A CRISPR activation screen identifies FBXO22 supporting targeted protein degradation

Abstract

Targeted protein degradation (TPD) represents a potent chemical biology paradigm that leverages the cellular degradation machinery to pharmacologically eliminate specific proteins of interest. Although multiple E3 ligases have been discovered to facilitate TPD, there exists a compelling requirement to diversify the pool of E3 ligases available for such applications. In this study, we describe a CRISPR-based transcriptional activation screen focused on human E3 ligases, with the goal of identifying E3 ligases that can facilitate heterobifunctional compound-mediated target degradation. Through this approach, we identified a candidate proteolysis-targeting chimera (PROTAC), 22-SLF, that induces the degradation of FKBP12 when the transcription of FBXO22 gene is activated. Subsequent mechanistic investigations revealed that 22-SLF interacts with C227 and/or C228 in FBXO22 to achieve the target degradation. Finally, we demonstrated the versatility of FBXO22-based PROTACs by effectively degrading additional endogenous proteins, including BRD4 and EML4-ALK.

Introduction

Traditional small molecule drugs function by directly interfering with the activities of proteins. However, many proteins lack functional sites for rational drug design, presenting challenges in targeting them with small molecules. A promising alternative approach involves the use of small molecules that guide proteins to the proteasomal machinery, leading to the complete removal of the protein¹. This targeted protein degradation (TPD) strategy employs two types of small molecules: 1) heterobifunctional compounds, known as PROTACs (proteolysis-targeting chimeras)²; and 2) monofunctional compounds, referred to as molecular glues^{3, 4}. A rapidly growing number of proteins have demonstrated susceptibility to this strategy¹. Nonetheless, only a limited subset of the 600+ human E3 ligases has been identified as supportive of TPD^{5, 6}. This underscores the imperative to uncover additional ligandable E3 ligases, which would unlock the full potential of TPD as a valuable pharmacological strategy.

Various strategies have been employed to identify E3 ligases for TPD. One approach involves structure-based rational design, such as the discovery of the VHL ligand⁷. These ligands can then be adapted into PROTACs to degrade target proteins. Another strategy involves chemical proteomics, aiming to identify small molecules that bind to E3 ligases. This approach has led to the identification of CRBN as a target of thalidomide⁸, RNF114 as a target of nimbolide⁹, and DCAF1 as a target of MY-1B¹⁰. Subsequently, these ligands can be transformed into PROTACs for TPD applications. An alternative approach entails target-degradation screening of a collection of heterobifunctional compounds. These compounds consist of a ligand that specifically binds to the desired target coupled with broad-spectrum electrophilic fragments capable of potentially binding to endogenously expressed E3 ligases. This method resulted in the discovery of E3 ligases DCAF16 and DCAF11, which facilitate TPD by interacting with covalent PROTACs^{11, 12}. Nonetheless, a limitation of this approach is that distinct cell lines exhibit varying E3 expression profiles, indicating potential missed opportunities for exploitable E3 ligases that remain poorly expressed. While transcriptional and protein expression profiles of E3 ligases are readily accessible through public resources^{13, 14}, the low expression of some E3 ligases may mask their functionality in facilitating TPD. For example, a recent study revealed that in platelets, although VHL is expressed, its expression level is insufficient to facilitate ligand-induced degradation of BCL-X_L¹⁵.

Here, we present a novel strategy employing a CRISPR transcriptional activation screen to evaluate the candidate PROTAC’s capability for TPD. This method has the potential to assess the interaction potentials between candidate PROTACs and a wide array of human E3 ligases, therefore streamlining the discovery of E3 ligases for TPD. Through this method, we identified F-box protein 22 (FBXO22) as an E3 ligase capable of supporting the degradation of multiple protein substrates when engaged by electrophilic PROTACs.

Results

CRISPR activation screen identifies FBXO22 for TPD

Evaluating target-directed bifunctional compounds has been demonstrated as an effective approach to discovering functional degraders that may engage novel endogenous E3 ligases^{11, 12}. These E3 ligases can be identified through CRISPR knockout screens^{16, 17} or Affinity Purification-Mass Spectrometry (AP-MS)^{11, 12}. Nevertheless, a limitation of this approach arises when endogenous E3 ligases, which can be engaged by candidate degraders, have low or minimal expression. Consequently, these bifunctional compounds may exhibit partial target degradation, making the E3 deconvolution process challenging. In such scenarios, utilizing CRISPR activation screens has the potential to transform these hypoactive degraders into hyperactive ones by enhancing the expression of desired E3 ligases. For our initial study, we chose FKBP12 as the substrate protein, a prolyl isomerase commonly employed to investigate ligand-induced protein degradation^{18, 19}. To establish CRISPR-Cas9 transcriptional activation cells, we first generated HEK293T cells expressing FKBP12-EGFP (Fig. 1a and Extended Data Fig. 1a). We confirmed the degradation of FKBP12-EGFP using Lenalidomide-SLF (Len-SLF) (Extended Data Fig. 1b,c), an established PROTAC targeting FKBP12 through recruiting CRBN¹². In contrast, the FKBP12 binder alone, SLF (synthetic ligand of FKBP), did not induce degradation of FKBP12-EGFP (Extended Data Fig. 1b,c). Subsequently, in FKBP12-EGFP-expressing cells, we stably expressed both dCas9-VP64 and MS2-P65-HSF1 (Fig. 1a and Extended Data Fig. 1d). This second-generation CRISPR-Cas9 transcriptional activation system incorporates the transcriptional activator VP64 along with additional coactivators p65 and HSF1, yielding enhanced transcriptional activation²⁰. To validate the efficacy of gene transcriptional activation, we selected a well-studied gene IL1B and generated two constructs with single-guide RNA (sgRNA) sequences targeting IL1B promoter regions²⁰. Transduction of FKBP12-EGFP, dCas9-VP64, and MS2-P65-HSF1-expressing cells with both IL1B sgRNAs resulted in substantially increased IL1B gene expression (Extended Data Fig. 1e).

Fig. 1. An E3 ligase focused CRISPR-Cas9 transcriptional activation screen identifies FBXO22 that supports 22-SLF-induced reduction in FKBP12-EGFP expression levels.

a, Schematic representation of the steps in the CRISPR-Cas9 transcriptional activation screen. FACS, fluorescence-activated cell sorting. NGS, next-generation sequencing. b, Structure of 22-SLF. c, Volcano plot showing the E3 ligase focused CRISPR-Cas9 transcriptional activation screen for FKBP12-EGFP degradation after treatment of 2 μM of 22-SLF for 24 hours (n = 3 biological independent samples). LFC, log2 fold change. P values were calculated by two-sided t test without adjustment. d, Quantitative PCR analysis of FBXO22 mRNA levels subsequent to the transduction of sgRNAs targeting the promoter regions of FBXO22 gene. The bar graph (n = 3 technical replicates) is representative of two independent experiments. e, Fluorescence quantification of FKBP12-EGFP levels in HEK293T CRISPR-Cas9 transcriptional activation cells transduced with FBXO22-activating sgRNAs, subsequent to the treatment of 2 μM of 22-SLF for 24 hours. Data are presented as mean values +/− SEM (n = 3 biological independent samples). The statistical significance was evaluated through unpaired two-tailed Student’s t-tests, comparing cells treated with 22-SLF to DMSO. Statistical significance denoted as ***P < 0.001, ns: not significant. P values are 0.00057 (sgRNA#1) and 0.00011 (sgRNA#2).

With the CRISPR-Cas9 transcriptional activation cells, we aimed to conduct a CRISPR-Cas9 transcriptional activation pool screen to identify E3 ligases that support FKBP12 degradation by FKBP12-directed heterobifunctional compounds. To this end, we generated a focused sgRNA library containing 3,520 sgRNAs targeting promoter regions of 680 human E3 ligases (5 sgRNAs per E3 ligase, Supplementary Data 1) and packaged this library into lentivirus (Fig. 1a). The CRISPR-Cas9 transcriptional activation cells were transduced with lentivirus containing the sgRNA library at a multiplicity of infection (MOI) of approximately 0.3. Following transduction, cells underwent puromycin selection and were treated with FKBP12-directed candidate PROTACs. Cells from the bottom 15% of the GFP population were isolated using fluorescence-activated cell sorting (FACS) (Fig. 1a and Extended Data Fig. 2). The relative abundance of sgRNAs in the sorted cells was compared to that in the input cells before sorting to identify the enriched sgRNAs.

