Administration of a small molecule drug often leads to systemic pharmacological effects that contribute to both efficacy and toxicity and define its therapeutic index. While off-target effects may be mitigated by chemical modifications that improve the specificity of the drug, on-target-off-tissue toxicity represents a unique challenge that requires precise control over tissue partitioning. On-target-off-tissue toxicities affect commonly used drugs such as statins (myopathy caused by HMG CoA reductase inhibition in skeletal muscle)1 and first-generation antihistamines (drowsiness caused by H1 receptor blockade in the brain)2 and can sometimes preclude the safe usage of otherwise effective therapeutic agents. Chemical inhibitors of the mechanistic target of rapamycin (mTOR) provide a case in point. Although mTOR inhibitors have shown efficacy in treating a number of central nervous system (CNS) diseases including tuberous sclerosis complex (TSC)3,4, glioblastoma (GBM)5,6, and alcohol use disorder (AUD)7–9, systemic mTOR inhibition is associated with a variety of dose-limiting adverse effects: immune suppression, metabolic disorders, and growth inhibition in children10,11. If the pharmacological effects of these mTOR inhibitors could be confined to the CNS, their therapeutic window could be substantially widened. Here we present a chemical strategy that allows brain-specific mTOR inhibition through the combination of two pharmacological agents – a brain-permeable mTOR inhibitor (RapaLink-1) whose function requires the intracellular protein FK506-binding protein 12 (FKBP12) and a brain-impermeant ligand of FKBP12 (RapaBlock). When used in a glioblastoma xenograft model, this drug combination drove tumor regression without detectable systemic toxicity. We further demonstrate that this strategy can be adapted to achieve brain-specific inhibition of other kinase targets by developing a method to convert known kinase inhibitors into FKBP12-dependent formats.
Therapeutic targeting of mTOR kinase can be achieved both allosterically and orthosterically. First-generation mTOR inhibitors rapamycin and its analogs (rapalogs) bind to the FK506 rapamycin binding (FRB) domain of mTOR as a complex with the intracellular protein FKBP12, resulting in substrate-dependent allosteric inhibition of mTOR complex 112–14. Second-generation mTOR kinase inhibitors (TORKi) directly bind in the ATP pocket of mTOR and inhibit the activity of both mTOR complex 1 and complex 215–17. The third generation mTOR inhibitor RapaLink-1 is a bitopic ligand that simultaneously engages the ATP pocket and an allosteric pocket (the FRB domain) of mTOR, achieving potent and durable inhibition of its kinase activity18. Despite its large molecular weight (1784 Da), RapaLink-1 is cell- and brain-permeant and has shown enhanced in vivo efficacy in driving glioblastoma regression compared to earlier mTOR inhibitors6.
Because RapaLink-1 contains a TOR kinase inhibitor (TORKi) moiety (Fig. 1a, grey shades) that may directly bind in the ATP pocket, we wondered whether the interaction of RapaLink-1 with FKBP12 is essential for its inhibition of mTOR. We performed in vitro kinase assay with purified mTOR protein and found that RapaLink-1 exhibited identical IC50 values as MLN0128 (the TORKi portion of RapaLink-1) and that inclusion of 10 µM FKBP12 had no effect on its activity (Fig. 1b). Surprisingly, when we tested RapaLink-1 in cells, we observed a strong dependency on FKBP12 for mTOR inhibition. In (K562/dCas9-KRAB cells), CRISPRi-mediated knockdown of FKBP12 with two distinct single guide RNAs (sgRNAs) drastically impeded its cellular activity (Fig. 1c). This effect was even more pronounced when we used FK506 (a high affinity natural ligand of FKBP12) to pharmacologically block the ligand binding site of FKBP12. Similar FKBP12-dependence of RapaLink-1 was observed when we monitored mTOR signaling by Western Blot (Fig. 1d, see also ref18) – whereas 10 nM RapaLink-1 reduced P-S6 (S240/244) and P-4EBP(T37/46) to almost undetectable levels, combination of 10 nM RapaLink-1 and 10 µM FK506 showed no effect on either marker.
These results highlight that while FKBP12 is not essential for RapaLink-1 to bind and inhibit the active site of mTOR in vitro, it is required for cellular activity perhaps serving as an intracellular sink for RapaLink-1 to accumulate in the cell. We probed this possibility using a structural analog of RapaLink-1, where the TORKi moiety had been replaced with tetramethylrhodamine (RapaTAMRA, Extended Data Fig. 1). This fluorescent analog of RapaLink-1 allowed us to quantify intracellular compound concentration by flow cytometry. Consistent with previous findings with RapaLink-118 and other FKBP-binding compounds19,20, RapaTAMRA showed high cellular retention (10-fold increase in median fluorescence intensity) even after extensive washout, but this effect was diminished by FKBP12 knockdown (Extended Data Fig. 1c). While other factors may contribute, we believe that FKBP12-mediated cellular partitioning is partly responsible for the exceptional potency of RapaLink-1.
