Integrated Analysis of Transcriptomes of Cancer Cell Lines and Patient Samples Reveals STK11/LKB1–Driven Regulation of cAMP Phosphodiesterase-4D

The limited success of chemotherapeutics has led to the development of cancer therapies that block the growth and progression of cancer by targeting and interfering with specific molecules involved in tumorigenesis. However, many targeted therapeutics that specifically target tumor-associated components have not measured up to expectations in clinical studies (1). DNA abnormalities and genetic alterations contribute to all aspects of cancer development, including tumor initiation, progression, and metastasis (2). Moreover, mutational status of specific cancer lineages can affect the sensitivity to or resistance against drugs (3). Thus, the variation in genotype of different cancer lineages is important in the context of the heterogeneity of the anticancer drug response.

Recently, there has been an increase in the availability of large-scale human cell line and tissue data in multilevel experiments such as sequencing, gene expression, protein regulation, and compound response. These diverse datasets can be used to identify mutation- and lineage-specific cancer signatures that provide new insights into targeted cancer therapies. For example, a public multiple cell line high-throughput screening dataset, called NCI-60 project, was developed by the U.S. National Cancer Institute (NCI; Bethesda, MD), to analyze 60 cancer cell lines as a predictive model system of gene/protein expression and drug response (4). The GlaxoSmithKline dataset, which contains microarray and drug screening data for more than 300 cancer cell lines, has been made available through the NCI's cancer Bioinformatics Grid (5, 6). A collection known as the Cancer Cell Line Encyclopedia (CCLE) furnishes gene expression data for 947 cancer cell lines as well as chemical screening data, and has expanded the understanding of preclinical cancer cell line models in the context of mutational status (7). The mutational genotypes and drug sensitivity of human cancer cell lines have been well characterized in the Sanger COSMIC project (8).

We have previously used the GlaxoSmithKline cell line datasets to assess mutation-oriented integration of multilevel “omics” and chemical screening data, in the context of mutations that might be the source of oncogenic signaling (9). If a mutation is the direct driver of cancer progression, blocking the functional consequences of the mutation or regulation of downstream markers may have better chance to inhibit cancer progression. For example, driver mutations such EGFR, PIK3CA, BRAF, ALK, HER2, AKT1, MAP2K1, and MET in NSCLC have been reported as potential targets of anticancer agents with some clinical relevance (10). Thus, it is important to distinguish mutation-driven markers from mutation-associated markers. Analysis of DNA microarray data of cancer cell lines has generated a large number of transcription signatures that may have direct or indirect relationships with a cancer lineage or oncogenic mutation (11).

In the present study, we applied two orthogonal metrics to independently measure gene expression and relations with specific mutations in the context of cellular lineage among a large number of samples. With this approach, we were able to identify mutation-driven signatures in specific cellular lineages. To validate the identified transcriptional signatures, we then applied the signature to The Cancer Genome Atlas (TCGA) datasets (12), which provide mutational genotypes and RNA expression levels for hundreds of human cancer tissues in major cancer lineages. Especially, we focused on the mutation of serine/threonine kinase 11 (STK11), which was known as a major tumor suppressor and the upstream kinase of adenine monophosphate–activated protein kinase (AMPK). Its significance for cancer and cell metabolism has been well known (13, 14). In this study, phosphodiesterase-4D (PDE4D) expression showed significant association with STK11 mutation in lung cancer. PDE4D is one of PDE4 subfamilies (A/B/C/D). PDE4 enzymes break down cAMP and sequestered species underpin cAMP signal compartmentalization in cells, such that each produces a series of isoforms with distinct functional roles (15–17). As a key candidate mutational target, PDE4D was demonstrated experimentally to be directly regulated by the predicted mutation and lineage characteristics and potentially plays a role in mutation-specific cancer regulation. The ability to identify and validate candidate targets of tissue-specific mutational events supports the power of the proposed data mining strategy.

