The flaw of averages: Bayes factors as posterior means of the likelihood ratio.

As an alternative to the Frequentist p-value, the Bayes factor (or ratio of marginal likelihoods) has been regarded as one of the primary tools for Bayesian hypothesis testing. In recent years, several researchers have begun to re-analyze results from prominent medical journals, as well as from trials for FDA-approved drugs, to show that Bayes factors often give divergent conclusions from those of p-values. In this paper, we investigate the claim that Bayes factors are straightforward to interpret as directly quantifying the relative strength of evidence. In particular, we show that for nested hypotheses with consistent priors, the Bayes factor for the null over the alternative hypothesis is the posterior mean of the likelihood ratio. By re-analyzing 39 results previously published in the New England Journal of Medicine, we demonstrate how the posterior distribution of the likelihood ratio can be computed and visualized, providing useful information beyond the posterior mean alone.

Epistemic uncertainty in Bayesian predictive probabilities.

Bayesian predictive probabilities have become a ubiquitous tool for design and monitoring of clinical trials. The typical procedure is to average predictive probabilities over the prior or posterior distributions. In this paper, we highlight the limitations of relying solely on averaging, and propose the reporting of intervals or quantiles for the predictive probabilities. These intervals formalize the intuition that uncertainty decreases with more information. We present four different applications (Phase 1 dose escalation, early stopping for futility, sample size re-estimation, and assurance/probability of success) to demonstrate the practicality and generality of the proposed approach.

Sample size formula for a win ratio endpoint.

The win ratio composite endpoint, which organizes the components of the composite hierarchically, is becoming popular in late-stage clinical trials. The method involves comparing data in a pair-wise manner starting with the endpoint highest in priority (eg, cardiovascular death). If the comparison is a tie, the endpoint next highest in priority (eg, hospitalizations for heart failure) is compared, and so on. Its sample size is usually calculated through complex simulations because there does not exist in the literature a simple sample size formula. This article provides a formula that depends on the probability that a randomly selected patient from one group does better than a randomly selected patient from another group, and on the probability of a tie. We compare the published 95% confidence intervals, which require patient-level data, with that calculated from the formula, requiring only summary-level data, for 17 composite or single win ratio endpoints. The two sets of results are similar. Simulations show the sample size formula performs well. The formula provides important insights. It shows when adding an endpoint to the hierarchy can increase power even if the added endpoint has low power by itself. It provides relevant information to modify an on-going blinded trial if necessary. The formula allows a non-specialist to quickly determine the size of the trial with a win ratio endpoint whose use is expected to increase over time.

Relationship between tumor biomarkers and efficacy in TH3RESA, a phase III study of trastuzumab emtansine (T-DM1) vs. treatment of physician's choice in previously treated HER2-positive advanced breast cancer.

In the phase III TH3RESA study (NCT01419197), 602 patients with HER2-positive advanced breast cancer who received prior taxane therapy and ≥2 HER2-directed regimens, including trastuzumab and lapatinib (advanced setting), were randomized to trastuzumab emtansine (T-DM1) or treatment of physician's choice (TPC). A statistically significant progression-free survival (PFS) benefit favoring T-DM1 was demonstrated. Here, we examine the relationship between HER2-related biomarkers and PFS in an exploratory analysis. Biomarkers assessed included HER2 (n = 505) and HER3 (n = 505) mRNA expression, PIK3CA mutation status (n = 410) and PTEN protein expression (n = 358). For biomarkers with continuous data (HER2, HER3, PTEN), subgroups were defined using median values (>median and ≤median). For all biomarker subgroups, median PFS was longer with T-DM1 vs. TPC. The PFS benefit favoring T-DM1 vs. TPC was numerically greater in the HER2 mRNA >median subgroup (7.2 vs. 3.4 months; unstratified hazard ratio [HR], 0.40; 95% CI, 0.28-0.59; p < 0.0001) vs. ≤median subgroup (5.5 vs. 3.9 months; HR, 0.68; 95% CI, 0.49-0.92; p = 0.0131). The PFS benefit with T-DM1 was similar among HER3, PIK3CA and PTEN subgroups. Consistent with other reports, benefit was seen with T-DM1 regardless of PIK3CA mutation status. In a multivariate analysis including an interaction term (treatment group by log2-transformed HER2 mRNA), patients with higher HER2 mRNA levels benefited more from receiving T-DM1 (HR, 0.84; 95% CI, 0.75-0.94; interaction p value = 0.0027). In summary, T-DM1 prolonged median PFS in all biomarker subgroups analyzed, including activating PIK3CA mutations, with numerically greater benefit in patients with tumors expressing HER2 mRNA >median vs. ≤median.

