Membrane transporters and channels: role of the transportome in cancer chemosensitivity and chemoresistance

Membrane transporters and channels (collectively the transportome) govern cellular influx and efflux of ions, nutrients, and drugs. We used oligonucleotide arrays to analyze gene expression of the transportome in 60 human cancer cell lines used by the National Cancer Institute for drug screening. Correlating gene expression with the potencies of 119 standard anticancer drugs identified known drug-transporter interactions and suggested novel ones. Folate, nucleoside, and amino acid transporters positively correlated with chemosensitivity to their respective drug substrates. We validated the positive correlation between SLC29A1 (nucleoside transporter ENT1) expression and potency of nucleoside analogues, azacytidine and inosine-glycodialdehyde. Application of an inhibitor of SLC29A1, nitrobenzylmercaptopurine ribonucleoside, significantly reduced the potency of these two drugs, indicating that SLC29A1 plays a role in cellular uptake. Three ABC efflux transporters (ABCB1, ABCC3, and ABCB5) showed significant negative correlations with multiple drugs, suggesting a mechanism of drug resistance. ABCB1 expression correlated negatively with potencies of 19 known ABCB1 substrates and with Baker's antifol and geldanamycin. Use of RNA interference reduced ABCB1 mRNA levels and concomitantly increased sensitivity to these two drugs, as expected for ABCB1 substrates. Similarly, specific silencing of ABCB5 by small interfering RNA increased sensitivity to several drugs in melanoma cells, implicating ABCB5 as a novel chemoresistance factor. Ion exchangers, ion channels, and subunits of proton and sodium pumps variably correlated with drug potency. This study identifies numerous potential drug-transporter relationships and supports a prominent role for membrane transport in determining chemosensitivity. Measurement of transporter gene expression may prove useful in predicting anticancer drug response.

Predicting drug sensitivity and resistance: profiling ABC transporter genes in cancer cells

For analysis of multidrug resistance, a major barrier to effective cancer chemotherapy, we profiled mRNA expression of the 48 known human ABC transporters in 60 diverse cancer cell lines (the NCI-60) used by the National Cancer Institute to screen for anticancer activity. The use of real-time RT-PCR avoided artifacts commonly encountered with microarray technologies. By correlating the results with the growth inhibitory profiles of 1,429 candidate anticancer drugs tested against the cells, we identified which transporters are more likely than others to confer resistance to which agents. Unexpectedly, we also found and validated compounds whose activity is potentiated, rather than antagonized, by the MDR1 multidrug transporter. Such compounds may serve as leads for development.

Mistaken identifiers: gene name errors can be introduced inadvertently when using Excel in bioinformatics

Background

When processing microarray data sets, we recently noticed that some gene names were being changed inadvertently to non-gene names.

Results

A little detective work traced the problem to default date format conversions and floating-point format conversions in the very useful Excel program package. The date conversions affect at least 30 gene names; the floating-point conversions affect at least 2,000 if Riken identifiers are included. These conversions are irreversible; the original gene names cannot be recovered.

Conclusions

Users of Excel for analyses involving gene names should be aware of this problem, which can cause genes, including medically important ones, to be lost from view and which has contaminated even carefully curated public databases. We provide work-arounds and scripts for circumventing the problem.

Transcriptomic analysis of the NCI-60 cancer cell lines

Pharmacogenomics aims at molecular subsetting of patients for more effective therapy. Transcriptomic profiling of the 60 human cancer cell lines (the NCI-60) used by the US National Cancer Institute serves that aim because the cells have been treated with > 100,000 chemical compounds over the last 13 years. Patterns of potency can be mapped into molecular structures of the compounds or into molecular characteristics of the cells. We discuss conceptual and experimental aspects of the profiling, as well as a number of bioinformatic computer programs that we have developed for biological interpretation of the profiles.

