding of VVD-065 to KEAP1_Cys¹⁵¹. B, Expression of NRF2 target genes in KYSE70 tumors treated with VVD-065 at indicated doses measured by qPCR. C, Expression of NRF2 protein in vehicle-treated or VVD-065–treated tumors. Chemoproteomic determination of KEAP1 Cys151 engagement by VVD-065 in KYSE70 lysates. From left, lanes 1-2 - vehicle, 3-4 5 mg/kg VVD-065 once daily. D, PK–PD time-course analysis in the KYSE70 xenograft model. KYSE70 tumor−bearing animals were orally dosed once (QD × 1) or for 7 days (QD × 7) with VVD-065 at 5 mg/kg. Plasma exposure, KEAP1_Cys¹⁵¹ target engagement, NRF2 protein expression, and expression of NRF2 targets at RNA (AKR1B10, AKR1C1, ALDH3A1, CYP4F11, GPX2, NR0B1, and SLC7A11) and protein (AKR1B10, AKR1C1, ALDH3A1, CYP4F11, and NR0B1) levels were measured at indicated timepoints. E, Antitumor efficacy of VVD-065 in the KYSE70 xenograft model. Data are shown as mean ± SEM; n = 9–10 animals/group. Mice were dosed orally with VVD-065 at indicated doses. F, Antitumor efficacy of VVD-065 in the HCC95 xenograft model. Data are shown as mean ± SEM; n = 10 animals/group. Mice were dosed orally with VVD-065 at indicated doses. For E and F, statistical significance was calculated by two-way ANOVA (ns, P > 0.05; *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001). BID, twice a day; QD, once a day.

Treatment of a KYSE70 xenograft mouse model with doses of VVD-065 ranging from 0.3 to 25 mg/kg was well tolerated and resulted in robust tumor growth inhibition (TGI; Fig. 4E; Supplementary Fig. S12A and S12B). Genetic depletion of NRF2 in KYSE70 resulted in similar pharmacodynamic (PD) and TGI effects (Supplementary Fig. S12C–S12E). TGI studies in additional cell lines revealed that responses in the 3D spheroid assay were generally predictive of in vivo efficacy. HCC95 [NRF2^Amp squamous non–small cell lung cancer (sqNSCLC)] showed strong responses to VVD-065 treatment in both 3D sphere and in vivo settings (Fig. 4F; Supplementary Figs. S10D and S12F), whereas KYSE180 (NRF2^D77V; KEAP1^P278Q ESCC) and NCI-H23 [KEAP1^Q193H lung adenocarcinoma (LUAD)] did not show growth inhibition in either setting (Supplementary Figs. S10D and S12G–S12J) despite exhibiting PD responses to VVD-065 (Supplementary Fig. S12K).

We expanded the in vivo evaluation of VVD-065 to genetically and histologically diverse patient-derived xenografts (PDX; n = 109), with a focus on sqNSCLC, ESCC, head and neck squamous cell carcinoma (HNSCC), and LUAD PDXs, given the high frequency of pathway mutations in these cancer types (Fig. 5A–D; Supplementary Table S3; refs. 35–38). TGI responses were seen across both WT and KEAP1/CUL3/NRF2 mutant models, with the highest responses being observed in tumors enriched for mutations in the NRF2 pathway. A subset of PDXs were analyzed for PD responses, with representative examples shown below, which identified three distinct groups: (i) tumors that exhibited both PD responses and TGI following treatment with VVD-065 (Fig. 5E; Supplementary Fig. S13A); (ii) tumors that did not show PD responses or TGI (Fig. 5F; Supplementary Fig. S13B); and (iii) tumors that showed significant VVD-065–dependent NRF2 degradation, but with no impact on growth (Fig. 5G; Supplementary Fig. S13C). These observations are consistent with nonadherent cell growth characterization of tumor lines, in which models varied in both their intrinsic NRF2 dependence and their PD response to VVD-065. In a model in which a strong response (partial regression) to VVD-065 was observed, the response was durable for at least 90 days with continued treatment (Fig. 5H; Supplementary Fig. S13D). VVD-065 treatment was safe and well tolerated across these PDX models (Supplementary Fig. S13E–S13L).

Figure 5.

VVD-065 exhibits antitumorigenic effects in PDX models. A–D, Percent change in tumor volume on response calling day (days 20–28 after treatment initiation) in indicated sqNSCLC (A), ESCC (B), HNSCC (C), and LUAD (D) PDX models after VVD-065 administration. VVD-065 was administered orally at 50 mg/kg once a day (QD) or twice a day (BID) to mice bearing the indicated PDX models (n = 2–3/group). Bottom, amino acid changes in NRF2/KEAP1/CUL3 are shown for PDX models. WT models are indicated by green filled boxes. Dotted line represents the 50% TGI threshold. E, Antitumor efficacy of VVD-065 in LU1258, LU6403, LU6407, and LU6917 PDX models. Data are shown as mean ± SEM; n = 2–3 animals/group. Mice were dosed orally with VVD-065 at indicated doses. F, Antitumor efficacy of VVD-065 in the ES3864 PDX model. Data are shown as mean ± SEM; n = 2 animals/group. Mice were dosed orally with VVD-065 at indicated doses. G, Antitumor efficacy of VVD-065 in ES3882 and LU2639 PDX models. Data are shown as mean ± SEM; n = 2–3 animals/group. Mice were dosed orally with VVD-065 at indicated doses. H, Antitumor efficacy of VVD-065 in the HN0696 PDX model. Data are shown as mean ± SEM; n = 3 animals/group. Mice were dosed orally with VVD-065 at indicated doses. For E–H, statistical significance was calculated by two-way ANOVA (ns, P > 0.05; *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001).

VVD-065 Exhibits Strong Antitumor Effects in Syngeneic Orthotopic Settings

We were intrigued by the relatively high rate of both cell lines and PDX models with NRF2 pathway mutations showing strong PD responses to VVD-065 in the absence of effects on growth. Although it is not unusual for a subset of cell lines or tumors harboring known oncogenic drivers to show no dependence on that driver (39, 40), we considered that the complexity of NRF2 biology may not be fully recapitulated by heterotopic xenografts in immunocompromised mice. Given the high oxygenation of the microenvironment, lung tissues and tumors are particularly vulnerable to oxidative stress, potentially contributing to the high rates of NRF2 pathway activation observed in lung cancers (1, 41). To explore potential contributors to NRF2 dependence in a more histologically relevant context, we tested VVD-065 effects on the growth of the KLN-205 (WT murine sqNSCLC) cell line introduced in a syngeneic orthotopic setting (42). Despite being insensitive to VVD-065 in a 3D sphere assay (Supplementary Fig. S14A), VVD-065 treatment resulted in a potent antitumor benefit with a 37% increase in median overall survival compared with vehicle-treated animals (Fig. 6A and B; Supplementary Fig. S14B and S14C). Modest but significant benefit was also seen in the syngeneic heterotopic setting (Fig. 6C).

Figure 6.

VVD-065 exhibits strong antitumor effects in syngeneic settings. A and B, Antitumor efficacy of VVD-065 in the KLN-205 syngeneic orthotopic model. KLN-205 cells were injected into the tail vein of DBA/2 animals and then dosed with vehicle or VVD-065 50 mg/kg once a day (QD). A, Representative hematoxylin and eosin images of lungs from non-tumor bearing animals, vehicle-treated animals, and VVD-065–treated animals. B, Kaplan–Meier survival analysis comparing VVD-065 with vehicle (n = 30 animals/group). Statistical significance was calculated by the Gehan–Breslow–Wilcoxon test (****, P < 0.0001). C, Antitumor efficacy of VVD-065 in the KLN-205 syngeneic heterotopic model. Data are shown as mean ± SEM; n = 10 animals/group. Mice were dosed orally with VVD-065 at indicated doses. Statistical significance was calculated by two-way ANOVA (****, P ≤ 0.0001). OS, overall survival.

VVD-065 Shows Combination Benefit with Chemotherapy and RT

As the upregulation of NRF2 is associated with chemo- and radio-resistance (7, 43–47), we tested VVD-065 in combination with these treatment modalities. In two squamous models (VMRC-LCP for sqNSCLC and KYSE70 for ESCC), coadministration of VVD-065 and cisplatin exhibited a robust combination benefit without any significant additional weight loss (Fig. 7A and B; Supplementary Fig. S14D and S14E). Coadministration of VVD-065 with other agents such as nab-paclitaxel, 5-fluorouracil, and gemcitabine also demonstrated robust benefits of combination therapy (Fig. 7C–E; Supplementary Fig. S14F–S14H). Modest combination efficacy was observed with pemetrexed (Supplementary Fig. S14I and S14J). Finally, we tested whether VVD-065 enhanced the radiosensitivity of tumors. In immunocompetent mice with subcutaneously implanted MOC1 (WT murine HNSCC) cells (48), VVD-065 or RT as monotherapy treatment achieved only modest TGI, whereas the combination of VVD-065 and RT showed nearly complete inhibition of tumor growth (Fig. 7F). These findings collectively indicate that pharmacologic degradation of NRF2 can sensitize tumors to a range of chemotherapies and RTs.

Figure 7.

VVD-065 sensitizes tumors to chemotherapy and RT. A, Antitumor efficacy of VVD-065 and cisplatin combination in the KYSE70 xenograft model. Data are shown as mean ± SEM; n = 9–15 animals/group. Mice were dosed with VVD-065, cisplatin, or with combination of VVD-065 and cisplatin. VVD-065 was dosed orally, and cisplatin was dosed intraperitoneally at indicated doses. B, Antitumor efficacy of VVD-065 and cisplatin combination in the VMRC-LCP xenograft model. Data are shown as mean ± SEM; n = 10 animals/group. Mice were dosed with VVD-065, cisplatin, or with combination of VVD-065 and cisplatin. VVD-065 was dosed orally, and cisplatin was dosed intraperitoneally at indicated doses. C, Antitumor efficacy of VVD-065 and nab-paclitaxel combination in the LU3345 PDX model. Data are shown as mean ± SEM; n = 3 animals/group. Mice were dosed with VVD-065, nab-paclitaxel, or with combination of VVD-065 and nab-paclitaxel. VVD-065 was dosed orally, and nab-paclitaxel was dosed intravenously at indicated doses. D, Antitumor efficacy of VVD-065 and 5-fluorouracil (5-FU) combination in the HN2174 PDX model. Data are shown as mean ± SEM; n = 2 animals/group. Mice were dosed with VVD-065, 5-FU, or with combination of VVD-065 and 5-FU. VVD-065 was dosed orally, and 5-FU was dosed intraperitoneally at indicated doses. E, Antitumor efficacy of VVD-065 and gemcitabine combination in the BL9215 PDX model. Data are shown as mean ± SEM; n = 2 animals/group. Mice were dosed with VVD-065, gemcitabine, or with combination of VVD-065 and gemcitabine. VVD-065 was dosed orally, and gemcitabine was dosed intraperitoneally at indicated doses. F, Antitumor efficacy of VVD-065 radiation combination at indicated doses in the MOC1 syngeneic heterotopic model. Data are shown as mean ± SEM; n = 10 animals/group. For A–F, statistical significance was calculated by two-way ANOVA (*, P ≤ 0.05; ***, P ≤ 0.001; ****, P ≤ 0.0001). BID, twice a day; QD, once a day; QW, once a week; Q3D, every 3 days; Q7D, every 7 days.