We next considered whether the dependence of RapaLink-1 on FKBP12 could be harnessed to achieve CNS-restricted mTOR inhibition. Specifically, selective blockade of FKBP12 in peripheral tissues with a potent, brain-impermeant small molecule ligand would allow RapaLink to accumulate in the brain but not peripheral tissues, resulting in brain-specific inhibition (Fig. 2a). To identify a candidate FKBP12 ligand with the suitable permeability profile (i.e. cell-permeable but BBB-impermeable), we considered known natural and synthetic high-affinity FKBP12 ligands (e.g., FK50621,22, rapamycin23,24, SLF/Shield-125,26), but all these compounds readily cross the blood-brain barrier (BBB). Therefore, we synthesized a panel of derivatives of SLF and FK506 where polar substituents (particularly hydrogen bond donors and acceptors) had been attached to the solvent exposed parts of these two molecules, a design principle directly against the empirical rules for designing BBB-permeable drugs27–30 (Extended Data Fig. 2). In addition, modification of the C21-allyl of FK506 has the added advantage that it abolishes the binding to calcineurin31, the natural target of FK506, whose inhibition causes suppression of Nuclear factor of activated T-cells (NFAT) signaling and hence immunosuppression (Fig. 2b). The majority of Shield-1 and FK506 derivatives we synthesized maintained potent FKBP12 binding (Extended Data Fig . 2), as measured by a competition fluorescence polarization assay32. We then screened these compounds (10 µM) in a cell-based assay, evaluating whether they can protect mTOR from inhibition by RapaLink-1 (10 nM) by monitoring phospho-S6 level by Western Blot. While none of the SLF derivatives were found to be effective, most FK506 analogs attenuated the potency of RapaLink-1 (rescuing phospho-S6 levels), with some of them completely blocking RapaLink-1 from inhibiting mTOR (Extended Data Fig. 2). Four of the most effective RapaLink-1 blocking compounds were subjected to an additional in vivo screen, where mTOR signaling was separately analyzed in the brain and skeletal muscle tissues of mice treated with a combination of RapaLink-1 and the candidate compound (1 mg/kg and 40 mg/kg, respectively). A pyridine N-oxide derivative of FK506 protected peripheral tissues from RapaLink-1 but allowed potent inhibition of mTOR in the brain (Extended Data Fig. 3). We therefore chose this compound for our further study and refer to it as “RapaBlock”.
RapaBlock and FK506 bind to FKBP12 with comparable affinity (Fig. 2d, Ki’s: 3.1 nM and 1.7 nM, respectively), but unlike FK506, RapaBlock does not exhibit any inhibitory activity for calcineurin (Fig. 2e). In cultured cells, RapaBlock does not affect mTOR signaling by itself (up to 10 µM) but attenuates the pharmacological effects of RapaLink-1 in a dose-dependent fashion (Fig. 3a). Interestingly, RapaBlock appears more effective at blocking rapamycin, restoring phospho-S6 signal to the same level as untreated cells at 100:1 stoichiometry (Fig. 3d). Because rapamycin and rapalogs exert immunosuppressive effects by inhibiting mTOR activity and thus suppressing cell proliferation in response to costimulatory and cytokine-mediated signals33,34, we asked whether RapaBlock can prevent RapaLink-1-mediated immunosuppression. We stimulated human peripheral mononuclear blood cells (PBMC) with anti-CD3 and anti-CD28 antibodies in the presence of different concentrations of RapaLink-1 and RapaBlock and assessed cell proliferation after 5 days. Whereas RapaLink-1 potently inhibited PBMC proliferation at nanomolar concentrations, addition of RapaBlock abolished this effect, shifting the IC50 by more than 100-fold (Fig. 3b). Though mTOR inhibition does not directly control cytokine production, we observed higher cumulative IL-2 release in cells co-treated with RapaLink-1 and RapaBlock compared to cells treated with RapaLink-1 alone, presumably as a result of greater cell proliferation (Fig. 3c). Consistent with our observation of mTOR signaling, RapaBlock appeared more effective at protecting PBMC from rapamycin-mediated proliferation inhibition (Figs. 3e, 3f). Together, these data demonstrate that by competitively binding to intracellular FKBP12, RapaBlock renders RapaLink-1 and rapamycin incapable of inhibiting mTOR, and in so doing diminishes the immunosuppressive effects of these two drugs.
To examine whether the combination of RapaLink-1 and RapaBlock allows brain-specific inhibition of mTOR in vivo, we treated healthy BALB/cnu/nu mice with RapaLink-1 (1 mg/kg) or a combination of RapaLink-1 (1 mg/kg) and RapaBlock (40 mg/kg), stimulated mTOR activity with insulin (250 mU) after 4 h or 24 h, and analyzed dissected tissues by immunoblot (Fig. 4a). RapaLink-1, when used as a single agent, potently inhibited mTOR signaling in both skeletal muscle and brain tissues, as revealed by the reduced phosphorylation of S6 and 4EBP1. The combination of RapaLink-1 and RapaBlock, however, exhibited remarkable tissue-specific effects: while mTOR activity in the brain was inhibited at a comparable level to mice treated with RapaLink-1 only, mTOR activity in skeletal muscle was not affected.