Properties of switch-like bioregulatory networks studied by simulation of the hypoxia response control system

A complex bioregulatory network could be more easily comprehended if its essential function could be described by a small "core" subsystem, and if its response characteristics were switch-like. We tested this proposition by simulation studies of the hypoxia response control network. We hypothesized that a small subsystem governs the basics of the cellular response to hypoxia and that this response has a sharp oxygen-dependent transition. A molecular interaction map of the network was prepared, and an evolutionarily conserved core subsystem was extracted that could control the activity of hypoxia response promoter elements on the basis of oxygen concentration. The core subsystem included the hypoxia-inducible transcription factor (HIFalpha:ARNT heterodimer), proline hydroxylase, and the von Hippel-Lindau protein. Simulation studies showed that the same core subsystem can exhibit switch-like responses both to oxygen level and to HIFalpha synthesis rate, thus suggesting a mechanism for hypoxia response promoter element-dependent responses common to both hypoxia and growth factor signaling. The studies disclosed the mechanism responsible for the sharp transitions. We show how parameter sets giving switch-like behavior can be found and how this type of behavior provides a foundation for quantitative studies in cells.

Molecular interaction maps--a diagrammatic graphical language for bioregulatory networks

Molecular interaction maps (MIMs) use a clear, accurate, and versatile graphical language to depict complex biological processes. Here, we discuss the main features of the MIM language and its potential uses. MIMs can be used as database resources and simulation guides, and can serve to generate new hypotheses regarding the roles of specific molecules in the bioregulatory networks that control progression through the cell cycle, differentiation, and cell death.

Karyotypic complexity of the NCI-60 drug-screening panel

We used spectral karyotyping to provide a detailed analysis of karyotypic aberrations in the diverse group of cancer cell lines established by the National Cancer Institute for the purpose of anticancer drug discovery. Along with the karyotypic description of these cell lines we defined and studied karyotypic complexity and heterogeneity (metaphase-to-metaphase variations) based on three separate components of genomic anatomy: (a) ploidy; (b) numerical changes; and (c) structural rearrangements. A wide variation in these parameters was evident in these cell lines, and different association patterns between them were revealed. Analysis of the breakpoints and other specific features of chromosomal changes across the entire set of cell lines or within particular lineages pointed to a striking lability of centromeric regions that distinguishes the epithelial tumor cell lines. We have also found that balanced translocations are as frequent in absolute number within the cell lines derived from solid as from hematopoietic tumors. Important similarities were noticed between karyotypic changes in cancer cell lines and that seen in primary tumors. This dataset offers insights into the causes and consequences of the destabilizing events and chromosomal instability that may occur during tumor development and progression. It also provides a foundation for investigating associations between structural genome anatomy and cancer molecular markers and targets, gene expression, gene dosage, and resistance or sensitivity to tens of thousands of molecular compounds.

Proteomic profiling of the NCI-60 cancer cell lines using new high-density reverse-phase lysate microarrays

Because most potential molecular markers and targets are proteins, proteomic profiling is expected to yield more direct answers to functional and pharmacological questions than does transcriptional profiling. To aid in such studies, we have developed a protocol for making reverse-phase protein lysate microarrays with larger numbers of spots than previously feasible. Our first application of these arrays was to profiling of the 60 human cancer cell lines (NCI-60) used by the National Cancer Institute to screen compounds for anticancer activity. Each glass slide microarray included 648 lysate spots representing the NCI-60 cell lines plus controls, each at 10 two-fold serial dilutions to provide a wide dynamic range. Mouse monoclonal antibodies and the catalyzed signal amplification system were used for immunoquantitation. The signal levels from the >30,000 data points for our first 52 antibodies were analyzed by using p-scan and a quantitative dose interpolation method. Clustered image maps revealed biologically interpretable patterns of protein expression. Among the principal early findings from these arrays were two promising pathological markers for distinguishing colon from ovarian adenocarcinomas. When we compared the patterns of protein expression with those we had obtained for the same genes at the mRNA level by using both cDNA and oligonucleotide arrays, a striking regularity appeared: cell-structure-related proteins almost invariably showed a high correlation between mRNA and protein levels across the NCI-60 cell lines, whereas non-cell-structure-related proteins showed poor correlation.