Discussion

The discovery of VVD-065 represents a new mechanistic paradigm for modulating the function of the KEAP1–NRF2 transcriptional circuit. By focusing primary screens on KEAP1_Cys¹⁵¹ engagement, followed by functional evaluation of confirmed binders, we discovered molecules with the unexpected ability to promote KEAP1-mediated NRF2 degradation through an allosteric molecular glue mechanism. These molecules enhance KEAP1 function by increasing KEAP1–CUL3 complexation, leading to enhanced NRF2 degradation in both WT and a significant subset of cancer-associated NRF2 pathway mutations. This contrasts with previous findings reporting KEAP1 inhibition through Cys¹⁵¹-reactive compounds and presents a compelling and potentially generalizable case study for inducing degradation of a protein of interest through boosting the activity of its cognate E3 ligase. Furthermore, we posit a new model wherein dynamic pharmacologic regulation of the KEAP1–CUL3 interaction, as opposed to the KEAP1–NRF2 interaction, can govern NRF2 degradation. It is intriguing to observe that covalent ligation of KEAP1_Cys¹⁵¹ can produce a range of agonistic, silent, and antagonistic effects. Although it is rare for a minor structural modification to reverse the mode of action, modulators with distinct scaffolds have been described to bind and stabilize pockets differently (49). The unusually low affinity between KEAP1 and CUL3 among characterized E3–CUL interactions may also be important for permitting regulation of NRF2 levels through bidirectional modulation of the KEAP1–CUL3 PPI.

Hyperactivation of NRF2 is a prevalent feature in various solid tumors. Increased NRF2 activity promotes aggressive tumor growth by disrupting redox homeostasis, amino acid metabolism, and autophagy while suppressing antitumor immune responses. Tumors such as sqNSCLC and ESCC, which frequently exhibit NRF2 pathway activation, often lack other actionable oncogenic drivers, making NRF2 an attractive therapeutic target. Despite robust preclinical validation of NRF2 as a therapeutic target, no direct NRF2-targeting agent has been available for clinical evaluation. Previous attempts to exploit metabolic vulnerabilities in NRF2-hyperactivated tumors, using agents like the glutaminase inhibitor CB-839 and the mTOR1/2 inhibitor sapanisertib, have yielded limited clinical benefit (50), underscoring the need for alternative approaches. Consistent with genetic studies, VVD-065–mediated NRF2 degradation elicited strong antitumor activity in cell line–derived xenograft (CDX) and PDX models. A related compound, VVD-130037, which shares the same mechanism of action, is currently being evaluated in a phase I clinical trial (NCT05954312).

In addition to its effects as monotherapy, VVD-065 demonstrates synergistic effects when combined with either chemotherapy or RT, suggesting that NRF2 depletion may have therapeutic relevance even in tumor types in which NRF2 is not the primary oncogenic driver. Emerging evidence also implicates NRF2 activation in resistance to targeted therapies such as KRAS^G12C inhibitors, EGFR inhibitors, and tyrosine kinase inhibitors like sorafenib. Although not the focus of the current study, these combination strategies warrant further investigation.

A key limitation of the approach described in this article is that certain mutations in KEAP1, NFE2L2, and CUL3 can render tumors mechanistically insensitive to KEAP1 activators. Additionally, KEAP1 promoter hypermethylation may contribute to reduced efficacy if KEAP1 levels are dramatically reduced. Ongoing efforts are focused on cataloging these genetic and epigenetic alterations to better define responsive patient populations and optimize clinical application of KEAP1 activators.

In summary, we report the discovery of a novel NRF2 inhibitor that acts via KEAP1 activation, establishing a new class of molecular glue NRF2 degraders. These findings highlight the broad therapeutic potential of this strategy in NRF2-hyperactivated tumors.

Methods

ARE Luciferase Reporter Assay

KYSE70, LK2, EBC1, NCI-H520, PC9, NCI-H2228, and HEK293T cells were engineered with the PiggyBac transposon system to contain NRF2-dependent ARE luciferase reporter cassettes. KYSE70, LK2, EBC1, NCI-H520, PC9, and NCI-H2228 cells were plated on Corning 96-well white/clear flat-bottom microplates at 25,000 cells/well in 100 μL of appropriate cell culture media on day 0. HEK293T cells were plated on Corning 384-well white, clear bottom microplates at 20,000 cells/well in 50 μL media. All cells were incubated overnight (for a minimum of 18 hours) at 37°C and 5% CO₂. On day 1, cells were treated with compounds. After 6 hours, to compound-treated HEK293T cells only, 50 μL of reconstituted room temperature One-Glo EX reagent was added to each well, and the contents were mixed for 2 minutes on a plate shaker at <200 rpm to lyse cells. Plates were then incubated for 3 to 5 minutes at room temperature to allow for stabilization of luciferase signal prior to measuring and reading Firefly luciferase signal on a CLARIOstar Plus plate reader. On day 2, after 18 to 20 hours of treatment time of KYSE70, LK2, EBC1, NCI-H520, PC9, and NCI-H2228 cells, 80 μL of reconstituted room temperature One-Glo EX reagent was added to each well, and the contents were mixed for 2 minutes on a plate shaker at <200 rpm to lyse cells. Plates were incubated for 3 to 5 minutes at room temperature to allow for stabilization of luciferase signal prior to measuring and reading Firefly luciferase signal on a CLARIOstar Plus plate reader.

Compound Treatment

Cells were treated with a dose response of Vividion compounds in 0.05% DMSO (except HEK293 ARE luciferase in which it was 0.1%) using an HP digital dispenser. Each concentration was tested in duplicate.

For in vivo studies, VVD-065 was formulated in 10% N-methyl-2-pyrrolidone/90% Labrafil m1944CS.

For the MG-132/TAK-243/MLN-4924 experiment, cells were plated on Corning 96-well white/clear flat-bottom microplates at 25,000 cells/well in 100 μL media on day 0 and incubated overnight (for a minimum of 18 hours) at 37°C and 5% CO₂. On day 1, cells were pretreated with 10 μmol/L MG-132/TAK-243/MLN-4924 for 5 hours prior to the addition of compound for 1 hour prior to harvesting for NRF2 expression analysis by Western blotting.

Protein Simple Western Blot Analysis

In Vitro

Cells were harvested for protein analysis by the addition of 25 μL of prepared lysis buffer (RIPA lysis buffer, Halt protease/phosphatase inhibitor cocktail, and benzonase nuclease; 25–29 U/μL) directly to cell culture plate wells. Eight microliters of each cell lysate was mixed with 2 μL of 5× Jess sample buffer (provided in EZ Standard Pack and prepared as per the manufacturer’s instructions). Ten microliters of Jess samples was boiled for 10 minutes at 95°C. The remaining lysate samples were used to determine the protein concentration using the Bio-Rad RC DC protein assay. After boiling, using the determined protein concentration, Jess lysates were normalized using 1× Jess sample buffer (5× diluted with prepared RIPA lysis buffer) to a final concentration of 1.0 μg/μL. Three microliters of each normalized lysate was loaded into each sample well on a prefilled plate provided in the Jess Separation module; all corresponding reagent wells were loaded as per the manufacturer’s instructions, following the RePlex protocol. The loaded plate was then centrifuged at 1,000 × g for 5 minutes to bring all contents to the bottom of each well. Both the sample-loaded plate and a capillary cassette were loaded onto the ProteinSimple machine as per the manufacturer’s instructions.

In Vivo

One microliter of cold Pierce IP lysis buffer (Thermo Fisher Scientific, PN 87788) was added to frozen tissues in bead beater tubes, and tissues were homogenized by bead beating (Omni Bead Ruptor Elite) at 4°C for 30 seconds. Homogenized samples were centrifuged for 3 minutes at 7,400 × g. Supernatants were transferred to cluster tubes (Axygen, PN MTS-11-12-C) and treated with 24 μL of 1:1 solution of Benzonase (MilliporeSigma, PN 70746-3) and 100 mmol/L MgSO₄ (Sigma-Aldrich, PN M2643-500g) for 10 minutes at room temperature with shaking at 600 rpm. Lysates were cleared by centrifugation at 1,000 × g, 4°C for 10 minutes, and the supernatants were transferred to a 96-well plate (Lab Force, PN 1149J81). Eight microliters of tissue homogenate was mixed with 2 μL of 5× Jess sample buffer (provided in EZ Standard Pack and prepared as per the manufacturer’s instructions). Ten microliters Jess samples was boiled for 10 minutes at 95°C. The remaining lysate samples were used to determine the protein concentration using Bio-Rad RC DC protein assay (cat. #5000122). Rest of the protocol follows the in vitro method described above.

For data analysis, raw chemiluminescence detection data were exported from the Jess instrument within Compass for Simple Western software.

The following antibodies were used:

• NRF2 (1:100 dilution; Cell Signaling Technology, # 20733);

• β-actin (1:50 dilution; Cell Signaling Technology, # 4970); and

• KEAP1 (1:50 dilution; Cell Signaling Technology, # 8047).

Cell Lines and Culture

Cell Line Source

The source of cell lines and their catalog number are provided below. All cell lines were confirmed negative for Mycoplasma and viruses by h-IMPACT1 testing (IDEXX BioAnalytics).

• KYSE70 cells (DSMZ; cat. #ACC363);

• HCC95 cells (Sigma-Aldrich; cat. #SCC483);

• VMRC-LCP cells (JCRB; cat. #JCRB0103);

• NCI-H520 cells (ATCC; cat. #HTB-182);

• NCI-H1793 cells (ATCC; cat. #CRL-5896);

• EBC1 cells (AcceGen; cat. #ABL-TC0170);

• NCI-H1623 cells (ATCC; cat. #CRL-5881);

• NCI-H2228 (ATCC; cat. #CRL-5935);

• LK2 (RIKEN; cat. #RCB1970);

• NCI-H23 (ATCC; cat. #CRL-5800);

• KYSE180 (DSMZ; cat. #ACC 379);

• SK-MES-1 (ATCC; cat. #HTB-58);

• LN-18 (ATCC; cat. #CRL-2610);

• NCI-H2122 (ATCC; cat. #CRL-5985);

• A549 (ATCC; cat. #CCL-185);

• NCI-H460 (ATCC; cat. #HTB-177);

• NCI-H2023 (ATCC; cat. #CRL-5912);

• BICR16 (Sigma-Aldrich; cat. #06031001);

• PC9 (Sigma-Aldrich; cat. #90071810); and

• KLN-205 (ATCC; cat. #CRL-1453).

Cell Culture Conditions

Cell lines were maintained according to the vendor handling instructions.

CRISPR−Cas9-Mediated Knock-in of KEAP1 C333G in A549

A549 has a homozygous KEAP1^G333C mutation. Using proprietary methods, Synthego generated a knock-in cell line, in which both mutant KEAP1 alleles were reverted to WT.

CRISPR−Cas9-Mediated KO of NRF2 in Cell Lines

Gene Knockout Kit was purchased from EditCo Bio, containing the following guides:

• NFE2L2: guide_#1 GCGACGGAAAGAGUAUGAGC, guide_#2 AUUUGAUUGACAUACUUUGG, and guide_#3 UAGUUGUAACUGAGCGAAAA.

