Automated Literature Mining for Drug–Gene Relationships Compares Favorably with Curated Databases
The identification of genes important in determining response to drug treatment is a central concept in pharmacogenomics, and there are several approaches to this task. One method, developed by Professor Russ Altman (Stanford University, CA, USA) was the generation of an algorithm for the automated genome-scale ranking of potential gene–drug interactions.
Despite being an overall success, this method relies upon a carefully maintained and structured database of drug–gene interactions and associated information, which is both costly and time consuming to keep ordered and current. Although a useful tool, the significant generation and maintenance costs have meant that it is still weighed on the same scale as traditional experimental methods for the investigation of gene–drug interactions. While it avoided the problems of frequent false positives (high-throughput screening), extremely costly set-up (candidate gene lists), or the potential duplication of pre-existing experimental data (both methods), the reliance of the algorithm-database method meant that there was still potential for improvement.
With his latest project, Professor Altman has attempted to remove that final dependency by creating a data-gathering tool for use with the genome-scale drug–gene interaction ranking algorithm mentioned above. This enables the use of raw scientific literature as an input, automatically mining texts for the joint mention of genes and drugs.
Such an approach is not without its risks – relying solely on rapid, automated searches necessarily limits the discrimination which can be applied to associations, and clearly such an algorithm cannot bring to bear any of the depth of knowledge that a human expert could. Still, as the process is far faster and cheaper than employing human experts, it appears to be an ideal method for preliminary examinations, and for cases in which there is insufficient evidence to warrant the expenditure of resources on expensive testing.
In terms of performance relative to the curated-database algorithm, the text-mining shapes up well. In the words of a paper by Professor Altman and colleagues, “a knowledge base derived from text-mining the literature performs as well as, and sometimes better than, a high-quality, manually curated knowledge base.”
It should be noted, however, that this text-mining utility is very much dependent on other methods. Being a tool to search the results of previous experimental assays, it is likely to be very helpful in minimizing the repetition of previous drug–gene interaction analysis, but it certainly cannot replace such analyses as a primary testing method. Rather, it offers a means by which the targets of such resource-intensive assays can be better selected. Without a large base of experimental investigation (and shared results), there would be nothing to search!
Nevertheless, given the cost of the alternatives at present, this development seems both timely and valuable. By allowing an initial low-cost examination of previous efforts in an area, it may well lower the barrier to beginning drug–gene association studies, potentially allowing interest and backing to be attracted without a substantial initial investment.
Source: Garten Y, Tatonetti NP, Altman RB: Improving the prediction of pharmacogenes using text-derived drug–gene relationships. Pac. Symp. Biocomput. 305–314 (2010)
Gene-Network Analysis Reveals Mechanism of Oxidative Stress-Induced Drug Sensitivity in Ovarian Cancer Cells
Resistance to drug therapy can be a major problem in almost all types of cancer, mediated as it is by complicated sets of genetic and molecular pathways. In ovarian carcinoma, tumor cells will usually become resistant to single-agent chemotherapy before complete eradication of the cancer is achieved, necessitating combined therapy, or a switch to an alternative treatment. In the case of the alkylating agent chlorambucil (Cbl), up to tenfold increases in resistance have been observed. Along with this resistance, a decrease in endogenous reactive oxygen species (ROS) has been previously noted, as has the tendency for changes in ROS levels (via antioxidants or exogenous ROS generation, decreasing and increasing ROS levels, respectively) to cause a similar variation in Cbl sensitivity among these resistant cells.
In light of this, the research of Professor Amit Maiti (Oklahoma Medical Research Foundation, OK, USA) into the genetic basis of this method for overcoming Cbl resistance may well be very useful in developing treatments that better exploit the weakness of ovarian carcinoma cells to elevated ROS levels in combination with this drug. Using a microarray to profile gene expression followed by gene network analysis, Professor Maiti discovered that ARHGEF6, a rho guanine nuclear exchange factor, played a major role in Cbl/oxidative stress-related apoptosis, along with p53 and the catalytic subunit of DNA-dependent protein kinase (DNA-PKc). Professor Maiti concludes: “These results suggest that low doses of Cbl and very low doses of chronic oxidative stress together kill Cbl-resistant ovarian carcinoma cells”.
As well as identifying the key interactions in the apoptosis process, this analysis also uncovered differences in interaction location. Under normal conditions, DNA-PKc interacted with ARHGEF6 and p53 in the nucleus of examined cells, but when exposed to chronic oxidative stress this interaction occurred more in the cytoplasm. This difference may well have a bearing on the greatly increased sensitivity to chlorambucil found in such stressed ovarian carcinoma cells.
Source: Maiti AK: Gene network analysis of oxidative stress-mediated drug sensitivity in resistant ovarian carcinoma cells. Pharmacogenomics J. (2009) (Epub ahead of print).