Comparing cDNA and oligonucleotide array data: concordance of gene expression across platforms for the NCI-60 cancer cells

Microarray gene-expression profiles are generally validated one gene at a time by real-time RT-PCR. A different approach is described, based on simultaneous mutual validation of large numbers of genes using two different expression-profiling platforms.

Microarray gene-expression profiles are generally validated one gene at a time by real-time RT-PCR. We describe here a different approach based on simultaneous mutual validation of large numbers of genes using two different expression-profiling platforms. The result described here for the NCI-60 cancer cell lines is a consensus set of genes that give similar profiles on spotted cDNA arrays and Affymetrix oligonucleotide chips. Global concordance is parameterized by a 'correlation of correlations' coefficient.

Human apurinic/apyrimidinic endonuclease (Ape1) and its N-terminal truncated form (AN34) are involved in DNA fragmentation during apoptosis

We previously isolated a 34-kDa nuclease (AN34) from apoptotic human leukemia cells. Here, we identify AN34 as an N-terminally truncated form of human AP endonuclease (Ape1) lacking residues 1-35 (delta35-Ape1). Although Ape1 has hitherto been considered specific for damaged DNA (specific to AP site), recombinant AN34 (delta35-Ape1) possesses significant endonuclease activity on undamaged (normal) DNA and in chromatin. AN34 also displays enhanced 3'-5' exonuclease activity. Caspase-3 activates AN34 in a cell-free system, although caspase-3 cannot cleave Ape1 directly in vitro. We also found that Ape1 itself preferentially cleaves damaged chromatin DNA isolated from cells treated with apoptotic stimuli and that silencing of Ape1 expression decreases apoptotic DNA fragmentation in DFF40/CAD-deficient cells. Thus, we propose that AN34 and Ape1 participate in the process of chromatin fragmentation during apoptosis.

Diagnostic markers that distinguish colon and ovarian adenocarcinomas: identification by genomic, proteomic, and tissue array profiling

Colon and ovarian cancers can be difficult to distinguish in the abdomen, and the distinction is important because it determines which drugs will be used for therapy. To identify molecular markers for that differential diagnosis, we developed a multistep protocol starting with the 60 human cancer cell lines used by the National Cancer Institute to screen for new anticancer agents. The steps included: (a) identification of candidate markers using cDNA microarrays; (b) verification of clone identities by resequencing; (c) corroboration of transcript levels using Affymetrix oligonucleotide chips; (d) quantitation of protein expression by "reverse-phase" protein microarray; and (e) prospective validation of candidate markers on clinical tumor sections in tissue microarrays. The two best candidates identified were villin for colon cancer cells and moesin for ovarian cancer cells. Because moesin stained stromal elements in both types of cancer, it would probably not have been identified as a marker if we had started with mRNA or protein profiling of bulk tumors. Villin appears at least as useful as the currently used colon cancer marker cytokeratin 20, and moesin also appears to have utility. The multistep process introduced here has the potential to produce additional markers for cancer diagnosis, prognosis, and therapy.

Impact of p53 knockout and topotecan treatment on gene expression profiles in human colon carcinoma cells: a pharmacogenomic study