Microarray gene expression data for 318 cell lines were obtained from Bioinformatics Grid (caBIG) of the NCI (6). This caArray_GSK dataset has 950 arrays performed in triplicate for each of 318 cell lines. The experiments were carried out with the Affymetrix U133 Plus 2.0 Array chip, which includes 54,613 probes. However, for the refined analysis, we extracted 22,357 probes, whose expression values existed over 150 cell lines. For validation, the GSE6135 dataset constructed from a total of 12 STK11 [liver kinase B1 (LKB1)] wild-type and mutant samples was obtained from the NCBI Gene Expression Omnibus (GEO) ftp. Gene expression levels were compared between parental A549 and H2126 lung cancer cell lines, which express STK11 homozygous-mutant proteins, and are null for STK11, respectively, and the same cell lines expressing wild-type STK11 (18), using the GEO2R tool. To analyze tissue-based gene expression profile, data from patients' lung adenocarcinoma (LUAD) samples were obtained from the TCGA. These data (level 3) have 355 and 490 (updated on 2013) tumor samples analyzed with the Illumina HiSeq 200 RNA-Sequencing platform. The expression signal of each gene is represented with upper quartile normalized RNA-Seq by expectation maximization (RSEM) count estimates. Fold change values per sample of a gene were calculated by median value of total samples.

To compare gene expression levels of various genotypes and/or lineages, fold change and t test P values were calculated using microarray gene expression data. The log₂ fold change of a probe is given by the difference between the average of cell lines for each category and median value of total cell lines. To determine statistical significance, we calculated P value from t statistic. To describe the selective association between gene expression and genotype, we also calculated enrichment score using an odds ratio between the observed odds and expected odds. The observed odds is the ratio for the number of differentially expressed (log₂ fold change>1 and P < 0.01) cell lines with specific genotype via the number of cell lines with specific genotype. The expected odds is the ratio for the number of differentially expressed (log₂ fold change>1 and P < 0.01) cell lines versus the number of total cell lines except missing values. In addition, the probability of an odds ratio was calculated by Fisher exact test using the R open-source computing language, version 2.15. Fisher exact test uses hypergeometric distribution to determine the statistical significance of the agreement between individual question pairs (19).

NCI-60 lung cancer cell lines (NCI-H460 and A549) were obtained from Developmental Therapeutics Program NCI/NIH (DTP) on January 13, 2010. NCI-60 lung cancer cell lines (NCI-H322M and NCI-H226) were obtained from the U.S. NCI (DTP) on February 7, 2012. NCI-H1993, NCI-H1395, NCI-H82, and NCI-H524 were obtained from American Type Culture Collection (ATCC) on January 6, 2012. All cells were grown in RPMI medium (GIBCO) with 10% FBS (GIBCO) and 1% penicillin/streptavidin (GIBCO), and maintained at 37°C in a humidified atmosphere at 5% CO₂. For siRNA transfection, 2 × 10⁵ cells per well were plated in a 6-well plate. After adhering for 24 hours, 45 nmol/L of target siRNA (Santa Cruz Biotechnology) were added in transfection medium for 6 hours at 37°C in a CO₂ incubator. After transfection, cells were supplemented with RPMI containing FBS and cultured at 37°C/5% CO₂ for up to 48 hours before harvest and RNA extraction.

Lung normal cell lines (LL24 and LL86) were obtained from the Korean Cell Line Bank (KCLB) on December 11, 2013. CCD-19Lu was obtained from ATCC on November 14, 2013. All cells were grown in RPMI medium (GIBCO) with 10% FBS (GIBCO) and 1% penicillin/streptavidin (GIBCO), and maintained at 37 °C in a humidified atmosphere at 5% CO₂. For reverse siRNA transfection, 1,500 cells per well were plated in a 384-well plate, which contained 10 nmol/L target siRNAs of PLK1, UBB1 and PDE4D (Thermo Fisher Scientific, Dharmacon Products) separately. After cultured for 72 hours at 37°C/5% CO₂, the cell viability was detected using the CellTiter-Blue Cell Viability Assay (Promega).

Total RNA was extracted in 6-well plates using TRIzol. Synthesis of cDNA and PCR amplification was carried out with the SuperScript One-step RT-PCR Platinum Taq Kit (Invitrogen). Quantitative real-time PCR for PDE4D (Hs01579625_m1, TaqMan) and STK11 (Hs00176092_m1, TaqMan) was carried out on an Applied Biosystems 7500 using a TaqMan real-time detection protocol.

