Development of a RSK Inhibitor as a Novel Therapy for Triple-Negative Breast Cancer

Metastatic breast cancer remains incurable, with therapy limited to slowing disease progression (1). In particular, triple- negative breast cancer (TNBC) patients have increased probability of death due to metastasis compared with other breast cancer subtypes (2). TNBC is characterized by its lack of currently available targeted markers (3). However, the MEK-ERK1/2 cascade is now considered as a viable drug target for TNBC (4–7). In genetic analysis of basal-like breast cancers, which includes ∼70% of TNBCs, activated MEK-ERK1/2 signaling is thought to occur in ∼80% of the tumors (4, 8, 9). In addition, numerous TNBC cell lines possess an activated RAS-transcriptional program and enhanced sensitivity to MEK inhibition (10, 11). In support of these preclinical observations, a complete response was observed in a phase Ib trial using a combination of trametinib, a MEK inhibitor, and gemcitabine, a nucleotide analogue, in a TNBC patient who had failed multiple therapies (5). Based on these data various MEK inhibitors are being tested in clinical trials, which include TNBC patients (12).

However, treating patients with drugs that inhibit “global regulators” such as MEK causes a number of side effects that result in limited efficacy (12). We postulate that inhibiting downstream effectors of MEK like the Ser/Thr protein kinase, p90RSK (RSK), will have fewer side effects because it controls a more limited set of targets. RSK phosphorylates various substrates that control diverse cellular processes, including metastasis (13–18). Approximately, 85% of TNBC patient samples have activated RSK, which is identified by the presence of phosphorylated residues critical for its activity (19). Taken together, these observations suggest that RSK is a viable target for TNBC.

RSK contains two nonidentical functional kinase domains referred to as the N-terminal (NTKD) and C-terminal (CTKD) (13). The CTKD functions to regulate RSK activation, whereas the NTKD, which belongs to the AGC kinase family, is responsible for substrate phosphorylation (13). In a screen of botanical extracts we identified the first RSK inhibitor, SL0101 (1a), which was isolated from Forsteronia refracta (20). SL0101 is an extremely specific allosteric inhibitor for the NTKD (14, 20–22).

In addition to SL0101, other RSK inhibitors have been described. However, the currently available NTKD inhibitors are not RSK specific (21, 23–26) or demonstrate poor pharmacokinetics (27, 28). Covalent inhibitors of the RSK CTKD (29–31), targeting autoactivation, are also available and have limited off-target effects. However, CTKD inhibitors do not inhibit an activated kinase and the autoactivation mechanism can be bypassed (29), suggesting that the clinical utility of CTKD inhibitors is limited.

Because of the selectivity of SL0101 for RSK we continue to improve its drug-like properties through extensive structure–activity–relationship (SAR) analysis. We have now identified a SL0101 analogue, C5″-n-propyl cyclitol SL0101 (1b), which retains specificity for RSK1/2 and is more potent in in vitro and cell-based assays than the parent compound. This improved analogue inhibits proliferation, survival in a nonadherent environment, and migration of TNBC lines but, unlike MEK inhibitors, does not activate the AKT pathway. Inhibition of RSK1/2 using (1b) or silencing RSK1 or RSK2 inhibited TNBC metastatic colonization in vivo. Moreover, (1b) was as effective as the FDA-approved MEK inhibitor, trametinib. Taken together, these results indicate that RSK1/2 are viable drug targets for TNBC metastasis.

Animal procedures had approval of the Vanderbilt University Institutional Animal Care and Use Committee. For in vivo metastatic models, NOD-SCID-IL2Rγ_c (NSG) mice (6–8 weeks; Jackson Laboratory) were injected in the left cardiac ventricle with 1 × 10⁵ cells/100 μL PBS. Mice injected with MCF-7 cells received a 17β-estradiol pellet (0.36-mg 60-day release; Innovative Research of America). Mice bearing MCF-7 metastasis were injected intraperitoneally with vehicle [10% (2-hydroxypropyl)-β-cyclodextrin (HPBCD) in 10% DMSO] or (1b) (40 mg/kg) 2 hours prior to euthanasia (day 50). Mice injected with HDQ-P1-Luc were randomized and at 2 hours after injection were treated for 5 days with HPBCD, (1b) (40 mg/kg) intraperitoneally Q12 hours or trametinib (2 mg/kg; Santa Cruz Biotechnology, Inc.) intraperitoneally Q24 hours. For bioluminescence imaging, mice were injected intraperitoneally with RediJect d-Luciferin (1.5 mg; PerkinElmer, Inc.) and imaged with a Xenogen IVIS using Living Image acquisition software (Xenogen Corp.). For ex vivo imaging, organs were placed in d-Luciferin (150 μg/mL PBS). After imaging, tissue was fixed in 4% buffered formalin and paraffin-embedded.