To uncover transcriptional stress responses related to p53, we used cDNA microarrays (National Cancer Institute Oncochips comprising 6500 different genes) to characterize the gene expression profiles of wild-type p53 HCT-116 cells and an isogenic p53 knockout counterpart after treatment with topotecan, a specific topoisomerase I inhibitor. The use of the p53 knockout cells had the advantage over p53-overexpressing systems in that p53 activation is mediated physiologically. RNA was extracted after low (0.1 microM)- and high (1 microM)-dose topotecan at multiple time points within the first 6 h of treatment. To facilitate simultaneous study of the p53 status and pharmacological effects on gene expression, we developed a novel "cross-referenced network" experimental design and used multiple linear least squares fitting to optimize estimates of relative transcript levels in the network of experimental conditions. Approximately 10% of the transcripts were up- or down-regulated in response to topotecan in the p53+/+ cells, whereas only 1% of the transcripts changed in the p53-/- cells, indicating that p53 has a broad effect on the transcriptional response to this stress. Individual transcripts and their relationships were analyzed using clustered image maps and by a novel two-dimensional analysis/visualization, gene expression map, in which each gene expression level is represented as a function of both the genotypic/phenotypic difference (i.e., p53 status) and the treatment effect (i.e., of topotecan dose and time of exposure). Overall, drug-induced p53 activation was associated with a coherent genetic program leading to cell cycle arrest and apoptosis. We identified novel p53-induced and DNA damage-induced genes (the proapoptotic SIVA gene and a set of transforming growth factor beta-related genes). Genes induced independently of p53 included the antiapoptotic cFLIP gene and known stress genes related to the mitogen-activated protein kinase pathway and the Fos/Jun pathway. Genes that were negatively regulated by p53 included members of the antiapoptotic protein chaperone heat shock protein 70 family. Finally, among the p53-dependent genes whose expression was independent of drug treatment was S100A4, a small Ca(2+)-binding protein that has recently been implicated in p53 binding and regulation. The new experimental design and gene expression map analysis introduced here are applicable to a wide range of studies that encompass both treatment effects and genotypic or phenotypic differences.

Apoptotic susceptibility of cancer cells selected for camptothecin resistance: gene expression profiling, functional analysis, and molecular interaction mapping

To study the molecular mechanisms by which drug resistance develops, we compared DU145 humanprostate cancer cells with a subline selected for resistance to camptothecin. Differences in gene expression level were assessed by hybridizing the two cell types against each other using quadruplicate "Oncochip" cDNA microarrays that included 1648 cancer-related genes. Expression levels differing by a factor of >1.5 were detected for 181 of the genes. These differences were judged statistically reliable on the basis of a stratum-adjusted Kruskal-Wallis test, after taking into account a dye-dependent variable. The 181 expression-altered genes included a larger than expected number of the "apoptosis-related" genes (P = 0.04). To assess whether this observation reflected a generalized resistance of RCO.1 to apoptosis, we exposed the cells to a range of stresses (cisplatin, staurosporine, UV, ionizing radiation, and serum starvation) and found greatly reduced apoptotic responses for RC0.1 (relative to DU145) using flow cytometric Annexin V and terminal deoxynucleotidyl transferase-mediated nick end labeling assays. We next examined the apoptosis-related genes in the context of a molecular interaction map and found expression differences in the direction "expected" on the basis of the apoptosis-resistance of RC0.1 for BAD, caspase-6, and genes that signal via the Akt pathway. Exposure of the cells to wortmannin, an inhibitor of the Akt effector phosphatidylinositol 3-kinase, provided functional support for involvement of the Akt pathway. However, closer examination of the molecular interaction map revealed a paradox: many of the expression differences observed for apoptosis-related genes were in the direction "contrary" to that expected given the resistance of RC0.1. The map indicated that most of these unexpected expression differences were associated with genes involved in the nuclear factor kappa B and transforming growth factor beta pathways. Overall, the patterns that emerged suggested a two-step model for the selection process that led to resistance in RC0.1 cells. The first hypothesized step would involve a decrease in apoptotic susceptibility through changes in the apoptosis-control machinery associated with the Bcl-2 and caspase gene families, and also in antiapoptotic pathways operating through Akt/PKB. The second step would involve changes in multifunctional upstream genes (including some genes in the nuclear factor kappa B and transforming growth factor beta pathways) that can facilitate apoptosis but that would also tend to contribute to cell proliferation in the presence of drug. Thus, we propose that a downstream blockade of apoptosis was "permissive" for the selection of upstream pathway changes that would otherwise have induced apoptosis. This model is analogous to one suggested previously for the relationship between oncogene function and apoptosis in carcinogenesis.

