Phthalimide conjugation as a strategy for in vivo target protein degradation

See allHide authors and affiliations

Science  19 Jun 2015:
Vol. 348, Issue 6241, pp. 1376-1381
DOI: 10.1126/science.aab1433

A degrading game plan for cancer therapy

Certain classes of proteins that contribute to cancer development are challenging to target therapeutically. Winter et al. devised a chemical strategy that, in principle, permits the selective degradation of any protein of interest. The strategy involves chemically attaching a ligand known to bind the desired protein to another molecule that hijacks an enzyme whose function is to direct proteins to the cell's protein degradation machinery. In a proof-of-concept study, they demonstrated selective degradation of a transcriptional coactivator called bromodomain-containing protein 4 and delayed the progression of leukemia in mice.

Science, this issue p. 1376


The development of effective pharmacological inhibitors of multidomain scaffold proteins, notably transcription factors, is a particularly challenging problem. In part, this is because many small-molecule antagonists disrupt the activity of only one domain in the target protein. We devised a chemical strategy that promotes ligand-dependent target protein degradation using as an example the transcriptional coactivator BRD4, a protein critical for cancer cell growth and survival. We appended a competitive antagonist of BET bromodomains to a phthalimide moiety to hijack the cereblon E3 ubiquitin ligase complex. The resultant compound, dBET1, induced highly selective cereblon-dependent BET protein degradation in vitro and in vivo and delayed leukemia progression in mice. A second series of probes resulted in selective degradation of the cytosolic protein FKBP12. This chemical strategy for controlling target protein stability may have implications for therapeutically targeting previously intractable proteins.

Phthalimide-based drugs emerged in the 1950s. Among the most notable was thalidomide, developed initially as a sedative but infamously withdrawn from human use owing to catastrophic teratogenicity (1). More recently, the phthalimides have been successfully repurposed for erythema nodosum leprosum, multiple myeloma (MM), and myelodyspasia. The efficacy of thalidomide, lenalidomide, and pomalidomide in MM (Fig. 1A) has prompted investigation into the mechanism of action of phthalimide immunomodulatory drugs (IMiDs). By ligand-affinity chromatography, cereblon (CRBN)—a component of a cullin-RING ubiquitin ligase (CRL) complex—was identified as the target of thalidomide (2). Recently, our group and others reported that phthalimides prompt CRBN-dependent proteasomal degradation of transcription factors (TFs) IKZF1 and IKZF3 (3, 4). Crystallographic studies now establish that IMiDs bind CRBN to form a cryptic interface that promotes recruitment of IKZF1 and IKZF3 (5).

Fig. 1 Design and characterization of dBET1.

(A) Chemical structure of JQ1(S), phthalimides, and dBET1. (B) Vehicle-normalized BRD4 displacement by AlphaScreen (triplicate means ± SD). (C) Selective displacement of phage-displayed BETs by dBET1 (BromoScan at 1 μM; n = 1). (D) Crystal structure of dBET1 bound to BRD4 bromodomain 1 (E) Docking of (D) into the published DDB1-CRBN structure (F) dBET1-induced ternary complex formation of recombinant BRD4(1) and CRBN-DDB1 by AlphaScreen [quadruplicate means ± SD; normalized to dimethyl sulfoxide (DMSO)]. (G) Competition of 111 nM dBET1-induced proximity as in (F) in the presence of vehicle (DMSO), JQ1, thal-(–), JQ1(R), and thal-(+) all at 1 μM. Values represent quadruplicate means ± SD, normalized to DMSO. (H) Immunoblot for BRD4 and vinculin (VINC) after 18 hours of treatment of MV4;11 cells with the indicated concentrations of dBET1. (I) Immunoblot for BRD4 and vinculin after treatment of MV4;11 cells with 100 nM dBET1 for the indicated exposures. (J) Cell count–normalized BRD4 levels as determined by high-content imaging in SUM149 cells treated with the indicated concentrations of dBET1 and dBET1(R) for 18 hours. Values represent triplicate means ± SD, normalized to DMSO-treated cells and baseline-corrected using immunoblots seen in fig. S2C.

