Discovery of a potent GLUT inhibitor using rapafucin 3D microarrays

Abstract

Glucose transporters play an essential role in cancer cell proliferation and survival and have been pursued as promising cancer drug targets. Using microarrays of a new macrocycle library known as rapafucins that was inspired by rapamycin, we screened for new inhibitors of GLUT1. We identified multiple hits from the rapafucin 3D microarray and confirmed one hit as a bona fide GLUT1 ligand, named Rapaglutin A (RgA). We demonstrated that RgA is a potent inhibitor of GLUT1 as well as GLUT3 and GLUT4 with an IC₅₀ value of low nanomolar for GLUT1. RgA was found to inhibit glucose uptake, leading to a decrease in cellular ATP synthesis, activation of AMP-dependent kinase, inhibition of mTOR signaling, and induction of cell cycle arrest and apoptosis in cancer cells. Moreover, RgA was capable of inhibiting tumor xenografts in vivo without obvious side effects. RgA is a new chemical tool to study GLUT function and a promising lead to develop anticancer drugs.

Graphical abstract

A rapafucin 3D small molecule microarray was developed and 3,918 rapafucins were screened against human glucose transporter 1 (GLUT1) in cell lysate. A potent pan-GLUT inhibitor, named Rapaglutin A (RgA), was discovered. RgA was found to inhibit glucose uptake, induce cell apoptosis, and inhibit tumor xenografts in vivo.

Glucose is a universal cellular fuel that serves as both an energy source and building blocks of a variety of macromolecules. In comparison to normal cells, cancer cells have a higher demand for glucose due to their faster proliferation rate and aerobic glycolysis as a consequence of the Warburg effect.^[1] Several common cancer driver mutations such as p53 and KRAS as well as hypoxia have been shown to upregulate the expression of glucose transporters, prominent among which are members of the facilitative glucose transporter family including GLUT1 and GLUT3.^[2] Inhibition of GLUTs has been shown to not only block cancer cell growth but can also sensitize cancer cells to other drugs.^[3] Extensive efforts have been made to discover new inhibitors of GLUTs, particularly GLUT1, as leads for developing novel anticancer drugs.^[4] Although a number of GLUT inhibitors have been reported, including BAY-876,^[5] a potent and isoform-specific GLUT1 inhibitor, none has entered the clinic to date.

We recently generated a library of macrocycles named rapafucins that were inspired by the natural products rapamycin and FK506. The premise of the rapafucin design is to exploit the FKBP-binding domain of rapamycin and FK506 that confers favorable cellular and pharmacokinetic advantages to macrocycles and use it as a key scaffold to display non-natural oligopeptides in place of the effector domains of rapamycin and FK506. The ability of rapafucins to bind FKBP proteins to form a tight complex confers a number of advantages as drug leads, including greater stability, higher intracellular accumulation, larger size and superior pharmacokinetic and pharmacodynamic properties.^[6] We have designed and synthesized a 45,000-compound rapafucin library and identified promising hits against several targets including a potent and isoform-specific inhibitor of the human equilibrative nucleoside transporter (hENT)1 that showed in vivo efficacy in an animal model of ischemic kidney reperfusion injury.^[7] Given that hENT1 and GLUT belong to the same superfamily of solute carrier transporters, we were prompted to screen the rapafucin library for new GLUT1 inhibitors. In the current study, we developed a 3D small molecule microarray by immobilizing 3,918 rapafucins on a single chip and screened cell lysates containing stably expressed GLUT1. We identified a potent inhibitor, named rapaglutin A (RgA), that inhibited GLUT1 as well as GLUT3 and GLUT4. We also demonstrated that RgA inhibited glucose uptake, induced cell apoptosis, and inhibited the growth of tumor xenografts of breast cancer cells in vivo.

Small molecule microarray has been shown to be a powerful platform for high-throughput screening.^[8] Among the different methods of small molecule immobilization, we chose to use pre-assembled diazirine that upon activation by UV light, generates a reactive carbene species to covalently react with and capture small molecules. As macrocycles, rapafucins are particularly suitable for this platform as multiple sites for immobilization exist around the periphery of the macrocycles. Given the stochastic nature of the carbene-mediated crosslinking reaction, there is a high probability that a fraction of a given rapafucin species will be covalently immobilized via positions that would not interfere with its binding to target protein. To develop a rapafucin microarray for high-throughput screening, we explored both 2D and 3D surface structures prefabricated on glass slides (Fig. 1a). Unlike the 2D surface structure,^[9] the 3D surface structure was fabricated by growing polymers on the glass surface with each polymeric chain carrying up to hundreds of trifluoromethylphenyl diazirine moieties, significantly increasing the number of sites for rapafucin immobilization on the 3D surface and at the same time providing a biocompatible environment for rapafucin-protein interactions (Fig. 1a, Fig.S1). ^[10]

Figure 1.

