FBW7-Dependent Mcl-1 Degradation Mediates the Anticancer Effect of Hsp90 Inhibitors

Heat shock protein 90 (Hsp90) is a molecular chaperone that mediates the stability and function of a variety of client proteins involved in cell proliferation and survival (1). Hsp90 is constitutively overexpressed in a variety types of cancers (2). Among more than 200 Hsp90 client proteins identified, many are oncoproteins associated with hallmarks of cancer (3), such as mutant p53, Raf-1, and Akt (2), suggesting that it represents an attractive target for developing new anticancer therapy.

A number of small-molecule Hsp90 inhibitors have been developed (4, 5). Most of these inhibitors, including the natural product geldanamycin and its derivative 17-allylamino-17-demethoxygeldanamycin (17-AAG; telatinib), bind to the amino-terminal ATP-binding pocket of Hsp90 and disrupt its complex formation with client proteins (2, 6). Inhibitors for targeting the carboxyl-terminal binding sites of Hsp90 have also been identified (7, 8). More than 17 Hsp90 inhibitors have entered phase I/II clinical trials, such as 17-AAG and its analog 17-DMAG (17-dimethylaminoethylamino-17-demethoxygeldanamycin; Alvespimycin) with improved pharmaceutical properties (2, 5). Previous studies showed that geldanamycin and 17-AAG induce apoptosis and inhibit tumor metastasis and angiogenesis (2, 9). Several anticancer mechanisms of Hsp90 inhibitors have been proposed, such as downregulation of Akt (10), inhibition of NF-κB activation due to IKK destabilization (11), endoplasmic reticulum stress (12), and more recently, activation of p53 and its downstream targets (13–15). However, the mechanisms by which Hsp90 inhibitors selectively kill tumor cells are still not well-understood. Biomarkers associated with differential sensitivity and resistance to Hsp90 inhibitors remain to be identified.

Apoptosis in mammalian cells is governed by the Bcl-2 family proteins via a cascade of cellular events, including permeabilization of outer mitochondrial membrane, release of the mitochondrial proteins such as cytochrome c, and activation of caspases (16, 17). Myeloid cell leukemia 1 (Mcl-1) is a prosurvival Bcl-2 family member frequently overexpressed or amplified in various human tumors (18). Mcl-1 suppresses apoptosis by binding to the BH3 domains of proapoptotic proteins through a surface groove containing hydrophobic pockets (16). Distinctive from other Bcl-2 members, Mcl-1 is quite unstable and has a short half-life (19). Defective Mcl-1 degradation allows tumor cells to evade death and become resistant to anticancer therapies. Recent studies identified F-box/WD repeat containing protein 7 (FBW7) as an E3 ubiquitin ligase that targets phosphorylated Mcl-1 through glycogen synthase kinase 3β (GSK3β) for destruction (20, 21). FBW7 (FBXW7; CDC4) is a tumor suppressor frequently mutated in human cancers, including 10% to 20% of colorectal tumors (22, 23). Heterozygous FBW7 missense mutations are often detected in three arginine residues (R465, R479, and R505), which are critical for binding to a conserved CDC4 phosphodegron motif of its substrates (23).

In this study, we identified a critical role of GSK3β/FBW7-mediated Mcl-1 degradation in apoptosis induced by Hsp90 inhibitors in colorectal cancer cells. Importantly, colorectal cancer cells containing FBW7 mutations are insensitive to Hsp90 inhibitors due to blocked Mcl-1 degradation. Our results suggest FBW7 mutational status as a potential biomarker, and the use of Mcl-1–selective inhibitors for overcoming resistance to Hsp90-targeted therapy.