Initially, our study set out to explore 27 FKBP12-directed bifunctional compounds, including both previously reported and newly synthesized ones^{11, 12}. We employed aldehydes and isocyanides as starting materials and applied the Ugi reaction²¹ to construct this targeted, structurally diverse library that encompasses aliphatic saturated hydrocarbons, saturated heterocycles, heteroaromatics, electron-poor aromatics, and electron-rich aromatics. We first conducted a cell viability screen to exclude compounds exhibiting cytotoxicity (< 50% cell viability) at 10 μM. Our rationale was to focus on compounds with minimal cytotoxicity, thereby allowing exploration of the chemical space within the E3 ligase family across a broader concentration range. Following this initial screening, we selected five compounds, along with the positive control degrader Len-SLF (Extended Data Fig. 3a,b). Subsequently, we established two cell systems expressing FKBP12-EGFP: one driven by a robust SFFV promoter and the other by a weaker hPGK promoter (Extended Data Fig. 3c). This yielded a highly expressed FKBP12-EGFP model and a model with lower FKBP12-EGFP expression (Extended Data Fig. 3d). The five compounds and Len-SLF were then tested at two concentrations (2 and 5 μM) in both models. We identified one compound, 22-SLF (1) (Fig. 1b), which had no effect on FKBP12-EGFP levels in the SFFV promoter system but induced a mild yet significant reduction of FKBP12 (~30%) in the hPGK promoter system (Extended Data Fig. 3e). These results suggest that 22-SLF might facilitate FKBP12-EGFP degradation by recruiting potential E3 ligases. However, the observed modest degradation could pose challenges in utilizing conventional methods to identify the E3 ligase^{11, 12, 16, 17}. Therefore, we employed a CRISPR activation screen to enhance the efficacy of candidate degraders with low activity, facilitating the discovery of the E3 ligase supporting 22-SLF-mediated FKBP12 degradation (Fig. 1a). During the cell sorting process, we observed a fraction of GFP⁻ cells following compound treatment. This silenced GFP population was also observed without compound treatment (Extended Data Fig. 3f), suggesting that these cells likely did not express FKBP12-GFP, possibly due to ineffective cell sorting or the rapid proliferation of GFP⁻ cells, which may dominate a portion of the entire cell population during culturing. Therefore, we gated out this GFP⁻ population and sorted cells with low 15% GFP expression, which may provide a more accurate reflection of compound-mediated degradation of FKBP12.

This screen led to a substantial enrichment of two distinct sgRNAs activating the same E3 ligase gene, FBXO22 (Fig. 1c and Supplementary Data 1). This suggests that upon activation of FBXO22, 22-SLF transforms into a potent degrader, promoting the reduction of FKBP12-EGFP. To validate this result, we individually introduced two hit sgRNAs into the CRISPR-Cas9 transcriptional activation cells and confirmed the activation of FBXO22 mRNA expression (Fig. 1d). Subsequently, we subjected the FBXO22-activated cells to treatment with 22-SLF and observed the loss of FKBP12-EGFP expression (Fig. 1e). To further demonstrate the potential of utilizing this method to discover additional E3 ligases, we generated HEK293T cells expressing nucleus-localized FKBP12-EGFP_NLS along with the CRISPR activation components (Extended Data Fig. 4a). We applied the same screening method to KB02-SLF, a previously reported FKBP12-directed PROTAC through recruiting DCAF16¹². This screen identified two DCAF16-targeting sgRNAs as the most significantly enriched (Extended Data Fig. 4b and Supplementary Data 1). This result underscores the generalizability of our strategy in different contexts and its potential for uncovering novel E3 ligases for targeted protein degradation.

22-SLF promotes FBXO22-dependent degradation of FKBP12

Next, we stably overexpressed FLAG-tagged FKBP12 and HA-tagged FBXO22 in HEK293T cells, followed by subjecting the cells to 22-SLF. The results revealed that 22-SLF-induced loss of FLAG-FKBP12 expression was FBXO22-dependent, as no reduction was observed in cells lacking HA-FBXO22 (Fig. 2a). Notably, the loss of FLAG-FKBP12 expression was blocked by the proteasome inhibitor MG132, the neddylation inhibitor MLN4924 and the FKBP12 ligand SLF (Fig. 2a). This indicates the proteasomal degradation of FKBP12 via a Cullin-RING ligase (CRL)²², consistent with FBXO22’s involvement in the SKP1-CUL1-F-box protein (SCF) ubiquitin ligase complexes, a subfamily of CRLs²³. The rescue by SLF demonstrates its competitive engagement with the FKBP12 binding site, indicating that 22-SLF binds to FKBP12 for the degradation process. To provide a more comprehensive insight into 22-SLF-induced FKBP12 degradation, we conducted a dose-dependent degradation experiment using 11 concentrations of 22-SLF ranging from 0.025 to 15 μM. This experiment allowed us to determine two critical parameters: the D_max value, which reached 89%, and the DC₅₀ value, measured at 0.5 μM (Fig. 2b). Furthermore, the kinetics of 22-SLF-induced degradation of FLAG-FKBP12 are rapid, with nearly complete substrate degradation occurring within 2 hours (Fig. 2c). Collectively, this array of data supports the mechanism that 22-SLF recruits FBXO22 to FKBP12, resulting in the proteasomal degradation of FKBP12.

Fig. 2. 22-SLF promotes FBXO22-dependent proteasomal degradation of FKBP12.

a, 22-SLF-induced FKBP12 degradation is dependent on FBXO22 and blocked by MG132 (1 μM), MLN4924 (1 μM) and SLF (25 μM) (n = 3 biological independent samples). The bar graph represents quantification of the FLAG-FKBP12/HSP90 protein content. Data are presented as mean values +/− SEM. b, Dose-dependent degradation of FKBP12 by 22-SLF. HEK293T cells expressing HA-FBXO22 were treated with 0.025 – 15 μM of 22-SLF for 8 hours. The graph represents quantification of the FLAG-FKBP12/HSP90 protein content. Data are presented as mean values (n = 2 biological independent samples). c, Time-dependent degradation of FKBP12 by 22-SLF. The bar graph represents quantification of the FLAG-FKBP12/HSP90 protein content. Data are presented as mean values (n = 2 biological independent samples). d, 22-SLF (2 μM) induced FKBP12 degradation in A549 wildtype, but not FBXO22 knockout cells (n = 3 biological independent samples). The bar graph represents quantification of the FLAG-FKBP12/HSP90 protein content. Data are presented as mean values +/− SEM. e, Global proteomic analysis in A549 wildtype and FBXO22 knockout cells treated with 22-SLF (2 μM) for 24 hours (n = 2 biological independent samples for DMSO treatment, n = 3 biological independent samples for 22-SLF treatment). f, Structure of 22-biotin. g, 22-biotin-conjugated streptavidin pull-down with lysates of A549 cells pretreated with 22-SLF (2, 10 or 50 μM, 2 hours) followed by Western blot analysis of endogenous FBXO22. The result is representative of two independent experiments with similar results. WCL, whole cell lysates.

FBXO22 has been demonstrated as a pivotal player in tumor progression²⁴. Its oncogenic functions are attributed to ubiquitination and degradation of various substrates, including KDM4A/B^{25, 26}, methylated p53²⁷, p21²⁸, PTEN²⁹, KLF4³⁰, and LKB1³¹. Further corroborating its relevance, FBXO22 demonstrates amplified expression in tumor tissues compared to normal counterparts (Extended Data Fig. 5a). To establish the role of endogenous FBXO22 in facilitating 22-SLF-induced FKBP12 degradation, we generated FBXO22 knockout in a panel of cancer cell lines, including A549 (lung cancer), MDA-MB-231 (breast cancer), and PC3 (prostate cancer). The efficacy of FBXO22 knockout was verified through genomic PCR and global proteomic analysis (Extended Data Fig. 5b,c and Supplementary Data 1). Notably, augmented expression of KDM4A and KDM4B – two substrates of FBXO22 – was evident in FBXO22 knockout cells, suggesting the functional disruption of FBXO22 in these cellular contexts (Extended Data Fig. 5c). Across all three cell lines, 22-SLF promoted degradation of endogenous FKBP12, which was abolished upon FBXO22 knockout (Fig. 2d and Extended Data Fig. 5d). Furthermore, we conducted MS-based global proteomic analysis in A549 parental and FBXO22 knockout cells, following the treatment with 2 μM of 22-SLF for 24 hours. This experiment revealed the selective degradation of endogenous FKBP12 in FBXO22 wildtype, but not knockout cells (Fig. 2e, Extended Data Fig. 5e, and Supplementary Data 1). Another protein, CUTA, was also observed to undergo reduction upon 22-SLF treatment in A549 parental cells, but not in FBXO22 knockout cells (Fig. 2e and Extended Data Fig. 5e). CUTA is an acetylcholinesterase (AChE)-associated protein and plays a crucial role in the folding and secretion of AChE³². However, it remains unclear why the level of CUTA decreases upon compound treatment in FBXO22 wildtype cells. One possibility is that 22-SLF may have an off-target effect, binding to CUTA and inducing its degradation in the presence of FBXO22. Moreover, we noted a moderate upregulation in KDM4A expression following 22-SLF treatment in FBXO22 wildtype cells, whereas no such effect was observed in FBXO22 knockout cells (Extended Data Fig. 5e,f). In contrast, KDM4B expression remains unchanged upon compound treatment. One possibility is that 22-SLF might exert a partial influence on FBXO22’s native function, resulting in the accumulation of its endogenous substrate, such as KDM4A. To further substantiate the engagement of FBXO22 by the electrophilic portion of 22-SLF, we synthesized a biotin-conjugated probe 22-biotin (2) (Fig. 2f). The probe was immobilized onto streptavidin-coated beads, which were subsequently subjected to incubation with cell lysates from HEK293T cells expressing HA-tagged FBXO22. Post bead washing, trypsin digestion, and MS analysis, we identified FBXO22 as one of the binding targets for 22-biotin (Extended Data Fig. 6a and Supplementary Data 1). Our use of 22-biotin also led to the enrichment of endogenous FBXO22 in A549 cells (Fig. 2g). Furthermore, 22-SLF competed with 22-biotin-enriched FBXO22 in a dose-dependent manner, thereby further confirming its direct binding to FBXO22 (Fig. 2g).