A total of five PDE4D inhibitors were screened on STK11 wide-type (NCI-H82 and NCI-H524) and mutational (NCI-H1993 and NCI-H1395) cell lines. Rolipram and roflumilast were purchased from Selleck Chemicals. L-454560, cilomilast, and lirimilast were purchased from Axon Medchem BV. Cells were seeded in a 96-well plate with density of the optimized cell number (1,000 cells/well for 3-day treatment, 330/well for 7-day treatment, and 165/well for 10-day treatment). After 24 hours of seeding, cells were treated with diluted chemicals at 20 μmol/L working concentration. Cells were incubated for another 3, 7, and 10 days, respectively, and then measured for the viability using the CellTiter-Blue Cell Viability Assay (Promega). We kept the culture medium fresh by replacing every 3 or 4 days. Apoptotic cells were identified using a terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL)-based cell detection kit, POD (Roche Applied Science). Cells (A549 and H226) were fixed in slides with 4% paraformaldehyde (Ducsan) for 1 hour, washed with PBS (Welgene), and then incubated for 10 minutes with 3% H₂O₂ in methanol (trisodium citrate dihydrate; Junsei) at room temperature. After washing with PBS again, cells were incubated 2 minutes on ice with 0.1% Trion X-100 in 0.1% sodium citrate. Then, slides were incubated with the TUNEL reaction mixture for 60 minutes at 37°C, and then rinsed with FBS. Stained slides were examined under light microscopy.

After 48-hour target siRNA treatment, cell samples were harvested and the protein supernatants were isolated using cell lysis buffer (Cell Signaling Technology; #9803) with added phenylmethylsulfonyl fluoride (PMSF). The total protein content (30 μg) from cell lysates was separated using SDS–PAGE (8%) and transferred to a 0.45-μm nitrocellulose membrane (Millipore) for 1 hour. The membranes were washed with TBS-T containing 5% (w/v) BSA. The membranes were incubated overnight with specific AMPK-α (Cell Signaling Technology), GAPDH (Cell Signaling Technology), PDE4D (Abcam plc.), and α-actin (Cell Signaling Technology) were exposed to secondary antibodies coupled to horseradish peroxidase for 2 hours at room temperature. The membranes were then washed three times with TBS-T at room temperature. Immunoreactivity was detected using an enhanced chemiluminescent substrate from Thermo and analyzed with an LAS3000 Luminescent image analyzer from Fuji Film.

The availability of DNA microarray data from a large well-characterized collection of cancer cell lines provides the opportunity to identify markers allowing classification of cancers into “categories” driven by their combined lineage and mutational status. We applied two complementary, orthogonal metrics to determining the sensitivity and specificity of transcriptome markers in these different categories. Fold change reflects the sensitivity of a transcriptional probe in each category against all other samples, while the enrichment score quantifies the specificity of expression of a transcript within a specific category (see Materials and Methods for details). Because these two metrics for gene expression were not correlated (Fig. 1A), they allow us to prioritize category-associated markers for further investigation.

In this study, cancer categories were defined by 19 lineages and 12 mutated cancer genes with sufficient frequency to allow a robust analysis with sufficient power to identify potential cancer drivers (Supplementary Table S1). Through combining lineage and mutation across the sample sets, a total of 228 categories were generated, each of which included at least three cell lines. Among genes that were significantly over- or underexpressed in a given category, only 10% to 20% were specific to each category (Table 1). Combined categories (two mutations or lineage-mutation) included expression markers with higher sensitivity and specificity than those in single categories (Fig. 1B). In particular, categories defined by a combination of lineage and mutation status identified markers with high sensitivity and specificity (Fig. 1B and Table 1). These selected markers (a total of 1,455 genes) provide a starting point to integrate the interaction of lineage and mutational status on gene expression as a functional readout for mutation-specific cancer progression (complete lists available in Supplementary Table S2).

Combining cancer lineage with mutational status revealed a varied distribution of expression markers (Fig. 1C). Importantly, the effects of particular mutations were most clearly evident in specific lineages. For example, RB1- and STK11-associated gene expression changes were exclusive to lung cancer cell lines, while BRAF-associated markers were dominant in skin cancer cell lines. Other mutations such as those in CDKN2A were associated with expression markers across multiple cell lineages. Lung cell lines showed the greatest number of markers, most of which were found in cells harboring RB1, STK11, or PTEN mutations, prompting us to further investigate interactions between lineage and mutation using lung cancer as a model.