• CUL3: guide_#1 ACAAUGGUUUUGCAUAAACA, guide_#2 AUCCAGCGUAAGAAUAACAG, and guide_#3 UACCUUAUUUAUGAGAUGUU.

• KEAP1: guide_#1 ACCAACGGGCUGCGGGAGCA, guide_#2 UGGGCCAUGAACUGGGCGGC, and guide_#3 CCGUGUAGGCGAAUUCAAUG.

A guide pool was reconstituted using Nuclease-Free Duplex Buffer [Integrated DNA Technologies (IDT), #11-01-03-01] to a final concentration of 100 μmol/L. NFE2L2 was knocked down using CRISPR editing via Ribonucleoprotein (RNP) electroporation. First, the RNP was complexed by mixing 3 μL of Synthego NRF2 Gene Knockdown Kit V2 (pooled single-guide RNAs) with 2 μL Cas9 (IDT; #1081059). The complex was flicked to mix and incubated for 10 to 20 minutes at room temperature. While the RNP incubated, 500,000 cells were pelleted at 5,000 rpm for 1 minute, resuspended and washed in 1 mL Dulbecco’s Phosphate-Buffered Saline (DPBS), and re-pelleted at 5,000 rpm for 1 minute. Cell pellets were resuspended in 20 μL prepared electroporation buffer (16.4 μL SE or P3 + 3.6 μL supplement buffer per reaction) provided by the Lonza 4D-Nucleofector X kit (SE, Thermo Fisher Scientific, #NC0828118; or P3, Lonza; #V4XP-3032). Twenty microliters of the cell solution was then transferred to the 5 μL RNP solution and gently mixed by pipetting. The cell + RNP mixture was then transferred to an 8-strip electroporation cuvette. Cells were electroporated using the CM-137 setting. One hundred microliters of warm serum-containing media was added to the cuvette to immediately recover cells. Nucleofected cells were immediately transferred to a 6-well plate containing 2 mL of prewarmed media.

qPCR Analysis

In Vitro

Cells were processed for qPCR by (i) lysis in 50 μL/well of the Cells-to-CT lysis buffer and (ii) the addition of 5 μL stop solution, as per the manufacturer’s instructions. Two microliters of each prepared cDNA lysate was then used directly for qPCR analysis by mixing with 1 μL of each PrimeTime assay probe, 5 μL TaqMan 1-Step qRT-PCR master mix, and 12 μL nuclease-free water for a total 20 μL reaction mix. Prepared qPCR plate(s) were sealed with an adhesive cover before vortexing to mix samples, followed by a brief centrifugation to bring the well contents to the bottom of each well. The standard RT-qPCR cycling, as outlined in the manufacturer’s instructions, was followed. For mRNA expression analysis, raw amplification data were exported into Microsoft Excel from the QuantStudio 6 instrument via the Thermo Fisher Scientific Connect Cloud functionality. Then, employing a ΔCt method of analysis, ΔCt values for each sample were first calculated using housekeeping genes. As all calculations are in the logarithm base 2, expression fold change (FC) was next calculated using the formula 2^−ΔCt. Expression FC values were then used to calculate % inhibition relative to DMSO-treated samples.

In Vivo

Tissue harvested from mice was immersed in 800 μL DNA/RNA Shield (Zymo Research R1200-125). After mechanical homogenization, homogenized tissue was spun down at 10,000 × g for 3 minutes to pellet cell debris. Two hundred microliters of the supernatant was transferred to a 96-well deep-well plate. Next, 10 μL of proteinase K solution was added to each sample and then incubated at room temperature for 30 minutes. After incubation at room temperature, 200 μL DNA/RNA lysis buffer, 30 μL of homogenously suspended mag-beads (Zymo Research PN: R2130/R2131), and 400 μL of 100% ethanol (Acros Organics: 61509-0010) were added to each well sequentially. Then, RNA was eluted using the KingFisher Flex instrument. The RNA concentration was measured afterward using NanoDrop. For real-time RT-PCR analysis, total RNA was reverse-transcribed to cDNA using High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific PN: 4368814) followed by real-time PCR using either PerfeCTa FastMix II Low ROX (Quantabio PN: 97066-004) or TaqMan Fast Advanced Master Mix (Applied Biosystems PN: 4444557) with primers as described below. A total of 10 genes were probed for expression as technical duplicates—four housekeeping genes B2M, HPRT1, TBP, and 18S or ActinB and eight NRF2 target genes AKR1B10, AKR1C1, ALDH3A1, CYP4F11, GPX2, NR0B1, NQO1, and SLC7A11. Expression of 18S or ActinB was used to normalize mRNA levels of the gene of interest (ΔCt). Duplicate ΔCt values were then averaged, followed by normalization to the vehicle group to obtain 2^−ΔΔCt. The 2^−ΔΔCt value of each gene of interest was normalized to the averaged 2^−ΔΔCt of Actin, B2M, HPRT1, and TBP to obtain FC. Probe information is available in Supplementary Table S4.

RNA sequencing Analysis

An RNA sequencing experiment was conducted at Novogene. The Gene Expression Omnibus (GEO) accession number for the sequencing data reported in this article is GSE278482.

Plasmid Information

KEAP1: WT KEAP1 (Promega; #N140A).

KEAP1^C151S: Q5 mutagenesis (NEB) performed using forward primer GGGCGAGAAGaGTGTCCTCCA and reverse primer ATGGAGATGGAGGCCGTG.

Cell Proliferation Assay

2D Assay

Cells were plated on Corning 96-well white/clear flat-bottom microplates at 500 cells/well in 100 μL media on day 0 and incubated overnight (for a minimum of 18 hours) at 37°C and 5% CO₂. On days 1 and 3, cells were treated with compounds. On day 5, 100 μL of CellTiter-Glo (CTG) reagent was added to all wells. The contents were mixed for 2 to 3 minutes on a plate shaker and incubated for 10 minutes at room temperature to allow for stabilization of the CTG signal. Luminescence was then measured on the CLARIOstar plate reader.

3D Sphere Assay

Cells were plated on Corning 96-well black/clear round-bottom microplates at 1,000 cells/well in 100 μL media on day 0 and incubated overnight (for a minimum of 18 hours) at 37°C and 5% CO₂ to allow for 3D spheroid formation. On days 1, 3, 5, 7, and 9, cells were treated with compounds. On day 10, 100 μL of CTG 3D reagent was added to all the wells. The contents were mixed for 10 minutes on a plate shaker and incubated for 25 minutes at room temperature to allow for stabilization of CTG signal. Luminescence was then measured on the CLARIOstar plate reader.

Growth Kinetics Assay

For 2D culture, 1,000 cells were added to each well of a 96-well plate (Corning 96-well flat clear bottom white polystyrene tissue culture–treated microplates #3610) in a volume of 200 μL. For 3D culture, 500 cells were added to each well of a 96-well spheroid plate (Corning 96-well spheroid microplates #4520) in a volume of 200 μL. Cells were then grown for 7 days. On day 7, plates were imaged using Incucyte SX5. Analysis was done using Incucyte software version 2022B Rev2. Two-dimensional analysis used Basic Analyzer Whole Well settings. Percent confluence was measured using an adjusted pixel size of −9 and a minimum area of 800 μm². Three-dimensional analysis used Spheroid analysis settings. The largest bright-field object area (μm²) was determined with an adjusted pixel size of −10 and a minimum area of 8,000 μm².

In Vitro, In Situ, and In Vivo Target Engagement

In Vitro Target Engagement

Pellets derived from the indicated cell lines treated with a covalent fragment library or indicated compounds were lysed into DPBS by sonication, and the protein concentration was determined via Bio-Rad DC protein assay. Cell lysates were diluted to 2 mg/mL and treated with DMSO or compound for 1 hour at room temperature. Following incubation, iodoacetamide-desthiobiotin (IA-DTB) was added to a final concentration of 200 μmol/L, and lysates were incubated for 1 hour at room temperature. Acetone (8×) was then added to the lysates, and the mixture was incubated at −80°C for 2 hours. Precipitated proteins were then pelleted by centrifugation (4,200 rpm, 45 minutes, 4°C).

In Situ Target Engagement

For the indicated cell lines, cells were suspended at 2.5 × 10⁶ cells/mL of RPMI supplemented with 10% FBS. One microliter of cell suspension was transferred to each well of a 96-well plate. Cells were treated with DMSO or compound and incubated in a CO₂ incubator at 37°C for 2 hours. After 2 hours, cells were pelleted (700 × g, 10 minutes) and washed twice with DPBS. After washing, cells were resuspended in 200 μL DPBS and lysed by sonication using a Qsonica water bath sonicator (50% amplitude, 30 seconds on, 30 seconds off, 2 minutes total sonication). IA-DTB was added to 200 μmol/L, and lysates were incubated for 1 hour at room temperature. Acetone (8×) was then added to the lysates, and the mixture was incubated at −80°C for 2 hours. Precipitated proteins were then pelleted by centrifugation (4,200 rpm, 45 minutes, 4°C).

In Vivo Target Engagement

Frozen tissues were homogenized by bead beating at 4°C for 30 seconds in Pierce IP lysis buffer (Thermo Fisher Scientific). Homogenized tissues were centrifuged for 3 minutes at 7,400 × g, and the resulting supernatants were transferred to a 96-well plate and treated with 24 μL of a 1:1 solution of Benzonase:100 mmol/L MgSO₄ for 10 minutes at room temperature. Lysates were then cleared by centrifugation at 1,000 × g for 10 minutes at 4°C. Supernatants were transferred to a 96-well plate, and the protein concentration was determined via Bio-Rad DC assay. Lysates were diluted to 2 mg/mL with DPBS, and IA-DTB was added to a final concentration of 200 μmol/L. Samples were then incubated with IA-DTB for 1 hour at room temperature. Acetone (8×) was then added to the lysates, and the mixture was incubated at −80°C for 2 hours. Precipitated proteins were then pelleted by centrifugation (4,200 rpm, 45 minutes, 4°C).

Protein pellets generated from in vitro, in situ, and in vivo samples were further processed by resuspending the acetone pelleted material in 9 mol/L urea and 50 mmol/L ammonium bicarbonate. Free cysteine thiols were then reduced and alkylated by the addition of dithiothreitol (DTT) and iodoacetamide (10 and 30 mmol/L, respectively). After reduction and alkylation, samples were exchanged into 2 mol/L urea (Zeba spin desalting plates, Thermo Fisher Scientific) and digested with trypsin for 1 hour. IA-DTB–labeled peptides were then isolated by the addition of 300 μL of 5% high-capacity streptavidin resin (Thermo Fisher Scientific). Bound peptides were washed 3 times with wash buffer (WB; 0.1% NP-40 and 150 mmol/L NaCl in 1× PBS) followed by four washes with high-performance liquid chromatography (HPLC)–grade water. Isolated peptides were eluted from streptavidin resin by the addition of 50% acetonitrile (ACN) and dried in a speed vacuum concentrator.