Pharmacogenomic analysis: correlating molecular substructure classes with microarray gene expression data

Genomic studies are producing large databases of molecular information on cancers and other cell and tissue types. Hence, we have the opportunity to link these accumulating data to the drug discovery processes. Our previous efforts at 'information-intensive' molecular pharmacology have focused on the relationship between patterns of gene expression and patterns of drug activity. In the present study, we take the process a step further-relating gene expression patterns, not just to the drugs as entities, but to approximately 27,000 substructures and other chemical features within the drugs. This coupling of genomic information with structure-based data mining can be used to identify classes of compounds for which detailed experimental structure-activity studies may be fruitful. Using a systematic substructure analysis coupled with statistical correlations of compound activity with differential gene expression, we have identified two subclasses of quinones whose patterns of activity in the National Cancer Institute's 60-cell line screening panel (NCI-60) correlate strongly with the expression patterns of particular genes: (i) The growth inhibitory patterns of an electron-withdrawing subclass of benzodithiophenedione-containing compounds over the NCI-60 are highly correlated with the expression patterns of Rab7 and other melanoma-specific genes; (ii) the inhibitory patterns of indolonaphthoquinone-containing compounds are highly correlated with the expression patterns of the hematopoietic lineage-specific gene HS1 and other leukemia genes. As illustrated by these proof-of-principle examples, we introduce here a set of conceptual tools and fluent computational methods for projecting directly from gene expression patterns to drug substructures and vice versa. The analysis is presented in terms of the NCI-60 cell lines and microarray-based gene expression patterns, but the concept and methods are broadly applicable to other large-scale pharmacogenomic database sets as well. The approach (SAT for Structure-Activity-Target) provides a systematic way to mine databases for the design of further structure-activity studies, particularly to aid in target and lead identification.

Development of gene ontology tool for biological interpretation of genomic and proteomic data

We have designed and developed a Gene Ontology based navigation tool, GoMiner, which organizes lists of interesting genes from a microarray or a protein array experiment for biological interpretation. It provides quantitative and statistical output files and useful visualization (e.g., a tree-like structure) to map the list of genes to its biological functional categories. It also provides links to other resources such as pubmed, locuslink, and biological molecular interaction map and signaling pathway packages.

GoMiner: a resource for biological interpretation of genomic and proteomic data

We have developed GoMiner, a program package that organizes lists of 'interesting' genes (for example, under- and overexpressed genes from a microarray experiment) for biological interpretation in the context of the Gene Ontology. GoMiner provides quantitative and statistical output files and two useful visualizations. The first is a tree-like structure analogous to that in the AmiGO browser and the second is a compact, dynamically interactive 'directed acyclic graph'. Genes displayed in GoMiner are linked to major public bioinformatics resources.

MatchMiner: a tool for batch navigation among gene and gene product identifiers

MatchMiner is a freely available program package for batch navigation among gene and gene product identifier types commonly encountered in microarray studies and other forms of 'omic' research. The user inputs a list of gene identifiers and then uses the Merge function to find the overlap with a second list of identifiers of either the same or a different type or uses the LookUp function to find corresponding identifiers.

'Omic' and hypothesis-driven research in the molecular pharmacology of cancer

Given the layered, evolutionary complexity of biological systems, it will not be possible to understand them comprehensively on the basis of hypothesis-driven research alone. Likewise, it will not be possible to do so solely through 'omic' studies of genes, proteins and other molecules in aggregate. The two modes of research are complementary and synergistic. The twin complexities of cancer biology and drug discovery demand such synergy as a basis for the next generation of oncopharmacomic research.