Ligand-induced target protein destabilization has proven to be an efficacious therapeutic strategy, in particular for cancer, as illustrated by arsenic trioxide–mediated degradation of the PML protein in acute promyelocytic leukemia (6) and estrogen receptor degradation by fulvestrant (7). Historically, target-degrading compounds have emerged serendipitously or through target-specific campaigns in medicinal chemistry. Chemical biologists have devised elegant solutions to modulate protein stability using engineered cellular systems, but these approaches have been limited to nonendogenous fusion proteins (811). Others have achieved the degradation of endogenous proteins through the recruitment of E3 ligases, but these approaches have been limited by the requirement of peptidic ligands (1214), the use of nonspecific inhibitors (15), and by low cellular potency.

RING-domain E3 ubiquitin-protein ligases lack enzymatic activity and function as adaptors to E2 ubiquitin-conjugating enzymes. Inspired by the retrieval of CRBN using a tethered thalidomide (2), we hypothesized that rational design of bifunctional phthalimide-conjugated ligands could confer CRBN-dependent target protein degradation as chemical adapters. We selected BRD4 as an exemplary target. BRD4 is a transcriptional coactivator that binds to enhancer and promoter regions by recognition of acetylated lysines on histone proteins and TFs (16). Recently, we developed a direct-acting inhibitor of BET bromodomains (JQ1) (17) that displaces BRD4 from chromatin and leads to impaired signal transduction from TFs to RNA polymerase II (1820). Silencing of BRD4 expression by RNA interference in murine and human models of MM and acute myeloid leukemia (AML) elicited rapid transcriptional down-regulation of the MYC oncogene and a potent antiproliferative response (19, 21). These and other studies in cancer, inflammation (22), and heart disease (23, 24) establish a desirable mechanistic and translational purpose to target BRD4 for selective degradation.

Having shown that the carboxyl group on JQ1 (25) and the aryl ring of thalidomide (5) can tolerate chemical substitution, we designed the bifunctional dBET1 to have preserved BRD4 affinity and an inactive epimeric dBET1(R) as a stereochemical control (Fig. 1, A and B). Selectivity profiling confirmed potent and BET-specific target engagement among 32 bromodomains (BromoScan) (Fig. 1C and tables S1 and S2). A high-resolution crystal structure (1.0 Å) of dBET1 bound to BRD4(1) confirmed the mode of molecular recognition, comparable to JQ1 (Fig. 1D, fig. S1, and table S3). Using the dBET1-BRD4(1) crystal structure and the recently reported structure of CRBN bound to thalidomide (5), we have modeled the feasibility of ternary complex formation in silico. An extended conformation of dBET1 was capable of bridging ordered BRD4(1) and CRBN without destructive steric interactions (Fig. 1E). To experimentally assess the chemical adaptor function of dBET1, we established a homogeneous proximity assay for recombinant human CRBN–DNA damage–binding protein 1 (CRBN-DDB1) and BRD4(1) (Fig. 1F). Luminescence arising from proximity of CRBN-DDB1- and BRD4-bound acceptor-donor beads increases with low concentrations of dBET1 and decreases at higher concentrations, consistent with saturation of CRBN and BRD4 binding sites by excess ligand. Competitive coincubation with free JQ1 or thalidomide inhibits ternary complex formation in a stereo-specific manner (Fig. 1G).