Cell lines were obtained and cultured as directed by ATCC or by German Collection of Microorganisms and Cell Culture. Stocks were prepared within one to two passages after receipt and new stocks thawed frequently and passaged <6 months. Authentication was based on growth rate, morphology, and absence of mycoplasma. Serum-starved cells were pretreated for 2 hours with vehicle or inhibitor. MCF-7 cells were treated with phorbol 12-myristate 13-acetate (PMA; Sigma) for 20 minutes. Cells were lysed as described previously (32).

For motility assays 2.5 × 10⁵ cells were plated on fibronectin-coated (5 μg/mL; Corning) 2-chamber Lab Teks (Thermo Fisher Scientific). After 48 hours, cells were pretreated with vehicle or inhibitors for 2 h and scratched with a P200 pipette tip. After washing, HEPES (50 mmol/L; Thermo Fisher Scientific)-buffered media with vehicle or inhibitor was added and images taken every 20 minutes using Nikon Eclipse Ti microscope and an Orca R2 digital CCD camera (Hamamatsu). Migration velocity was quantified using Volocity software (PerkinElmer, Inc.). Additional details are in Supplementary Data.

For two-dimensional (2D) proliferation assays, 2 × 10⁵ cells/well in 24-well or 10³ cells/well in 96-well were seeded. For 3D proliferation, 1.5 × 10³ cells/well in 96-well were plated in 2% matrigel (MG; Corning, Inc.) onto 100% MG. Inhibitor or vehicle was added and proliferation was measured at 48 to 72 hours using CellTiterGlo reagent (Promega Corp.) with a GloMax Discover luminometer (Promega Corp.).

For survival assays, cells were seeded at 1.5 × 10³ cells/well in 96-well poly-HEMA–coated plates (Corning, Inc.) and vehicle or inhibitors added and bioluminescence measured at 48 to 72 hours.

The IC₅₀ values for proliferation and survival were determined using nonlinear regression analysis (GraphPad Prism version 6.0a).

Section preparation and antibodies are listed in Supplementary Data. Fluorescent images were obtained with a laser-scanning microscope (510/Meta/FCS Carl Zeiss, Inc.). Objectives were: mouse tissue 40× Plan-Neofluar oil NA 1.3 (zoom 0.7×); human tissue 20× NA 0.8. Images were acquired using LSM-FCS software (Carl Zeiss, Inc.), quantitated using Openlab 5.5.0 (PerkinElmer, Inc.), and processed in Photoshop version CS6 version 13.0 (Adobe).

RSK2 kinase assays performed as described previously (32). The kinase screen was performed using the ZLYTE screen (Thermo Fisher Scientific, Inc.).

Statistical analyses (GraphPad Prism 6.0a) using the Mann–Whitney test (two-sided) unless indicated. *P < 0.05 was statistically significant.

