ISSN 00268933, Molecular Biology, 2011, Vol. 45, No. 4, pp. 600–607. © Pleiades Publishing, Inc., 2011. Original Russian Text © U.A. Boyarskikh, Yu.V. Kondrakhin, I.S. Yevshin, R.N. Sharipov, A.V. Komelkov, E.A. Musatkina, E.M. Tchevkina, M.A. Sukoyan, F.A. Kolpakov, K.N. Kashkin, M.L. Filipenko, 2011, published in Molekulyarnaya Biologiya, 2011, Vol. 45, No. 4, pp. 652–661.
MOLECULAR BIOLOGY OF THE CELL UDC 577.29
Prediction of Nonsmall Cell Lung Cancer Sensitivity to Cisplastin and Paclitaxel Based on Marker Gene Expression U. A. Boyarskikha, b, Yu. V. Kondrakhinc, d, I. S. Yevshinc, e, R. N. Sharipovc, e, A. V. Komelkovf, E. A. Musatkinaf, E. M. Tchevkinaf, M. A. Sukoyana, F. A. Kolpakovc, d, K. N. Kashking, and M. L. Filipenkoa, b, h a
Institute of Chemical Biology and Fundamental Medicine, Siberian Branch, Russian Academy of Sciences, Novosibirsk, 630090 Russia b Novosibirsk State University, Novosibirsk, 630090 Russia; email:
[email protected] c Institute of System Biology, Novosibirsk, 630090 Russia d Design Technological Institute of Digital Techniques, Siberian Branch, Russian Academy of Sciences, Novosibirsk, 630090 Russia e Institute of Cytology and Genetics, Siberian Branch, Russian Academy of Sciences, Novosibirsk, 630090 Russia f Blokhin Cancer Research Center, Russian Academy of Medical Sciences, Moscow, 115478 Russia g Shemyakin–Ovchinnikov Institute of Bioorganic Chemistry, Russian Academy of Sciences, Moscow, 117997 Russia Received October 22, 2010; in final form, December 12, 2010
Abstract—The goal of the present study was to define gene expression signatures that predict a chemosensi tivity of nonsmall cell lung cancer (NSCLC) to cisplatin and paclitaxel. To generate a set of candidate genes likely to be predictive, current knowledge of the pathways involved in resistance and sensitivity to individual drugs was used. Fortyfour genes coding proteins belonging to the following categories—ATPdependent transport proteins, detoxification system proteins, reparation system proteins, tubulin and proteins responsi ble for its synthesis, cell cycle, and apoptosis proteins—were considered. Eight NSCLC cell lines (A549, Calu1, H1299, H322, H358, H460, H292, and H23) were used in our study. For each NSCLC cell line, a cis platin and paclitaxel chemosensitivity, as well as an expression level of 44 candidate genes, were evaluated. To develop a chemosensitivity prediction model based on selected genes’ expression level, a multiple regression analysis was performed. The model based on the expression level of 11 genes (TUBB3, TXR1, MRP5, MSH2, ERCC1, STMN, SMAC, FOLR1, PTPN14, HSPA2, GSTP1) allowed us to predict the paclitaxel cytotoxic concentration with a high level of correlation (r = 0.91, p < 0.01). However, no model developed was able to reliably predict sensitivity of the NSCLC cells to cisplatin. DOI: 10.1134/S0026893311030034 Keywords: nonsmall cell lung cancer, cisplatin, paclitaxel, resistance prediction, gene expression
Currently available therapies for nonsmall cell lung cancer (NSCLC) include chemotherapy courses. Dif ferent protocols imply treatment with platinumcon taining drugs used as monotherapy or in combination with paclitaxel, ethoposide, hemcitabine, and some other substances. However, positive clinical response is achieved only in 20–30% of patients, when conven tional therapy approaches are used [1]. In most cases, the developed tumors are resistant to chemotherapy. The opportunity to apply an individual approach in the selection of combination and dosage of chemo therapeutic drugs is highly important for clinical oncology. One of the current directions in this field is the search for molecular markers that would predict the Abbreviations: MDR—multiple drug resistance; NSCLC— nonsmall cell lung cancer.