To determine the relevance of markers identified in lung cancer cell lines, we extended our analysis to clinical tumor samples. Together with KRAS (20) and CDKN2A (21), STK11 (18) was consistently found to be one of three major mutations in non–small cell lung cancer (NSCLC) subtype of tissue and cell line samples (Fig. 2A). RB1 (22) and PTEN (23) mutations were dominantly found in small cell lung cancer (SCLC) subtype. In addition, NSCLC type is histologically classified into LUAD and lung squamous cell carcinoma (LUSC) subtypes. Thus, the relative frequency of mutations was also comparatively analyzed on LUAD and LUSC subtypes of tissue samples (Fig. 2A). Cancer cell lines did not include enough LUSC subtype samples in datasets. Among three major mutations (i.e., KRAS, CDKN2A, and STK11) in NSCLC, KRAS and STK11 mutations were dominantly found in LUAD, while CDKN2A was dominant in LUSC subtype of cancer tissues. Although the importance of STK11 mutations in lung cancer has been addressed previously (24), therapeutic targets specific to or driven by STK11 mutations have not been identified. In the present analysis, a total of 32 expression gene probes showed alterations with significant sensitivity and specificity in lung cell lines harboring STK11 mutations (Fig. 2B). Because TP53 mutations are widely observed in cancer cell lines including lung cancer cell lines (25), we analyzed whether or not the 32 gene expression changes associated with STK11 mutation were also associated with TP53 mutational status. We found that the expression of 16 of 32 gene probes (16 probes representing 11 unique genes) associated with the STK11 mutations was independent of TP53 genotype, and is likely driven by STK11 (Fig. 2C).

To confirm the direct association of selected markers with the STK11 genotype, we analyzed transcriptome data from two STK11-defective lung cancer cell lines, A549 and H2126, as well as cells stably expressing wild-type STK11 (18). We determined the fold change of genes in parental cells expressing mutated STK11, as compared with wild-type STK11 (Fig. 2D). Among the 11 STK11 category-specific genes identified above, only the expression of PDE4D showed the expected pattern in response to the expression of wild-type STK11. PDE4D expression was significantly downregulated by transfection of wild-type STK11 into STK11-mutant lines, while it was upregulated in STK11-mutant lung cell lines (Fig. 2D). The overexpression of PDE4D has been observed in multiple types of cancer cells originating from the central nervous system (CNS), lung, breast, and melanomas (26). But, so far, there has been no report that the elevated expression of PDE4D is mutation-specific in cancer. In the present study, a total of 318 cancer cell lines were analyzed to find association between STK11 and PDE4D. Among them, there were 17 cancer cell lines harboring STK11 mutation. Most of STK11-mutant cell lines were lung cell lines and only four cell lines originated from leukemia, cervical, and ovarian cancer (Fig. 2E). Although the overexpression of PDE4D was generally associated with STK11 mutation in most cell lines, only lung category was appropriate for the association study with statistical confidence. PDE4D has been implicated in cancer cell proliferation and metastasis (27, 28). The inhibition of PDE4 activity suppressed cell growth and induced apoptosis in human leukemia cells (29). Furthermore, the PDE4 was considered as a target to regulate the malignancies of cancers including leukemia and colon (30, 31). Thus, we explored whether the previously undetected link between STK11 and PDE4D could contribute to lung cancer progression and provide a tractable therapeutic target in lung cancer cell lines.

Lung cell lines with STK11 mutations showed higher levels of PDE4D expression than cell lines without STK11 mutations, and cells from other lineages with STK11 mutations (Figs. 2E and 3A). The association of PDE4D expression with STK11 mutation was conserved in patient lung cancer samples (Fig. 3B), implying that PDE4D plays a critical role in STK11-dependent progression of lung cancers. Furthermore, PDE4D have 11 types of isoforms (1–11). The expressional change of isoforms was confirmed using the TCGA RNA-sequencing data (Fig. 3C and D). Generally, the expression of all PDE4D isoforms was significantly enriched and increased in STK11-mutant lung cancers except PDE4D4 and PDE4D7.