A Phase I Dose-Escalation Study Evaluating the Safety Tolerability and Pharmacokinetics of CUDC-427, a Potent, Oral, Monovalent IAP Antagonist, in Patients with Refractory Solid Tumors.

PURPOSE: To determine the dose-limiting toxicities (DLT), adverse events (AE), pharmacokinetics, and preliminary evidence of antitumor activity of CUDC-427 (formerly GDC-0917), a selective antagonist of inhibitor of apoptosis (IAP) proteins. EXPERIMENTAL DESIGN: Patients with advanced solid malignancies were treated with escalating doses of CUDC-427 orally on a daily 14-day on/7-day off schedule in 21-day cycles using a modified continuous reassessment method design. Blood samples were assayed to determine the pharmacokinetic properties, pharmacodynamic alterations of cellular IAP levels in peripheral blood mononuclear cells (PBMC), and monocyte chemoattractant protein-1 (MCP-1) levels. RESULTS: Forty-two patients received 119 cycles of CUDC-427. Overall, the most common treatment-related toxicities were fatigue, nausea, vomiting, and rash. One DLT (grade 3 fatigue) occurred in a patient at 450 mg dose level during cycle 1, and 5 patients experienced AEs related to CUDC-427 that led to discontinuation and included grade 3 pruritus, and fatigue, and grade 2 drug hypersensitivity, pneumonitis, rash, and QT prolongation. The maximum planned dose of 600 mg orally daily for 2 weeks was reached, which allometrically scaled to exceed the IC90 in preclinical xenograft studies. Significant decreases in cIAP-1 levels in PBMCs were observed in all patients 6 hours after initial dosing. Responses included durable complete responses in one patient with ovarian cancer and one patient with MALT lymphoma. CONCLUSIONS: CUDC-427 can be administered safely at doses up to 600 mg daily for 14 days every 3 weeks. The absence of severe toxicities, inhibition of cIAP-1 in PBMC, and antitumor activity warrant further studies. Clin Cancer Res; 22(18); 4567-73. ©2016 AACR.

Blinding properties of methods for supplying drug kits to investigational sites.

Most blinded, late stage, randomized clinical trials package study drug, active or placebo, into drug kits for distribution to investigational sites. Drug kits enable investigators to administer study drug to subjects in a blinded manner without the assistance of an unblinded pharmacist. Supply methods determine when and how many kits to send to sites. If not properly designed, these methods can partially unblind investigators, i.e., investigators can conclude that two subjects are (1) on the same treatment arm with certainty or (2) on different treatment arms with certainty. Partial unblinding can bias the way investigators provide patient care, report adverse events and assess efficacy endpoints, and can lead to full unblinding when the other subject is unblinded. In this paper, we describe several examples of partial unblinding in the supply methods commonly used by many Randomization and Trial Supply Management (RTSM) systems, propose a new criterion for evaluating the blinding properties of supply methods, and prove that two alternative supply methods do not permit full or partial unblinding, even after the investigator is unblinded up to a certain number of other subjects.

The other randomization - methods for labeling drug kits.