Transcriptional regulation of mitotic genes by camptothecin-induced DNA damage: microarray analysis of dose- and time-dependent effects

cDNA microarray technology can be used to establish associations between characteristic gene expression patterns and molecular responses to drug therapy. In this study, we used cDNA microarrays of 1694 cancer-related genes to monitor the gene expression consequences of the treatment of HCT116 colon cancer cells with the topoisomerase I inhibitor camptothecin (CPT). To obtain a more homogeneous cellular response, we synchronized the cells in S-phase using aphidicolin (APH) before CPT treatment. Brief incubation with 20 and 1000 nM CPT caused reversible and irreversible G(2) arrest, respectively, and the patterns of gene expression change (with reference to untreated controls) were strikingly different at the two concentrations. Thirty-three genes, mainly divided into three groups, showed characteristic changes in the first 20 h as a consequence of treatment. Northern blots performed for five of these genes (each under eight experimental conditions) were quite consistent with the microarray results (average correlation coefficient, 0.86). Several p53-activated stress response genes were up-regulated after treatment with 1000 nM CPT or prolonged exposure to APH, but it seemed that the up-regulation did not directly cause cell cycle arrest because the up-regulation induced by prolonged treatment with APH did not prevent cell cycle progression after removal of APH. In contrast, cell cycle-dependent up-regulation of a group of mitosis-related genes was delayed or blocked after CPT treatments. The interrupted up-regulation of this group of genes was directly associated with G(2) arrest. In addition, we observed down-regulation of gene expression in cells that were recovering from cell cycle delay. The observations reported here suggest a fundamental difference at the gene expression level between the molecular mechanism of reversible G(2) delay that follows mild DNA damage and the mechanism of permanent G(2) arrest that follows more extensive DNA damage.

Quantification of neural tissue injury in a rat radiculopathy model: comparison of local deformation, behavioral outcomes, and spinal cytokine mRNA for two surgeons

Clinical and experimental work indicate that a variety of factors contribute to radicular pain mechanisms, including mechanical injury. While it has been qualitatively suggested that the magnitude of nerve root mechanical injury affects the nature of the pain response, no study has quantified the local in vivo injury biomechanics in these models. Therefore, it was the purpose of this study to develop and implement an in vivo method to quantify compressive nerve root injury strain severity and characterize its effect on the resulting responses in an existing lumbar radiculopathy rat model. Male Holtzman rats were divided into a sham group with only nerve root exposure or a ligation group with the nerve root tightly ligated using silk suture. Using image analysis, nerve root radial strains were calculated at the time of injury for two surgeons. Mechanical allodynia was continuously assessed throughout the study and spinal cord cytokine mRNA levels were assayed on postoperative day 7. The degree of intersurgeon variability for imposing a ligation injury in this model was also assessed. Mean compressive injury strains in the nerve root were 32.8+/-14.2% and were not different for the two experimenters. Animals undergoing more severe ligation strains exhibited significantly heightened allodynia following injury and greater upregulation of the inflammatory cytokines IL-1alpha/beta, IL-6, and IL-10. Results indicate a direct correlation of local nerve root injury severity with the ensuing physiologic responses associated with nociception.

Repeated injury to the lumbar nerve roots produces enhanced mechanical allodynia and persistent spinal neuroinflammation

STUDY DESIGN

A lumbar radiculopathy model investigated pain behavioral responses after nerve root reinjury.

OBJECTIVES

To gain a further understanding of central sensitization and neuroinflammation associated with chronic lumbar radiculopathy after repeated nerve root injury.

SUMMARY OF BACKGROUND DATA

The pathophysiologic mechanisms associated with chronic radicular pain remain obscure. It has been hypothesized that lumbar root injury produces neuroimmunologic and neurochemical changes, sensitizing the spinal cord and causing pain responses to manifest with greater intensity and longer duration after reinjury. However, this remains untested experimentally.