FBXO22 C227/C228 mediate 22-SLF-induced FKBP12 degradation

The electrophilic α-chloroacetamide group in 22-SLF indicates a potential interaction with cysteine residues in FBXO22. To identify the specific cysteine in FBXO22 engaged with 22-SLF, we treated HA-FBXO22-expressing HEK293T cells with 2 μM of 22-SLF for 2 hours. Using cysteine-directed activity-based protein profiling (ABPP), a chemical proteomics technology for the measurement of target cysteine engagement^{33, 34}, we sought to quantify the degree of blockade of iodoacetamide-desthiobiotin (IA-DTB)-modified cysteines on FBXO22. From a pool of 5,650 quantified IA-DTB-modified peptides, we identified 4 cysteines in FBXO22 (C83, C117, C228, and C365) (Fig. 3a and Supplementary Data 1), with approximately 20% of FBXO22 C228 being blocked by 22-SLF treatment (Fig. 3b). Human FBXO22 contains a total of 14 cysteines, with 10 being conserved between humans and rodents. In the ABPP experiment, we quantified 4 cysteines, including C228 that potentially interacts with 22-SLF. The observation of low fractional engagement of 22-SLF on FBXO22 raises a question: whether this low engagement is driven by low cell permeability of 22-SLF or low affinity of 22-SLF to FBXO22. To address this, we assessed the engagement of FKBP12 by 22-SLF through measuring the rescue of FKBP12 degradation by Len-SLF. In HEK293T FBXO22 knockout cells pretreated with 22-SLF, we observed a dose-dependent blockage of Len-SLF-induced degradation of FKBP12 (Extended Data Fig. 6b). The EC₅₀ value for this blockage effect is 1.96 μM, indicating that at this concentration, 22-SLF likely occupies 50% of FKBP12. Considering our ABPP data suggests that at a similar concentration (2 μM), 22-SLF engages approximately 20% of FBXO22 C228, these findings suggest that the low fractional engagement is likely due to 22-SLF’s low affinity to FBXO22 rather than the low cell permeability.

Fig. 3. FBXO22 C227 and C228 are involved in 22-SLF-mediated degradation of FKBP12.

a, Competitive cysteine-directed ABPP measured the degree of blockade of IA-DTB-modified cysteines by 22-SLF. Data are presented as mean values (n = 2 biological independent samples). b, Quantification of four IA-DTB-modified peptides in FBXO22 treated with DMSO or 22-SLF. Data are presented as mean values (n = 2 biological independent samples). c, Single mutation of C227 or C228 to alanine in FBXO22 partially blocked 22-SLF-induced degradation of FKBP12, while mutating both C227 and C228 to alanine in FBXO22 completely abolished the degradation. The bar graph represents quantification of the FLAG-FKBP12/HSP90 protein content. Data are presented as mean values +/− SEM (n = 3 biological independent samples). d, Global proteomic analysis in HEK293T cells expressing HA-FBXO22 wildtype or C227AC228A double mutant treated with 22-SLF (0.5 μM, 24 hours). Data are presented as mean values (n = 2 biological independent samples for DMSO treatment, n = 3 biological independent samples for 22-SLF treatment). e, Modeling study reveals several hydrogen bond interactions between the electrophilic portion of 22-SLF and a pocket in FBXO22 involving C227 and C228. f, Sequence alignment of human FBXO22 (aa 219–230) and mouse FBXO22 (aa 218–229). g, Mouse FBXO22 supported 22-SLF-induced FKBP12 degradation. The bar graph represents quantification of the FLAG-FKBP12/HSP90 protein content. Data are presented as mean values (n = 2 biological independent samples). h, Mutation of both C226 and C227 in mouse FBXO22 abolished 22-SLF-induced degradation of FKBP12. The bar graph represents quantification of the FLAG-FKBP12/HSP90 protein content. Data are presented as mean values (n = 2 biological independent samples).

Next, we mutated C228 to alanine and generated HA-FBXO22 C228A-expressing HEK293T cells. Interestingly, mutating C228 to alanine did not fully abolish 22-SLF-induced degradation of FKBP12 (Fig. 3c). Notably, in close proximity to C228 lies another cysteine, C227, implying the potential redundancy of these two cysteines in facilitating 22-SLF-induced FKBP12 degradation. We then generated another single mutant (C227A) and a double mutant (C227AC228A) in FBXO22 and expressed these variants in HEK293T cells to assess 22-SLF-induced FKBP12 degradation. The data revealed that both single mutants, C227A and C228A, partially impeded 22-SLF-induced FKBP12 degradation, and the double mutant, C227AC228A, completely blocked the degradation of FKBP12 induced by 22-SLF (Fig. 3c). This suggests that C227 and C228 likely have overlapping roles in supporting 22-SLF-mediated FKBP12 degradation. To address the concern regarding potential protein conformational changes caused by mutations affecting the degradation, we conducted an AP-MS experiment comparing the interactome of FBXO22 wildtype and C227AC228A mutant. The results indicated that both forms were incorporated into the SKP1-CUL1-RBX1 E3 complex (Extended Data Fig. 7a and Supplementary Data 1), suggesting the likelihood of the proper folding of FBXO22 C227AC228A mutant.

Expanding on this, we performed global proteomic analysis in FBXO22 wildtype- and C227AC228A-expressing HEK293T cells treated with 22-SLF. The findings revealed that among 7,936 quantified proteins, FKBP12 was the only protein degraded by 22-SLF in FBXO22 wildtype expressing cells, with no degradation observed in FBXO22 C227AC228A expressing cells (Fig. 3d, Extended Data Fig. 7b and Supplementary Data 1). Although the three-dimensional structure of FBXO22 remains undetermined, AlphaFold predictions demonstrate high confidence (pLDDT > 90) across most of the protein sequence, including the peptide encompassing C227 and C228. To gain deeper insights into the binding model, we conducted a docking study focusing on the FBXO22 binding portion with the predicted structure generated by AlphaFold. The resulting binding model indicated a feasible fit of 22-SLF on the surface of FBXO22 according to the docking score (−4.476) and predicted binding energy (Glide energy = −52.853 Kcal/mol and Glide Emodel = −61.272 Kcal/mol) (Fig. 3e). Notably, the α-chloroacetamide established a covalent bond with C228, consistent with findings from ABPP analysis. Furthermore, hydrogen bonds were observed between the benzothiazole moiety and N257, the carbonyl group of the α-chloroacetamide and C227, and the oxygen atom of the PEG linker and V230. The benzothiazole group also engages in a cation-pi interaction with K125. To further investigate these amino acids, we introduced three mutations in FBXO22 (N257A, V230A, and K125A). Subsequently, we assessed whether mutating any of these residues could abolish the degradation of FKBP12 by 22-SLF. Interestingly, all three mutants still supported 22-SLF-induced FKBP12 degradation (Extended Data Fig. 8). We interpret these data to indicate that 22-SLF-induced FKBP12 degradation may primarily be driven by the formation of covalent bond between C227/228 and the α-chloroacetamide group. As demonstrated in previous studies^{11, 12}, a low fractional engagement with the E3 ligases can be sufficient to effectively drive target degradation. Therefore, these mutations may result in a partial loss of binding to 22-SLF but can still be catalytically sufficient to induce FKBP12 degradation.