PURPOSE: Most phases 2 and 3 blinded randomized clinical trials package study drug, e.g., active and placebo, into drug kits for distribution to investigational sites. Kits are made so that it is not possible to determine the type of drug in the kit. This enables investigators to administer drug to patients in a manner that blinds investigators and patients. Kits are labeled with unique kit IDs that code for the drug type. Kit lists contain the assignment of kit IDs to drug. Algorithms for making kit lists, like algorithms for randomizing patients, incorporate randomness to ensure investigators and patients are blind to the process. This paper reviews three types of kit list: blocked, double randomized, and scrambled, and discusses the operational benefits of what a pharmaceutical company might obtain using scrambled lists along with an overview of the challenges of generating and extending the lists for large trials. METHODS: We reviewed the operational characteristics of three types of kit list: blocked, double randomized, and scrambled. RESULTS: Blocked kit lists were a popular choice until their unblinding and operational deficiencies became well known. The many difficulties associated with blocked kit lists are unnecessary and can be avoided by using a double randomized kit list or a scrambled kit list. Compared to double randomized kit lists, scrambled kit lists can be more easily extended due to their advantage of having randomly sized gaps between kit IDs. CONCLUSIONS: Among the three types of kit list, scrambled kit lists offer the most flexibility. The adoption of scrambled kit lists has in practice provided the many operational benefits described in this paper.

Trastuzumab emtansine versus treatment of physician's choice for pretreated HER2-positive advanced breast cancer (TH3RESA): a randomised, open-label, phase 3 trial.

BACKGROUND: Patients with progressive disease after two or more HER2-directed regimens for recurrent or metastatic breast cancer have few effective therapeutic options. We aimed to compare trastuzumab emtansine, an antibody-drug conjugate comprising the cytotoxic agent DM1 linked to trastuzumab, with treatment of physician's choice in this population of patients. METHODS: This randomised, open-label, phase 3 trial took place in medical centres in 22 countries across Europe, North America, South America, and Asia-Pacific. Eligible patients (≥18 years, left ventricular ejection fraction ≥50%, Eastern Cooperative Oncology Group performance status 0-2) with progressive HER2-positive advanced breast cancer who had received two or more HER2-directed regimens in the advanced setting, including trastuzumab and lapatinib, and previous taxane therapy in any setting, were randomly assigned (in a 2:1 ratio) to trastuzumab emtansine (3·6 mg/kg intravenously every 21 days) or physician's choice using a permuted block randomisation scheme by an interactive voice and web response system. Patients were stratified according to world region (USA vs western Europe vs other), number of previous regimens (excluding single-agent hormonal therapy) for the treatment of advanced disease (two to three vs more than three), and presence of visceral disease (any vs none). Coprimary endpoints were investigator-assessed progression-free survival (PFS) and overall survival in the intention-to-treat population. We report the final PFS analysis and the first interim overall survival analysis. This study is registered with ClinicalTrials.gov, number NCT01419197. FINDINGS: From Sept 14, 2011, to Nov 19, 2012, 602 patients were randomly assigned (404 to trastuzumab emtansine and 198 to physician's choice). At data cutoff (Feb 11, 2013), 44 patients assigned to physician's choice had crossed over to trastuzumab emtansine. After a median follow-up of 7·2 months (IQR 5·0-10·1 months) in the trastuzumab emtansine group and 6·5 months (IQR 4·1-9·7) in the physician's choice group, 219 (54%) patients in the trastuzumab emtansine group and 129 (65%) of patients in the physician's choice group had PFS events. PFS was significantly improved with trastuzumab emtansine compared with physician's choice (median 6·2 months [95% CI 5·59-6·87] vs 3·3 months [2·89-4·14]; stratified hazard ratio [HR] 0·528 [0·422-0·661]; p<0·0001). Interim overall survival analysis showed a trend favouring trastuzumab emtansine (stratified HR 0·552 [95% CI 0·369-0·826]; p=0·0034), but the stopping boundary was not crossed. A lower incidence of grade 3 or worse adverse events was reported with trastuzumab emtansine than with physician's choice (130 events [32%] in 403 patients vs 80 events [43%] in 184 patients). Neutropenia (ten [2%] vs 29 [16%]), diarrhoea (three [<1%] vs eight [4%]), and febrile neutropenia (one [<1%] vs seven [4%]) were grade 3 or worse adverse events that were more common in the physician's choice group than in the trastuzumab emtansine group. Thrombocytopenia (19 [5%] vs three [2%]) was the grade 3 or worse adverse event that was more common in the trastuzumab emtansine group. 74 (18%) patients in the trastuzumab emtansine group and 38 (21%) in the physician's choice group reported a serious adverse event. INTERPRETATION: Trastuzumab emtansine should be considered as a new standard for patients with HER2-positive advanced breast cancer who have previously received trastuzumab and lapatinib. FUNDING: Genentech.