effect of a chemotherapeutic drug on cancer cells. The amount of mRNA transcribed from genes that are dif ferentially expressed in cells sensitive and resistant to the chemotherapeutic drug might be used as one of such markers. Marker genes are usually selected with one of the following methods. The first method implies screening of tens of thousands of genes with hybridized microarrays, determination of mRNA level of 10–30 thousand genes in samples with known sen sitivity to the drug, and selection of genes that better discriminate sensitive and resistant samples. The sec ond approach implicated selection of candidate genes based on the known mechanisms of chemoresistance. Here, we have used the second approach. The work was aimed to define a group of candidate marker genes and determine their potential to predict sensitivity of eight NSCLC cell lines (A549, Calu1, H1299, H322,
600
PREDICTION OF NONSMALL CELL LUNG CANCER SENSITIVITY TO CISPLASTIN
601
Candidate genes (44) that might be used as informative markers for chemoressitance prediction * Genes TUBB3, STMN1 TXR, TSP1
ERCC1, XPC, DDB2, MLH1, MSH2, BRCA1** YB1
Resistance to paclitaxel Maintenance of balance between tubulin polymerization and depolymerization [3, 4]. TXR prevents taxaneinduced apoptosis by downregulation of TSP1 (thrombospondin1) tran scription [5]. Resistance to cisplatin Genes of reparation proteins [6–9].
DNAbinding protein, binds predominantly to cisplatinmodified DNA regions, is presumably in volved in the recognition of DNAadducts by reparation complex [10]. TXN, TXNIP, CYB5R3 Adaptive antioxidant response [11]. AKR1B10** Indirectly inhibits retinoic acid receptor signal pathway [12]. Nonspecific resistance MDR, MRP1, MRP3, ATPdependent transport proteins [13, 14]. MRP5, BCRP AQP1, FOLR1, SLC34 Membrane transport proteins [15]. GSTP1 Detoxication of xenobiotics [16]. TP53, PML, BRCA1**, Proapoptotic factors [17–19]. DAPK1, RIPK2, BAX, SMAC BIRC5, BCL2 Antiapoptotic factors [18, 19]. HSPA2 Heatshock protein of Hsp70 family, contributes to cell survival and ensures crosstolerance to var ious stress factors [20]. EGFR, ERBB2, ERBB3, Tyrosinekinase receptors, involved in regulation of proliferation [21]. ERBB4, IGFR PTPN14 βcathenin dephosphorylation, loss fo intracellular contacts and activation of wntpathway [22]. TIMP1, FBLN1 Extracellular matrix components [23]. CAV1 Major component of specialized domains in plasma membrane (caveolae); regulates different sig nal pathways associated with chemoresistance by modulation of membrane receptor activity [24, 25]. AKR1B10**, AKR1B1 Differentially expressed in NSCLC cell isolated from smoking patients [26]. SFN Expressed in differentiated epithelial cells, repressed in tumor cells, mediates G2arrest of cell cy cle in response to genotoxicity [27]. * Genes were selected based on their role in chemoresistance development in tumor cells. ** Indicated genes are mentioned twice in this table.
H358, H460, H292, H23) to cisplatin and paclitaxel regarding the level of gene expression. Furthermore, we focus here on some aspects of data analysis that are critical for the creation of a reliable model for predic tion of cell response to chemotherapeutic drugs. MATERIALS AND METHODS Cultivation of NSCLC cell lines. Eight NSCLC cell lines (A549, Calu1, H1299, H322, H358, H460, H292, H23) were used in our study. Cells were grown in 25ml flasks in DMEM/F12 medium supple mented with 10% FBS, 0.03% glutamine, and 100 μg/ml kanamycin at 37°С and 5% СО2. Culture medium was refreshed every 24 h. Cells were 4fold reseeded when they reached 70–80% confluency. Determination of NSCLC sensitivity to chemothera peutic drugs. Cell sensitivity to cisplatin and paclitaxel MOLECULAR BIOLOGY
Vol. 45
No. 4
2011
(both from Sigma) was estimated by the value of IC50 (inhibitory concentration; drug concentration that inhibits growth of 50% cells). Cells in the exponential growth phase were plated into 96well plates (2000– 5000 cells per well), and, after 24 h, the medium was substituted for a fresh one containing cisplatin or paclitaxel taken at different concentrations. Paclitaxel concentration was tested in the range from 0.0001 to 1 μM and cisplastin was tested from 0.5 to 40 μM. The viability of cells was analyzed with MTTtest 48 h after drug addition [2]. Estimation of mRNA level of candidate genes with realtime RTPCR. Relative level of 44 candidate gene expression was determined for each NSCLC cell line (see table). Untreated cells in the exponential growth phase were plated into 6well plates and grown to 80% confluency. Then, culture medium was aspirated and
602
BOYARSKIKH et al.