FBXO22 exhibits great conservation across mammals, with human and mouse FBXO22 sharing 93% identity. Intriguingly, the peptide sequence containing the two consecutive cysteines (C227 and C228) is identical between human and mouse (Fig. 3f). To explore whether mouse Fbxo22 can facilitate 22-SLF-induced FKBP12 degradation and if the conserved cysteines (C226 and C227 in mouse FBXO22) are pivotal, we generated HEK293T cells expressing mouse FBXO22 wildtype, C226A, C227A, and C226AC227A. The results showed that 22-SLF induced FBKP12 degradation in mouse FBXO22 wildtype-expressing cells, which was blocked by MG132, SLF, and MLN4924 (Fig. 3g). Similar to human FBXO22, the single mutants, C226A and C227A, partially impeded degradation, while the double mutant, C226AC227A, fully blocked the degradation (Fig. 3h). This data underlines that mouse FBXO22 can integrate effectively into the human SKP1-CUL1-RBX1 complex and function as an active E3 ligase supporting 22-SLF-induced FKBP12 degradation.

22-SLF induces a ternary complex between FKBP12 and FBXO22

It has been well established that the formation of ternary complex by degraders with their target protein and E3 ligase is a key step driving the subsequent target degradation³⁵. To demonstrate the ternary complex involving 22-SLF, FKBP12 and FBXO22, we treated HEK293T cells expressing HA-FBXO22 and FLAG-FKBP12 with 22-SLF and MG132. This experiment reveals that HA-FBXO22 co-immunoprecipitated with FLAG-FKBP12 in the presence of 22-SLF and MG132 (Fig. 4a), supporting the formation of a ternary complex. Notably, the assembly of this ternary complex was partially hindered upon expression of FBXO22 C227A or C228A single mutant, and nearly completely abolished when the FBXO22 C227AC228A double mutant was introduced (Fig. 4a). To comprehensively assess the impact of 22-SLF on the FKBP12 interactome landscape, we employed an AP-MS approach to identify proteins co-immunoprecipitating with FLAG-FKBP12 from HEK293T cells treated with 22-SLF. This analysis revealed several protein components of the FBXO22 complex, including FBXO22, SKP1, CUL1, RBX1 and NEDD8, as proteins recruited by 22-SLF (Fig. 4b and Supplementary Data 1). Furthermore, we conducted an enrichment proteomics study comparing FKBP12 interactome landscape in HEK293T cells expressing FBXO22 wildtype and C227AC228A mutant treated with 22-SLF. This experiment revealed the enrichment of the FBXO22 complex in FBXO22 wildtype, but not in FBXO22 C227AC228A-expressing cells (Fig. 4c and Supplementary Data 1). Collectively, these findings indicate that 22-SLF effectively drives the formation of a ternary complex involving 22-SLF, FBXO22 and FKBP12 by engaging C227/228 in FBXO22.

Fig. 4. 22-SLF induces the formation of a ternary complex involving 22-SLF, FKBP12 and FBXO22.

a, A co-immunoprecipitation assay reveals that HA-FBXO22 wildtype, but not HA-FBXO22 C227AC228A double mutant, co-immunoprecipitated with FLAG-FKBP12 in the presence of 22-SLF and MG132. The bar graph represents quantification of the immunoprecipitated HA-FBXO22 protein content compared to HA-FBXO22 protein in WCL. Data are presented as mean values (n = 2 biological independent samples). b, c, A FKBP12-directed enrichment proteomic analysis revealed that HA-FBXO22 wildtype and its associated components in the SKP1-CUL1-RBX1 E3 complex, but not HA-FBXO22 C227AC228A double mutant, co-immunoprecipitated with FLAG-FKBP12 in the presence of 22-SLF (2 μM) and MG132 (5 μM). Data are presented as mean values (n = 2 biological independent samples).

Harnessing FBXO22 to degrade additional protein targets

Next, we sought to investigate the potential of FBXO22 to facilitate the degradation of additional proteins. We first investigated BRD4 due to its significant physiological role and the availability of a selective ligand, JQ1, which has found extensive application in degrader development^{18, 36}. We synthesized a compound, 22-JQ1 (3), by coupling the FBXO22 binding moiety to JQ1 (Fig. 5a). 22-JQ1 promoted degradation of BRD4 in A549 wildtype cells, which was abolished upon FBXO22 knockout (Fig. 5b). Subsequently, we performed a global proteomic analysis in A549 wildtype and FBXO22 knockout cells treated with 22-JQ1. Our findings revealed that 22-JQ1 induced a potent degradation of BRD4 in A549 wildtype, but not FBXO22 knockout cells (Fig. 5c and Supplementary Data 1). To further demonstrate the potential of utilizing FBXO22 for degrading additional targets, we synthesized another PROTAC, 22-TAE (4), which incorporates the FBXO22 binding moiety coupled with a defined ALK ligand (Fig. 5d)³⁷. We generated FBXO22 knockout in H2228 (Extended Data Fig. 9 and Supplementary Data 1), a non-small cell lung cancer cell line expressing the EML4-ALK fusion protein³⁸. We found that 22-TAE induced degradation of EML4-ALK fusion protein in H2228 wildtype cells, while no such degradation was observed in FBXO22 knockout H2228 cells (Fig. 5e). Taken together, these results underscore the potential of FBXO22 in the creation of PROTACs capable of degrading multiple protein targets. Finally, considering that α-chloroacetamide is a highly reactive electrophilic warhead, there may be challenges in advancing to the therapeutic stage due to its potential engagement of off-target cysteines, such as catalytic cysteines^{33, 34}. To this end, we synthesized an acrylamide variant of 22-SLF, 22a-SLF (5) (Extended Data Fig. 10a). Intriguingly, 22a-SLF facilitated the degradation of FKBP12 with similar potency to 22-SLF (Extended Data Fig. 10b). These results suggest that embarking on a future medicinal chemistry campaign utilizing the acrylamide warhead could be a promising strategy to pursue.

Fig. 5. Harnessing FBXO22 for the degradation of BRD4 and EML4-ALK.

a, Structure of 22-JQ1. b, 22-JQ1 (1 and 2 μM, 24 hours) induced BRD4 degradation in A549 wildtype, but not FBXO22 knockout cells. The bar graph represents quantification of the BRD4/β-actin protein content. Data are presented as mean values (n = 2 biological independent samples). c, Global proteomic analysis in A549 wildtype and FBXO22 knockout cells treated with 2 μM of 22-JQ1 for 24 hours. Data are presented as mean values (n = 2 biological independent samples). d, Structure of 22-TAE. e, 22-TAE (1 and 2 μM, 24 hours) induced EML4-ALK degradation in H2228 FBXO22 wildtype, but not knockout cells. The bar graph represents quantification of the EML4-ALK/β-actin protein content. Data are presented as mean values (n = 2 biological independent samples).

Discussion

In this study, we employed a CRISPR-Cas9 transcriptional activation screen to identify the E3 ligase FBXO22 in supporting TPD when engaged by electrophilic PROTACs. Given the distinct gene expression profiles of various cell lines, candidate heterobifunctional compounds might exhibit as hypoactive or silent degraders due to the absence or low expression of compatible E3 ligases. Our method offers an unbiased means of significantly boosting E3 ligase expression, facilitating assessment of candidate heterobifunctional compounds. Projecting forward, we are intrigued by the prospect of utilizing this screening approach to uncover additional E3 ligases suitable for TPD applications.

A commonly employed technique for identifying E3 ligases supportive of ligand-induced protein degradation is CRISPR-Cas9 knockout screens^{16, 17}. However, in instances where candidate compounds display moderate activity leading to partial target degradation, CRISPR-Cas9 knockout screens may yield high background noise, obscuring the deconvolution of E3 ligases. In such scenarios, our approach may offer an alternative solution. By activating E3 ligases, we can observe more pronounced target degradation, allowing for the enrichment of sgRNAs associated with the relevant E3 genes. Additionally, this technique holds value in cases where certain protein degraders exhibit degradation solely in specific cell lines or primary cells, which can pose challenges for establishing CRISPR-Cas9 knockout screens in hard-to-transduce cell models. In these scenarios, a CRISPR-Cas9 transcriptional activation screen in transducible cells, such as HEK293T cells, can transform these otherwise silent degraders into active degraders by activating the paired E3 ligases. Nevertheless, it’s crucial to recognize that our CRISPR activation approach may not significantly enhance the function of E3 ligases already highly expressed. In such cases, existing levels of the E3 ligase might be sufficient for targeted protein degradation. Consequently, we foresee identifying functional degraders without gene expression manipulation, as demonstrated previously^{11, 12}. These functional degraders can undergo analysis using CRISPR-Cas9 knockout or AP-MS strategy to deconvolute the relevant E3 ligases.