A Phase Ib study evaluating MNRP1685A, a fully human anti-NRP1 monoclonal antibody, in combination with bevacizumab and paclitaxel in patients with advanced solid tumors.

PURPOSE: MNRP1685A is a human monoclonal antibody that blocks binding of vascular endothelial growth factor (VEGF), VEGF-B, and placental growth factor 2 to neuropilin-1 resulting in vessel immaturity and VEGF dependency. The safety of combining MNRP1685A with bevacizumab, with or without paclitaxel, was examined. METHODS: Patients with advanced solid tumors received escalating doses of MNRP1685A (7.5, 15, 24, and 36 mg/kg) with bevacizumab 15 mg/kg every 3 weeks in Arm A (n = 14). Arm B (n = 10) dosing consisted of MNRP1685A (12 and 16 mg/kg) with bevacizumab 10 mg/kg (every 2 weeks) and paclitaxel 90 mg/m(2) (weekly, 3 of 4 weeks). Objectives were to determine safety, pharmacokinetics, pharmacodynamics, and the maximum tolerated dose of MNRP1685A. RESULTS: Infusion reactions (88 %) and transient thrombocytopenia (67 %) represent the most frequent study drug-related adverse events (AEs). Drug-related Grade 2 or 3 proteinuria occurred in 13 patients (54 %). Additional study drug-related AEs occurring in >20 % of patients included neutropenia, alopecia, dysphonia, fatigue, and nausea. Neutropenia occurred only in Arm B. Grade ≥3 study drug-related AEs in ≥3 patients included neutropenia (Arm B), proteinuria, and thrombocytopenia. Two confirmed and three unconfirmed partial responses were observed. CONCLUSIONS: The safety profiles were consistent with the single-agent profiles of all study drugs. However, a higher than expected rate of clinically significant proteinuria was observed that does not support further testing of MNRP1685A in combination with bevacizumab.

A phase I study of the human monoclonal anti-NRP1 antibody MNRP1685A in patients with advanced solid tumors.

The human monoclonal antibody MNRP1685A targets the VEGF binding domain of neuropilin-1 (NRP1), a multi-domain receptor necessary for neural development and blood vessel maturation. In nonclinical studies, MNRP1685A prevents vascular maturation by keeping blood vessels in an immature, highly VEGF-dependent state. We explored the safety and tolerability of MNRP1685A in patients with advanced solid tumors. Patients were treated with MNRP1685A given intravenously every 3 weeks using a 3 + 3 dose-escalation design with 7 dose-escalation cohorts. Twenty-four of 35 patients (69 %) experienced drug-related adverse events (AEs) of infusion-related reaction on the day of MNRP1685A administration. With premedication including dexamethasone, infusions were well-tolerated with main symptoms of pruritus and rash. Outside the day of infusion, most common (≥ 2 patients) related AEs were fatigue (17 %), pruritus (9 %), myalgia and thrombocytopenia (both 6 %) (all were Grade 1-2). MNRP1685A-related Grade ≥ 3 AEs consisted of one dose-limiting toxicity of Grade 3 upper gastrointestinal bleeding and one related Grade 3 thrombocytopenia, coinciding with unrelated Grade 3 fungemia and duodenal obstruction. MNRP1685A showed nonlinear PK with more-than-dose proportional increases in exposure, consistent with broad target expression. Transient platelet count reductions (≥ 30 % from predose) were observed in 56 % of evaluable patients. Nine patients were on study for ≥ 4 cycles, one colorectal cancer patient for one year. MNRP1685A was generally well-tolerated. The primary MNRP1685A-related AE was infusion-related reaction, which were attenuated by premedication including dexamethasone. Transient platelet count reductions were frequent but did not impact MNRP1685A dosing.