As predicted by the expression patterns shown in Fig. 2D, the expression of wild-type STK11 in lung cancer cells with aberrant STK11 function resulted in markedly decreased PDE4D levels (Fig. 4A). The raw DNA microarray data were obtained from Ji and colleagues (18). Furthermore, knockdown of STK11 in wild-type, but not STK11-mutant, lung cell lines resulted in increased PDE4D expression (Fig. 4B). In addition, knockdown of AMPK, which is a direct downstream target of STK11, also increased PDE4D expression in wild-type, but not STK11-mutant, cell lines (Fig. 4C). In the Western blot analysis of PDE4D protein expression, the efficacy of AMPK knockdown was also reproduced at protein expression level (Fig. 4D). AMPK knockdown was selectively effective in increasing the PDE4D protein expression in STK11 wild-type cell lines. These results indicate that the regulation of PDE4D expression is at the downstream of STK11 and AMPK signaling.

Previously, PDE4D has been known as a tumor-promoting factor (26). We investigated whether inhibition of PDE4D would selectively inhibit the growth of cancer cells with aberrant STK11. When we knocked down PDE4D in a series of lung cell lines, cell growth was significantly decreased in STK11-mutant cell lines but not wild-type cancer lines (Fig. 5A). Thus PDE4D appears to play an important role in cancer survival or proliferation in STK11-mutated lung cancer cells. In addition, three lung normal cell lines (LL24, LL86, and CCD-19Lu) were tested for treatment with siPDE4D, no significant decreased cell growth were observed in comparison with positive control siRNAs (Fig. 5B). Together with Fig. 5A, this result shows that PDE4D knockdown is selectively effective in STK11-mutant cancer cells, while it is less effective in STK11 wild-type cancer cells or noncancer normal cells. Furthermore, the induction of apoptosis by siRNA treatment showed similar patterns to those of cell growth assay (Fig. 5C), that is, siRNA knockdown of PDE4D induced apoptosis more effectively in STK11 mutants than in wild-type cells.

A number of PDE4D inhibitors are in clinical trials (32) offering a potential therapeutic opportunity for STK11-mutant lung cancers. For example, rolipram, an anti-inflammatory drug, has anticancer efficacy (33, 34), but it is not currently used because of the difficulty of identifying patients likely to benefit. In the present study, a total of five selective PDE4D inhibitors—cilomilast, roflumilast, rolipram (35), lirimilast (36), and L454560 (37)—were tested at 20 μmol/L concentration. Generally, all the five PDE4D inhibitors showed differential efficacy on the proliferation of STK11-mutant cell lines in a time-dependent manner (Fig. 5D). The result of 10-day viability assays showed that the sensitivity of STK11-mutant cell lines was significantly increased for cilomilast, lirimilast, rolipram, and L454560 more than STK11 wild-type cell lines (Fig. 5D). We thus expect that PDE4D inhibition with effective drugs may provide efficacious treatment of patient tumors harboring STK11 mutations.

In this study, we demonstrated that classification of cell lines by a combination of lineage and mutational status provides a powerful approach to identify sensitive and specific transcriptional markers. Furthermore, these approaches were able to reveal nonintuitive downstream mediators and elicit potential mutation-specific anticancer therapies. Thus, an integrative analysis of highly characterized cell lines and human tissue samples has the potential to uncover novel mechanisms of carcinogenesis, and therapeutic strategies.

High-throughput cancer cell line screening provides a strategy to identify transcription markers to help guide targeted cancer therapy (38, 39). Here, we present a computational strategy to profile unique transcriptional signatures associated with somatic mutations in combination with cancer lineage, using two metrics reflecting sensitivity and specificity independently.

