The impact of pharmacokinetic gene profiles across human cancers

Abstract

Background: The right drug to the right patient at the right time is one of the ideals of Individualized Medicine (IM) and remains one of the most compelling promises of the post-genomic age. The addition of genomic information is expected to increase the precision of an individual patient's treatment, resulting in improved outcomes. While pilot studies have been encouraging, key aspects of interpreting tumor genomics information, such as somatic activation of drug transport or metabolism, have not been systematically evaluated.

Methods: In this work, we developed a simple rule-based approach to classify the therapies administered to each patient from The Cancer Genome Atlas PanCancer dataset (n = 2858) as effective or ineffective. Our Therapy Efficacy model used each patient's drug target and pharmacokinetic (PK) gene expression profile; the specific genes considered for each patient depended on the therapies they received. Patients who received predictably ineffective therapies were considered at high-risk of cancer-related mortality and those who did not receive ineffective therapies were considered at low-risk. The utility of our Therapy Efficacy model was assessed using per-cancer and pan-cancer differential survival.

Results: Our simple rule-based Therapy Efficacy model classified 143 (5%) patients as high-risk. High-risk patients had age ranges comparable to low-risk patients of the same cancer type and tended to be later stage and higher grade (odds ratios of 1.6 and 1.4, respectively). A significant pan-cancer association was identified between predictions of our Therapy Efficacy model and poorer overall survival (hazard ratio, HR = 1.47, p = 6.3 × 10^{- 3}). Individually, drug export (HR = 1.49, p = 4.70 × 10^{- 3}) and drug metabolism (HR = 1.73, p = 9.30 × 10^{- 5}) genes demonstrated significant survival associations. Survival associations for target gene expression are mechanism-dependent. Similar results were observed for event-free survival.

Conclusions: While the resolution of clinical information within the dataset is not ideal, and modeling the relative contribution of each gene to the activity of each therapy remains a challenge, our approach demonstrates that somatic PK alterations should be integrated into the interpretation of somatic transcriptomic profiles as they likely have a significant impact on the survival of specific patients. We believe that this approach will aid the prospective design of personalized therapeutic strategies.

Keywords: Cancer treatment protocols; Genomic interpretation; Individualized medicine; Pharmacokinetics; Transcriptome profiling.

Fig. 1

Human tumors may up- or down-regulate PK genes. Each gene was scored relative to a composite-normal reference to generate conservative estimates of aberrant somatic gene expression. RSEM normalized gene expression data were used. The expression score of each gene in each tumor sample is the signed Z-score relative to normal tissue samples. Example probability density distributions of gene expression for two genes are shown: (Left) drug importer SLC16A2 and (Right) drug exporter ABCC5

Fig. 2

Clinical decision making is augmented by interpreted tumor genomics. a The clinical decision making cycle of therapy selection is multifaceted and can be informed by properly interpreted tumor genomics profiling. Developing high-confidence and mechanism-based algorithms for properly interpreting genomic information remains a clinical challenge. b We considered a set of rules based on tumor genomics features for interpreting somatic PK gene expression. From these rules, we developed a simple Therapy Efficacy model for interpreting if an administered therapy may be ineffective due to somatic PK gene expression

Fig. 3

Our Therapy Efficacy model identified high-risk patients who received therapies with specific PK mechanisms activated within their tumors. a We first plot the Kaplan-Meyer survival curves of the pan-cancer cohort classified by our genomics-based efficacy model based on high expression of the exporters or metabolizers of the drugs administered. Shaded bands indicate 95% confidence intervals. Analogous per-cancer survival curves are shown for (b) KIRC and (d) OV. In this retrospective study, cohorts were not balanced with respect to disease state or clinical characteristics (Additional file 1: Figure S1). Thus, we plot survival curves adjusted to a uniform cohort of 50 year olds with stage-3 grade-3 (c) KIRC or (e) OV. Analogous plots are shown for all cancer types in Additional file 1: Figure S3. We assessed statistical significance using Cox regression with results shown in Table 3 and Figs. 4, 5 and 6

Fig. 4

High expression of drug export genes for administered therapies is associated with poorer overall survival. a Our pan-cancer analysis revealed a statistically significant association between high expression levels of genes known to export therapies administered to each patient and overall survival. Pan-cancer cohort size is indicated (N) along with the number of patients affected (M) Varying the expression threshold used identified a tradeoff between the number of patients affected and HR magnitude. A distribution diagram highlights the analogous region considered for each model. Models are summarized by p-value, HR, and bounds of the 95% confidence interval. Meta-analyses are summarized by a diamond centered on the HR and its width extending to the confidence interval bounds. b For the Z ≥ 2 criteria, per-cancer models are summarized in a Forest plot. For each cancer type, the HR is marked and scaled by M; a line extends to the confidence interval bounds

Fig. 5

Low expression of drug target genes for administered therapies is associated with patient survival in a mechanism-dependent manner. For specific therapies and per-cancer, examples with at least five patients in each group are shown. Data are presented as in previous figures; N, the number of patients administered the therapy; M, the number of patients with low levels of at least one of the therapy’s targets

Fig. 6

Genes involved in drug metabolism of one therapy may be the target of another therapy. High drug metabolism gene expression in BRCA was protective. This is the opposite association as expected and observed in other cancer types. The upper panel shows the survival association with high expression levels of drug metabolizing genes in BRCA, across a series of threshold values. Analogous data for UCEC is shown below, for comparison. We believe this association can be explained by interactions between the molecular mechanisms of cytotoxic therapies and anti-hormone therapies; see Discussion