In our study, we employed two filters to process the screening data in order to identify candidate hits: 1) Log2 fold change of enriched versus input gRNA reads number ≥ 0.8; and 2) Log10 fold change of P-value comparing enriched versus input gRNA reads number ≥ 2.5. For future applications, employing more stringent filters can statistically reduce the likelihood of false positive hits. Regarding false negatives, we acknowledge the possibility that a portion of sgRNAs in the library may not efficiently activate E3 gene expression, potentially resulting in false negatives. Nonetheless, our inclusion of five sgRNAs for each E3 gene in the library helps mitigate the risk of ineffectiveness associated with only one or two sgRNAs being designed. We posit that if one or more sgRNAs pass the above filter and exhibit statistical differences in the screen, they are likely high-confidence hits for validation. Furthermore, we anticipate future advancements in CRISPR activation technology, including approaches that more effectively activate E3 gene expression. These advancements can seamlessly integrate into our approach to uncover new E3 ligases for TPD. In terms of treatment duration, we initially selected a 24-hour treatment for the screen. Subsequent investigations indicated that shorter treatments (2–8 hours) also led to efficient target degradation (Fig. 2c). For future screens, treatment duration is another critical factor to consider, as shorter treatment times could improve throughput. Additionally, shorter treatment times may be more appropriate for targets that could elicit substantial changes in cell signaling or induce cell death upon compound induced degradation.

22-SLF, as identified in this study, features a more elaborate structure compared to our previously reported scout fragment-based bifunctional compounds¹². Nonetheless, we believe that both scout fragments and elaborated bifunctional compounds hold value in the pursuit of discovering E3 ligases for TPD. In contrast to protein inhibitors, which typically require high target occupancy in the initial screening phase, our previous studies suggest that the identification of new ligands targeting E3 ligases may require only partial target engagement^{11, 12}. From this perspective, scout fragments generally exhibit the ability to bind to a broader range of protein targets compared to elaborated compounds, thus increasing the likelihood of capturing potentially active E3 ligases. Conversely, elaborated compounds might display a narrower target scope, but their relatively higher affinity for E3 ligases could enhance their potential for facilitating degradation. Therefore, for future screens, incorporating both types of compounds in the initial screening process can be beneficial. Additionally, since the electrophilic component of PROTACs contains a chiral center, future investigations will determine if these compounds display stereoselective degradation activity when evaluated as individual stereoisomers.

FBXO22 is a well-studied E3 ligase known for its involvement in tumor progression²⁴. Given its elevated expression in tumors compared to normal tissue, FBXO22 could offer an advantageous avenue for the degradation of cancer targets with potentially minimized toxicity in normal cells. Despite the extensive research into FBXO22’s biology, there remains a scarcity of studies focused on developing inhibitors or ligands for FBXO22. Our investigation establishes a foundational step in the development of FBXO22-based degraders with therapeutic potential. In addition, given the pivotal role of FBXO22 in tumor progression, inhibiting FBXO22 could be promising as a cancer treatment strategy²⁴. Thus, for cell contexts where FBXO22 inhibition could confer therapeutic benefits, potent ligands exhibiting high engagement with FBXO22 for degrader development could potentially yield dual pharmacological effects – degradation of target proteins alongside FBXO22 inhibition.

Materials and Methods

Reagents.

The anti-HA (clone#: C29F4, cat#: 3724, dilution 1:5000), anti-FKBP12 (cat#: 55104, dilution 1:1000), HRP-linked anti-HSP90 (clone#: C45G5, cat#: 79641, dilution 1:5000), anti-BRD4 (clone#: E2A7X, cat#: 13440, dilution 1:1000), HRP-linked rabbit IgG (cat#: 7074, dilution 1:5000) and HRP-linked mouse IgG (cat#: 7076, dilution 1:5000) antibodies were purchased from Cell Signaling Technology. The anti-FLAG HRP antibody (clone M2, cat#: A8592, dilution 1:5000), anti-FLAG affinity gel (clone M2, cat#: A2220), dCas9-VP64-blasticidin SAM CRISPRa helper construct 1 plasmid DNA (cat#: SAMVP64BST), MS2-P65-HSF1-hygromycin SAM CRISPRa helper construct 2 plasmid DNA (cat#: SAMMS2HYG), and KB02-SLF (cat#: 914738) were purchased from Sigma-Aldrich. The anti-β-Actin (clone#: C4, cat#: sc-47778, dilution 1:5000) and anti-FBXO22 (clone#: FF-7, cat#: sc-100736, dilution 1:1000) antibodies were purchased from Santa Cruz Biotechnology. Puromycin (cat#: ant-pr-1), blastcidin (cat#: ant-bl-05) and hygromycin (cat#: ant-hg-1) were purchased from InvivoGen. MG132 (cat#: S2619) was purchased from Selleck Chemicals. MLN4924 (cat#: 15217) and SLF (cat#: 10007974) were purchased from Cayman Chemical. Polyethylenimine (PEI, MW 40,000, cat#: 24765–1) was purchased from Polysciences, Inc. Enzyme-linked chemiluminescence (ECL) (cat#: 32106), ECL plus (cat#: 32132) western blotting detection reagents, Streptavidin agarose (cat#: 20349) and Tandem Mass Tag (TMT) isobaric label reagent (cat#: 90110) were purchased from Thermo Scientific. PureLink genomic DNA mini kit (cat#: K182001) was purchased from Invitrogen. FuGene 6 (cat#: E2692) transfection reagent and sequencing grade modified trypsin (cat#: V5111) were purchased from Promega. Cas9 endonuclease was purchased from Integrated DNA Technologies (cat# 1081061).

Cell lines.

HEK293T, A549, MDA-MB-231, H2228 and PC3 cells were obtained from ATCC. HEK293T and MDA-MB-231 cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM, Corning, cat# 15013CV) with 10% (v/v) fetal bovine serum (FBS, Omega Scientific, cat# FB-01) and L-glutamine (2mM, Gibco, cat# 25030081). H2228 cells were cultured in RPMI-1640 medium (Corning, cat# 15040CV) with 10% (v/v) FBS and L-glutamine (2mM). A549 cells cultured in DMEM with 10% (v/v) FBS, L-glutamine (2mM) and MEM non-essential amino acids (Gibco, cat# 11140050). PC3 cells were cultured in F12K medium (Corning, cat# 10025CV) with 10% (v/v) FBS. All the cell lines were tested negative for mycoplasma contamination using Universal Mycoplasma Detection Kit (ATCC, cat# 30–1012K).

Cell viability assay.

HEK293T cells were plated in a 96-well clear bottom white plate (Corning, cat# 3610) at a density of 1×10⁴ cells per well in 100 μL of DMEM medium and incubated for 24 hours. Subsequently, the cells were treated with 10 μM of compound, with a final DMSO concentration of 0.1% (v/v)) in 100 μL of DMEM medium for an additional 24 hours. Following treatment, 50 μL of Cell Titer Glo reagent (Promega, cat# G7573) was added to each well and incubated for 10 minutes at room temperature. Luminescence was measured using a CLARIOstar Plus microplate reader (BMG LABTECH).

Generation of CRISPR-Cas9-mediated knockout cells.

A549, MDA-MB-231 and PC3 cells with FBXO22 CRISPR-Cas9 knockout were generated through electroporation of Cas9-sgRNA ribonucleoprotein (RNP) complex using 4D-Nucleofector (Lonza Bioscience). Two sgRNAs targeting FBXO22 gene (FBXO22 sgRNA#1: GGACCCAUCGGAGCGUAACC; FBXO22 sgRNA#2: UCAACACGAAGGUGCUCCGC) were mixed for the electroporation. To confirm the knockout of FBXO22 gene, genomic DNAs from the electroporated cells were extracted using PureLink genomic DNA mini kit. FBXO22 gene was amplified by PCR and confirmed by DNA gel electrophoresis and DNA sequencing. Sequencing primers for FBXO22: TCCGAGCGTATTACGGAACG (forward) and AGACTGAACCACCAACCTGC (reverse).

Cloning and Mutagenesis.