cells were lyzed with Trisol (Invitrogen). The obtained samples were stored at –70°С. Total RNA and subsequent cDNA synthesis were performed in the reverse transcription reaction simul taneously for all samples. Total RNA was isolated according to Invitrogen recommendations. Reverse transcription was performed at 42°С for 45 min in a 20 μl reaction mixture containing 10 mM TrisHCl (pH 8.5), 5 mM MgCl2, 10 mM DTT, 100 mM KCl, 0.2 mM dNTPs, randon 9nucleotide primer (10 ng/μl), 200 U MoMLV RNAdependent DNA polymerase, and from 200 ng to 1 μg of total RNA. Relative amount of cDNA corresponding to each of 44 genes was determined with realtime PCR. Ampli fication of cDNA samples was performed in 20 μl reac tion mixture containing 10 mM TrisHCl (pH 8.9), 2.5 mM MgCl2, 55 mM KCl, 0.2 mM dNTPs, 0.5 U Taqpolymerase and SybrGreen I (Invitrogen) taken at 1 : 25000 dilution, and a pair of appropriate forward and reverse primers (300 nM each) and cDNA (5–25 ng). cDNA content (expressed in arbitrary units, a.u.) was calculated using normalization coefficient calculated as mean cDNA level of POLR2A, GAPDH, RPL32 and ACTB genes. Data analysis. Data were analyzed with statistical software R (version 2.10.1) (http://www.rproject.org). NSCLC cell lines were ranked according to IC50 value of cisplatin and paclitaxel and divided into either sensi tive (n = 4) or resistant (n = 4) groups for each drug. For each gene, Pearson coefficient (r) was determined for correlation of mRNA content and IC50 value. Ttest was used to determine significance of variability between expression level in sensitive and resistant cell lines. The function predicting IC50 values regarding the expression of mRNA predictors was performed using multiple regression analysis. We tested two variants of inclusion of predictor genes in the regression equa tion. In the first case, the selection of genes for the regression model was performed using stepwise regres sion. In the other case, new variables were used as pre dictors in the regression equation, which were rede fined with the method of principal (five) components. For both variants of regression analysis, a small subset of genes (5–15) was used as variables. The indicated subset included genes described by maximal r or max imal t values. The number of genes in the subset was gradually reduced from 15 to 5, and each time the quality of resulting predictive function was monitored. The major criterion used to estimate predictive func tion was the invariance (reliability) of prediction, the reliability of prediction functions was assessed with the JackKnife method, a method of crossvalidation: each cell line was successively excluded from the analysis, and the predictive function was generated based on the remaining cell lines. Thus, the JackKnife method enabled us to estimate the efficacy of IC50 predictions for cell lines that were not included in generation of predictive function.
RESULTS Here, we have created a group of candidate genes (44 genes) and determined their expression level in eight NSCLC cell lines; moreover, we have assessed their sensitivity to generally used chemotherapeutic drugs (paclitaxel and cisplatin). Using the data con cerning expression level of 44 genes as predictive fac tors, we have generated a series of regression models that predicted sensitivity of NSCLC cell lines to the abovementioned drugs with different reliability. Creation of a candidate gene group. Candidate genes were selected based on the current knowledge about chemoresistance of tumor cells. A cancer cell can be resistant either to a single group of chemother apeutic drugs (specific resistance) or to a wide range of drugs (multiple drug resistance, MDR). Increase in membrane protein expression was shown to be one of the causes that led to the development of MDR, since these proteins ensure export of various chemical com pounds from the cytoplasm to the extracellular space. Too active transport prevents accumulation of chemo therapeutic drug inside the cell at therapeutically effi cient concentrations. Many proteins associated with MDR belong to ATPdependent transport proteins, and include Рgp (ABCB1, encoded by MDR1 gene), MRP1 (ABCC1), MRP3 (ABCC3), MRP5 (ABCC5), BCRP (ABCG2), and others [13, 14]. Another mecha nism of MDR implies the activation of intracellular detoxication systems, including glutathioneStrans ferases (GSTs)—a family of protein that metabolize