Dogs are more sensitive to antagonists of inhibitor of apoptosis proteins than rats and humans: a translational toxicokinetic/toxicodynamic analysis.

Inhibitor of apoptosis (IAP) proteins suppress apoptosis and are overexpressed in a variety of cancers. GDC-0152 is a potent and selective IAP antagonist being developed as an anticancer agent. In preclinical safety studies, dogs were particularly sensitive to GDC-0152 showing adverse signs of a tumor necrosis factor alpha (TNF-α) driven systemic inflammatory response, related to cellular IAP degradation and activation of NFκB signaling, at lower exposures compared with rat. In addition, downstream increases in systemic levels of cytokines and chemokines, such as monocyte chemotactic protein-1 (MCP-1), were observed. A semimechanistic population toxicokinetic/toxicodynamic (TK/TD) model incorporating transit compartments was used to fit MCP-1 plasma concentrations from rats or dogs given iv GDC-0152 doses. Estimated TD parameters inferred that lower GDC-0152 plasma concentrations triggered more severe increases in plasma MCP-1 in dogs compared with rats. Human simulations performed using dog TD parameters and human pharmacokinetics predicted 300-2400% increases of MCP-1 in humans at iv doses from 0.76 to 1.48mg/kg. Similar simulations using rat TD parameters suggest little or no change. Patients given iv doses of GDC-0152 up to 1.48mg/kg iv showed no substantial increases in systemic MCP-1 or signs of a severe TNF-α driven systemic inflammatory response. Emerging clinical data reported for other IAP antagonists are consistent with our observations. Taken together, the data suggest dogs are more sensitive to IAP antagonists compared with humans and rats. This study illustrates how TK/TD analysis can be utilized to quantitatively translate and context an identified preclinical safety risk in dogs to humans.

Discovery of a potent small-molecule antagonist of inhibitor of apoptosis (IAP) proteins and clinical candidate for the treatment of cancer (GDC-0152).

A series of compounds were designed and synthesized as antagonists of cIAP1/2, ML-IAP, and XIAP based on the N-terminus, AVPI, of mature Smac. Compound 1 (GDC-0152) has the best profile of these compounds; it binds to the XIAP BIR3 domain, the BIR domain of ML-IAP, and the BIR3 domains of cIAP1 and cIAP2 with K(i) values of 28, 14, 17, and 43 nM, respectively. These compounds promote degradation of cIAP1, induce activation of caspase-3/7, and lead to decreased viability of breast cancer cells without affecting normal mammary epithelial cells. Compound 1 inhibits tumor growth when dosed orally in the MDA-MB-231 breast cancer xenograft model. Compound 1 was advanced to human clinical trials, and it exhibited linear pharmacokinetics over the dose range (0.049 to 1.48 mg/kg) tested. Mean plasma clearance in humans was 9 ± 3 mL/min/kg, and the volume of distribution was 0.6 ± 0.2 L/kg.

Assessing therapeutic responses in Kras mutant cancers using genetically engineered mouse models.