Target Engagement Quantification by Parallel Reaction Monitoring MS

Target engagement quantification by parallel reaction monitoring (PRM) MS has been described in detail previously (27). Briefly, isolated probe-labeled peptides were reconstituted in 3% ACN and 0.1% formic acid (FA). Peptides were concentrated onto an Acclaim PepMap 100 C18 loading column (100 μm × 2 cm, 5-μm particle size, Thermo Fisher Scientific) and separated on a custom made C18 nanoviper analytic column (75 μm × 15 cm, 2-μm particle size, Thermo Fisher Scientific) using a Dionex UltiMate 3000 nano-LC (Thermo Fisher Scientific). Peptides were separated using a 12.7-minute gradient from 6% to 32.5% solvent B (96.4% ACN, 3.5% DMSO, and 0.1% FA). Peptides were analyzed by PRM MS using product ion scan mode on a Exploris 240 Orbitrap mass spectrometer (Thermo Fisher Scientific) operated in positive ion mode. Detection of precursor ions was scheduled with 2-minute acquisition windows. Product ion scan properties were as follows: 1.5 m/z Q1 resolution, normalized collision energy of 30%, 30,000 Orbitrap resolution, 70% RF lens, normalized automated gain control (AGC) of 200%, and dynamic injection time set to provide seven points across a peak.

PRM data were analyzed using Skyline v.21.2.0.369 (MacCoss Lab, University of Washington). Peptide quantification was performed by summing the peak areas corresponding to six fragment ions preselected from an in-house reference spectral library. Retention time standard peptides were used to normalize for sample variability. KEAP1^C151 target engagement is determined by monitoring the loss of IA-DTB probe labeling of the C151-containing peptide, CVLHVMNGAVMYQIDSVVR. Target engagement is estimated by comparing the average peptide area under the curve (AUC) for each dose group/test article to the average peptide AUC of the DMSO/vehicle group.

Global Chemoproteomics Profiling

KYSE70 and MDA-MB-468 cell lysates were treated with 10, 2, 0.4, 0.08, 0.016, 0.0032, and 0.00064 μmol/L VVD-065 for 1 hour at room temperature and prepared as described for target-engagement (TE) determination. After streptavidin enrichment and elution, the resulting dried peptides were resuspended in 70 μL of 0.2 mol/L EPPS pH 8.5, 30 μL of anhydrous ACN, and 55 μg of TMTpro 16plex reagent (Thermo Fisher Scientific, A44520). After a 2-hour incubation at room temperature, tandem mass tag (TMT) labeling was quenched by the addition of 6.5 μL of 5% hydroxylamine. Samples were then combined and desalted on a Biotage EVOLUTE EXPRESS ABN plate. The resulting peptides were then separated into 96 fractions by high-pH reversed-phase chromatography. The 96 fractions were then recombined into 24 fractions, and 12 of them were then subjected to MS analysis.

Peptides were reconstituted in 5% ACN and 0.1% FA and concentrated onto an Acclaim PepMap 100 C18 loading column (100 μm × 2 cm, 5 μm particle size, Thermo Fisher Scientific) and then resolved on an Acclaim PepMap 100 analytic column (75 μm × 25 cm, 2 μm particle size, Thermo Fisher Scientific) on a Dionex UltiMate 3000 nano-LC (Thermo Fisher Scientific). Peptides were separated using a 171-minute gradient from 6% to 30% solvent B (96.4% acentonitrile, 3.5% DMSO, and 0.1% FA). The gradient then went from 30% to 40% B over 1 minute, 40% to 50% B over 1 minute, and 50% to 60% B over 1 minute. Quantitative TMT measurements were acquired on an Orbitrap Lumos Tribrid mass spectrometer (Thermo Fisher Scientific) using standard MS3 SPS data acquisition settings.

Raw files were searched with the MasSPIKE software package (GFY development team and the President and Fellows of Harvard University, GFY Core Version 3.4). Data were searched against the human FASTA database with a peptide mass tolerance of 30 ppm, a fragment ion tolerance of 0.8, trypsin digestion with a maximum of one internal cleavage site, and a maximum of three differential modifications per peptide. Modification settings were as follows: static carboxyamidomethylation (+57.0214637236); static modification of lysine and peptide N-terminus (+304.2071); differential oxidation of methionine (+15.99); and differential IA-DTB labeling (+267.1945) of cysteines.

Searched data were exported from MasSPIKE and further analyzed using our in-house data analysis algorithms. In brief, data are filtered such that all quantified peptides have an MS3 isolation specificity of ≥0.5, a sum signal for control wells of ≥40, control well coefficient of variation (CV) <0.5, and a max group standard deviation (SD) of <25. Data are represented as % TE relative to controls. Data for all potential off-targets were manually inspected, and sites were excluded from off-target consideration if replicate data were discordant or TE did not exhibit a logical dose response.

Protein Selectivity Analysis

KYSE70 cells were lysed by sonication in PBS. The protein concentration was adjusted to 10 mg/mL. DMSO (n = 2) or VVD-065 (2 μmol/L final, n = 2) was added to 5 mg of KYSE70 lysate and incubated at room temperature for 1 hour. After 1 hour, the alkyne probe, VVD-369, was added to all samples at a final concentration of 5 μmol/L, and samples were again incubated for 1 hour at room temperature. Following incubation, biotin was clicked onto VVD-369 by the addition of Biotin–PEG3–azide, CuSO₄, Tris (2-carboxy-ethyl)-phosphine-HCl (TCEP), and tris((1-benzyl-4-triazolyl)methyl)amine (TBTA) [1.7 mmol/L in DMSO tBuOH (1:4 v/v)] at final concentrations of 100 μmol/L, 2 mmol/L, 1 mmol/L, and 100 μmol/L, respectively. The click reaction was allowed to proceed for 1 hour at room temperature. Proteins were then precipitated by the addition of methanol:chloroform. Precipitated proteins were resuspended in 8 mol/L urea containing 0.2% SDS. After resuspension, samples were reduced with DTT (15 minutes at 65°C) and alkylated (room temperature for 30 minutes). To isolate proteins engaged by the biotinylated alkyne probe (VVD-369), lysates were incubated with streptavidin agarose beads for 1.5 hours at room temperature. The resin was then washed 2 times with PBS, 1 time with water, and 1 time with 200 mmol/L EPPS (pH 8.0). The resin was then resuspended in 2 mol/L urea and 200 mmol/L EPPS. To digest isolated proteins off streptavidin beads, 2 μg trypsin was added to each sample, and samples were incubated overnight at 37°C. The resulting peptides were TMT-labeled and analyzed by MS as described in the global chemoproteomics profiling methods.

Raw files were searched using the MasSPIKE software package (GFY development team and the President and Fellows of Harvard University, GFY Core Version 3.4). Data were searched against the human FASTA database with a peptide mass tolerance of 30 ppm, a fragment ion tolerance of 0.8, trypsin digestion with a maximum of one internal cleavage site, and a maximum of three differential modifications per peptide. Modification settings were as follows: static carboxyamidomethylation (+57.0214637236); static modification of lysine and peptide N-terminus (+304.2071); and differential oxidation of methionine (+15.99).

Searched data were exported from MasSPIKE and further analyzed using our in-house data analysis algorithms. In brief, data are filtered such that all quantified peptides have an MS3 isolation specificity of ≥0.5, a sum signal for control samples of ≥30, and control sample CV < 0.2. Protein identification required ≥3 unique peptides.

Crystallization, Data Collection, and Structure Determination

Purified KEAP1 BTB domain was incubated with VVD-065 in 50 mmol/L Tris pH 8.0, 150 mmol/L NaCl, 1 mmol/L TCEP, and 2% DMSO, and the reaction was monitored by intact protein MS. Upon completion, the protein sample was buffer exchanged into 20 mmol/L Tris pH 8.0, 150 mmol/L NaCl, and 1 mmol/L TCEP using a PD-10 desalting column and concentrated to ∼10 mg/mL. Crystals were grown in a 1:1 drop of protein to reservoir solution and equilibrated against 24% PEG 3350 and 100 mmol/L CHES (N-cyclohexyl-2-aminoethanesulfonic acid) pH 9.2 at 20°C. Crystals were cryoprotected by rapid transfer into reservoir solution supplemented with 25% glycerol before flash freezing in LN2. Diffraction data were collected on Advanced Light Source Beamline 5.0.2 and processed using XDS. The structure was determined by molecular replacement in Phaser using a single chain of 4CXT as a search model. The structure was refined using iterative rounds of refinement in REFMAC5 with manual inspection and model building in COOT. Ligand restraints for VVD-065 and the covalent bond with Cys¹⁵¹ were generated in JLigand. Waters were automatically added in COOT and REFMAC5 and manually inspected. Supplementary Table S5 shows the crystallography parameters (values in parenthesis include only reflections from the highest resolution shell). The PDB validation report is available in Supplementary Method S1.

Recombinant Protein Expression and Purification

The gene for the KEAP1 BTB domain for crystallization (residues 48–180 with the S172A mutation) was synthesized and cloned into pET28b with an N-terminal hexa-histidine tag and a thrombin cleavage site. The gene WT full-length KEAP1 was synthesized and cloned into pFastBac1 with an N-terminal FLAG-6xHis tag followed by a TEV protease cleavage site. The gene encoding the N-terminal fragment of CUL3 (residues 1–388) was synthesized cloned into pGEX-6P-1 with an N-terminal GST tag and PreScission Protease (GenScript) cleavage site. For Escherichia coli expression, plasmids were transformed into BL21 DE3 Star. Large-scale cultures were grown in 2× YT medium at 37°C to an OD600 of 0.6 and induced with 0.4 mmol/L IPTG at 18°C overnight. Full-length KEAP1 was expressed in ExpiSf9 cells using baculovirus-mediated expression in the EmBacY viral genome. The KEAP1 BTB domain for crystallization was purified using the following procedure. Cells were resuspended in 50 mmol/L HEPES pH 8, 500 mmol/L NaCl, 10% glycerol, 10 mmol/L imidazole, and 1 mmol/L TCEP and lysed using a Microfluidizer. The lysate was cleared by centrifugation at 25,000 × g for 45 minutes before being loaded onto 5 mL of pre-equilibrated Ni-NTA resin. The resin was washed with 500 mL lysis buffer and eluted with lysis buffer supplemented with 250 mmol/L imidazole. The 6×His tag was removed with thrombin protease, and the sample was dialyzed against lysis buffer. Uncleaved KEAP1 was removed with an additional Ni-NTA purification before being diluted 1:10 in 50 mmol/L HEPES pH 8 and loading onto a POROS HQ ion-exchange column. The protein was eluted with a 20 CV linear gradient from 50 mmol/L to 1 mol/L NaCl, and fractions containing KEAP1 were pooled, concentrated, and injected onto a Superdex S200 column equilibrated in 25 mmol/L HEPES, pH 8, 150 mmol/L NaCl, and 1 mmol/L TCEP. Full-length KEAP1 was purified in a similar manner, omitting the proteolytic cleavage and IEX steps. The N-terminal fragment of Cul3 was captured on Glutathione Sepharose resin and eluted with 10 mmol/L glutathione in lysis buffer, concentrated, and further purified with size-exclusion chromatography.