To assess the effect of dBET1 in cells, we treated a human AML cell line (MV4;11) for 18 hours with increasing concentrations of dBET1 and assayed endogenous BRD4 levels by immunoblot. Pronounced loss of BRD4 (>85%) was observed with concentrations of dBET1 as low as 100 nM (Fig. 1H). The epimeric control dBET1(R) was inactive (Fig. 1B), which demonstrated that BRD4 degradation requires target protein engagement (fig. S2, A and B). The kinetics of BRD4 degradation were next determined using 100 nM dBET1 in MV4;11 cells. Marked depletion of BRD4 was observed at 1 hour, and complete degradation was observed at 2 hours of treatment (Fig. 1I). A partial recovery in BRD4 abundance at 24 hours establishes the possibility of compound instability, a recognized liability of phthalimides. To quantify dose-responsive effects on BRD4 protein stability, we developed a cell-count normalized, high-content assay using adherent SUM149 breast cancer cells (Fig. 1J). Depletion of BRD4 was observed for dBET1 [half-maximal effective dose (EC50) = 430 nM] without apparent activity for dBET1(R), confirmed by immunoblot (fig. S2, C and D). Additional cultured adherent and nonadherent human cancer cell lines showed comparable response (SUM159, MOLM13) (fig. S3).

To explore the mechanism of dBET1-induced BRD4 degradation, we studied requirements on proteasome function, BRD4 binding, and CRBN binding using chemical genetic and gene-editing approaches. First, we confirmed that treatment with either JQ1 or thalidomide alone was insufficient to induce BRD4 degradation in MV4;11 cells (fig. S4A). Degradation of BRD4 by dBET1 was rescued by the proteasome inhibitor carfilzomib, which established a requirement for proteasome function (Fig. 2A). Pretreatment with excess JQ1 or thalidomide abolished dBET1-induced BRD4 degradation, consistent with a requirement for both BRD4 and CRBN engagement (Fig. 2A). Pretreatment with the NAE1 inhibitor MLN4924 (26) rescued BRD4 stability and established its dependence on CRL activity, as expected for cullin-based ubiquitin ligases that require neddylation for processive E3 ligase activity (4, 27) (fig. S4B). Finally, to definitively confirm the cellular requirement for CRBN, we used a previously published CRBN-deficient human MM cell line (MM1.S-CRBN−/−) (4). Whereas treatment of wild-type MM1.SWT cells with dBET1 promoted degradation of BRD4, exposure of MM1.S-CRBN−/− cells to dBET1 was ineffectual (Fig. 2B). These data provide mechanistic evidence for CRBN-dependent proteasomal degradation of BRD4 by dBET1.

Fig. 2 Chemical and genetic rescue of dBET1- and dFKBP-1–mediated degradation.

(A) Immunoblot for BRD4 and vinculin after a 4-hour pretreatment with DMSO, carfilzomib (400 nM), JQ1 (10 μM), or thalidomide (10 μM), followed by a 2-hour dBET1 treatment (100 nM) in MV4;11 cells. (B) Immunoblot for BRD4, CRBN, and tubulin after treatment of MM1SWT or MM1SCRBN−/− cells with dBET1 for 18 hours at the indicated concentrations. (C) Structures of dFKBP-1 and dFKBP-2. (D) Immunoblot for FKBP12 and vinculin after 18 hours of treatment with the indicated compounds. (E) Immunoblot for FKBP12 and vinculin after a 4-hour pretreatment with DMSO, carfilzomib (400 nM), MLN4924 (1 μM), SLF (20 μM), or thalidomide (10 μM), followed by a 4-hour dFKBP-1 treatment (1 μM) in MV4;11 cells. (F) Immunoblot for FKBP12, CRBN, and tubulin (Tub) after treatment of 293FTWT or 293FTCRBN−/− cells with dFKBP12 at the indicated concentrations for 18 hours.

To assess the feasibility of extending this strategy to other protein targets, we designed and synthesized phthalimide-conjugated ligands to the cytosolic signaling protein FKBP12 (FKBP1A). FKBP12 has been extensively studied in the chemical biology literature, which includes studies of engineered target degradation and a growing literature on chemical dimerizers. At a known permissive site on the FKBP12-directed ligand steel factor (SLF), we positioned two chemical spacers to create the conjugated phthalimides dFKBP-1 and dFKBP-2 (Fig. 2C). Both potently decreased FKBP12 abundance in MV4;11 cells (Fig. 2D). As with dBET1, destabilization of FKBP12 by dFKBP-1 was rescued by pretreatment with carfilzomib, MLN4924, free SLF, or free thalidomide (Fig. 2E). We established CRBN-dependent degradation using previously published isogenic 293FT cell lines that are wild type (293FT-WT) or deficient (293FT-CRBN−/−) (4) for CRBN (Fig. 2F).