In prior SAR studies of the flavonoid glycoside, SL0101 (1a), we determined that replacement of the C5-methyl group on the pyranose with an n-propyl moiety (1c), improved the IC₅₀ by >25-fold but that the compound had limited aqueous solubility (32). In addition, we determined that exchanging the rhamnose with a cyclitol (1d), improved the cell-based efficacy for inhibition of proliferation but this compound was not RSK specific (33). We hypothesized that combining the modifications would improve the potency for RSK inhibition while maintaining specificity for RSK. Consistent with our hypothesis C5″-n-propyl cyclitol SL0101 (1b) has a six-fold improved IC₅₀ in an in vitro kinase assay for RSK inhibition compared to SL0101 (1a; Fig. 1A). Furthermore, (1b) inhibited the proliferation in 2D culture of the estrogen receptor α–positive (ER⁺) breast cancer line, MCF-7, with an IC₅₀ of approximately 8 μmol/L versus approximately 50 μmol/L for SL0101 (Fig. 1B; ref. 20). Previously, we found that the proliferation of the immortalized but untransformed breast line, MCF-10A, is less dependent on RSK for proliferation than the MCF-7 line (33). Consistent with these observations, only a slight decrease in MCF-10A proliferation occurred at the highest concentrations of (1b) (Fig. 1B). The efficacy of SL0101 diminishes after >48 hours in in vitro culture (20). One advantage in replacing the rhamnose with a cyclitol moiety is that the cyclitol should be resistant to acid catalyzed anomeric bond hydrolysis, which should increase stability. To test this possibility we incubated MCF-7 cells with (1b) (25 μmol/L) for varying lengths of time. In agreement with our rationale, only a minor increase in proliferation over a 96-hour time course was observed when MCF-7 cells were incubated with (1b) (Fig. 1C). In contrast, there was a 100% increase in proliferation from 48 to 96 hours in the presence of SL0101 (100 μmol/L). These data indicate that the modifications to generate the SL0101 analogue (1b) resulted in a more potent RSK inhibitor than the parent compound.

SL0101 (1a) is highly selective for RSK (14, 21), which is most likely due to the fact that SL0101 inhibits RSK by an allosteric mechanism (22). Therefore, to evaluate the specificity of (1b), we compared its ability to inhibit RSK substrates in comparison to SL0101 (1a). In agreement with previous results, SL0101 induces an increase in the phosphorylation of eukaryotic elongation factor 2 (p-eEF2) in MCF-7 cells, which also occurred with (1b) (Fig. 1D). This increase is due to the activation of eEF2 kinase, which is inhibited by RSK (34). Furthermore, both RSK inhibitors decreased the phosphorylation of Ser167-ERα, an important marker for anti-estrogen responsiveness (35). SL0101 and (1b) also decreased the phosphorylation of the ribosomal protein, S6 (pS6), a known RSK downstream effector (Fig. 1E) (36). Previously, we identified that silencing RSK2 reduced cyclin D1 levels (37), and consistent with these results RSK inhibition decreased cyclin D1 levels. In a more global analysis, in vitro kinase assays were performed against a panel of 247 purified kinases, which contained representatives from all kinase families (Supplementary Fig. S1). At 10 μmol/L of (1b), RSK1 and RSK2 were the top hits, with colony stimulating factor 1 receptor (CSF1R) and mitogen-activated protein kinase kinase kinase kinase (MAP4K4) being inhibited by approximately 37% compared with RSK2 (Fig. 1F). CSF1R regulates macrophage function, and inhibitors are currently in development as cancer therapies (38). MAP4K4 is an endothelial protein kinase, and inhibitors are being developed as antidiabetic drugs (39). Thus, the off-target effects of (1b) are very limited. Neither of these off-target effects is viewed as problematic for further drug development. Taken together, these data demonstrate that (1b) is very specific for RSK1/2.

The overall goal of our studies is to develop a RSK inhibitor for in vivo use. To evaluate the ability of (1b) to inhibit RSK1/2 in vivo, we used an MCF-7 metastatic model because most of our prior characterization of SL0101 was performed using this line. MCF-7 cells that stably express luciferase (MCF-7-Luc) were introduced by intracardiac (IC) injection into NSG mice, and metastasis were established for ∼50 days. Before treatment we determined that the tumor burden between animals was equivalent (Fig. 1G). Two hours after treatment with (1b) or vehicle, the animals were euthanized and the tibia isolated, as ER⁺ tumors frequently metastasize to the bone. Moreover, MCF-7 cells within the tibia were easily identified by their positive staining with cytokeratin 8 (K8) (Fig. 1H). The levels of the RSK target, pS6, were decreased by >2.5-fold with (1b). These results demonstrate that (1b) is able to attain a sufficient concentration to induce pharmacodynamic changes in vivo.

RSK has been proposed as a drug target for TNBC based on observations that ∼85% of TNBC tumors have activated RSK (19). In agreement with these observations we found that the levels of activated RSK (pRSK) were higher in TNBC tumors than normal tissue (Fig. 2A and B and Supplementary Table S1). The levels of activated RSK varied considerably within and between tumors. Moreover, in TNBC tumor tissue-activated RSK could be present in the nucleus, cytoplasm, or both whereas it was mainly cytoplasmic in normal breast cells. The differences in subcellular localization suggest that the substrates regulated by RSK differ between normal and TNBC tissue. Taken together, these results are consistent with RSK as a viable drug target for TNBC.