Development of rapafucin 3D microarray. (a) The construction of the rapafucin 2D and 3D microarrays. Ab, antibody; POI, protein of interest; red star, positive POI binder; green star, negative POI binder. (b) Optimization of 3D microarray. Small molecule array screening against purified GST-FKBP12 on 3D copolymer diazirine surface with different ratios of monomers, PEGMA: DMAEMA= 0:10, 2:8, 5:5, 8:2 or 10:0. (c) Comparison of the interaction between rapafucins and FKBP12 on 2D and 3D surfaces. Microarray images of the rapafucin 2D and 3D microarrays probed by purified GST-FKBP12. All of the compounds were spotted in duplicate.

To develop the 3D surface for rapafucin microarray fabrication, we attempted to optimize the polymer density both horizontally and vertically to achieve the highest sensitivity. Horizontal density was controlled by mixing 2-bromoisobutyryl bromide and propionyl bromide (as the spacer) at different molar ratios to control the density of active atom transfer radical polymerization (ATRP) initiation sites as previously described.^[10] We applied 1:100 ratio of 2-bromoisobutyryl bromide and propionyl bromide to our 3D surface without further optimization as this ratio is commonly used to achieve high sensitivity of 3D surface.^{[10a, 10d]} Vertical density was manipulated using poly-(PEGMA-co-DMAEMA) matrix to maximize the binding of proteins to adjacent ligands displayed on the same polymer. We optimized the vertical density by adjusting the gradient ratio of monomer PEGMA and DMAEMA followed by determination of binding of FKBP12 to the resultant 3D surface as all rapafucins contain an embedded FKBP-binding domain. The highest signal-to-background ratio (SBR) was obtained at a PEGMA-to-DMAEMA ratio of 8:2 (Fig. 1b, Fig. S2). Using this optimized diazirine-containing 3D surface grafted on a glass slide, we robotically arrayed a rapafucin library^[7] containing 3,918 individual compounds to the glass. As a comparison, we also arrayed the same 3,918 individual rapafucins on a 2D surface grafted slide displaying the same diazirine as previously described.^[9] Once stock solutions of the rapafucin library were arrayed on the 2D or 3D surface and most of solvent carrier was evaporated, the crosslinking reaction was initiated by irradiating the surfaces with 365-nm wavelength UV light.^[11] We next compared the 2D and the newly developed 3D surface for their capacity to bind FKBP12 under the same conditions, the SBR of the binding of FKBP12 on the 3D surface is on average 6-fold greater than that on the 2D surface (Fig. 1c, Fig. S3), indicating that the 3D microarray of rapafucins is a superior platform for screening target proteins.

To screen for GLUT1-interacting rapafucins, we stably overexpressed GLUT1 in HEK293T cells and generated cell lysates containing detergent-solubilized recombinant GLUT1 (Fig. S4). We next incubated the GLUT1-containing cell lysate on both 2D and 3D rapafucin microarrays. After washing the slides to remove unbound proteins, the bound GLUT1 protein was detected with anti-GLUT1 antibodies, followed by visualization with Cy5-labeled secondary antibodies using a microarray scanner. Rapafucins were scored as positive hits when the corresponding SBR was greater than 3. Based on this criterion, a total of 17 rapafucin hits and one positive control BAY-876, were identified on the 3D rapafucin microarray. In contrast, only one hit (WL13-F11, also among the 17 hits identified from the 3D rapafucin microarray) was identified on the 2D rapafucin microarray (Table S1, Fig. S5).