Human colorectal cancer cell lines, including HCT116, DLD1, RKO, LoVo, Lim2405, SW48, and HCT-8, were obtained from the ATCC. Isogenic FBW7-KO HCT116 and DLD1 cell lines were obtained from Horizon Discovery (22). HCT116 cells with knock-in of the Mcl-1 phosphorylation site mutant S121A/E125A/S159A/T163A (Mcl-1-KI) were generated by homologous recombination as described (24). Cells were authenticated by genotyping and analysis of protein expression by Western blotting throughout the study and routinely checked for Mycoplasma contamination by PCR. All cell lines were maintained at 37°C in 5% CO₂ and cultured in McCoy's 5A modified media (Invitrogen) supplemented with 10% defined FBS (HyClone), 100 units/mL penicillin, and 100 μg/mL streptomycin (Invitrogen). For drug treatment, cells were plated in 12-well plates at 20% to 30% density 24 hours before treatment. DMSO (Sigma) stocks of agents, including 17-AAG, 17-DMAG, MG132, TW-37, SB216763 (ref. 25; Selleck Chemicals), 5-fluorouracil (5-FU; Sigma), necrosulfonamide (NSA; Calbiochem), and Mcl-1 inhibitors, including UMI-77 (26), UMI-212 (compound 21), and UMI-36 (compound 36) (27), were diluted to appropriate concentrations with cell culture medium.

Cells seeded in 96-well plates at a density of 1 × 10⁴ cells per well were treated with 17-AAG or 17-DMAG for 72 hours. MTS assay was performed using the MTS Assay Kit (Promega) according to the manufacturer's instructions. Chemiluminescence was measured by a Wallac Victor 1420 Multilabel Counter (Perkin Elmer). Each assay was conducted in triplicate and repeated three times.

Western blotting was performed as previously described (28), with antibodies for phospho-Mcl-1 (Ser159/Thr163), cleaved caspases-3, -8, and -9, ERK, phospho-ERK (Thr202/Tyr204), AKT, phospho-AKT (Ser473), GSK3β, phospho-GSK3β (Ser9) (Cell Signaling), cytochrome oxidase subunit IV (Invitrogen), HA (Santa Cruz), Mcl-1 (BD Biosciences), β-actin, cytochrome c, Flag (Sigma), p62 (Novus), LC3 (MBL), and FBW7 (Abcam).

WT FBW7 expression construct was a gift from Dr. Wenyi Wei at Harvard Medical School (Boston, MA). Mutant FBW7 and HA–ubiquitin expression constructs were previously described (24, 29). Transfection was performed using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. siRNA transfection was done 24 hours before drug treatment using 200 pmol of control scrambled siRNA, human Mcl-1 siRNA (CGCCGAATTCATTAATTTATT-dTdT) (GE Dharmacon) or human GSK3β siRNA (sc-35527; Santa Cruz).

After treatment, cells were harvested and resuspended in 1 mL of EBC buffer (50 mmol/L Tris-HCl, pH 7.5, 100 mmol/L NaCl, 0.5% Nonidet P-40) supplemented with a protease inhibitor cocktail (Roche). Cell suspensions were sonicated and spun at 10,000 × g for 10 minutes to prepare cell lysates. For immunoprecipitation (IP), 1 to 2 μg of IP antibodies were mixed with protein G/A-agarose beads (Invitrogen) for 20 minutes at room temperature. The beads were washed twice with PBS containing 0.02% Tween-20 (pH 7.4), incubated with cell lysates on a rocker for 6 hours at room temperature, and then washed three times with PBS (pH 7.4). Beads were then boiled in 2× Laemmli buffer and subjected to SDS-PAGE and Western blot analysis.

Total RNA was isolated using the Mini RNA Isolation II Kit (ZYMO Research) according to the manufacturer's protocol. One microgram of total RNA was used to generate cDNA by the SuperScript II Reverse Transcriptase (Invitrogen). Real-time PCR was carried out for Mcl-1 using the primer pair 5′-ATGCTTCGGAAACTGGACAT-3′/5′-TGGAAGAACTCCACAAACCCA-3′; and for β-actin using the primer pair 5′-GACCTGACAGACTACCTCAT-3′/5′-AGACAGCACTGTGTTGGCTA-3′ as described (30).

Apoptosis was measured by counting condensed and fragmented nuclei after nuclear staining with Hoechst 33258 (Invitrogen) as previously described (30). At least 300 cells were analyzed for each sample. Colony formation assays were performed by plating treated cells in 12-well plates at appropriate dilutions, followed by crystal violet staining 14 days after plating as described (30). Each experiment was performed in triplicate and repeated at least twice. Cytochrome c release was analyzed by Western blotting of cytoplasmic and mitochondrial fractions isolated from treated cells as described (28).