Intact Protein MS and Rate Determination

To observe covalent ligand engagement of recombinant KEAP1, samples were analyzed following FA quenching on an LC1290 Infinity II instrument coupled to a 6545 Q-TOF LC/MS (Agilent Technologies). A sample volume of 10 μL, equivalent to approximately 1.5 pmol of KEAP1 protein, was injected. The protein was desalted and separated on an Aeris 3.6-μm-wide-bore XB-C8 liquid chromatography column (50 × 2.1 mm², Phenomenex) at 60°C at a flow rate of 0.5 mL/ minute. In liquid chromatography, solvent A comprised 0.1% FA in 99.9% water and solvent B comprised 0.1% FA in 99.9% ACN. The column was equilibrated in 10%  B for 30 seconds, followed by a 3.5-minute gradient from 10% to 70%  B to separate the analytes. This was followed by a 15-second gradient from 70% to 95%  B, a 15-second gradient from 95% to 10%  B, a 15-second gradient from 10% to 95%  B, and finally a 15-second gradient from 95% to 10%  B to clean the column before re-equilibration. Mass spectra were acquired from 700 to 1,700  Da at a resolution of 25,000. A Dual Agilent Jet Stream Electrospray Ionization Source was used for ionization. The gas temperature was set to 325°C, with a flow rate of 10 L/ minute. The nebulizer was set to 45 pounds per square inch, and sheath gas temperature and flow were set to 375°C and 12 L/ minute, respectively. One spectrum was acquired per second with a collision energy of 10 V. The capillary voltage was set to 5,000  V and the nozzle voltage to 2,000  V. The fragmenter, skimmer, and octopole radiofrequency peaks were set at 250, 65, and 750  V, respectively. The resulting data files were deconvoluted to protein masses using Agilent MassHunter BioConfirm software, v.11.0. The biomolecule table containing protein mass and peak intensities was used to quantify the percentage of compound modification relative to the unmodified protein peak, by dividing modified protein intensity by the sum of the unmodified and modified protein intensities. k_obs/[I] was calculated using assumptions for pseudo first-order reaction kinetics [d(VVD-065)/dt = −k  × (VVD-065), (VVD-065)_t  = (VVD-065)_t0 × e^−kt] and averaged for each inhibitor concentration; k_obs was determined by dividing these values by (I) for each concentration.

KEAP1–CUL3 and KEAP1–NRF2 HTRF Assays

KEAP1 diluted to 12 nmol/L in KEAP1 buffer was plated at 5 μL/well. Compound was added to the plate via compound printer and incubated for 15 minutes at room temperature. Next, CUL3 was diluted in KEAP1 buffer and plated at 5 μL/well. The plate was spun and pulsed to 1,000 × g, followed by incubation for 1 hour. Donor/acceptor solution (anti–His-Tb gold and anti–GST-d2 both diluted 1:100 in PPI Tb detection buffer) was added at 10 μL/well and incubated for 1 hour. Luminescence reading was measured in CLARIOstar. Same general protocol was used for KEAP1–NRF2 HTRF assay. Final concentrations of 2 nmol/L FL FLAG-KEAP1 and 1 nmol/L GST-NRF2 (Abnova H00004780-P01) were used, and the dyes used were anti–FLAG-Tb and anti–GST-d2. The KEAP1 buffer composition is available in Supplementary Tables S5 and S6.

NRF2 ELISA

Cells were seeded and treated as indicated with either DMSO or the indicated concentrations of VVD-065 for 18 to 24 hours at 37°C. Following treatment, cells were washed and lysed with minimal volumes of Pierce IP lysis buffer (+Protease Inhibitor Cocktail, MgSO₄, and Benzonase) and incubated for 1 hour at room temperature. Then, lysate was diluted with 100 μL of dilution buffer to the indicated concentrations, centrifuged for 5 minutes at 4,122 × g and stored at −80°C.

The NRF2-specific capture antibody (Cell Signaling Technology 12721S) was diluted in plating buffer and added to the assay plate (20 μL/well at 5 ng/μL) overnight at 4°C. The plate was washed 2 times in Phosphate-Buffered Saline with Tween 20 (PBST), and the blocking buffer was added at 100 μL/well and incubated for 1 hour at room temperature. Following incubation, the blocking buffer was removed, and lysate diluted in dilution buffer was added at 25 μL/well and incubated for 1.5 hours at room temperature. The assay plate was washed 3 times with 100 μL/well of PBST, and biotinylated detection antibody (Abcam, Ab62352) was diluted (1:100 in dilution buffer) and added at 20 μL/well, followed by incubation for 1 hour at room temperature. The assay plate was washed 3 times with 100 μL/well of PBST, and horseradish peroxidase (HRP)–conjugated NeutrAvidin diluted in dilution buffer was added at 20 μL/well and incubated for 1 hour at room temperature. The assay plate was washed 4 times with PBST at 100 μL/well. HRP substrate was added to the plate at 20 μL/well, prior to the CLARIOstar luminescence read. Data were normalized to untreated wells (100%).

RNA sequencing Data Analysis

All statistical analyses were done in Python (3.9.15) and R (4.3.2). Differential expression was done using DESeq2 (1.40.2) and limma (3.56.2; ref. 51). Gene set enrichment analysis was done using GSEApy (0.14.0; refs. 52, 53). Visualizations were made using Matplotlib (3.6.2), Seaborn (0.13.2), and ComplexHeatmap (2.16.0; refs. 54, 55). Principal component analysis was done using scikit-learn (1.1.3).

Plasma VVD-065 Measurement

VVD-065 was solvated in DMSO (ACS reagent ≥99.9%, Sigma-Aldrich) to prepare 0.5 mg/mL stock, which was stored at −20°C until use. Reagents used for bioanalysis included ACN and water (HPLC grade, Thermo Fisher Scientific) and FA (Optima grade, Thermo Fisher Scientific), whereas consumables included 96-well polypropylene plates (Nunc-Thermo) and 96-well silicone plate maps (Axygen). Prior to bioanalysis, working solutions were serially prepared in 75:25 ACN:water and spiked (20 μL) into commercially obtained mouse plasma containing K2-EDTA anticoagulant (BioIVT) for preparation of standards with concentrations ranging from 0.500 to 5,000 ng/mL. Study samples were removed from frozen storage and thawed, and 20-μL plasma aliquots (2-μL aliquots combined with 18 μL of blank commercial mouse plasma for 10-fold dilutions and 10-μL aliquots combined with 10 μL of blank commercial mouse plasma for 2-fold dilutions) were transferred to clean 96-well deep-well plates for extraction by protein precipitation via addition of 200 μL 95:5 ACN:methanol (HPLC grade, Thermo Fisher Scientific) containing carbamazepine and verapamil as internal standard followed by sealing, vortex mixing (11 minutes), and centrifugation (4,121 × g at 4°C for 10 minutes). For the final sample preparation step, 180 μL of the supernatant was transferred to a clean 96-well plate, sealed with a silicone mat, and submitted for liquid chromatography/tandem MS (LC/MS-MS) analysis. Upon injection of the sample (9 μL), chromatographic separation was achieved using a Kinetex biphenyl HPLC column (2.1 × 50 mm, 2.6 μm, 10 Å, Phenomenex) maintained in a column oven at 39°C using a binary gradient 5% to 95% B, in which pump A delivered water containing 0.1% FA and pump B delivered ACN with 0.1% FA for a total flow rate of 0.65 mL/minute and total run time of 2.1 minutes. A linear ion trap mass spectrometer (Sciex 5500 or Sciex 6500+) was operated in either positive ion mode electrospray ionization or positive ion mode atmospheric pressure chemical ionization with detection by multiple reaction monitoring. The acquired raw data were processed using Analyst software (version 1.7.1), in which sample concentrations were determined using a linear calibration curve comprising a minimum of eight non-zero standards and appropriate weighting. The minimum batch acceptance criteria included calibrator accuracy ±20%, correlation coefficient (r) >0.99, and 2 of 3 quality control samples within ±25% the theoretical value.

In Vivo Studies

Animals were maintained in accordance with the guidelines for the care and use of laboratory animals, as approved by the appropriate Institutional Animal Care and Use Committee. All animal procedures were conducted under Institutional Animal Care and Use Committee protocols in a facility overseen by Explora BioLabs, with accreditation by Association for Assessment and Accreditation of Laboratory Animal Care International.

Human CDX Studies

Following the acclimation period (3–7 days), immunodeficient mice, NSG (The Jackson Laboratory) for KYSE70 and VMRC-LCP inoculation and NCG (GemPharmatech Co.) for HCC95 inoculation, were inoculated subcutaneously into the dorsal flank with a cell suspension of 90% viable tumor cells in 0.2 mL of serum-free RPMI-1640: high-concentration Matrigel or regular Matrigel (Corning, cat. #354262; Corning, cat. #356237; 1:1 v/v) delivering tumor cells (KYSE70 and VMRC-LCP: 5 × 10⁶ cells/mouse; HCC95: 1 × 10⁷ cells/mouse). Mice were randomized based on tumor volume and were enrolled into different groups. Tumor volumes were measured twice per week after randomization in two dimensions using a caliper, and the volume was determined using the formula V = (L × W × W)/2, where V is tumor volume, L is tumor length (the longest tumor dimension), and W is tumor width (the longest tumor dimension perpendicular to L). TGI was calculated using the formula TGI% = [1 − (Ti − T0)/(Ci − C0)] × 100, where Ti is the mean tumor volume of the treatment group on the measurement day, T0 is the mean tumor volume the treatment group on the initiation day, Ci is the mean tumor volume of the control group on the measurement day, and C0 is the mean tumor volume of the control group on the initiation day. VVD-065 was administered at the indicated dose level by oral gavage. HCC95 CDX TGI was conducted at Crown Bioscience. Tumor tissue for ex vivo analysis was collected at 14 hours after the final dose in a twice-a-day regimen and 24 hours after the final dose in a once-a-day regimen.

Mouse Syngeneic TGI Study

KLN-205 syngeneic TGI study was conducted at Champions Oncology. A total of 2 × 10⁵ KLN-205 cells per mouse were subcutaneously implanted in DBA/2 mice (The Jackson Laboratory). Rest of the study was conducted as described above.

Radiation Combination Study

MOC1 cells were generously provided by Dr. Ravindra Uppaluri (Dana Farber Cancer Institute, Boston, MA) and cultured in a medium with 5% FCS, 40 ng/mL hydrocortisone (cat. #H0135, Sigma-Aldrich), 5 ng/mL EGF (cat. #01–107, Sigma-Aldrich), 5 mg/mL insulin (cat. #I0516, Sigma-Aldrich), and 1% penicillin–streptomycin. Eight- to ten-week-old female C57BL/6 mice were purchased from Taconic Biosciences. A total of 5 × 10⁶ MOC1 cells were subcutaneously injected into the dorsal flanks of the mice. Tumor-bearing mice were randomized after 7 days to either vehicle control (Labrafil M 1944 CS in 5% DMSO), VVD-065 (25 mg/kg once a day) on days 7 to 28, RT alone on days 14 to 18, or VVD-065 + RT. RT was performed using the Small Animal Radiation Research Platform 200 from Xstrahl for a tumor dose of 2 Gy daily on five consecutive days (10 Gy in total).

PDX Studies

PDX studies were conducted at Crown Bioscience. Fresh tumor tissues from mice bearing established primary human cancer PDX model were harvested and cut into small pieces (approximately 2–3 mm in diameter). PDX tumor fragment, harvested from donor mice, was inoculated subcutaneously at the upper right dorsal flank into immunocompromised mice for tumor development. Mice were randomized based on tumor volume (100–200 mm³). After randomization, tumor-bearing mice were allocated into two or four groups with two to three mice per group. Tumor tissue for ex vivo analysis was collected at 14 hours after the final dose in a twice-a-day regimen and 24 hours after the final dose in a once-a-day regimen.