Human FBXO22 and mouse Fbxo22 cDNAs with N-terminal HA tag were purchased as gene block from Integrated DNA Technologies and cloned into pCDH-CMV-MCS-EF1-Blast vector (modified from pCDH-CMV-MCS-EF1-Puro vector by replacing PuroR with BlastR) via XbaI and EcoRI sites. FLAG-FKBP12 construct was generated as previously described¹². FKBP12-EGFP and FKBP12-EGFP_NLS with nuclear localization sequence (NLS, PKKKRKV) constructs were generated by cloning FKBP12 gene into a modified Artichoke plasmid with SFFV promoter (Addgene plasmid#: 73320) in which the selection marker puromycin was replaced with neomycin. FKBP12 gene was also cloned into a Cilantro plasmid with hPGK promoter (Addgene plasmid#: 74450). FBXO22 mutants were generated using Q5 site-directed mutagenesis kit (New England Biolabs) following the manufacturer’s instruction.

Library preparation of sgRNAs targeting human E3 ligases.

sgRNAs targeting 680 human E3 ligases were designed using CRISPick (Broad Institute). 5 sgRNAs per E3 ligase were selected. sgRNAs with BsmBI restriction sites were ordered as oligonucleotide pool from Twist Bioscience. sgRNA library was amplified by PCR and purified using QIAquick gel extraction kit (Qiagen, cat# 28706). CRISPRa vector was modified from LentiGuide-Puro (Addgene plasmid#: 52963) by inserting a tetraloop and stem loop sequences after the BsmBI restriction site. Pooled DNA was cloned into CRISPRa vector in the presence of Esp3I (Thermo Scientific, cat# ER0452) and T7 ligase (New England Biolabs, cat# M0318L). The construct library was transformed into ElectroMAX Stbl4 competent cells (Thermo Scientific, cat# 11635018) using gene pulser Xcell electroporation system (Bio-Rad). The electroporated competent cells were cultured on Nunc Bioassay dish (Thermo Scientific, cat# 240835). The construct library was extracted using plasmid Maxi kit (Qiagen, cat# 12163).

Generation of FKBP12 and FBXO22 stably expressed cells by lentivirus transduction.

Lentivirus containing FLAG-FKBP12 and HA-FBXO22 were generated by co-transfection of FLAG-FKBP12/HA-FBXO22, psPAX2 and pMD2.G into HEK293T cells using FuGene 6 transfection reagent. Medium containing lentiviral particles were collected 48 hours post transfection, filtered with 0.45 μM Millex-HV sterile syringe filter unit (MilliporeSigma, cat# SLHV013SL), and used to transduce HEK293T cells in the presence of 10 μg/mL polybrene. 48 hours post transduction, puromycin (2 μg/mL) or blastcidin (10 μg/mL) was added and incubated with the cells for 7 days.

CRISPR-Cas9 transcriptional activation screen.

Lentivirus containing sgRNA library targeting human E3 ligases was added to FKBP12-GFP, dCas9-VP64, and MS2-P65-HSF1-expressing HEK293T cells at a multiplicity of infection (MOI) of 0.3. 24 hours after transduction, cells were treated with 2 μg/mL puromycin for 4 days. After puromycin selection, cells were treated with 2 μM 22-SLF or 2 μM KB02-SLF for 24 hours. Cells were sorted from the bottom 15% of the GFP population in BD FACSAria III cell sorter. BD FACSDiva software (version 9.0.1) was used to collect flow cytometry data. Total DNA was extracted using NucleoSpin blood mini kit (MACHEREY-NAGEL, cat# 740951.250). The illumina run manager and sequencer control software were used to quantify sgRNA. Read counts for sgRNAs targeting each gene were used to calculate fold changes and P values.

Cell lysis and Western blot.

Cells were lysed using RIPA lysis buffer (Thermo Scientific, cat# 89900) containing 25 mM Tris-HCl, pH 7.6, 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate, and 0.1% SDS. Lysis buffer was supplemented with the cOmplete protease inhibitor cocktail (MilliporeSigma, cat# 11873580001) before use. The cell suspension was subjected to sonication in 5 cycles of 40% power for 4 pulses each. Following sonication, the resulting mixture was centrifuged at 16,000g for 10 minutes at 4°C to obtain the supernatant. The protein concentration in the supernatant was quantified using the BCA assay (Thermo Scientific, cat# 23225). The protein lysate was mixed with Laemmli sample buffer (Bio-Rad, cat# 1610737EDU) and heated at 95°C for 5 minutes. The proteins were separated using 4–20% Novex Tris-Glycine mini gels (Thermo Scientific, cat# XP04205BOX) and transferred onto a 0.2 μM polyvinylidene fluoride (PVDF) membrane (Bio-Rad, cat# 1620177). The PVDF membrane was blocked with a solution containing 5% non-fat milk in TBST buffer (0.1% Tween 20, 20 mM Tris-HCl at pH 7.6, and 150 mM NaCl) for 1 hour at room temperature. Primary antibodies were diluted in 5% non-fat milk in TBST buffer (at a dilution of 1:5000 for FLAG and HA, and 1:1000 for others) and applied to the membrane. Incubation times were 1 hour at room temperature for FLAG, HA and β-Actin, and overnight at 4°C for others. Following antibody incubation, the membrane was washed three times with TBST buffer and incubated with a secondary antibody (diluted 1:5000 in 5% non-fat milk in TBST) for 1 hour at room temperature. After another three washes with TBST buffer, the chemiluminescence signal on the membrane was developed using ECL western blotting detection reagent, and the resulting signal was captured using ChemiDoc MP (Bio-Rad). Relative band intensities were quantified using ImageJ (version 1.51h).

Real-time PCR analysis.

HEK293T CRISPR transcriptional activation cells after sgRNA transduction were collected and total RNAs were extracted using the RNeasy mini kit (Qiagen, cat# 74104). cDNAs were obtained by reverse transcription using the iScript cDNA kits (Bio-Rad, cat# 1708890). For real-time PCR analysis, SYBR Green real-time PCR master mix (Thermo Scientific, cat# 4472908) was used. Gene expression was quantified using QuantStudio 3 Real-Time PCR System (Applied Biosystems). sgRNA sequences for IL1B gene transcriptional activation are: IL1B sgRNA#1 TGGCTTTCAAAAGCAGAAGT, and IL1B sgRNA#2 AAAAACAGCGAGGGAGAAAC. Primers for real-time PCR analysis are:

Forward primer of IL1B: ATGATGGCTTATTACAGTGGCAA

Reverse primer of IL1B: GTCGGAGATTCGTAGCTGGA

Forward primer of FBXO22: CGGAGCACCTTCGTGTTGA

Reverse primer of FBXO22: CACACACTCCCTCCATAAGCG

Forward primer of GAPDH: CTGGGCTACACTGAGCACC

Reverse primer of GAPDH: AAGTGGTCGTTGAGGGCAATG

Immunoprecipitations.

Cells were suspended and lysed in NP-40 lysis buffer (25 mM Tris-HCl at pH 7.4, 150 mM NaCl, 10% glycerol, 1% Nonidet P-40) supplemented with cOmplete protease inhibitor cocktail. The cell suspension was incubated on ice for 10 minutes. Subsequently, the mixture was centrifuged at 16,000 g for 10 minutes at 4°C, and the resulting supernatant was extracted for use in immunoprecipitation. For immunoprecipitation, FLAG or HA affinity gel (25 μL slurry per sample) was added to the protein lysates and rotated at 4°C for 2 hours. The affinity gel was then washed four times with immunoprecipitation washing buffer consisting of 0.2% NP-40, 25 mM Tris-HCl at pH 7.4, and 150 mM NaCl. Subsequently, the affinity gel was mixed with Laemmli sample buffer and heated at 95°C for 10 minutes. The resulting supernatant, containing the eluted proteins, was collected and used for subsequent western blot analysis.

Mass spectrometry-based whole proteome analysis.