The described animal experiments were approved by the University of Pittsburgh Institutional Animal Care and Use Committee. Female 5- to 6-week-old nu/nu mice (Charles River) were housed in micro-isolator cages in a sterile environment and allowed access to water and chow ad libitum. Xenograft tumors were established by subcutaneously injecting mice with 4 × 10⁶ of each cell line. After tumor growth for 7 days, mice were treated daily on days 1 to 4 and 7 to 11 with 17-DMAG (15 mg/kg/d) or the control vehicle. Tumor volumes were measured by calipers and calculated according to the formula 0.5 × length × width². After tumors reached 1.0 cm³ in size, mice were euthanized and tumors were dissected and fixed in 10% formalin and embedded in paraffin. TUNEL (Millipore) and active caspase-3 (Cell Signaling) immunostaining was performed on 5-mm paraffin-embedded tumor sections as previously described (28). Signals were detected by Alexa Fluor 488–conjugated (for TUNEL) or Alexa Fluor 594–conjugated (for active caspase-3) secondary antibodies (Invitrogen) with nuclear counter staining by DAPI.

Statistical analysis was carried out using GraphPad Prism IV software. P values were calculated by the Student t test and were considered significant if P < 0.05. Means ± one SD are displayed in the figures.

Upon analyzing colorectal cancer cell lines with different mutations in common tumor suppressors and oncogenes, we identified a correlation between 17-AAG sensitivity and FBW7 mutational status (Supplementary Table S1). FBW7-mutant colorectal cancer cell lines, including SW48, HCT-8, and LoVo, were substantially less sensitive to 17-AAG, compared with FBW7 wild-type (WT) cell lines, including RKO, HCT116, Lim2405 (Fig. 1A). The IC₅₀ values of 17-AAG ranged from 1.28 to 3.07 μmol/L in the FBW7-mutant cell lines, compared with 0.30 to 0.46 μmol/L in the WT cell lines (Supplementary Table S1). Compared with WT cells, FBW7-mutant cells had lower levels of 17-AAG–induced apoptosis determined by nuclear fragmentation (Fig. 1B). Analysis of the expression of Bcl-2 family proteins revealed marked depletion of the antiapoptotic Mcl-1 in WT cells, but not in FBW7-mutant cells (Fig. 1C and D). In contrast, proapoptotic members, including PUMA, Bax, and Bim, were similarly upregulated in FBW7 WT and FBW7-mutant colorectal cancer cells (ref. 15; Supplementary Fig. S1). These results demonstrate that FBW7 mutations are associated with blocked Mcl-1 depletion and apoptosis in response to Hsp90 inhibition in colorectal cancer cells.

To investigate the role of FBW7 in regulating sensitivity to Hsp90 inhibitors, we analyzed isogenic HCT116 cells with knockout (KO) of FBW7 (FBW7-KO; ref. 22). Compared with the parental cells, FBW7-KO HCT116 cells were substantially more resistant to apoptosis induced by 17-AAG (Fig. 2A) and were deficient in 17-AAG–induced Mcl-1 degradation (Fig. 2B). Similar results were obtained with another Hsp90 inhibitor, 17-DMAG (Fig. 2A and B). Reconstituting WT FBW7 expression in FBW7-KO cells by transient transfection restored 17-AAG sensitivity, as well as Mcl-1 degradation (Fig. 2C). Knocking down Mcl-1 by siRNA also restored 17-AAG sensitivity in FBW7-KO cells (Fig. 2C). Similar observations were made using WT and FBW7-KO DLD1 cells (ref. 22; Supplementary Fig. S2A–S2C). Furthermore, transfection of WT FBW7, which did not affect the expression of mutant FBW7 (Supplementary Fig. S2D), and Mcl-1 knockdown also restored 17-AAG sensitivity and apoptosis induction in FBW7-mutant SW48 cells (Fig. 2D). In contrast to WT FBW7, transfection with tumor-derived FBW7 mutants, including R465C, R479Q, and R505C, failed to restore 17-AAG sensitivity and Mcl-1 degradation in SW48 and FBW7-KO HCT116 and DLD1 cells (Fig. 2E; Supplementary Fig. S2E and S2F). These results demonstrate that the functional FBW7 is required for 17-AAG sensitivity and Mcl-1 degradation in colorectal cancer cells and that FBW7-inactivating mutations abrogate Mcl-1 degradation and apoptosis in response to Hsp90 inhibition.