KLN-205 Disseminated Model

KLN-205 cells were harvested in the exponential growth phase and suspended in 100 μL PBS to deliver 0.5 × 10⁶ cells via a tail vein injection in DBA/2 mice (The Jackson Laboratory). Seven days after inoculation, mice were randomized based on initial body weight, and the indicated treatment was dosed once a day via oral gavage. Health observations and body weights were recorded daily. Mice reached their endpoint upon body weight loss of 15% compared with treatment initiation, which was validated to capture mice prior to health deterioration. Upon euthanasia, lung weights were recorded, followed by inflation with 10% NBF (neutral buffered formalin) for subsequent formalin-fixed paraffin embedding. Tumor nodules were quantified on hematoxylin and eosin–stained lung sections.

Affinity Purification of Biotinylated Proteins

HEK293T cells stably expressing empty-miniTurbo (mT-V5) or miniTurbo-tagged nuclear export signal (mT-NES) or miniTurbo-tagged KEAP1 protein (mT-KEAP1; ref. 56) were treated with 100 nmol/L VVD-065 (mT-KEAP1) or DMSO (mT-V5, mT-NES, or mT-KEAP1) for 4 hours. Cells were then treated with 50 μmol/L biotin (Sigma-Aldrich, cat. #B4639-1G) for last 1 hour. Cells were washed with cold PBS and harvested on ice. Cells were sonicated at 4°C in RIPA buffer (10% glycerol, 50 mmol/L HEPES, 150 mmol/L NaCl, 2 mmol/L EDTA, 0.1% SDS, 1% Triton X-100, and 0.2% sodium deoxycholate) containing protease inhibitor (Thermo Fisher Scientific, cat. #78429), phosphatase inhibitor (Thermo Fisher Scientific, cat. #78426), and Benzonase (Sigma-Aldrich, cat. #E1014-5KU). Protein lysates were incubated with 30 μL of packed prewashed Streptavidin Sepharose beads (Cytiva, cat. #17511301) overnight on a rotator at 4°C. Beads were washed two times with WB1 (2% SDS), once with WB2 (500 mmol/L NaCl, 0.1% deoxycholate, 1% Triton X-100, 1 mmol/L EDTA, and 50 mmol/L HEPES, pH 7.5), once with WB3 (250 mmol/L LiCl, 0.5% Triton X-100, 0.5% deoxycholate, 1 mmol/L EDTA, and 50 mmol/L HEPES. pH 8.1), and once with WB4 (150 mmol/L NaCl and 50 mmol/L HEPES, pH 7.4). After washing, biotinylated proteins were eluted in 2× LDS sample buffer (Invitrogen, cat. #NP0007) with 2 mmol/L biotin and 50 mmol/L DTT (Sigma-Aldrich, cat. #D0632-5G) for 10 minutes at 70°C.

Immunoblotting

Proteins were separated in 4% to 12% Bis–Tris gel (Invitrogen, cat. #NP0321BOX) using 1× MOPS buffer (Thermo Fisher Scientific/Boston BioProducts, cat. #NC0778891) and transferred to nitrocellulose membranes (Thermo Fisher Scientific, cat. #PI88018). Membranes were blocked in 5% milk in Tris-Buffered Saline with Tween 20 (TBST) and incubated with primary antibodies diluted in 5% BSA overnight at 4°C. Membranes were then washed in TBST and incubated with secondary antibodies at a 1:10,000 dilution in 5% milk in TBST (Thermo Fisher Scientific/LI-COR, cat. #NC0250903 and #NC0250902) at room temperature for 1 hour. Imaging was performed in an Odyssey CLx imager and analyzed in Image Studio Lite version 5.2 software. Primary antibodies used were as follows: NRF2 (Cell Signaling Technology, cat. #20733S, 1:1,000), CUL3 (Cell Signaling Technology, cat. #2759, 1:1,000), KEAP1 (Cell Signaling Technology, cat. #8047, 1:1,000), SQSTM1 (Bethyl, cat. #A302-856A-T, 1:1,000), TSC22D4 (Bethyl, cat. #A303-222A, 1:200), PGAM5 (Abcam, cat. #ab126534, 1:1,000), MCM3 (Bethyl, cat. #A300-192A, 1:2,000), SLK (Bethyl, cat. #A300-500A, 1:1,000) and β-actin (Sigma-Aldrich, cat. #A5316, 1:5,000).

LC/MS-MS Sample Preparation, Data Acquisition, Raw Data Processing, and Analysis

Sample Preparation for MS Analysis

HEK293T cells stably expressing N-terminus and C-terminus miniTurbo biotin ligase–tagged KEAP1 were treated with 50 nmol/L VVD-065 or DMSO for 4 hours and 50 μmol/L biotin during the last 1 hour. Affinity purification of biotinylated proteins was performed as described earlier. For MS, protein lysates collected from a 15-cm tissue culture plate were incubated with 30 μL packed Streptavidin Sepharose beads. Beads were then washed in WB1–4 as described earlier. The streptavidin beads were further washed 3 times in 50 mmol/L ammonium bicarbonate (ABC). Beads were then resuspended in 100 μL of 50 mmol/L ABC containing 1 μg trypsin (Promega, cat. #V5113) and 0.1 mAu Lys-C (Wako Chemicals, cat. #129-02541) and incubated overnight at 37°C with shaking for on-bead digestion. On the following day, 0.5 μg trypsin and 0.05 mAu Lys-C were added to the beads and incubated for 2 hours. Digested peptides in the supernatant were collected into a fresh tube, and the beads were washed 2 times with HPLC-grade water and pooled with the peptides. Pooled peptides were centrifuged at 16,000 × g for 10 minutes and filtered using BioPureSPN columns (The Nest Group, cat. #C100500), prewetted with 0.1% trifluoroacetic acid (TFA), and spun at 3,000 × g for 2 minutes. Filtered peptides were acidified to 2% FA, dried using a SpeedVac, and stored at −80°C. Peptides were resuspended in 13 μL of 98:2 buffer A (water + 0.1% FA):buffer B (99.9% ACN + 0.1% FA), and 5 μL peptides were injected for MS analysis.

MS Analysis, Chromatographic Separation, and Label-free Quantification

Tryptic peptides were separated by reverse-phase nano-HPLC using an UltiMate 3000 RSLCnano system (Thermo Fisher Scientific) with a μPAC trapping column (Thermo Fisher Scientific) and a 50 cm μPAC Neo HPLC column (Thermo Fisher Scientific). For peptide separation and elution, mobile phase A comprised 0.1% FA in water and mobile phase B comprised 0.1% FA in ACN. Peptides were injected onto the trap column at 10 μL/minute for 3 minutes using the loading pump. Initially, the nanoflow rate was set at 0.75 μL/minute and 2% mobile phase B, while the peptides were loaded onto the trap column, at 2.8 minutes the solvent composition was changed to 10% mobile phase B. At 5 minutes, the flow rate was dropped to 0.300 μL/minute at 12% mobile phase B. A two-step gradient was used from 12% to 20% mobile phase B for 41.8 minutes followed by 20% to 40% mobile phase for 15.9 minutes. The flow rate was then increased to 0.750 μL/minute for column washing using seesaw gradients and re-equilibration.

MS analysis was performed on Orbitrap Eclipse (Thermo Fisher Scientific) operated in data-dependent acquisition mode. The MS1 scans were acquired in Orbitrap at 240k resolution, with a 1 × 10⁶ AGC target, auto max injection time, and a 375 to 2,000 m/z scan range. MS2 targets were filtered for charge states 2 to 7, with a dynamic exclusion of 60 seconds, and were accumulated using a 0.7 m/z quadrupole isolation window. MS2 scans were performed in the ion trap at a turbo scan rate following higher energy collision dissociation with a 35% normalized collision energy. MS2 scans used a 1 × 10⁴ AGC target and 35 ms max injection time.

MS Data Analysis: Protein Identification and Data Filtering for Sample Comparison

Raw MS data files were processed for protein identification and label-free quantification by MaxQuant (version 2.5.1.0; ref. 57) using the human Swiss-Prot canonical sequence database (UP000005640, downloaded February 2024). The following parameters were used: specific tryptic digestion up to four missed cleavages; variable modification search for up to 5 modifications per peptide including carbamidomethyl cysteine, protein N-terminal acetylation, and methionine oxidation; and default match between run parameters and label-free quantification with minimum ratio count of 1. Only unmodified, oxidized, or N-term acetylated unique peptides were used for protein quantification.

Quantified protein intensity data from MaxQuant were imported and analyzed using Perseus (version 2.0.11.0; ref. 58). First, the data were filtered based on a categorical column to remove proteins labeled as “only identified by site,” “reverse,” and “potential contaminant.” Then, intensity values were log₂-transformed, and replicates were grouped in categorical annotation rows. For this grouping, the N-terminus and C-terminus miniTurbo-tagged KEAP1 samples were segregated into either VVD-065–treated or DMSO-treated groups to detect most robust changes consistent across alternatively tagged KEAP1 cell lines. Data were further filtered to remove proteins without quantitative data (valid values) in all eight replicates (four from mT-KEAP1 and four from KEAP1-mT) in at least one group (VVD-065–treated or DMSO-treated). Missing values were replaced based on normal distribution with a width of 0.3 and a downshift of 1.8 from the SD for the total matrix. A comparison of compound (VVD-065)-treated group versus control (DMSO) group was done using a two-sided t test with an FDR of 0.05 and S0 of 0 with the default permutation-based FDR correction for multiple t tests. A volcano plot was generated using GraphPad Prism 10.

SILAC Protein Turnover Assay

Jurkat cells were grown in SILAC (Stable Isotope Labeling by Amino Acids in Culture) media containing light isotopes of Arg and Lys. Cells were treated with 2 μmol/L VVD-065 for 2 hours. After compound treatment, cells were collected by centrifugation and resuspended in SILAC media containing 2 μmol/L VVD-065 and heavy isotopes of Arg and Lys. Aliquots of cells were harvested after 0, 2, 6, 11, and 24 hours of incubation in heavy SILAC media. Cell pellets were stored at −80°C until all samples were collected. Cell pellets were then resuspended in PBS and lysed by sonication. The resulting lysates were probe-labeled by the addition of 200 μmol/L IA-DTB for 1 hour at room temperature followed by acetone protein precipitation. After precipitation, proteins were digested and probe-labeled peptides were enriched as described for in vitro, in situ and in vivo target engagement determination.

The resulting peptides were analyzed by PRM as described for target engagement quantification with slight modification. To determine protein turnover, both the light and heavy forms of peptides containing KEAP1 C196, C151, C288, C273, C297, C319, C434, C368, and C613 were monitored. Raw peptide AUCs were calculated by dividing the raw heavy AUC by the summed light and heavy peptide AUCs for each KEAP1 peptide monitored.