An unbiased, proteome-wide approach was selected to assess the cellular consequences of dBET1 treatment on protein abundance, in a quantitative and highly parallel format (28). We compared the immediate impact of dBET1 treatment (250 nM) to JQ1 and vehicle controls in MV4;11 cells. A 2-hour incubation was selected to capture primary, immediate consequences of small-molecule action and to mitigate expected, confounding effects on suppressed transactivation of BRD4 target genes. We prepared three biological sample replicates for each treatment condition using isobaric tagging that allowed the detection of 7429 proteins. After JQ1 treatment, few proteomic changes were observed (Fig. 3A and table S4). Only MYC was significantly depleted more than twofold after 2 hours of JQ1 treatment, which confirmed the reported rapid and selective effect of BET bromodomain inhibition on MYC abundance in AML (Fig. 3, A and C) (21). JQ1 treatment also down-regulated the oncoprotein PIM1 (Fig. 3, A and C).

Fig. 3 Selective BET bromodomain degradation established by expression proteomics.

MV4;11 cells were treated for 2 hours with DMSO, 250 nM dBET1, or 250 nM JQ1. (A) Fold-change in abundance of 7429 proteins comparing JQ1 to vehicle (DMSO) treatment, versus P value (t-test; triplicate analysis). (B) As for (A), but comparing 250 nM dBET1 to vehicle treatment. (C) Selected proteins from (A) and (B) normalized to vehicle. Values represent triplicate means ± SD. (D) Immunoblot of BRD2, BRD3, BRD4, MYC, PIM1, and vinculin (VINC) after a 2-hour treatment of MV4;11 cells with DMSO, 250 nM dBET1, or 250 nM JQ1. (E) Quantitative reverse transcription–polymerase chain reaction analysis of transcript levels of BRD2, BRD3, BRD4, MYC, and PIM1 after a 2-hour treatment of MV4;11 cells with DMSO, 250 nM dBET1, or 250 nM JQ1. Values represent triplicate means ± SD. *P = 0.01 to 0.05; **P = 0.001 to 0.01; ****P < 0.0001.

Treatment with dBET1 elicited a comparable, modest effect on MYC and PIM1 expression. Remarkably, only three additional proteins were identified as significantly (P < 0.001) and markedly depleted (to one-fifth) in dBET1-treated cells: BRD2, BRD3, and BRD4 (Fig. 3, B and C). These findings are consistent with the anticipated, BET-specific bromodomain target spectrum of the JQ1 bromodomain-biasing element on dBET1 (17). The remaining BET-family member BRDT is not detectable in MV4;11 cells. To validate these findings, we measured BRD2, BRD3, BRD4, MYC, and PIM1 levels by immunoblot after compound treatment, as above. BET family members were degraded only by dBET1, whereas MYC and PIM1 abundance was decreased by both dBET1 and JQ1 (and to a lesser degree) (Fig. 3D). No effect on Ikaros TF expression was observed in either treatment condition (fig. S5). MYC and PIM1 are often associated with massive adjacent enhancer loci by epigenomic profiling (18, 20), which suggested a transcriptional mechanism of down-regulation. We therefore measured mRNA transcript abundance for each depleted gene product (Fig. 3E). Treatment with either JQ1 or dBET1 down-regulated MYC and PIM1 transcription, suggestive of secondary transcriptional effects. Transcription of BRD4 and BRD3 were unaffected, consistent with posttranscriptional effects. Note that transcription of BRD2 was affected by JQ1 and dBET1, whereas protein stability of the BRD2 gene product was only influenced by dBET1, suggestive of transcriptional and posttranscriptional consequences. These data establish a highly selective effect of dBET1 on target protein stability, proteome-wide.