METHODS

Male Holtzman rats were divided into two groups: a sham group having only nerve root exposure, and a chromic group in which the nerve root was ligated loosely with chromic gut suture. Animals underwent a second procedure at 42 days. The chromic group was further divided into a reinjury group and a chromic-sham group, in which the lumbar roots were only re-exposed. Bilateral mechanical allodynia was continuously assessed throughout the study. Qualitative assessment of spinal cord glial activation and IL-beta expression was performed.

RESULTS

Mechanical allodynia was significantly greater on both the ipsilateral and contralateral sides after reinjury (P < 0.001), and the response did not return to baseline after reinjury, as it did with the initial injury. There were also persistent spinal astrocytic and microglial activation and interleukin-1beta expression.

CONCLUSIONS

The bilateral responses support central modulation of radicular pain after nerve root injury. An exaggerated and more prolonged response bilaterally after reinjury suggests central sensitization after initial injury. Neuroinflammatory activation in the spinal cord further supports the hypothesis that central neuroinflammation plays an important role in chronic radicular pain.

Quantitative structure-antitumor activity relationships of camptothecin analogues: cluster analysis and genetic algorithm-based studies

Topoisomerase 1 (top1) inhibitors are proving useful against a range of refractory tumors, and there is considerable interest in the development of additional top1 agents. Despite crystallographic studies, the binding site and ligand properties that lead to activity are poorly understood. Here we report a unique approach to quantitative structure-activity relationship (QSAR) analysis based on the National Cancer Institute's (NCI) drug databases. In 1990, the NCI established a drug discovery program in which compounds are tested for their ability to inhibit the growth of 60 different human cancer cell lines in culture. More than 70 000 compounds have been screened, and patterns of activity against the 60 cell lines have been found to encode rich information on mechanisms of drug action and drug resistance. Here, we use hierarchical clustering to define antitumor activity patterns in a data set of 167 tested camptothecins (CPTs) in the NCI drug database. The average pairwise Pearson correlation coefficient between activity patterns for the CPT set was 0.70. Coherence between chemical structures and their activity patterns was observed. QSAR studies were carried out using the mean 50% growth inhibitory concentrations (GI(50)) for 60 cell lines as the dependent variables. Different statistical methods, including stepwise linear regression, principal component regression (PCR), partial least-squares regression (PLS), and fully cross-validated genetic function approximation (GFA) were applied to construct quantitative structure-antitumor relationship models. For our data set, the GFA method performed better in terms of correlation coefficients and cross-validation analysis. A number of molecular descriptors were identified as being correlated with antitumor activity. Included were partial atomic charges and three interatomic distances that define the relative spatial dispositions of three significant atoms (the hydroxyl hydrogen of the E-ring, the lactone carbonyl oxygen of the E-ring, and the carbonyl oxygen of the D-ring). The cross-validated r(2) for the final GFA model was 0.783, indicating a predictive QSAR model.

Chemosensitivity prediction by transcriptional profiling

In an effort to develop a genomics-based approach to the prediction of drug response, we have developed an algorithm for classification of cell line chemosensitivity based on gene expression profiles alone. Using oligonucleotide microarrays, the expression levels of 6,817 genes were measured in a panel of 60 human cancer cell lines (the NCI-60) for which the chemosensitivity profiles of thousands of chemical compounds have been determined. We sought to determine whether the gene expression signatures of untreated cells were sufficient for the prediction of chemosensitivity. Gene expression-based classifiers of sensitivity or resistance for 232 compounds were generated and then evaluated on independent sets of data. The classifiers were designed to be independent of the cells' tissue of origin. The accuracy of chemosensitivity prediction was considerably better than would be expected by chance. Eighty-eight of 232 expression-based classifiers performed accurately (with P < 0.05) on an independent test set, whereas only 12 of the 232 would be expected to do so by chance. These results suggest that at least for a subset of compounds genomic approaches to chemosensitivity prediction are feasible.