Glutathione Reactivity Assay

In brief, GSH was diluted to a final concentration of 50 μmol/L in buffer comprising 0.1 mol/L Tris pH 8.8 and 30% ACN. In triplicate, 100 μL of GSH solution was added to a clear 384-well plate (Greiner, no. 781101). Next, 5 μL of 10 mmol/L electrophilic compounds was added to the GSH solution to achieve a final concentration of 500 μmol/L, and the reaction was incubated for 2- and 6-hour timepoints at room temperature; 5 μL of 100 mmol/L Ellman’s reagent was then added to the plate and absorbance read at 440 nm. The concentration of GSH remaining was derived from a standard curve, and the observed rate [kobs/(I)] was calculated assuming pseudo first-order reaction kinetics using the equations: d(GSH)/dt = −k  × (GSH) and (GSH)t = (GSH)_t0  ×  e^−kt).

NanoBiT Proximity Assay to Examine Protein–Protein Interaction

A pair of NanoBiT plasmids including KEAP1-C’Term LgBiT/CUL3-C’Term smBiT and N’Term smBiT-NRF2/N’Term-LgBiT TUBE were cotransfected into Hek293Lx cells for 24 hours using FuGENE 6 (Promega). Transfected cells were trypsinized and seeded into a 384-well plate at 30,000 cells per well for NanoBiT proximity assay. VVD-065 (10 μmol/L) was serially diluted at 1:3 and dispensed into the 384-well plate using HP dispenser after cells were seeded for 1 hour. The restored NanoBiT luciferase enzyme activity was recorded by adding Nano-Glo Vivazine Live Cell Substrates (Promega) and read at CLARIOstar at 4 hours after compound treatment. Dose–response curve data were analyzed by normalizing luminescent signal counts from each compound-treated well to vehicle (DMSO) control.

To generate the NanoBiT plasmid for studying PPI, we used PCR strategy and Gibson Assembly technique to clone TUBE gBlock or PCR-amplified gene fragments of KEAP1, CUL3, and NRF2 open reading frame (ORF) into the pCDH vector carrying smBiT or LgBit at the N-terminal or C-terminal of the insertion site.

Split TurboID Approach for Proximity Labeling Study

Pair of split TurboID plasmids Tb(N)-tagged KEAP1 (59) and Tb(C)-tagged TUBE were cotransfected into the HEk293Lx line for 24 hours using FuGENE 6 (Promega). Transfected cells were treated with 2 μmol/L carfilzomib for 30 minutes prior to compound treatment. Cells were treated with biotin (50 μmol/L) and VVD-064 or VVD-065 (2 μmol/L) after proteosome activity was inhibited. Cells were harvested and pelleted after 2 hours of compound and biotin treatment. Cell pellets were lysed by IP lysis buffer (Pierce) for immunoprecipitation.

Biotinylated proteins were immunoprecipitated by incubating the cell lysate with streptavidin magnetic beads (Pierce) for 18 hours at 4°C. The streptavidin magnetic beads were washed with PBST and PBS before elution. To elute the biotinylated interactomes, washed beads were suspended in Jess sample buffer, boiled at 95°C for 10 minutes, and then subjected to Jess Western validation of proximity-labeled interactomes using CUL3, KEAP1, NRF2, and K48 linkagespecific polyubiquitin antibodies (Cell Signaling Technology).

To generate the split TurboID plasmids for TurboID proximity labeling study, the TUBE gBlock or PCR-amplified gene fragments of KEAP1 ORF were cloned into the vector plasmid carrying the split TurboID fragment Tb(N) or Tb(C) tagged at the N-terminal of the insertion site. The sequence of synthesized TUBE gblock (IDT) contain four RAD23 UBA1 domains placed in tandem with Gly/serine linker between each domain and an additional 3× Gly/serine linker at either end when linked to N-terminal or C-terminal tag.

TUBE gBlock sequence:

• AAAGCCACTATGGTCACAGGAAGTGAGTATGAAACTATGCTCACGGAGATTATGTCTATGGGCTATGAACGCGAACGAGTTGTCGCGGCGCTGAGGGCTTCTTACAATAATCCACACCGCGCGGTCGAGTACCTGCTGACTGGTATTCCAGGGTCTCCTGAACCAGAGCATGGAAGCGGAGGGGGAGGTTCTGGGGGTGGCGTCACAGGCTCAGAGTACGAGACCATGCTTACTGAGATTATGTCAATGGGATACGAGAGAGAGCGGGTAGTAGCTGCACTTAGAGCATCTTACAACAACCCGCATAGGGCTGTCGAATATCTCCTCACTGGTATCCCTGGGTCCCCGGAACCTGAGCACGGGAGCGGTGGGGGCGGATCTGGCGGCGGGGTGACTGGAAGCGAATATGAGACGATGCTTACAGAAATTATGAGCATGGGATATGAAAGAGAAAGAGTCGTAGCAGCGTTGCGAGCATCCTATAATAACCCGCATCGAGCTGTCGAATATTTGCTTACGGGAATCCCTGGTTCTCCGGAACCCGAACACGGCTCAGGTGGAGGAGGGTCAGGCGGAGGAGTCACGGGGAGTGAATACGAAACCATGTTGACCGAGATTATGAGTATGGGTTATGAAAGGGAGAGGGTGGTCGCCGCCTTGCGAGCATCTTATAACAACCCACACCGGGCTGTGGAGTATCTGCTGACAGGCATCCCAGGGTCACCTGAACCGGAGCACGGTAGT.

CRISPR−Cas9-Mediated Knock-in of KEAP1 C151S in HCC95

CRISPR RNA (crRNA) was purchased from IDT (guide sequence: TACACGGCCTCCATCTCCAT) and was reconstituted using nuclease-free water to a final concentration of 100 μmol/L. First, 1.5 μL of crRNA was combined with 100 μmol/L of trans-activating CRISPR RNA (tracrRNA) (IDT, cat. #1072532), heated at 95°C for 5 minutes, and cooled to room temperature. The RNP complex was formed by combining 3 μL RNA with 1 μL Cas9 (IDT, cat. #1081059). The complex was flicked to mix and incubated at room temperature for 10 to 20 minutes. Meanwhile, recovery media was prepared by adding the homology-directed repair (HDR) enhancer (IDT, cat. # 10007921) to a final concentration of 30 μmol/L in warm serum-containing media and incubated at 37°C. Next, cells were trypsinized, recovered in warm serum-containing media, and counted. A total of 1 × 10⁶ cells/electroporation were transferred to an Eppendorf tube and pelleted at 5,000 rpm for 1 minute at room temperature. Cell pellet was resuspended in 1 mL DPBS to wash and re-pelleted at 5,000 rpm for 1 minute. Washed pellet was then resuspended in 20 μL electroporation buffer, prepared according to Lonza SE kit (cat. #V4XC-1032) instructions: 16.4 μL SE buffer + 3.6 μL supplement buffer/reaction. Cell suspension was then combined with 4 μL RNP complex by gentle pipetting, and 1.5 μL of 100 μmol/L single-stranded HDR donor oligo was added for knock-in (IDT, sequence: TGGAGCGCCTCATTGAATTCGCCTACACGGCCTCCATTAGTATGGGCGAGAAGTCTGTCTTGCACGTCATGAACGGTGCTGTCATGTACCAGATCGACAGCGTT). Cell + RNP mix was then transferred to an electroporation cuvette and electroporated using program setting CM-137. Electroporated cells were transferred to prewarmed plate containing HDR enhancer (IDT cat. #10007921) and allowed to recover for 72 hours. To isolate single-cell clones, cells were plated at a density of 0.5 cells/well in a 384-well plate. Expanded clones were eventually graduated to a 96-well plate. Upon reaching confluence, half of each well was sent for Sanger sequencing to identify clones harboring successful C151S knock-in. For validated clones, remaining half was expanded for downstream analysis.

Click-iT Plus OPP Alexa Fluor 488 Assay (Protein Synthesis)

KYSE70 cells were plated at 18,000 cells/well in a Corning 96-well white/clear flat-bottom microplate and incubated overnight at 37°C and 5% CO₂. The next day, cells were treated with 20 μmol/L cycloheximide (Thermo Fisher Scientific, cat. #J66901-03) or 250 nmol/L VVD-065 (n ≥ 5) for 6 hours. After compound treatment, the Click-iT Plus OPP Alexa Fluor 488 assay kit (Life Technologies, cat. #C10456) was used as per the manufacturer’s instructions with the exception of the permeabilization step which was carried out overnight at 4°C rather than 15 minutes at room temperature.

Renilla Luciferase Assay

A total of 1 × 10⁶ KYSE70 ARE luciferase–expressing cells were electroporated with 20 ng pCMV-Green Renilla Luc vector (Thermo Fisher Scientific, cat. #16153) in triplicate using Lonza SE kit instructions (cat. #V4XC-1032). Briefly, each cell pellet was resuspended in 20 μL electroporation buffer (16.4 μL SE buffer + 3.6 μL supplement), combined with 20 ng final Renilla vector, suspension transferred to an electroporation cuvette, and electroporated using the CM-137 setting. One hundred microliters of prewarmed serum media was then added to each well of the cuvette to immediately recover cells, triplicates were combined, cells were recounted, and 25,000 cells/well were plated in a Corning 96-well white/clear flat-bottom microplate. Cells were allowed to recover by incubating overnight at 37°C and 5% CO_2. The following day, cells were treated with VVD compounds for 18 to 20 hours. After compound treatment, Dual-Glo luciferase reagent (Promega, cat. #PRE2920) was added to cells, and both Firefly and luciferase signals were detected on a CLARIOstar plate reader, following the manufacturer instructions.

ROS Assay (CellROX Green)

KYSE70 cells were plated at 20,000 cells/well in a Corning 96-well white/clear flat-bottom microplate and incubated overnight at 37°C and 5% CO_2. The following day, cells were treated with 50 μmol/L menadione (Selleck Chemicals, cat. #S1949) for 2 hours either without or with 250 nmol/L VVD-065. After compound treatment, all conditions were incubated with 5 μmol/L CellROX Green reagent (Invitrogen, cat. #C10444) for 30 minutes as per the manufacturer’s instructions prior to reading fluorescence signal on a CLARIOstar plate reader. A duplicate plate was set up and treated as stated above except for the addition of CellROX Green reagent, and samples were prepared for in vitro NRF2 protein analysis.

For menadione experiment, KYSE70 cells plated at 5 × 10⁶ cells/10-cm dish in serum media and incubated overnight at 37°C and 5% CO_2. The following day, duplicate dishes were treated with 50 μmol/L menadione (Selleck Chemicals, cat. #S1949) at 0, 30, 60, 90, and 120 minutes (120 minutes × 4 dishes for ± DTT to be completed after lysis). After treatment, cells were collected in DPBS through scraping before pelleting by centrifugation at 5,000 rpm for 5 minutes. Cell pellets were washed once with DPBS and re-pelleted before freezing at −80°C.

KEAP1 Mutation Sensitivity Analysis

HEK293T cells were transfected with a mixture containing 100 ng/well of the ARE luciferase reporter plasmid [pGL4.37(luc2P/ARE/Hygro), Promega cat. #E3641], 100 ng/well of a control Renilla luciferase plasmid [pGL4.75(hRluc/CMV), Promega cat. #E6931], and varying doses of a KEAP1 overexpression plasmid [pRP(Exp)-EGFP/Puro-CAG>hKEAP1(NM_203500.2), VectorBuilder cat. #VB900009-2802tpq] for a final amount of 50, 25, or 12.5 ng per well. Transfections were performed using FuGENE HD reagent (Promega) at a 3:1 FuGENE:DNA ratio in Opti-MEM (Gibco) in a total transfection volume of 50 μL. The DNA/FuGENE mixture was added directly to the cells and gently mixed by pipetting to ensure even distribution.