Previous work has reported that genetic variants of the STK11–AMPK signaling pathway can influence therapeutic effects in diseases such as type II diabetes (40). STK11 encodes a serine/threonine kinase that functions as a tumor suppressor and regulates cell polarity. Loss-of-function somatic mutations of STK11 have been found in approximately 30% of lung cancer, and have been proposed to promote metastasis (41). In the present study, we searched for transcriptional markers directly associated with STK11 mutations in lung cancer cells. Among several STK11-specific signatures we identified, PDE4D was up- or downregulated depending on whether the STK11 gene was wild-type or mutant. Although STK11 is a tumor suppressor with a mutational frequency of about 20% to 30% in NSCLC (42), PDE4D was deleted in only 3% of lung cancer cell lines (43). Further analysis of somatic copy number alterations showed that about 4% of solid tumor primary specimens from the TCGA project had homozygous deletion. Whether PDE4D functions as to increase or decrease tumorigenesis is likely context dependent and potentially dependent on whether STK11 is intact. Indeed, PDE4 has previously been implicated in cancer pathophysiology by altering proliferation and angiogenesis in lung cancer (44). The association of PDE4D expression and cancer progression has been reported in multiple types of cancer (26). Recently, PDE4D7, one of PDE4D isoforms, has been reported as an androgen-independent marker with a role in regulation of prostate cancer cell proliferation (45). Also, it has been demonstrated that regulation of PDE4D9 by a group of kinases, of which AMPK is a key species, could influence cell-cycle progression, particularly through mitosis. This study showed a connection between PDE4D and AMPK in cancer/cell-cycle dynamics (46). Further depletion of PDE4D can cause apoptosis (10%–40%) in multiple types of cancer cells which have elevated expression level of PDE4D (26).

We experimentally confirmed that the knockdown of PDE4D gene expression suppressed proliferation of STK11-mutated cell lines. Conversely, we also found that rescue of STK11 activity in mutant and siRNA-silenced cells decreased the expression of PDE4D. These results demonstrated that PDE4D can be a transcriptional biomarker capable of detecting STK11 mutation in lung cancers. It further represents a possible route to therapeutic application for regulating lung cancer progression. In addition, our results are consistent with STK11-mediated activation of AMPK contributing to carcinogenesis (47). We found the knockdown of AMPK and STK11 resulted in similar effects on PDE4D expression.

The knockdown of PDE4D gene expression induced selective reduction of cell growth in STK11 mutants (Fig. 4A). This indicated that PDE4D expression was important for the growth of cells, when normal STK11 signaling was lost. These observations were confirmed in apoptosis assays (Fig. 4C). In addition to PDE4D expression, STK11-dependant AMPK signaling seems to be important in the survival/growth of STK11-mutant cancer cells. Our study has demonstrated that mutant STK11 is closely associated with enhanced PDE4D expression in NSCLC. Previous studies give some clues about how STK11–AMPK signaling is linked to PDE4D regulation as well as the potential mechanism of apoptosis induction. PDE4D inhibitors could competitively inhibit cAMP-degrading phosphodiesterases, leading to elevated cAMP level (48). cAMP is known to activate protein kinase A (PKA)—an AMP-dependent enzyme, then activate enzymes related to metabolic response and regulate gene expression by the activation of transcription factors (such as CREB) (49). An increased intracellular cAMP has also been shown to increase the activity of AMPK (50), which is the direct downstream target of STK11 (LKB1) (47). Further mechanistic studies will be required to understand the relationship among STK11, AMPK, and PDE4D in cancer progression. Most of PDE4D inhibitors were more active in STK11-mutant cell lines than STK11 wild-type cells (Fig. 4D and E). Collectively, our findings suggest that the therapeutic potential of PDE4D-targeted therapy should be further assessed in STK11 mutant–type lung cancer. Therefore, in addition to describing a method to identify therapeutic biomarkers, our approach may provide useful clues for effective novel targeted therapies.

No potential conflicts of interest were disclosed.

Conception and design: S. Yoon

Development of methodology: N. He, S. Yoon

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Song, C. Park, S. Kim, H.Y. Yim, J.H. Park, K.I. Kim

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. He, N. Kim, K. Kim, F. Zhang, S. Yoon

Writing, review, and/or revision of the manuscript: N. He, N. Kim, G.B. Mills, S. Yoon

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E.Y. Park

Study supervision: J.H. Park, G.B. Mills, S. Yoon

The authors thank Hyun-Young Lee for her professional work in improving the graphic quality and visibility of the figures.

This work was supported by the National Research Foundation of Korea (KRF) grants, Bio and Medical Technology Development Program (NRF-2012M3A9B6055398) and National Leading Research Lab (NLRL) program (NRF-2011-0028816), funded by Korea government (MEST). S. Yoon received the NRF-2012M3A9B6055398 and NRF-2011-0028816 grants. G.B. Mills receives sponsored research support from GlaxoSmithKline.

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.