RapaLink-1 has been shown to be efficacious in orthotopic mouse models of glioblastoma but inhibition of mTOR in the periphery does not contribute to efficacy. We asked whether our combination regimen, lacking the ability to inhibit mTOR activity in peripheral tissues, could retain the efficacy of RapaLink-1 in treating glioblastoma. We established orthotopic intracranial xenografts of U87MG cells expressing firefly luciferase in nude mice and treated these mice with i.p. injections of RapaLink-1 (1 mg/kg), RapaBlock (40 mg/kg), or a combination of both every 5 days. All treatments were well tolerated, and no significant changes of body weight were observed (Fig. 4b). RapaLink-1, both as a single-agent or in combination with RapaBlock, significantly suppressed tumor growth and improved survival, whereas RapaBlock alone had no significant effect on either.
Having established a binary therapeutic approach to achieve brain-specific inhibition of mTOR, we wondered whether the same strategy could be adapted to other drugs for which brain-restricted pharmacology would be desirable. One critical challenge is that for this approach to be generalizable, the “active” component (e.g. RapaLink-1) must be dependent on the availability FKBP12. Drugs with this property are rare, few if any besides rapamycin and FK506 are known. Our earlier investigation of RapaLink-1 (Fig. 1) and RapaTAMRA (Extended Data Fig. 1) led us to hypothesize that other bifunctional molecules consisting of a FKBP12-binding moiety and a kinase inhibitor moiety could be conditionally active and amenable to modulation with RapaBlock.
We first tested our hypothesis with GNE7915, a potent and specific inhibitor of leucine-rich repeat kinase 2 (LRRK2) a pre-clinical agent under investigation for the treatment of Parkinson’s disease35–37. As gain-of-function mutations of LRRK2 are strongly associated with hereditary and sporadic forms of Parkinson’s diseases, LRRK2 kinase inhibition has been pursued as a potential therapeutic strategy38,39. However, recent studies demonstrating that systemic LRRK2 inhibition with small molecule inhibitors induced cytoplasmic vacuolation of type II pneumocytes suggests a potential safety liability for these compounds40,41. We synthesized FK-GNE7915 by chemically linking the pharmacophores of FK506 and GNE7915 with a piperazine group (Fig. 5a). In in vitro LRRK2 kinase assays, FK-GNE7915 was an inferior inhibitor to GNE7915 (Fig. 5c, IC50s of 107 nM and 3.0 nM, respectively), though its activity was potentiated by including 10 µM FKBP12 in the assay (IC50: 21 nM). However, in a cellular assay where we quantified phospho-LRRK2 (S935) levels as a marker for LRRK2 inhibition, FK-GNE7915 was more potent than the parent compound GNE7915 by more than 10-fold (Fig 5d, IC50s of 6.7 nM and 81 nM, respectively). The juxtaposition of these two results suggested a role of the FK506 moiety in the enhanced cellular potency of FK-GNE7915. We reasoned that the high-affinity FK506-FKBP12 interaction can promote the intracellular accumulation of FK-GNE7915 similar to the case of RapaLink-1 and confirmed this using a bifunctional fluorescent probe FK-TAMRA (Extended Data Fig. 4). This feature allowed us to program LRRK2 inhibition by controlling the availability of FKBP12: whereas 100 nM FK-GNE7915 treatment reduced phosphor-LRRK2 (S935) level to 15% of basal level, the combination of 100 nM FK-GNE7915 and 1 µM RapaBlock had no effect on LRRK2 activity (Fig. 5e).
The successful conversion of GNE7915 into a FKBP-dependent LRRK2 inhibitor by simply linking it to an FK506 fragment prompted us to evaluate the generality of this strategy. We explored a number of kinase inhibitors that are being investigated for CNS-diseases: dasatinib (Src family kinase inhibitor, glioblastoma)42–44, lapatinib (EGFR/HER2 inhibitor, glioblastoma)45,46, and prostetin (MAP4K4 inhibitor, amyotrophic lateral sclerosis and Alzheimer’s disease)47. In all three cases, chemically linking the kinase inhibitor to FK506 yielded bifunctional molecules that retained the kinase inhibitory activities of the parent molecules (Extended Data Figs. 5-7). For FK-dasatinib, we also examined its target specificity using both biochemical (Invitrogen SelectScreen Kinase Profiling) and live-cell kinase profiling48 and observed an identical spectrum of kinase targets as dasatinib with the exception of DDR2, a known off-target of dasatinib (Extended Data Fig. 5). Despite their large molecular weights, these bifunctional molecules are active in cells with comparable potencies to their parent compounds and sensitive to deactivation by RapaBlock. While these examples represent a limited set of kinase inhibitors with potential CNS disease indications, it is conceivable more drugs can be similarly configured with programmable pharmacology without losing cellular potency.