The low rate of approval of novel anti-cancer agents underscores the need for better preclinical models of therapeutic response as neither xenografts nor early-generation genetically engineered mouse models (GEMMs) reliably predict human clinical outcomes. Whereas recent, sporadic GEMMs emulate many aspects of their human disease counterpart more closely, their ability to predict clinical therapeutic responses has never been tested systematically. We evaluated the utility of two state-of-the-art, mutant Kras-driven GEMMs--one of non-small-cell lung carcinoma and another of pancreatic adenocarcinoma--by assessing responses to existing standard-of-care chemotherapeutics, and subsequently in combination with EGFR and VEGF inhibitors. Standard clinical endpoints were modeled to evaluate efficacy, including overall survival and progression-free survival using noninvasive imaging modalities. Comparisons with corresponding clinical trials indicate that these GEMMs model human responses well, and lay the foundation for the use of validated GEMMs in predicting outcome and interrogating mechanisms of therapeutic response and resistance.

[Cytokine contents in single donor platelets during storage].

This study was aimed to investigate the changes of cytokine contents in single donor platelets (SDPs) collected by using MCS(+), Trima, Amicus blood cell separators during storage. 18 portions of SDPs were collected by MCS(+), Trima, Amicus blood cell separators, were preserved in standard condition of blood bank, the levels of cytokines such as IL-8, RANTES, CD154, TGF-beta(1) and VEGF were detected by ELISA at 1, 3, 5, 7 days during storage. The results showed that the levels of IL-8, RANTES, CD154, TGF-beta(1) and VEGF in SDPs collected by blood cell separators MCS(+), Trima, and Amicus gradually increased with prolonging of time during storage, but the increase of IL-8 level in SDPs collected by MCS(+) separator was significant difference from SDPs collected by Trime and Amicus separators (p < 0.05). It is concluded that the all collected SDPs mentioned above express IL-8, RANTES, CD154, TGF-beta(1) and VEGF during storage, and their cytokine levels show a tendency to increase with prolonging of time during storage, apheresis platelets with less leukocytes express IL-8 lower.

Identification of direct target genes using joint sequence and expression likelihood with application to DAF-16.

A major challenge in the post-genome era is to reconstruct regulatory networks from the biological knowledge accumulated up to date. The development of tools for identifying direct target genes of transcription factors (TFs) is critical to this endeavor. Given a set of microarray experiments, a probabilistic model called TRANSMODIS has been developed which can infer the direct targets of a TF by integrating sequence motif, gene expression and ChIP-chip data. The performance of TRANSMODIS was first validated on a set of transcription factor perturbation experiments (TFPEs) involving Pho4p, a well studied TF in Saccharomyces cerevisiae. TRANSMODIS removed elements of arbitrariness in manual target gene selection process and produced results that concur with one's intuition. TRANSMODIS was further validated on a genome-wide scale by comparing it with two other methods in Saccharomyces cerevisiae. The usefulness of TRANSMODIS was then demonstrated by applying it to the identification of direct targets of DAF-16, a critical TF regulating ageing in Caenorhabditis elegans. We found that 189 genes were tightly regulated by DAF-16. In addition, DAF-16 has differential preference for motifs when acting as an activator or repressor, which awaits experimental verification. TRANSMODIS is computationally efficient and robust, making it a useful probabilistic framework for finding immediate targets.

Molecular dynamics and free energy studies on the wild-type and double mutant HIV-1 protease complexed with amprenavir and two amprenavir-related inhibitors: mechanism for binding and drug resistance.