To determine which of the 17 rapafucin hits inhibited the transporter activity of GLUT1, we employed an orthogonal glucose uptake assay using 2-deoxy-D-[³H]glucose ([³H]-2DG), a nonhydrolyzable, radioactive glucose analog. Each hit was separately incubated with A549 cells for 10 min before the amount of [³H]-2DG taken up by the cells was measured using scintillation counting. Two of the 17 hits, JW11-D2 and HP17-C2 (Fig. 2a and 2b), were found to block the uptake of [³H]-2DG appreciably with IC₅₀ values of 11.6 nM and 243 nM, respectively (Fig. 2c, Table S1). In light of the potent inhibition of glucose uptake by JW11-D2, we named it rapaglutin A (RgA).

Figure 2.

Identification of Rapafucin JW11-D2 as a GLUT1 binder. (a) 3D Microarray images of two positive hits JW11-D2 and HP17-C2. (b) Chemical structures of JW11-D2 (Rapaglutin A, RgA) and HP17-C2. (c) Inhibition of 2-deoxy-D-[³H] glucose (³H-2DG) uptake in A549 cells by RgA and HP17-C2. (d) The competition profile of biotin-RgA binding to GLUT1 in HEK 293T cell lysate by RgA.

Next, we determined the binding affinity of RgA for GLUT1 using a RgA biotin pull-down assay. We synthesized a biotin-RgA conjugate by tethering the biotin moiety through carbon-carbon double bond in the FKBP-binding domain of RgA (Fig.S6). An [³H]-2DG uptake assay in A549 cells revealed that biotin-RgA retained inhibitory activity against GLUT1 with an IC₅₀ value of 211 nM (Fig. S6), suggesting that the biotin-RgA conjugate remained active against GLUT1 albeit with lower potency. Using the biotin-RgA conjugate, we performed a pull-down experiment with cell lysate containing detergent-solubilized GLUT1 protein prepared from HEK293T cells overexpressing GLUT1, followed by Western blot analysis with a GLUT1-specific antibody. The biotin-RgA conjugate was capable of pulling down GLUT1 (Fig. 2d), further supporting that RgA directly interacts with GLUT1. Importantly, binding of GLUT1 to the biotin-RgA probe is dose-dependently competed by free RgA, allowing for determination of the binding affinity of RgA for GLUT1 with an estimated K_d value of 78 nM (Fig. 2d).

GLUT1 is a basal glucose transporter expressed in almost all cell types, and is upregulated in many cancer cells.^[12] To determine whether RgA could inhibit glucose uptake in cancer cell lines in addition to A549, we measured the impact of RgA on [³H]-2DG uptake in six other cancer cell lines, including HCC1954, MCF-7, PANC10.05, Jurkat T, HeLa, and RKO (Fig. 3a,Table S2). We observed that RgA dose-dependently inhibited glucose uptake in all cell lines tested with IC₅₀ values ranging from 3 nM to 19 nM (Table S2). Among them, breast cancer cell MCF-7 was most sensitive to RgA with an IC₅₀ of 3.3 nM. These results demonstrated that RgA has a general inhibitory effect on glucose uptake in all cancer cell lines tested.

Figure 3.

RgA is a potent, isoform-nonspecific, and FKBD-independent inhibitor of glucose transporters. (a) Inhibition of 2-deoxy-D-[³H] glucose ([3H]-2DG) uptake in A549, HCC1954, and MCF-7 cells by RgA; (b) Inhibition of [³H]-2DG uptake in DLD1 wild type or GLUT1 knock out cells by RgA, BAY-876, and Cytochalasin B (CytoB); (c) The competition profiles of biotin-RgA binding to GLUT3 and GLUT4 by RgA. (d) Inhibition of [³H]-2DG uptake in MCF-7 cells by 100nM of RgA, 20 μM of FK506, 20 μM of Rapamycin and their combinations.