xenobiotics. Detoxication of chemical compounds medi ated by these enzymes involves conjugation with reduced glutathione. GSTP1 is one of the beststudied enzymes from this group [16]. Mechanisms of specific resistance of tumor cell depend on the mechanism of chemotherapeutic drug action. Thus, cisplatin is a platinumcontaining drug, which produces Pt2+ and Pt4+ ions that form coordi nate bonds with guanine base in DNA [28]. Formation of multiple DNAadducts and crosslinks results in an increase of replication errors, which arrests cell cycle progression and induces apoptosis. Increased activity of reparation systems in tumor cells neutralizes the cyto static effect of cisplatin. Thus, the alteration of expression level of genes responsible for DNA reparation might be used as a marker of tumor resistance to cisplatin. For analysis, we have chosen ERCC1, XPC, DDB2, MLH1, and MSH2 genes [6–9]. Cisplatin interacts not only with nuclear DNA but also with mitochondrial DNA and some potentialdependent mitochondrial proteins. Therefore, cisplatin is considered to induce apoptosis through activation of mitochondrial oxidative stress [29– 31]. Taking into account the abovementioned facts, we have chosen genes that encode antioxidant proteins TXN, TXNIP, and CYB5R3 as markers [14]. Another chemotherapeutic drug—paclitaxel (taxan group)—represents an antimitotic drug. In cells, paclitaxel binds to βtubulin, a major structural MOLECULAR BIOLOGY
Vol. 45
No. 4
2011
PREDICTION OF NONSMALL CELL LUNG CANCER SENSITIVITY TO CISPLASTIN
Estimation of NSCLC Cell Sensitivity to Cisplatin and Paclitaxel NSCLC cell sensitivity to paclitaxel and cisplatin was estimated by IC50 value for each of 8 cell lines (Figs. 1a, 1b). Based on the empirical IC50 values obtained for each drug, NSCLC cell lines were divided into sensitive and resistant groups. Average IC50 value was equal to 0.0097 ± 0.0038 μM for paclitaxelsensi tive cell lines (NCIH460, NCIH292, Calu1, NCI H322) and 0.0225 ± 0.0045 μM for resistant cell lines (NCIH23, NCIH358, NCIH1299, A549). In the case of cisplatin, average IC50 value was equal to 1.89 ± 0.35 μM for sensitive (NCIH358, A549, NCIH460, NCIH23) and 4.78 ± 2.34 μM for resistant (NCI H1299, NCIH292, NCIH322, Calu1) cell lines. No significant correlation between sensitivity of NSCLC cell lines to paclitaxel and cisplatin was observed. Expression Level of Candidate Genes in NSCLC Cells Relative amount of mRNA of 44 genes in NSCLC cells was estimated with realtime RTPCR. Then, the amount of mRNA was referred to IC50 values of drugs in each cell line, Pearson correlation coefficient was determined (r) for these values (separately for paclitaxel MOLECULAR BIOLOGY
Vol. 45
No. 4
2011
(a) 0.04
Resistant cell lines
Sensitive cell lines 0.02
IC50, µM
component of microtubules, and, thus, prevents their depolymerization. As a result, bundles of abnormal microtubules unable to assemble mitotic spindle are formed, and mitosis is arrested [32, 33]. One specific mechanism of tumor cell resistance to paclitaxel is over expression of βIIItubulin (TUBB3), one of βtubulin isotypes, which has lower affinity for paclitaxel com pared to other tubulin isotypes [3]. Another mecha nism of paclitaxelresistance is the increase of stath min expression (encoded by STMN1 gene), which binds to free tubulin heterodimers and destabilizes microtubules [4]. Moreover, along with direct antimi totic activity, taxan drugs stimulate thrombospondin (TSP) expression. Thrombospondin is an extracellular matrix protein, which binds to CD36 and CD47 receptors and induces apoptosis. Expression of TSP1 is suppressed by TXR protein (taxolresistance), which, therefore, mediates cell resistance to taxans [5]. Cytotoxic effect of most chemotherapeutic drugs finally results in the arrest of cell division and induc tion of apoptosis. Therefore, the defects in expression of genes responsible for cell cycle control, such as p53, BRCA1, PML, and SFN, and factors that directly stimulate (BAX, DAPK1, RIPK2, Smac/diablo) or inhibit apoptosis (BIRC5) might result in the develop ment of chemoresistance in tumor cells [17–19]. Along with the abovementioned marker genes, we considered genes of tyrosinekinase receptors of the EGF family (epidermal growth factor) that stimulate proliferation of epithelial cells, and some genes differen tially expressed in normal and tumor cells. As a result, a group of 44 candidate genes was created (see table).