We investigated the mechanism by which Hsp90 inhibitors induce FBW7-dependent Mcl-1 depletion. Treating the sensitive HCT116 cells with 17-AAG or 17-DMAG caused Mcl-1 depletion in a time- and dose-dependent manner (Fig. 1C; Supplementary Fig. S3A and S3B) and did not significantly affect the mRNA level of Mcl-1 (Supplementary Fig. S3C). 17-AAG–induced Mcl-1 depletion was blocked by the proteasome inhibitor MG132 (Fig. 3A), suggesting ubiquitin/proteasome-dependent protein degradation. In 17-AAG–treated cells, we detected phosphorylation of Mcl-1 at Ser159/Thr163 (Fig. 3B), binding of phosphorylated Mcl-1 (p-Mcl-1) to FBW7 (Fig. 3C), and ubiquitination of Mcl-1 (Fig. 3D), which was dependent on FBW7 and completely blocked in FBW7-KO cells (Fig. 3D).

GSK3β was previously shown to phosphorylate Mcl-1 to promote its degradation (31) and is inhibited by ERK through Ser9 phosphorylation (32). Analysis of upstream kinase signaling revealed that 17-AAG prevented Ser9 phosphorylation of GSK3β and inhibited ERK activity (Fig. 3B). Treating cells with the GSK3 inhibitor SB216763 (25) suppressed 17-AAG–induced Mcl-1 phosphorylation, binding to FBW7, ubiquitination, and degradation (Fig. 3E and F), which was verified by knocking down GSK3β using siRNA (Supplementary Fig. S3D). These results suggest that Hsp90 inhibitors promote GSK3β-dependent Mcl-1 phosphorylation, which facilitates its binding to FBW7 and subsequent ubiquitination and proteasomal degradation.

To determine the functional role of Mcl-1 degradation in apoptotic response to Hsp90 inhibitors, we sought to specifically inhibit Mcl-1 phosphorylation and degradation by a genetic approach. A knock-in (KI) strategy was used to target the phosphorylation sites of Mcl-1 involved in its degradation, including S121, E125, S159, and T163 (Fig. 4A), which have been shown to be phosphorylated by GSK3β and other kinases in response to therapeutic agents and other stresses (20, 21). HCT116 cell lines with KI of Mcl-1 phosphorylation site mutant (Mcl-1-KI) were identified (Fig. 4A; ref. 24). In stark contrast to WT cells, Mcl-1-KI cells were deficient in 17-AAG- and 17-DMAG–induced Mcl-1 degradation (Fig. 4B) and completely lacked Mcl-1 phosphorylation, binding to FBW7, and ubiquitination (Fig. 4C and D).

Mcl-1-KI cells were markedly resistant to 17-AAG and 17-DMAG, showing significantly increased viability (Fig. 5A), reduced apoptosis (Fig. 5B), and enhanced clonogenic survival (Fig. 5C), upon 17-AAG or 17-DMAG treatment. Mcl-1-KI and GSK3β-knockdown cells had reduced activation of caspases-9 and -3 (Fig. 5D; Supplementary Fig. S3D) and cytosolic release of cytochrome c (Fig. 5E), hallmarks of mitochondria-mediated apoptosis (33). The effect is Mcl-1–dependent, as knockdown of Mcl-1 restored apoptosis induced by 17-AAG and 17-DMAG in Mcl-1-KI cells (Fig. 5B). In contrast to apoptosis inhibition, necroptosis and autophagy induced by anticancer agents, such as 5-FU, were not affected in Mcl-1-KI cells (Supplementary Fig. S4A and S4B). Furthermore, treating cells with an Mcl-1–selective small-molecule inhibitor and analogs, including UMI-77 (26), UMI-212 (compound 21), and UMI-36 (compound 36; ref. 27), or the pan-Bcl-2 inhibitor TW-37 with strong Mcl-1 inhibitory activity (34), restored 17-AAG- and 17-DMAG–induced apoptosis in both FBW7-KO and Mcl-1-KI cells (Fig. 5F), suggesting that inhibiting Mcl-1 overcomes resistance to Hsp90 inhibitors in colorectal cancer cells.