To investigate whether activated RSK was functionally important in TNBC, we chose a panel of eight cell lines representing five different TNBC subtypes (40). We observed that activated RSK was present at different levels in these lines (Fig. 2C and D). In 2D culture, the proliferation of all the TNBC lines was inhibited at a lower concentration of (1b) than SL0101 (1a) (Supplementary Fig. S2A). The lines from the mesenchymal subtype, CAL-120 and MDA-MB-231, were relatively resistant whereas the basal-like 2 (BL2), HDQ-P1, and HCC70 were among the most sensitive (Fig. 3A). The BL2 lines are of interest clinically because this subtype is correlated with the poorest response to neoadjuvant chemotherapy (41). To better understand these observations, we compared the levels of activated RSK normalized to total RSK1 and RSK2, which should reflect RSK1/2 specific activity (Fig. 2E). The anti-RSK2 antibody is less sensitive than the anti-RSK1 antibody and this difference was accounted for by normalizing to recombinant proteins. We observed an inverse relationship between the IC₅₀ for (1b) and the specific activity of the combined isoforms (Supplementary Fig. S2B), consistent with the hypothesis that higher RSK specific activity increases sensitivity to the inhibitor. In a separate analysis, active RSK was normalized to RSK2 or RSK1 separately and a statistically significant inverse correlation was observed for RSK2 (Fig. 3B).

To evaluate specificity we investigated the efficacy of (1b) in the context of RSK1/2 silencing. As expected, loss of RSK1/2 decreased 2D proliferation by approximately 60% in MDA-MB-231 cells (Fig. 3C and D). Importantly, silencing RSK1/2 resulted in loss of sensitivity to (1b) (Fig. 3D). These results support the conclusion that (1b) is specific for RSK1/2 and also demonstrate that RSK1/2 are primarily responsible for regulating the proliferation of MDA-MB-231 cells.

Proliferation of the MDA-MB-231 line is reported to be more sensitive to RSK inhibition in 3D versus 2D (27). In agreement, we observed that the IC₅₀ for (1b) is approximately 8 μmol/L in 3D and approximately 50 μmol/L in 2D (Fig. 3E). Surprisingly, HDQ-P1 and HCC70 were unable to proliferate in 3D, suggesting that these lines have more stringent requirements for proliferation than MDA-MB-231.

The ability of cancer cells to survive in circulation is an important step in metastasis (42) and therefore, we analyzed survival in ultra-low adhesion plates. Survival of HDQ-P1, HCC70, and MDA-MB-231 was dependent on RSK and the IC₅₀ for inhibition of survival by (1b) was approximately 30, 15, and 3 μmol/L, respectively (Fig. 3F). Silencing RSK1/2 in MDA-MB-231 cells decreased survival by approximately 75% and was not further inhibited by (1b) (Fig. 3G). These results demonstrate that the survival of some TNBC lines depends on RSK and confirm that (1b) is a very specific RSK inhibitor.

RSK has been implicated in regulating motility (14) and we investigated this possibility using the scratch assay. In all lines tested, (1b) reduced cell velocity to the same extent as SL0101 but at lower concentrations (Fig. 3H–J and Supplementary Fig. S3A–S3C and Supplementary Movies S1–S3). The motility of HCC70 was reduced by approximately 50%, and in MDA-MB-231 and HDQ-P1 cells motility was decreased by at least 75%. Apoptosis was not detected with the doses and time course used in the scratch assay (Supplementary Fig. S3D). Taken together, our results demonstrate that inhibition of RSK by (1b) reduces proliferation, survival in a nonadherent environment and motility, which are essential components of the metastatic process.