The human GLUT family consists of 14 members that differ in substrate affinity, specificity, and tissue distribution.^[12] We determined whether RgA is specific for GLUT1 using a pair of isogenic cell lines. DLD-1 wild type (for GLUT1) and DLD-1 GLUT1 knock out (for GLUT3) ^[5] cells were treated in parallel with RgA, BAY-876, a reported GLUT1-specific inhibitor, and cytochalasin B, a non-specific GLUT inhibitor,^[13] followed by assessment of [³H]-2DG uptake. As expected, BAY-876 lost its inhibitory activity but cytochalasin B maintained its inhibitory activity in GLUT1 knock out cells (Fig. 3b). Similar to cytochalasin B, RgA inhibited glucose uptake in both wild type and GLUT1 knock out cells with IC₅₀ values of 17.5nM and 27.1nM respectively, suggesting that RgA, unlike BAY-876, is not specific for GLUT1 (Fig. 3b, Table S3). To further assess the isoform specificity, we overexpressed three other isoforms of glucose transporters, including GLUT2, 3, and 4, in HEK 293T GLUT1 knock out cells. The overexpression of GLUT2 in HEK293T cells did not succeed due to unknown reasons. Using the [³H]-2DG uptake assay, RgA was also found to block glucose uptake in GLUT4 overexpressed cells with IC₅₀ value of 25.9nM (Table S4). In addition, similar to the result of GLUT1 pull-down, biotin-RgA was able to pull down both GLUT3 and GLUT4 (Fig. 3c). RgA exhibited lower binding affinity against GLUT3 and GLUT4 with estimated K_d values of 330 nM and 98.2 nM, respectively (Fig. 3c, Table S4). These results suggested that RgA is a non-specific inhibitor of multiple isoforms of GLUTs, including at least GLUT1, 3, and 4.

Like FK506 and rapamycin, RgA contains an FKBP-binding domain. We determined the binding affinity of RgA to different isoforms of FKBP. Interestingly, RgA showed binding selectivity among different isoforms of FKBPs, with the highest affinity for FKBP12 (K_i = 1.5 nM for inhibition of the prolyl isomerase activity) (Table S5). The ability of RgA to form a complex with FKBP12 raised the question of whether FKBP is required for its interaction with GLUTs. A hallmark of FKBP dependence is that the cellular effects can be antagonized by other FKBP-binding ligands with no or orthogonal biological activity as has been shown for FK506 and rapamycin.^[14] High concentration of FK506 and rapamycin had negligible effect on the inhibitory activity of RgA in the [³H]-2DG uptake assay (Fig. 3d). To further determine the dependence of RgA on endogenous FKBP, we knocked out three major isoforms of FKBP, FKBP12, 51 and 52 using CRISPR-Cas9 in Jurkat T cells.^[7] Unlike ENT1 inhibitor rapadocin,^[7] knockout of the three FKBP isoforms showed negligible effects on the sensitivity of cells to RgA (Fig. S7). Taken together, these results strongly suggested that the inhibitory activity of RgA is independent of the endogenous FKBP.

To understand the metabolic impact of GLUT inhibition by RgA, we determined the steady-state levels of 272 metabolites using LC/MS^[15] in MCF-7 cells upon treatment with RgA for 30 min and 6 h, respectively. As shown in Fig. 4a, the most significant metabolic changes caused by RgA are related to glycolysis. Specifically, there were significant decreases in three upper glycolytic intermediates including glucose-6-phosphate (G6P), fructose 1,6-bisphosphate (F1,6-BP) and dihydroxy- acetone phosphate (DHAP), and three key pentose phosphate pathway intermediates including 6-phosphogluconic acid (6PGA), ribose 5-phosphate (R5P), and erythrose-4-phosphate (E4P) (Fig. 4a and 4b). In contrast, the TCA cycle and redox status were not significantly affected by RgA treatment of MCF-7 cells (p > 0.001) (Fig. S8). Together, these results suggested that metabolic effects of RgA were due almost exclusively to the inhibition of glucose uptake.

Figure 4.

RgA inhibits glycolysis pathway. (a) Volcano plots showing metabolite profiles of MCF-7 cells treated with RgA for 30 min or 6 h treatment periods compared to cells treated with vehicle (DMSO). Log2 fold change versus –log10 p value. Dotted lines along x-axis represent ± log2 (2) fold change and dotted line along y-axis represents –log10 (0.05). Metabolites ± log2 (2) fold change shown as red dots with metabolite names denoted. All other metabolites are black dots. (b) Upper glycolysis metabolites and pentose phosphate pathway decrease after 30min or 6h treatments with RgA; P value is from two-sided student t-test. *** means p < 0.0001; ** means p < 0.001.

A major consequence of inhibition of GLUT is the decrease in the level of cellular ATP and the corresponding increase in the AMP/ATP ratio, which was indeed observed upon treatment of MCF-7 cells with RgA (Fig. S8c). The increase in AMP/ATP ratio, in turn, is expected to activate AMPK, leading to the inhibition of the mTOR signaling pathway.^{[3b, 16]} We therefore determined the effect of RgA on AMPK and mTOR activity in MCF-7 cells. RgA activated AMPK and inhibited mTOR activity in both time- and dose-dependent manners (Fig. S9). These results suggested that AMPK was likely to act as the key link between the upper glycolysis inhibition and subsequent mTOR pathway inhibition.