We then used xenograft model to validate the role of FBW7 and Mcl-1 phosphorylation and degradation in Hsp90-targeted therapy. Xenograft tumors were established by subcutaneously injecting nude mice with WT, Mcl-1-KI, and FBW7-KO HCT116 cells. After tumor establishment, tumor-bearing mice were intraperitoneally injected with 15 mg/kg of water-soluble 17-DMAG or the control vehicle, followed by analysis of tumor size every 2 days for 3 weeks. WT tumors responded well to 17-DMAG treatment and were only about one fourth of the size of vehicle-treated tumors on day 21 (Fig. 6A and B). In contrast, Mcl-1-KI and FBW7-KO HCT116 tumors were significantly more resistant to 17-DMAG treatment (Fig. 6A and B). 17-DMAG–treated Mcl-1-KI and FBW7-KO tumors showed blocked Mcl-1 depletion but with intact GSK3β dephosphorylation (Fig. 6C). Apoptosis was significantly reduced in the Mcl-1-KI and FBW7-KO tumors compared with WT tumors as shown by TUNEL and active caspase-3 staining (Fig. 6D and E). Together, these findings demonstrate a critical role of FBW7 in mediating the in vivo antitumor effects of Hsp90 inhibition via the degradation of phosphorylated Mcl-1.

Inhibition of Hsp90, which buffers various stresses incurred by oncogenic transformation, promotes degradation of oncogenic client proteins and death of tumor cells (2, 3). Our results demonstrate for the first time that the apoptotic response to Hsp90 inhibitors is mediated by GSK3β/FBW7-dependent Mcl-1 degradation both in vitro and in vivo. Results from isogenic knock-in cells indicate that the four phosphorylation sites (S121/E125/S159/T163) are essential for Mcl-1 ubiquitination and degradation induced by Hsp90 inhibitors. Phosphorylation at these sites by GSK3β engages Mcl-1 to bind to FBW7, which likely recruits Mcl-1 to the SCF ubiquitin ligase complex composing of FBW7, CUL1, SKP1, and ROC1 (21). This complex then covalently links ubiquitin chains to Mcl-1, leading to its degradation in the 26S proteasome. In addition to GSK3β, several other kinases have also been implicated in regulating Mcl-1 turnover, including p38, JNK, CDK1, and casein kinase II (20, 21, 31). Mcl-1 stability can also be regulated by other E3 ubiquitin ligases such as Mule and β-TrCP (35, 36) and by the deubiquitinase USP9X (37). Depending on cell types, these proteins may also be involved in modulating sensitivity to Hsp90 inhibitors through their effect on Mcl-1 stability. Furthermore, siRNA screen identified the E3 ubiquitin ligase Cullin-5 as a mediator of response to Hsp90 inhibitors (38).

Unlike in hematopoietic cells, which are dependent on Mcl-1 for survival, depletion of Mcl-1 alone in colorectal cancer and most solid tumor cells is not sufficient for inducing cell death. Our recent study showed that the induction of the proapoptotic Bcl-2 family member PUMA by p53 is also a key event in apoptotic response to Hsp90 inhibitors in colorectal cancer cells (15). Mcl-1 binds to and sequesters BH3-only Bcl-2 family proteins to suppress apoptosis (18). It is possible that Mcl-1 degradation induced by Hsp90 inhibitors primarily exerts an effect on PUMA by reliving its full proapoptotic activity. In Mcl-1-KI and FBW7-mutant colorectal cancer cells with defective Mcl-1 degradation, Mcl-1 may maintain cell survival by trapping PUMA and preventing its binding to other prosurvival factors such as Bcl-X_L to activate caspases. Our in vivo data showed that tumor cells with or without deficiency in FBW7 and Mcl-1 degradation still grew, albeit at a slower rate, in response to 17-DMAG treatment, and tumor growth became accelerated following the last dose of treatment (Fig. 6A). These observations suggest that in addition to apoptosis, growth inhibition is also involved in the antitumor effect of Hsp90 inhibition, which may be explained by downregulation of the Hsp90 client proteins involved in regulating cell growth, such as Raf-1 and Akt (2).