The V82F/I84V double mutation is considered as the key residue mutation of the HIV-1 protease drug resistance because it can significantly lower the binding affinity of protease inhibitors in clinical uses. In the current work, the binding of amprenavir to both of the wild-type and the drug-resistant V82F/I84V mutant of the HIV-1 protease was investigated by molecular dynamics (MD) simulations and was compared to those of two inhibitors in development, TMC126 and TMC114. Absolute binding free energies were calculated by molecular mechanics/Poisson-Boltzmann surface area (MM/ PBSA) methodology. The predicted binding affinities give a good explanation of structure-affinity relationship (SAR) of three studied inhibitors. Furthermore, in the 18 ns MD simulations on the free wild-type and the mutated proteases, we observed that the free mutated protease shows similar dynamic characteristics of the flap opening and a little higher structural stability than the free wild-type protease. This suggests that the effect of the mutations may not significantly affect the equilibrium between the semiopen and the closed conformations. Finally, decomposition analysis of binding free energies and the further structural analysis indicate that the dominating effect of the V82F/I84V double mutation is to distort the geometry of the binding site and hence weaken the interactions of inhibitors preshaped to the wild-type binding site.

Endpoints for agents that slow tumor growth.

The endpoints of confirmed response and progression are widely used in oncology clinical trials. These endpoints originated at a time when agents used to treat cancer were primarily cytotoxic. Increasingly agents that are cytostatic are being investigated in combination with agents that are cytotoxic. Since tumor growth rates often depend on the volume of the tumor, combining cytotoxic agents that reduce tumor volume with cytostatic agents may mask the activity of the cytostatic agent. This paper explores the sensitivity of time to progression/progression free survival (TTP/PFS) as an endpoint for evaluating the efficacy of cytostatic drugs when combined with cytotoxic agents. Mathematical models of tumor growth are used to describe tumor growth over time. The models account for the impact of chemotherapy and an agent that slows tumor growth in the context of the Gompertzian growth kinetics. We use a clinical trial of an angiogenesis inhibitor in metastatic breast cancer as a motivating example. We demonstrate that the endpoint TTP/PFS may not be sensitive to the effects of active cytostatic agents when they are combined with chemotherapy, and measuring time to regrowth would be more sensitive to the activity of a cytostatic agent. We conclude that an active cytostatic agent may not appear effective when combined with chemotherapy and evaluated with TTP/PFS in a clinical trial.

Tumor necrosis factor-alpha-induced protein 3 as a putative regulator of nuclear factor-kappaB-mediated resistance to O6-alkylating agents in human glioblastomas.

PURPOSE: Pre-existing and acquired drug resistance are major obstacles to the successful treatment of glioblastomas. METHODS: We used an integrated resistance model and genomics tools to globally explore molecular factors and cellular pathways mediating resistance to O6-alkylating agents in glioblastoma cells. RESULTS: We identified a transcriptomic signature that predicts a common in vitro and in vivo resistance phenotype to these agents, a proportion of which is imprinted recurrently by gene dosage changes in the resistant glioblastoma genome. This signature was highly enriched for genes with functions in cell death, compromise, and survival. Modularity was a predominant organizational principle of the signature, with functions being carried out by groups of interacting molecules in overlapping networks. A highly significant network was built around nuclear factor-kappaB (NF-kappaB), which included the persistent alterations of various NF-kappaB pathway elements. Tumor necrosis factor-alpha-induced protein 3 (TNFAIP3) was identified as a new regulatory component of a putative cytoplasmic signaling cascade that mediates NF-kappaB activation in response to DNA damage caused by O6-alkylating agents. Expression of the corresponding zinc finger protein A20 closely mirrored the expression of the TNFAIP3 transcript, and was inversely related to NF-kappaB activation status in the resistant cells. A prediction model based on the resistance signature enabled the subclassification of an independent, validation cohort of 31 glioblastomas into two outcome groups (P = .037) and revealed TNFAIP3 as part of an optimized four-gene predictor associated significantly with patient survival (P = .022). CONCLUSION: Our results offer strong evidence for TNFAIP3 as a key regulator of the cytoplasmic signaling to activate NF-kappaB en route to O6-alkylating agent resistance in glioblastoma cells. This pathway may be an attractive target for therapeutic modulation of glioblastomas.