We next determined the effects of RgA on cell growth, survival, and cell death. Cell cycle analysis revealed that treatment of MCF-7 cells with RgA for 24 h led to G1 cell cycle arrest (Fig. S10a). In addition, 24-h treatment with RgA led to activation of both p53 and p21 in a dose-dependent manner (Fig.S10b). Prolonged treatment of MCF-7, which is deficient in caspase 3, with RgA for 72 h resulted in poly(ADP-ribose) polymerase (PARP) and caspase 7 cleavage that was inhibited by cotreatment with the pancaspase inhibitor Z-VAD (Fig. S10c), indicative of apoptosis. We also determined the IC₅₀ values of RgA against several human cancer cell lines using the alamar blue cell proliferation assay. RgA dose-dependently inhibited proliferation of all the cancer cell lines tested, including the lung cancer cell line A549 and two breast cancer cell lines HCC1954 and MCF-7 (Fig. 5a), with IC₅₀ values ranging from 87 to 281 nM (Table S2), validating the antiproliferative activity of RgA.

Figure 5.

Effect of RgA on the growth of human breast cancer xenografts in mice. (a) Inhibition of cell proliferation in A549, HCC1954, and MCF-7 cells by RgA. (b) Analysis of tumor volume index. P value is from two-sided student t-test.

Having demonstrated the anti-proliferative and apoptosis-inducing effects of RgA in vitro, we proceeded to determine whether RgA was capable of blocking tumor xenograft growth in vivo. Given that RgA inhibited multiple isoforms of GLUTs, including GLUT1, 3, and 4, it raised the question of whether animals could tolerate RgA. As breast cancer cell lines are more sensitive to RgA than other cell lines (Fig.5a and Table S2), we assessed the anti-breast cancer activity of RgA in vivo. We selected two breast cancer cell lines for the xenograft experiment–MCF-7, an ER⁺, HER2^- line, and HCC1954, an ER^-, HER2⁺ breast line. NSG mice bearing MCF-7 tumors were given daily vehicle or RgA at a dose of 2 mg/kg for 38 days. Compared to vehicle control group, RgA treatment significantly delayed the xenograft growth of MCF-7 cells (Fig. 5b) with tumor volume indexes on day 38 being 2.8 vs. 1.7 between vehicle- and RgA-treatment groups. In addition, RgA treatment significantly decreased tumor weight from 454 mg (vehicle group) to 285 mg (treatment group) (Fig. S11a). Similarly, daily intraperitoneal injection of RgA at 2 mg/kg also effectively inhibited HCC1954 xenograft growth in nude mice (Fig. S11c) with tumor volume indexes on day 38 of 26.1 vs. 13.0 between vehicle and RgA treatment groups. Importantly, we did not observe any significant weight loss or any signs of adverse effects in animals receiving RgA during the course of the experiments (Fig. S11), suggesting RgA at the efficacious dose was well tolerated in mice.

In conclusion, to facilitate the screening of the rapafucin libraries against new protein targets, we developed a microarray platform by immobilizing rapafucins on a chip surface. Using an optimized 3D microarray with a total of 3918 rapafucins on a single chip, we screened cell lysates containing stably expressed GLUT1. We identified several hits, two of which were confirmed as GLUT1 inhibitor in an orthogonal assay. The most potent inhibitor, named rapaglutin A (RgA), inhibited GLUT1 as well as GLUT3 and GLUT4 with an IC₅₀ value of low nanomolar for GLUT1. We demonstrated that RgA inhibited glycolysis and ATP biogenesis, causing activation of AMPK, inhibition of mTOR, and induction of cell cycle arrest and apoptosis. RgA also inhibited the growth of tumor xenografts of breast cancer cells in vivo without obvious side effects. Using the newly developed 3D rapafucin microarrays, we were able conduct a successful screen against a multi-pass trans-membrane protein target for the first time. It will be interesting to screen the rapafucin microarrays against other types of multi-pass membrane proteins ranging from GPCRs to ion channels.

Experimental Section

Experimental Details are available in Supporting Information section.