To identify the contributions of RSK1 and RSK2 to metastasis, we used an in vivo metastatic MDA-MB-231 model in which luciferase was stably expressed (MDA-MB-231-Luc). MDA-MB-231-Luc cells were transduced with control, RSK1- or RSK2-specific shRNAs (Supplementary Fig. S4A). The cells were quality controlled for their luciferase signal, and equal numbers of cells were introduced by IC injection into female NSG mice (Supplementary Fig. S4B). This model will identify whether RSK1 or RSK2 contribute to the metastatic processes that includes metastatic colonization and proliferation at the metastatic site. At day 19 silencing RSK1 or RSK2 decreased the total metastatic burden, as determined by bioluminescence, by more than three-fold (Fig. 4A and B and Supplementary Fig. S4C). This decrease in metastatic burden is further supported by the observations that silencing RSK1 or RSK2 increased survival by approximately 40% to 60% (Fig. 4C). Silencing RSK1 or RSK2 reduced the number of metastatic foci by nearly half (Fig. 4D), and remained constant over the duration of the experiment. The number of bioluminescent foci was linearly correlated with the number of metastatic foci as determined by histology (Supplementary Fig. S4D). Therefore, we conclude that the increased whole animal bioluminescence from day 5 onwards reflects proliferation at the metastatic sites (Fig. 4E). Thus, silencing RSK1 or RSK2 decreased proliferation from day 12 to day 19 more than three-fold compared to the control. We also conclude that the decrease in metastatic foci reflects that RSK1 or RSK2 is necessary for metastatic colonization. This decrease in metastatic foci was not organ dependent (Fig. 4F). Ex vivo analysis of bioluminescence was also performed as it improved the resolution for determining individual metastatic foci. These results of the ex vivo bioluminescence (Fig. 4G and H) and the histologic analysis (Supplementary Fig. S4E and S4F) were consistent. We conclude that RSK1/2 regulate numerous steps that comprise the metastatic process, which results in improved survival.

We investigated whether (1b) would be sufficiently efficacious to decrease metastatic colonization in vivo. For these experiments we used the HDQ-P1 model because of the clinical importance of the BL2 subtype. This model has not previously been used as an in vivo metastatic model. To validate the model, HDQ-P1 cells were transduced with luciferase (HDQ-P1-Luc) and introduced by IC injection into male NSG mice. At 24 hours after injection, the cells were widespread through the animal, but by day 5 the cells were primarily localized to the liver, adrenal glands, and testes (Supplementary Fig. S5). This model is in contrast to the widely used MDA-MB-231 IC metastatic model, in which the cells primarily metastasize to the bone. TNBC primarily metastasizes to lymph nodes and viscera more often than to the bone, and the HDQ-P1 model better recapitulates these clinical observations.

To evaluate the efficacy of (1b), we compared it to a drug that is in the same class as (1b) and therefore, would be expected to generate a similar phenotype. The MEK inhibitor, trametinib, is approved for melanoma and is currently in multiple clinical trials including those for breast cancer (43). MEK inhibition will decrease ERK1/2 activity and reduce RSK activation (13). HDQ-P1-Luc cells were introduced by IC injection into female NSG mice and treatment began 2 hours after injection. The animals were imaged just before treatment to ensure viability and distribution of the cells in vivo. This approach recapitulates the clinical scenario of tumor cells within the circulation, which have been proposed to act as a negative prognostic marker and demonstrate similar therapeutic responsiveness as the metastatic tumor (1). By 24 hours both (1b) and trametinib decreased the total in vivo bioluminescence by three-fold (Fig. 5A and B). Moreover, the number of metastatic foci in both the skeleton and the viscera was reduced by drug treatment (Fig. 5C). Treatments were stopped on day 5 and on day 6 the total in vivo bioluminescence was reduced three-fold by drug treatment in comparison to the control (Fig. 5D). To confirm these findings, we measured the bioluminescence of the liver and adrenal glands ex vivo and observed a five-fold reduction in metastatic burden in mice treated with either drug (Fig. 5E–H). We conclude that inhibition of RSK or its upstream activator, MEK, decreases metastatic colonization. Moreover, these observations with HDQ-P1 confirm those obtained with MDA-MB-231.