Clinical applications of Hsp90 inhibitors have been hampered by lack of patient stratification due to lack of biomarkers of sensitivity and resistance (2, 5). Our findings elucidate a critical functional role of FBW7 mutations and Mcl-1 stability in differential sensitivity and resistance of colorectal cancer cells to Hsp90 inhibition. Most of the FBW7 mutations in colorectal cancers are heterozygous point mutations (29). Mutant FBW7 may have altered protein stability or act as dominant-negative proteins upon heterodimerization with WT FBW7 (23). FBW7 mutations are found in various human cancers and likely play a broad functional role in intrinsic and acquired therapeutic resistance of cancer cells (23). These mutations affect responses to γ-secretase inhibitors in leukemia cells (39), to HDAC inhibitors in squamous tumor cells (40), and to antimitotic drugs in colorectal cancer cells (21). Several other factors have also been shown to modulate sensitivity to Hsp90 inhibitors (41). For example, improved response rates were observed in patients with HER2-positive metastatic breast cancer treated with 17-AAG and trastuzumab (42) and in patients with ALK-rearranged non–small cell lung cancer treated with the Hsp90 inhibitor IPI-504 (43). These effects were attributed to the degradation of the Hsp90 clients, which are driver oncoproteins in these tumor types. Studies by us and other groups showed that p53 mutational status influences 17-AAG sensitivity through the effects on p53 downstream targets including PUMA, Bax, Bim, and p21 (13–15). Our recent study showed that colon cancer cells contain a small fraction (∼0.1%) of pre-existing FWB7-mutant cells, which can be enriched upon treatment with the multi-kinase inhibitor regorafenib, leading to acquired drug resistance (29). It is possible that these cells can also escape from death induced by Hsp90 inhibitors and cause acquired resistance to these agents.

Our data suggest that Mcl-1–binding BH3 mimetics are potentially useful for overcoming resistance to Hsp90 inhibitors caused by FBW7 mutations and/or Mcl-1 stabilization in colorectal cancer cells. We recently identified a critical role of Mcl-1 and FBW7 mutations in intrinsic and acquired resistance of colorectal cancers to the FDA-approved multi-kinase inhibitor regorafenib (24, 29). Aberrant Mcl-1 expression is frequently detected in colorectal cancers and significantly correlated with advanced tumor stages, lymph node metastasis, resistance to chemotherapy, and poor patient survival (44, 45). These findings suggest that Mcl-1 is an attractive target for developing new anticancer agents to improve the efficacy of targeted therapies. Inhibiting prosurvival Bcl-2 family proteins in tumor cells has emerged as an attractive therapeutic strategy, highlighted by the recent FDA approval of the Bcl-2-selective inhibitor ABT-199 (Venetoclax) for the treatment of chronic lymphocytic leukemia (46). Several different chemical classes of Mcl-1 inhibitors have been described (47, 48). The potential use of these Mcl-1 inhibitors for improving the therapeutic effects of Hsp90-targeted agents needs to be further established.

In conclusion, we demonstrate that GSK3β/FBW7-mediated Mcl-1 degradation is critical for apoptosis induced by Hsp90 inhibitors in colorectal cancer cells. Our findings suggest FBW7 mutational status and Mcl-1 stability as key determinants of response to Hsp90 inhibitors, which provides a rationale for using FBW7 genotype for potential patient stratification, and for drug combinations with Hsp90 inhibitors that can effectively overcome Mcl-1–mediated resistance.

No potential conflicts of interest were disclosed.

Conception and design: J. Tong, F. Zou, L. Zhang

Development of methodology: J. Tong, J. Yu, L. Zhang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Tan, J. Yu, L. Zhang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Tong, S. Tan, L. Zhang

Writing, review, and/or revision of the manuscript: J. Tong, S. Tan, Z. Nikolovska-Coleska, J. Yu, L. Zhang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Zou, L. Zhang

Study supervision: L. Zhang

Other (provided key reagents): Z. Nikolovska-Coleska

We thank Ms. Xiao Tan for technical assistance and our lab members for critical reading and discussion.

This study was supported by U.S. NIH grants R01CA172136, R01CA203028 to L. Zhang; U01DK085570, U19AI068021 to J. Yu; R01CA149442 to Z. Nikolovska-Coleska; P30CA047904 to University of Pittsburgh Cancer Institute; and National Natural Science Foundation of China grant 81672942 to F. Zou.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.