Cells were seeded at 12,500 cells/well in 50 μL of DMEM (without penicillin–streptomycin) in 384-well white, tissue culture–treated plates (Corning) and incubated overnight at 37°C and 5% CO₂. On day 1 (24 hours after transfection), cells were treated with VVD-065. On day 2 (18 hours after treatment), 50 μL of reconstituted ONE-Glo EX reagent (Promega) at room temperature was added per well. Plates were mixed for 2 minutes at <200 rpm to lyse cells and then incubated at room temperature for 3 to 5 minutes to stabilize the luminescence signal. Firefly luciferase activity was measured using a CLARIOstar Plus plate reader (BMG Labtech).

SPR Assay

CUL3-NTD (at 10 μg/mL in 10 mmol/L sodium acetate pH 4.0) was amine-coupled onto a C1 sensor chip at four surface densities (from 500 to 1,800 RU) using standard NHS/EDC activation and ethanolamine blocking. Full-length KEAP1 (APO and +compound) was tested in a threefold concentration series up to 2 μmol/L over the Cul3 surfaces using single-cycle injections. Running buffer contained 25 mmol/L HEPES pH 7.5, 150 mmol/L NaCl, and 1 mmol/L TCEP, and data were collected at 25°C. The response data from all four surfaces were fit globally to a simple 1:1 interaction model using Scrubber 2 (BioLogic Software Pty Ltd). The APO protein data are shown by fit with the blue line. The +compound data are shown by the red lines.

Global Proteomic Profiling

KYSE70 parental, KYSE70 monoclonal KEAP1 KO, or NRF2 KO cells were kindly provided by Dr. Xiaoxin (Luke) Chen’s laboratory (Coriell Institute for Medical Research). Cells were cultured in a humidified incubator at 37°C and 5% CO₂ in RPMI-1640 supplemented with 10% FBS and 1% penicillin–streptomycin. Cells were seeded onto 10-cm plates and treated with DMSO or 100 nmol/L VVD-065 for 24 hours. Cells were washed in cold PBS two times and harvested by scraping. Cells were then lysed in preheated lysis buffer containing 5% SDS and 100 mmol/L Tris at 95°C with shaking at 850 rpm for 10 minutes. Lysates were centrifuged at 21,000 × g for 30 minutes at 4°C, and cleared lysates were transferred to low-retention microcentrifuge tubes. Protein concentrations were measured by bicinchoninic acid (BCA) assay. Protein (150 μg) was aliquoted, and lysate volume was balanced to 300 μL. Samples were reduced with 5 mmol/L DTT for 30 minutes at 56°C with shaking at 850 rpm. Samples were then cooled briefly to room temperature and alkylated with 5 mmol/L chloroacetamide for 20 minutes at room temperature in dark. Samples were further processed in the KingFisher Flex (Thermo Fisher Scientific) system using protein aggregation capture protocol adapted from Koenig and colleagues (60). Star protocols. Proteins in samples were precipitated in 700 μL 100% ACN and captured with 20 μL MegReSyn Hydroxyl beads (ReSyn Biosciences). Beads were washed in 1,000 μL 100% ACN 3 times followed by two washes in 1,000 μL 70% EtOH. On-bead digestion was performed in 250 μL 50 mmol/L ABC buffer with trypsin and LysC for 5 hours. Digested peptides were then filtered in BioPureSPN columns (The Nest Group) prewetted with 100 μL 0.1% TFA at 3,000 × g for 2 minutes. Digestion enzymes were quenched with a final concentration of 2% FA and dried down in SpeedVac. Dried peptides were reconstituted in 75 μL 98:2 buffer.

Peptides were reconstituted in 2% ACN and 0.1% FA and concentrated onto an μPAC trapping column (Thermo Fisher Scientific) and then resolved on an μPAC Neo column (50 cm, Thermo) on a Dionex UltiMate 3000 nano-LC (Thermo). Peptides were separated using a 120-minute two-step gradient from 10% to 20% solvent B (99.9% ACN and 0.1% FA) over 74 minutes and 20% to 40% B over 45.2 minutes, followed by 40% to 90% B over 0.8 minutes. The Orbitrap Eclipse Tribrid mass spectrometer (Thermo Fisher Scientific) was operated in data-independent mode. The MS1 scans were acquired in the Orbitrap at 30K resolution with an 4 × 10⁵ AGC target, 54 milliseconds max injection time, and a 390 to 1,010 m/z scan range. MS2 scans were performed in the Orbitrap at 30K resolution and 10 m/z window with a precursor mass range 400 to 1,000 m/z. Ions were fragmented with higher-energy collision dissociation with a 30% normalized collision energy. MS2 scans used a 5 × 10⁴ AGC target, 54 milliseconds max injection time, and 145 to 1,450 scan range (m/z). The scan cycle was set at 3 seconds.

Raw files were searched with DIA-NN 2.2.0 Academia software package (61). Data were searched against a predicted spectral library from human FASTA database (UniProt/Swiss-Prot reviewed, downloaded October 4, 2025). For global quantitative analysis, data were filtered to include only proteins identified by more than one peptide with signal detected in all four replicates of each condition. For NRF2 MS-based quantitation, log₂ FC was calculated per replicate regardless of the number of peptides detected.

Synthesis Routes of Compounds

The synthesis routes for the compounds are provided in Supplementary Method S1.

Supplementary Material

Supplementary Method 1

Shows synthesis routes of compounds

Supplementary Method 2

Shows PDB validation report

Supplementary Table 1

Shows chemo-proteomic analysis of VVD-065 selectivity

Supplementary Table 2

Shows global proteomic analysis of VVD-065 treated cells

Supplementary Table 3

Shows genotype of PDX models used

Supplementary Table 4

Shows q-PCR probe information

Supplementary Table 5

Shows crystallography data collection parameters

Supplementary Table 6

Shows buffer composition in HTRF assay

Supplementary Figure 1

Shows distribution of physicochemical properties for the covalent fragment library and initial chemoproteomic screening results

Supplementary Figure 2

Shows chemoproteomic determination of KEAP1 Cys151 engagement and subsequent downstream signaling effects of VVD-065 related compounds

Supplementary Figure 3

Shows cysteine-directed selectivity analysis of VVD-065 and related compounds

Supplementary Figure 4

Shows global proteomic analysis of VVD-065 treated cells with KEAP1 or NRF2 knock out

Supplementary Figure 5

Shows immunoblots revealing that the activity of VVD-065 is dependent on KEAP1/ CUL3 but independent of cellular redox status

Supplementary Figure 6

Shows further evaluation of VVD-065, including assessment of the acetamide analog's lack of activity, determination of VVD-065 reaction kinetics, and additional structural analysis of the compound

Supplementary Figure 7

Shows VVD-065 enhances NRF2 ubiquitination by increasing the KEAP1- CUL3 protein-protein interaction

Supplementary Figure 8

Shows VVD-065 increases the KEAP1-CUL3 protein-protein interaction without affecting the KEAP1 protein half-life

Supplementary Figure 9

Shows differential cellular response to VVD-065, showing activity in wild-type cells but inactivity or reduce activity in cells with certain KEAP1 or NFE2L2 mutations

Supplementary Figure 10

Shows the anti-proliferative activity of VVD-065, observed across various cell lines and potentiated in 3D culture, is fundamentally linked to the presence of KEAP1 Cys151

Supplementary Figure 11

Shows VVD-065 activity on NRF2 targets

Supplementary Figure 12

Shows TGI effects of VVD-065 and NRF2 genetic ablation in a variety of cell line derived xenograft models

Supplementary Figure 13

Shows pharmacodynamic effects and tolerability of VVD-065 evaluated in patient-derived xenograft (PDX) models

Supplementary Figure 14

Shows additional analysis of the tumor growth inhibition study in the KLN205 syngeneic model, and evaluation of the tolerability of VVD-065 in combination with chemotherapy in patient-derived xenograft models

Data Availability

The X-ray coordinates and structure factors for KEAP1 BTB with VVD-065 have been deposited in the PDB under accession code 9DU7. Sequencing data have been deposited to the GEO with accession number GSE278482. The MS proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository with the dataset identifier PXD056257 (global proteomics profiling of VVD-065 treatment), PXD071657 (chemoproteomics cysteine selectivity of VVD-065, VVD-330, and VVD-325), PXD071609 (chemoproteomics cysteine selectivity of VVD-446 and VVD-860), PXD071607 (chemoproteomics analysis of VVD-065 protein selectivity), and PXD071555 (chemoproteomics profiling of VVD-065 selectivity).

Authors’ Disclosures

N. Roy reports other support from Vividion Therapeutics during the conduct of the study, as well as a patent for 18/477,240 pending. J. Inloes reports other support from Vividion Therapeutics during the conduct of the study. B. Kuenzi reports other support from Vividion Therapeutics during the conduct of the study. M. Pariollaud reports other support from Vividion Therapeutics during the conduct of the study. Z. Rush reports other support from Vividion Therapeutics during the conduct of the study. G. Ambrus-Aikelin reports personal fees from Vividion Therapeutics during the conduct of the study, as well as personal fees from Vividion Therapeutics outside the submitted work. T.M. Kinsella reports other support from Vividion Therapeutics outside the submitted work. M.B. Major reports grants from Vividion Therapeutics during the conduct of the study. M.P. Patricelli reports a patent for 18/477,240 pending to Vividion Therapeutics. No disclosures were reported by the other authors.

Authors’ Contributions

N. Roy: Conceptualization, supervision, writing–original draft, writing–review and editing. T. Wyseure: Investigation. I.-C. Lo: Investigation. J. Lu: Investigation. C.L. Eissler: Investigation. S.M. Bernard: Investigation. I. Bok: Investigation. A.N. Snead: Investigation. A. Parker: Investigation. U.-G. Lo: Investigation. J.C. Green: Investigation, writing–review and editing. J. Inloes: Investigation. S.R. Jacinto: Investigation. B. Kuenzi: Investigation. M. Pariollaud: Investigation. K. Negri: Investigation. K. Le: Investigation. B.D. Horning: Investigation. N. Ibrahim: Investigation. S. Grabow: Investigation. H. Panda: Investigation. D.P. Bhatt: Investigation. E.M. Wilkerson: Investigation. S. Saeidi: Investigation. P. Zolkind: Investigation. Z. Rush: Investigation. H.N. Williams: Investigation. E. Walton: Investigation. M.K. Pastuszka: Investigation. J.J. Sigler: Investigation. E. Tran: Investigation. K. Hee: Investigation. J. McLaughlin: Investigation. G. Ambrus-Aikelin: Supervision. J. Pollock: Supervision. R.T. Abraham: Supervision. T.M. Kinsella: Conceptualization, supervision. G.M. Simon: Supervision, writing–review and editing. M.B. Major: Supervision, writing–review and editing. D.S. Weinstein: Supervision, writing–review and editing. M.P. Patricelli: Conceptualization, supervision, writing–review and editing.