Inhibiting “global regulators” such as MEK results in a number of side effects and their ability to induce an effective clinical response appears limited (12). MEK inhibition can result in activation of AKT (44) and based on these results there are clinical trials underway combining MEK inhibitors with an AKT or PI3K inhibitor (45). Consistent with the literature, we observed that treatment of MDA-MB-231 cells with trametinib enhanced the levels of phosphoSer 473 AKT (pAKT), which is necessary for AKT activity (Fig. 6A). In contrast, activation of AKT was not observed in response to (1b). In comparison to MDA-MB-231, HDQ-P1 have high basal levels of active AKT but consistent with the results observed in MDA-MB-231, (1b) did not increase AKT activity in contrast to trametinib (Fig. 6B). Taken together, these results indicate that RSK inhibition by itself will effectively target the TNBC metastatic process but not have the undesirable side effect of activating AKT.

The importance of RSK in regulating metastasis in vivo has not been thoroughly investigated. Kang and colleagues (46) reported that silencing RSK2 decreased metastatic colonization to the lymph nodes using a human head and neck squamous cell carcinoma line. They further followed up on these observations using the RSK CTKD inhibitor, FMK-MEA, which resulted in a modest decrease in metastatic tumor burden from 97% to 79% (47). In a screen Lara and colleagues (48) identified that loss of RSK1 increased motility in lung cancer lines but in contrast Zhou and colleagues (49) found that inhibition of RSK activity was associated with decreased motility in lung cancer lines. It is possible that the discrepancy between these studies results from the ability of RSK1 to act as a scaffold and regulate other signaling pathways. RSK has also been proposed as a drug target for TNBC based on observations that it decreased the levels of the surface marker CD44, which is reported to be associated with cancer stem cells (19). We demonstrated using genetic and pharmacologic approaches in vitro and in vivo that reducing RSK1/2 activity or RSK1 or RSK2 levels inhibits multiple steps within the metastatic program. Furthermore, we confirmed that activated RSK is present in the majority of TNBCs. Our results strongly suggest that RSK is a viable drug target for TNBC metastasis.

We also report the generation and validation of a novel SL0101 analogue, C5″-n-propyl cyclitol SL0101 (1b) that is specific for RSK1/2. The specificity of the inhibitor for RSK1/2 is demonstrated by our observations that silencing RSK1/2 eliminates responsiveness to (1b) in MDA-MB-231 proliferation and survival assays. In in vitro assays using multiple TNBC cell lines the new analogue inhibits the major steps involved in the metastatic process, which include motility, proliferation and survival in a nonadherent environment. In addition, (1b) was as effective at inhibiting metastatic colonization in vivo as the FDA-approved MEK inhibitor, trametinib. Activation of the AKT pathway is proposed as a mechanism to account for the lack of efficacy of selumetinib, a MEK inhibitor, in combination with the anti-estrogen, fulvestrant, in a phase II clinical trial (50). We propose that inhibitors of RSK will offer greater flexibility in designing combination cancer therapies than MEK inhibitors, as there will be no positive feedback loop that results in activation of AKT.

D.A. Lannigan has ownership interest (including patents) in US 9,040, 673 B2 Patent. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K.A. Ludwik, J.P. Campbell, G.A. O'Doherty, D.A. Lannigan

Development of methodology: J.P. Campbell, M. Li, G.A. O'Doherty, D.A. Lannigan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.A. Ludwik, J.P. Campbell, M. Li, Z.M. Sandusky, L. Pasic, M.E. Sowder, D.R. Brenin, J.A. Pietenpol, D.A. Lannigan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.A. Ludwik, J.P. Campbell, Z.M. Sandusky, L. Pasic, M.E. Sowder, D.A. Lannigan

Writing, review, and/or revision of the manuscript: K.A. Ludwik, J.P. Campbell, M. Li, D.R. Brenin, J.A. Pietenpol, D.A. Lannigan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.P. Campbell, Y. Li, G.A. O'Doherty, D.A. Lannigan

Study supervision: G.A. O'Doherty, D.A. Lannigan

Kinome illustrations reproduced courtesy of Cell Signaling Technology, Inc. (www.cellsignal.com). We thank the University of Virginia Biorepository and Tissue Procurement Facility, for human breast samples and P. Brulet and R. Kemler, Developmental Studies Hybridoma Bank, through the NICHD, Department of Biology, University of Iowa, for the cytokeratin Endo-A (K8) monoclonal antibody.

This work is supported by S.G. Komen #IIR12223770 (to D.A. Lannigan), NIH GM088839 (to G.A. O'Doherty), and NSF CHE-1565788 (to G.A. O'Doherty).

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.