Cells were lysed in 100 μL of PBS through sonication (10 pulses at 40% intensity, 3 rounds). Protein concentration was measured by a DC assay (Bio-Rad, cat# 5000112). 100 μg of proteins in 100 μL of lysis buffer were denatured using 8 M urea. For reduction, 5 μL of 200 mM DTT stock solution in water was added, and the mixture was heated to 65°C for 15 minutes. Alkylation was achieved by adding 5 μL of 400 mM iodoacetamide stock solution in water and incubating in dark at 37°C for 30 minutes. Subsequently, proteins were precipitated by adding 600 μL of MeOH, 200 μL of CHCl₃, and 500 μL of water. After precipitation, protein pellets were washed with 1 mL of MeOH. The resulting protein pellets were solubilized in 160 μL of EPPS buffer (200 mM). 2 μg of LysC enzyme was added to each sample, and the digestion was carried out at 37°C for 2 hours. This was followed by the addition of 5 μg of trypsin to each sample for another round of digestion, which was allowed to proceed at 37°C for 12 hours. For TMT labeling, 12.5 μg of resulting peptides in 35 μL of EPPS buffer were used. To each sample, 9 μL of CH₃CN was added, and TMT tags (3 μL per sample, with a concentration of 20 μg/μL in CH₃CN) were added. The samples were incubated at room temperature for 1 hour. The TMT labeling reaction was halted by adding 6 μL of a 5% hydroxylamine solution and 2.5 μL of formic acid. Subsequently, the samples were pooled together and subjected to fractionation followed by the analysis using liquid chromatography tandem mass-spectrometry (LC-MS) following the previously reported method³³. Subsequently, the peptides were separated into 12 distinct fractions utilizing the Thermo Vanquish UHPLC fractionator. These resultant peptide fractions were then subjected to liquid chromatography tandem mass spectrometry analysis using an Orbitrap Eclipse Tribrid Mass Spectrometer coupled with a Vanquish Neo UHPLC System. The peptides were introduced onto an EASY-Spray HPLC column (C18, 2 μm particle size, 75 μm inner diameter, 250 mm length) and eluted at a flow rate of 0.25 μL/min, following the gradient: 5% buffer B (80% acetonitrile with 0.1% formic acid) in buffer A (water with 0.1% formic acid) from 0 to 15 minutes, 5% to 45% buffer B from 15 to 155 minutes, and 45% to 100% buffer B from 155 to 180 minutes. The nano-LC electrospray ionization source was set to a voltage of 1.5 kV. The analysis commenced with an MS1 master scan (Orbitrap analysis, resolution 120,000, m/z range 375–1600, RF lens 30%, standard AGC target, auto maximum injection time). In the MS2 analysis, precursor ions were quadrupole-isolated (isolation window 0.7) and then subjected to HCD collision in the ion trap (standard AGC, collision energy 32%, maximum injection time 35 ms). Following each MS2 spectrum, synchronous precursor selection (SPS) enabled the selection of up to 10 MS2 fragment ions for MS3 analysis. These MS3 precursors were fragmented by HCD and analyzed using the Orbitrap (collision energy 55%, AGC 250%, maximum injection time 200 ms, resolution 50,000). The RAW data was collected in Xcalibur (version 4.5.445.18) and analyzed in Proteome Discoverer 2.5.

Chemical proteomic analysis of cysteine reactivity in 22-SLF-treated cells.

Cells were lysed in PBS through sonication (10 pulses at 40% intensity, 3 rounds). The protein concentration was determined using a DC assay and adjusted to 1 mg/mL. Subsequently, 500 μL of lysates were subjected to labeling with 100 μM iodoacetamide-desthiobiotin (IA-DTB) at room temperature for 1 hour. Protein precipitation was achieved by adding 500 μL of methanol and 100 μL of chloroform, followed by a methanol wash (1 mL). The resulting protein pellets were denatured using 90 μL of 9 M urea and 10 mM DTT in 50 mM tetramethylammonium bicarbonate. Alkylation was carried out using 50 mM iodoacetamide at 37°C for 30 minutes. Next, 350 μL of 50 mM tetramethylammonium bicarbonate was added to each sample, followed by the addition of 2 μg of trypsin. Digestion was allowed to proceed at 37°C for 12 hours. Subsequently, 50 μL of streptavidin-agarose beads were added to each sample and the mixture was gently rotated at room temperature for 2 hours. The beads were subjected to washing three times with 1 mL of washing buffer consisting of 0.2% NP-40, 25 mM Tris-HCl at pH 7.4, and 150 mM NaCl, three times with 1 mL of PBS, and two washes with 1 mL of water. Peptides were eluted using 300 μL of 50% acetonitrile containing 0.1% formic acid. The eluted peptides were subsequently dried using a SpeedVac vacuum concentrator (Thermo Scientific). The subsequent steps of TMT labeling and LC-MS analysis were carried out following the methodology described above.

Affinity purification–mass spectrometry.

Cells were lysed in NP-40 lysis buffer comprising 25 mM Tris-HCl at pH 7.4, 150 mM NaCl, 10% glycerol, 1% Nonidet P-40, and the cOmplete protease inhibitor cocktail. Subsequent to centrifugation at 16,000 g for 10 minutes at 4°C, the supernatant was collected for immunoprecipitation. FLAG affinity gel (25 μL slurry per sample) was mixed with the protein lysates and incubated for 2 hours at 4°C. The gel was then washed four times using an immunoprecipitation washing buffer containing 0.2% NP-40, 25 mM Tris-HCl at pH 7.4, and 150 mM NaCl, followed by two washes with PBS. Elution of FLAG-FKBP12 and its associated proteins was carried out by heating the affinity gel at 65°C for 10 minutes in the presence of 8 M urea dissolved in PBS. The eluted proteins were subsequently reduced using 12.5 mM DTT at 65°C for 15 minutes, followed by alkylation with 25 mM iodoacetamide at 37°C for 30 minutes. The protein solution was diluted with PBS to reach a urea concentration of 2 M, and then digested using 2 μg of trypsin at 37°C for 12 hours. Subsequently, 6 μL of TMT tags (20 μg/μL in dry CH₃CN) were added. The TMT labeling was allowed to proceed at room temperature for 1 hour, following which the reaction was quenched by the addition of 6 μL of a 5% hydroxylamine solution and 2.5 μL of formic acid. Subsequently, the samples were pooled and subjected to desalting using Sep-Pak C18 cartridge (Waters, cat# WAT054955). The eluted peptide solution was dried using a SpeedVac vacuum concentrator. Peptides were then subjected to LC-MS analysis following the methodology described above.

Streptavidin enrichment.

Cells were lysed using NP-40 lysis buffer, which contained 25 mM Tris-HCl at pH 7.4, 150 mM NaCl, 10% glycerol, and 1% Nonidet P-40, supplemented with cOmplete protease inhibitor cocktail. The cell suspension was maintained on ice for 10 minutes. Following this, the mixture underwent centrifugation at 16,000 g for 10 minutes at 4°C, leading to the extraction of the resulting supernatant for subsequent use in enrichment. 22-biotin was incubated with streptavidin-agarose beads at room temperature for 2 hours. The beads were subjected to washing three times with 1 mL of washing buffer consisting of 0.2% NP-40, 25 mM Tris-HCl at pH 7.4, and 150 mM NaCl. Cell lysates were added to the probe-conjugated streptavidin-agarose beads and incubated at 4°C for 2 hours. The streptavidin-agarose beads were subjected to four washes using washing buffer, followed by two additional washes using PBS. Elution of interacted proteins was achieved by heating the streptavidin-agarose beads at 65°C for 10 minutes in the presence of 8 M urea dissolved in PBS. After elution, the proteins were subjected to reduction using 12.5 mM DTT at 65°C for 15 minutes, and subsequently alkylated using 25 mM iodoacetamide at 37°C for 30 minutes. To adjust the urea concentration to 2 M, the protein solution was diluted with PBS. The digestion was carried out using 2 μg of trypsin at 37°C for 12 hours. Following digestion, 6 μL of TMT tags (20 μg/μL in dry CH₃CN) were added to the samples. The TMT labeling process was allowed to proceed at room temperature for 1 hour. Subsequently, the reaction was halted by adding 6 μL of a 5% hydroxylamine solution and 2.5 μL of formic acid. The samples were then pooled together and subjected to desalting using Sep-Pak C18 cartridge. The eluted peptide solution was dried using a SpeedVac vacuum concentrator. The dried peptides were subsequently prepared for LC-MS analysis following the methodology described above.

Modeling study.

The predicted structure of FBXO22 from AlphaFold Protein Structure Database was utilized for the covalent docking. The protein preparation workflow in Maestro 13.4 (Schrödinger) was used to preprocess, optimize H-bond assignment, minimize energy and delete waters. Compounds were prepared by the LigPrep module with the OPLS4 force field. CovDock module was used for the docking process. The reaction type of nucleophilic substitution was selected, and the virtual screening approach was applied, along with default parameters for other settings. The final pose with low-energy conformation and good hydrogen-bond and cation-pi interaction geometries was selected. The figure was generated by PyMOL software.

Statistical analysis.

Quantitative data were depicted using scatter plots, displaying the mean accompanied by the standard error of the mean (SEM) represented as error bars. Differences between two groups were assessed using an unpaired two-tailed Student’s t-test. Significance levels were denoted as follows: *P < 0.05, **P < 0.01 and ***P < 0.001. Statistical significance was defined for P values < 0.05.