We next explored the differential antiproliferative consequences of BET degradation with dBET1 to BET bromodomain inhibition with JQ1. Degradation of BRD4 by dBET1 was associated with an enhanced apoptotic response in MV4;11 AML and DHL4 lymphoma cells, as measured by caspase activation (Caspase-GLO) (Fig. 4A), cleavage of poly(ADP-ribose) polymerase (PARP), cleavage of caspase-3 (immunoblot) (Fig. 4B), and Annexin V staining (flow cytometry) (fig. S6A). Kinetic studies revealed an apoptotic advantage for dBET1 after pulsed treatment followed by washout in MV4;11 cells (Fig. 4, C and D). Indeed, dBET1 induced a potent and superior inhibitory effect on MV4;11 cell proliferation at 24 hours (Fig. 4E).

Fig. 4 The kinetic and antileukemic advantage of BET bromodomain degradation.

(A) Fold increase of apoptosis, assessed via Caspase-Glo assay relative to DMSO-treated controls, 24 hours of treatment in MV4;11 or DHL4 cells. Values represent quadruplicate means ± SD. (B) Immunoblot for cleaved caspase 3, PARP cleavage, and vinculin after treatment with dBET1 and JQ1 for 24 hours, as indicated. (C) Fold increase of apoptosis (Caspase-Glo) relative to DMSO-treated controls. MV4;11 cells were treated for 4 or 8 hours with JQ1 or dBET1 at the indicated concentrations. Drug was washed out with phosphate-buffered saline (3 times) before cells were plated in drug-free medium for a final treatment duration of 24 hours. Values represent quadruplicate means ± SD. (D) Immunoblot for cleaved caspase 3, PARP cleavage, and vinculin after treatment conditions as described in (C). (E) Dose-proportional effect of dBET1 and JQ1 (24 hours) on MV4;11 cellular viability as approximated by adenosine triphosphate–dependent luminescence. Values represent quadruplicate means ± SD. (F) Immunoblot for BRD4 and vinculin after treatment of primary patient cells with the indicated concentrations of dBET1 for 24 hours. (G) Annexin V–positive primary patient cells after 24 hours of treatment with either dBET1 or JQ1 at the indicated concentrations. Values represent the average of duplicates and the range as error bars (representative scatter plots in fig. S6). (H) Tumor volume (means ± SEM) of vehicle-treated mice (n = 5) or mice treated with dBET1 (50 mg/kg; n = 6) for 14 days. (I) Immunoblot for BRD4, MYC, and vinculin (VINC) by using tumor lysates from mice treated either once for 4 hours or twice for 22 hours and 4 hours, compared with a vehicle-treated control. (J) Immunohistochemistry for BRD4, MYC, and Ki67 of a representative tumor of a dBET1-treated and a control-treated mouse (quantification of three independent areas in fig. S8). (K) Percentage of mCherry+ leukemic cells (means ± SEM) in flushed bone marrow from disseminated MV4;11 xenografts after daily treatment with dBET1 (n = 8) and JQ1 (n = 8) (both at 63.8 μmol/kg) or formulation control (n = 7) for 19 days. ***P = 0.0001 to 0.001; other P values as in Fig. 3 legend.

The rapid biochemical activity and robust apoptotic response of cultured cell lines to dBET1 established the feasibility of assessing effects on primary human AML cells, where ex vivo proliferation is negligible in short-term cultures. Exposure of primary leukemic patient blasts to dBET1 elicited dose-proportionate depletion of BRD4 (Fig. 4F) and improved apoptotic response compared to JQ1 (Annexin V and propidium iodide staining) (Fig. 4G and fig. S6B). Together, these data exemplify that target degradation can elicit a more pronounced biological consequence than domain-specific target inhibition.