603
0
10 8 6 4 2 0
H460 H292 Calu1 H322 H23 H358 H1299 A549 (b) Resistant cell lines Sensitive cell lines
H358 A549 H460 H23 H1299 H292 H322 Calu1 Fig. 1. IC50 values (average value + SD) for (a) paclitaxel and (b) cisplatin in eight NSCLC cell lines.
and cisplatin), and significance of difference (t) between expression levels of each mRNA was determined with a t test for sensitive and resistant groups of cell lines. For most candidate genes analyzed in this work, neither correlation nor dispersion analysis defined sig nificant dependence between transcription level and sen sitivity of cells to chemotherapeutic drugs. Figures 2a and 2b illustrate expression profiles of 11 genes that exhibited maximal absolute values of r and t criteria for cisplatin and paclitaxel. Notably, significant correla tion between IC50 value for paclitaxel and mRNA con tent was observed only in the case of HSPA2, SMAC, TXR1, TUBB3, PTPN14, and MSH2 genes. Statisti cally significant difference between mRNA level in sen sitive and resistant to paclitaxel cell lines was defined in the same genes (except for SMAC) and for MRP5. In the case of cisplatin, significant correlation between IC50 and mRNA level was determined for AQP1, CAV1, YB1, FBLN1, CYB5R3, RIPK2, and EGFR genes, and mRNA level was different in cells sensitive and resistant to cisplatin only in the case of EGFR, AKR1B10, and GSTP1 genes. Prediction of NSCLC Cell Sensitivity to Cisplatin and Paclitaxel Regarding Transcription Level of Candidate Genes Regression analysis was used to generate several linear models that predicted IC50 values for paclitaxel and cisplatin depending on the mRNA level of marker genes. Parameters of regression equations were deter mined with a stepwise regression and/or regression analysis with principal components. It should be noted that a peculiar feature of our research was a rather small number of predictors (44 genes, whose level of expression was determined) that exceeded the number of observations (only eight
604
BOYARSKIKH et al. (a) 5 Relative level of gene expression, arb. units
4 3
NCIH460 NCIH292 Calu1 NCIH322
NCIH23 NCIH358 NCIH1299 A549
2 1 0 HSPA2 SMAC TXR TUBB3 PTPN14 MSH2 ERCC1 MRP5 FOLR1 STMN1 GSTP1 r 0.89** 0.76* 0.76* 0.72* –0.71* 0.71* 0.63^ 0.54 –0.53 0.53 0.35 t –2.53 –1.83^ –3.22** –2.87* 2.19* –2.46* –1.75^ –2.03* 1.44^ –1.50^ –1.81^ 6
(b)
5 4
NCIH358 A549 NCIH460 NCIH23
NCIH1299 NCIH292 NCIH322 Calu1
3 2 1 0 CAV1 AQP1 YB1 FBLN1 CYB5R3 EGFR RIPK2 MLH1 AKR1B10 GSTP1 ERBB3 r 0.87** 0.87** 0.82* 0.76* 0.74* 0.72* –0.72* 0.50 –0.49 –0.50 0.30 t –1.10 –0.97 –1.45^ –1.57^ –1.56^ –2.81* 1.86^ –1.11 2.28* 3.59** –1.14
Fig. 2. Expression profile of 11 genes (average value + SD) that exhibited maximal absolute values of r and t criteria for (a) pacli taxel and (b) cisplatin. The lines are shown regarding the increase of IC50 values for corresponding chemotherapeutic drug; genes are shown regarding to the increase of correlation coefficient r for mRNA content and IC50 value.
cell lines), whereas methods of regression analysis require a contrary ratio of these parameters. The prob lem might be solved by the reduction of sample size, which means reduction of predictors number. Prelim inary selection of a subset of genes enabled successive reduction of predictors to 5–15. Two variants of selec tion were used: the subset included genes with maxi mal absolute value or criterion r (О1 approach) or t (О2 approach).
components enabled more reliable prediction of IC50 values for both chemotherapeutic drugs. More precise (CORR1 = 0.99) and reliable (CORR2 = 0.91) predic tion of IC50 for paclitaxel was achieved using regression analysis of five principal components, when 11 genes selected with О2 were used as predictive parameters. Regression equation that defines the dependence of IC50 for paclitaxel and the amount of mRNA of 11 marker genes was as follows:
As a result, a set of predictive functions that varied in the number of genes, method of their selection, and regression method used to generate function was obtained. The efficacy of each a function was esti mated by two values: CORR1 and CORR2, which describe accuracy and reliability of the prediction, respectively. CORR1 value was estimated as correla tion coefficient for the observed and predicted IC50 values for each of eight cell lines. CORR2 value was calculated as correlation coefficient for the IC50 values observed and was predicted with the “JackKnife” method (JackKnife is one of the bootstrap methods of applied statistics that enables estimation of invariance of the obtained conclusions).
F = 7 + 1.6∗ TUBB3 + 0.3∗ TXR1 + 0.2∗ MRP5 + 2.4∗ MSH2 + 0.03∗ ERCC1 + 1.9∗ STMN
CORR1 values turned out to be rather high (from 0.90 to 0.99) for all obtained functions. CORR2 values significantly varied but, in most cases, they were statis tically significant (