Antipsychotic-Related Weight Gain in Schizophrenia Patients Associated with ADRA1A SNPs
The association between treatment with antipsychotics and increase in BMI is well known, as is the list of likely candidate genes for causing this behavior. A recent study headed by Dr Hsien-Jane Chiu (Jianan Mental Hospital, Tainan, Taiwan) has tested for 58 candidate SNPs of the α-1A adrenergic receptor gene (ADRA1A) and, following statistical analysis, isolated 44 SNPs that were associated with BMI in a population of patients treated with antipsychotics for schizophrenia.
The study examined a population of 401 patients with schizophrenia who had been treated with antipsychotics for more than 2 years, of whom 394 were genotyped for the 58 candidate ADRA1A SNPs. The BMI of these patients was recorded over the course of a year, and then subjected to statistical analysis in association with the SNPs, noting patient age, gender, diabetes status and type of antipsychotic used (typical vs atypical) as covariates.
A total of 44 minor allele-frequency ADRA1A SNPs were found to be associated with BMI in these patients, the majority of which were located in the gene promoter and intron regions of the gene. As the paper put it “our study suggests that the ADRA1A gene is involved in weight gain among schizophrenia patients treated with antipsychotics.”
Interestingly, a gender effect was observed on the ADRA1A/BMI increase association, with female patients experiencing an enhanced gene effect.
Weight gain associated with antipsychotic treatment is a serious issue, for both clinical (in terms of relation to the development of Type 2 diabetes), and patient-perception purposes (i.e., treatment adherence).
Now that part of the molecular mechanism behind this effect has been demonstrated (rather than just suspected), it is clear that, as the study concludes “further molecular dissection of the ADRA1A gene warrants better understanding on weight gain mechanisms in schizophrenia.”
Source: Liu YR, Loh EW, Lan TH et al.: ADRA1A gene is associated with BMI in chronic schizophrenia patients exposed to antipsychotics. Pharmacogenomics J. (2009) (Epub ahead of print).
CYP2D6 Variants Associated with Clinical Outcomes for Tamoxifen-Treated Women
Tamoxifen, a drug used as an adjuvant treatment for breast cancer, has long been known for its variable therapeutic effects (and side effects). A recent study headed by Professor Hiltrud Brauch, with Werner Schroth as first author (Dr Margarete Fischer-Bosch-Institute of Clinical Pharmacology, Stuttgart, Germany), claims to have uncovered an association between cytochrome P450 2D6 (CYP2D6) polymorphisms and clinical outcome in these patients.
CYP2D6 is the enzyme responsible for the conversion of tamoxifen into its active metabolites, 4-hydroxytamoxifen and endoxifen. It has been shown that polymorphisms in this enzyme can cause differing rates of metabolism, leading to interindividual variation in plasma concentrations of endoxifen assuming a consistent dose.
There are many known genetic variants of CYP2D6, but these translate into only four enzyme phenotypes, ranging from ultrarapid (high metabolic activity) to extensive (normal), intermediate, and poor (no enzyme activity). In terms of activity, both gene type and copy number are relevant, with high- or low-metabolism alleles having an additive effect on CYP2D6 activity.
Given this interindividual variation in tamoxifen metabolism and subsequent plasma concentrations, it would be reasonable to expect that such variation also leads to differing clinical outcomes for patients prescribed tamoxifen as a treatment for breast cancer. In order to test this, the recent study retrospectively examined a population of 1325 adjuvant tamoxifen-treated patients diagnosed with stage I–III breast cancer between 1986 and 2005, taking into account CYP2D6 metabolizer status, and the outcomes time to recurrence, event-free survival, disease-free survival and overall survival.
In terms of recurrence rates, those for extensive metabolizer were the lowest at 15%, rising to 21% for heterozygous extensive/intermediate metabolizer, and 29% for poor metabolizer. Relative to extensive metabolizer, decreased CYP2D6 activity of any kind led to 33% worse event-free survival and 29% disease-free survival, but not to significantly different overall survival.
As summarized in the paper, “among women with breast cancer treated with tamoxifen … the presence of 2 functional CYP2D6 alleles was associated with better clinical outcomes and the presence of nonfunctional or reduced-function alleles with worse outcomes.”
Since 7–10% of Caucasians are poor tamoxifen metabolizers, they would therefore be expected to benefit much less from treatment with tamoxifen, and have indeed been shown by this study to have a 90% increased risk of breast cancer recurrence over a 9-year period when compared with extensive metabolizer, therefore, there is a case to be made for genotyping potential tamoxifen recipients for CYP2D6 alleles. This would allow the identification of these poor metabolizers at an early stage so that they could be assigned to an alternative course of therapy. This could both increase their chances of event-free survival and reduce the inefficient use of resources on courses of tamoxifen treatment with little therapeutic effect. Given the recent interest in the side effects of breast cancer drugs, information that could be used to predict a patient‘s response to such therapy might aid the decision-making process with regard to their use in direct therapy or as risk-reduction treatments.
Source: Schroth W, Goetz MP, Hamann U et al.: Association between CYP2D6 polymorphisms and outcomes among women with early stage breast cancer treated with tamoxifen. JAMA 302(13), 1429–1436 (2009).