To model the therapeutic opportunity of BRD4 degradation in vivo, we first evaluated the tolerability and antitumor efficacy of dBET1 in a murine hind-limb xenograft model of human MV4;11 leukemia cells (29). As pharmacokinetic studies of dBET1 indicated adequate drug exposure in vivo (fig. S7A), tumor-bearing mice were treated with dBET1 administered by intraperitoneal injection (50 mg/kg body weight daily, ip) or vehicle control. After 14 days of therapy, a first tumor reached institutional limits for tumor size, and the study was terminated for comparative assessment of efficacy and modulation of BRD4 stability and function. Administration of dBET1 attenuated tumor progression, as determined by serial volumetric measurement (Fig. 4H), and decreased tumor weight was assessed post mortem (fig. S8A). Acute pharmacodynamic degradation of BRD4 was observed by immunoblot 4 hours after a first or second daily treatment with dBET1 (50 mg/kg ip) (Fig. 4I), accompanied by down-regulation of MYC (Fig. 4I). These findings were confirmed by immunohistochemistry at the conclusion of the 14-day efficacy study (Fig. 4J). A statistically significant destabilization of BRD4, down-regulation of MYC, and inhibition of proliferation (Ki67 staining) was observed with dBET1 compared with vehicle control in excised tumors (Fig. 4J and fig. S8B). Notably, 2 weeks of dBET1 was well tolerated with preservation of animal weight and normal complete blood counts (fig. S7, C and D). To compare the efficacy of BET inhibition to BET degradation, we selected an aggressive disseminated leukemia model (mCherry+ MV4;11) and treated animals with established disease using equimolar concentrations of JQ1 and dBET1 for 19 days. Post mortem analysis of leukemic burden in bone marrow by fluorescence-activated cell sorting revealed significantly decreased mCherry+ disease with dBET1 administration (Fig. 4K).

In summary, we present a mechanism-based chemical strategy for endogenous target protein degradation. Phthalimide conjugation to selective small molecules produces CRBN-dependent posttranslational degradation with exquisite target-specific activity. Although this approach is CRBN-dependent, CRBN is ubiquitously expressed in physiologic and pathophysiologic tissues, which supports its broad utility in developmental and disease biology. The increased apoptotic response of primary AML cells to dBET1, even compared with the efficacious tool compound JQ1, highlights the potential superiority of BET degradation over BET bromodomain inhibition and prompts consideration of therapeutic development. Pharmacologic destabilization of BRD4 in vivo also resulted in improved antitumor efficacy in a human leukemia xenograft compared with the effects of JQ1. The extension of this approach to new targets (here, FKBP12), low molecular mass, high cell permeability, potent cellular activity, and synthetic simplicity addresses limitations associated with prior pioneering systems. We anticipate dFKBP1 will also serve as a useful tool for the research community in control of fusion protein stability. A more general implication of this research is the feasibility of approaching intractable protein targets using phthalimide-conjugation of target-binding ligands that may or may not have target-specific inhibitory activity.

Supplementary Materials

Materials and Methods

Figs. S1 to S8

Tables S1 to S3

References (3036)

References and Notes

  1. Acknowledgments: We thank W. Kaelin, S. Orkin, and R. Mazitschek for engaging discussions and W. Kaelin for cellular reagents; S.-H. Seo and S. Deangelo for assistance in the purification of BRD4; C. Ott for help with in vivo model studies; and N. Thoma for CRBN expression reagents. This research was supported by generous philanthropic gifts from Marc Cohen and Alain Cohen, the William Lawrence and Blanche Hughes Foundation, and the NIH (R01-CA176745 and P01-CA066996 to J.E.B.). G.E.W. is supported by an EMBO long-term fellowship. D.L.B. is a Merck Fellow of the Damon Runyon Cancer Research Foundation (DRG-2196-14). Atomic coordinates and structure factors have been deposited in the Protein Data Bank (accession code 4ZC9). Quantitative proteomics studies were performed by R. Kunz of the Thermo Fisher Scientific Center for Multiplexed Proteomics at Harvard Medical School. Dana-Farber Cancer Institute has filed patent applications (62/096,318; 62/128,457; 62/149,170) that include the dBET and dFKBP compositions described in this manuscript. J.E.B. is a Founder of Tensha Therapeutics, a biotechnology company that develops druglike derivatives of JQ1 as investigational cancer therapies.
View Abstract

Stay Connected to Science

Navigate This Article