Extended Data

Extended Data Fig. 1. Generation of CRISPR-Cas9 transcriptional activation cells for the discovery of E3 ligases supporting targeted protein degradation.

a, The construct of FKBP12-EGFP and a schematic representation of the generation of FKBP12-EGFP expressing HEK293T cells. b, Structures of Len-SLF and SLF. c, Fluorescence quantification of FKBP12-EGFP levels in HEK293T cells treated with 2 μM of Len-SLF or 20 μM of SLF for 24 hours. Data are presented as mean values +/− SEM (n = 3 biological independent samples). The statistical significance was evaluated through unpaired two-tailed Student’s t-tests, comparing cells treated with Len-SLF or SLF to DMSO. Statistical significance denoted as ***P < 0.001 and ns: not significant. P value is 0.00026. d, The constructs used for the CRISPR-Cas9 transcriptional activation screen. e, Quantitative PCR analysis of IL1B mRNA levels subsequent to the transduction of sgRNAs targeting the promoter regions of the IL1B gene in HEK293T CRISPR-Cas9 transcriptional activation cells. The bar graph (n = 4 technical replicates) is representative of two independent experiments.

Extended Data Fig. 2. Gating strategy and procedure of fluorescence-activated cell sorting for the CRISPR-Cas9 transcriptional activation screen.

Cells were gated for singlets using forward and side scatter. GFP⁺ cells were gated for the subsequent sorting. Cells from the bottom 15% of the GFP population were sorted and harvested.

Extended Data Fig. 3. Compound screening to identify candidates for the CRIPSR activation screen.

a, HEK293T cell viability after treatment of FKBP12-directed bifunctional compounds (10 μM, 24 hours). Data are presented as mean values +/− SEM (n = 3 biological independent samples). b, Structures of five FKBP12-directed bifunction compounds that show no significant cytotoxicity (cell viability > 50%) at 10 μM. c, The constructs of FKBP12-EGFP with SFFV and hPGK promoters. d, Fluorescence quantification of FKBP12-EGFP and mCherry levels in HEK293T cells stably expressing FKBP12-EGFP with SFFV or hPGK promoter. The bar graph (n = 8 technical replicates) is representative of two independent experiments with similar results. e, Fluorescence quantification of FKBP12-EGFP/mCherry levels in HEK293T cells stably expressing FKBP12-EGFP with SFFV or hPGK promoter, treated with 2 or 5 μM of candidate bifunctional compounds for 24 hours. Data are presented as mean values +/− SEM (n = 3 biological independent samples for compound treatment, n = 8 biological independent samples for DMSO treatment). The statistical significance was evaluated through unpaired two-tailed Student’s t-tests, comparing cells treated with 22-SLF or Len-SLF to DMSO. Statistical significance denoted as *P < 0.05, ***P < 0.001 and ns: not significant. P values are 0.000011 (2 μM Len-SLF in SFFV), 0.000051 (5 μM Len-SLF in SFFV), 0.039 (2 μM 22-SLF in hPGK), 0.031 (5 μM 22-SLF in hPGK), 0.00011 (2 μM Len-SLF in hPGK) and 0.000014 (5 μM Len-SLF in hPGK). f. Flow cytometry analysis of DMSO-treated cells revealed a silenced GFP population. The gating strategy was the same as described in Extended Data Fig. 2. The result is representative of two independent experiments with similar results.

Extended Data Fig. 4. An E3 ligase focused CRISPR-Cas9 transcriptional activation screen identifies DCAF16 supporting KB02-SLF-induced degradation of FKBP12-EGFP_NLS.

a, The construct of FKBP12-EGFP_NLS and a schematic representation of the steps in the CRISPR-Cas9 transcriptional activation screen. b, Volcano plot showing the E3 ligase focused CRISPR-Cas9 transcriptional activation screen for FKBP12-EGFP_NLS degradation after treatment of 2 μM KB02-SLF in HEK293T CRISPR-Cas9 transcriptional activation cells for 24 hours (n = 3 biological independent samples). P values were calculated by two-sided t test without adjustment.

Extended Data Fig. 5. 22-SLF promotes FBXO22-dependent proteasomal degradation of FKBP12.

a, Gene expression ratio values of FBXO22 and CRBN between tumor and normal samples. Data is obtained from GEPIA (http://gepia.cancer-pku.cn/). Full names of the abbreviations are shown in Supplementary Table 1. b, Genomic PCR confirms FBXO22 knockout in A549, MDA-MB-231 and PC3 cells. The result is representative of two independent experiments with similar results. c, Global proteomic analysis confirms FBXO22 knockout in A549, MDA-MB-231 and PC3 cells. The result is representative of two independent experiments with similar results. d, 22-SLF promoted reduction in FKBP12 levels in MDA-MB-231 and PC3 wildtype, but not FBXO22 knockout cells. The bar graph represents quantification of the FLAG-FKBP12/HSP90 protein content. Data are presented as mean values (n = 2 biological independent samples). e, Global proteomic analysis in A549 wildtype and FBXO22 knockout cells treated with 22-SLF (2 μM, 24 hours) (n = 3 biological independent samples). P values were calculated by two-sided t test and adjusted using Benjamini-Hochberg correction for multiple comparisons. f, Bar graph quantification showing the change in KDM4A and KDM4B upon 22-SLF treatment in A549 wildtype and FBXO22 knockout cells. Data are presented as mean values +/− SEM (n = 2 biological independent samples for DMSO treated samples, n = 3 biological independent samples for 22-SLF samples). The statistical significance was evaluated through unpaired two-tailed Student’s t-tests, comparing cells treated with 22-SLF to DMSO. Statistical significance denoted as *P < 0.05 and ns: not significant. P value is 0.044.

Extended Data Fig. 6. 22-SLF rescues Len-SLF-induced FKBP12 degradation in HEK293T FBXO22 knockout cells.

a, 22-biotin-conjugated streptavidin pull-down with lysates of HA-FBXO22-expressing HEK293T cells followed by proteomic analysis revealed FBXO22 as one of the protein targets bound by 22-biotin. Data are presented as mean values (n = 2 biological independent samples). b, HEK293T FBXO22 knockout cells pretreated with 22-SLF (0.1–25 μM, 2 hours) were treated with 0.5 μM of Len-SLF for 4 hours. The graph represents quantification of the FLAG-FKBP12/HSP90 protein content. Data are presented as mean values (n = 2 biological independent samples).

Extended Data Fig. 7. FBXO22 C227 and C228 are involved in 22-SLF-mediated degradation of FKBP12.

a, Interactome studies of FBXO22 wildtype and C227AC228A double mutant revealed that both FBXO22 wildtype and C227AC228A double mutant were assembled into the SKP1-CUL1-RBX1 E3 complex. Data are presented as mean values (n = 2 biological independent samples). b, Global proteomic analysis in HEK293T cells expressing HA-FBXO22 wildtype versus C227AC228A double mutant treated with 22-SLF (0.5 μM, 24 hours) (n = 2 biological independent samples for DMSO treated samples, n = 3 biological independent samples for 22-SLF samples). P values were calculated by two-sided t test and adjusted using Benjamini-Hochberg correction for multiple comparisons.

Extended Data Fig. 8. Evaluation of the impact of FBXO22 K125, V230, and N257 on FKBP12 degradation induced by 22-SLF.

Single mutation of K125, V230 or N257 to alanine in FBXO22 did not block 22-SLF-induced degradation of FKBP12. The bar graph represents quantification of the FLAG-FKBP12/HSP90 protein content. Data are presented as mean values (n = 2 biological independent samples).

Extended Data Fig. 9. Global proteomic analysis in H2228 wildtype and FBXO22 knockout cells.

Proteome in H2228 wildtype and FBXO22 knockout cells was extracted, digested by LysC and trypsin, labeled by TMT tags, and analyzed via MS. Data are presented as mean values (n = 2 biological independent samples).

Extended Data Fig. 10. The acrylamide variant of 22-SLF, 22a-SLF, induced the degradation of FKBP12.

a, Structure of 22a-SLF. b, Comparison of FKBP12 degradation by 22-SLF and 22a-SLF. HEK293T cells expressing HA-FBXO22 were treated with 0.5, 1, or 2 μM of 22-SLF or 22a-SLF for 8 hours. The graph represents quantification of the FLAG-FKBP12/HSP90 protein content. Data are presented as mean values (n = 2 biological independent samples).

Data Availability

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE³⁹ partner repository with the dataset identifier PXD050270. Protein sequences were retrieved from UniProt (https://www.uniprot.org) under accession codes FBXO22 (human), Q8NEZ5; FBXO22 (mouse), Q78JE5. Predicated protein structure was retrieved from the AlphaFold Protein Structure Database (https://alphafold.ebi.ac.uk) under accession code FBXO22 (human), AF-Q8NEZ5-F1. The data supporting the findings of this study are available within the article and Supplementary Information. Source data are provided with this paper.