The Rise of Hypothesis-Driven Artificial Intelligence in Oncology.

Cancer is a complex disease involving the deregulation of intricate cellular systems beyond genetic aberrations and, as such, requires sophisticated computational approaches and high-dimensional data for optimal interpretation. While conventional artificial intelligence (AI) models excel in many prediction tasks, they often lack interpretability and are blind to the scientific hypotheses generated by researchers to enable cancer discoveries. Here we propose that hypothesis-driven AI, a new emerging class of AI algorithm, is an innovative approach to uncovering the complex etiology of cancer from big omics data. This review exemplifies how hypothesis-driven AI is different from conventional AI by citing its application in various areas of oncology including tumor classification, patient stratification, cancer gene discovery, drug response prediction, and tumor spatial organization. Our aim is to stress the feasibility of incorporating domain knowledge and scientific hypotheses to craft the design of new AI algorithms. We showcase the power of hypothesis-driven AI in making novel cancer discoveries that can be overlooked by conventional AI methods. Since hypothesis-driven AI is still in its infancy, open questions such as how to better incorporate new knowledge and biological perspectives to ameliorate bias and improve interpretability in the design of AI algorithms still need to be addressed. In conclusion, hypothesis-driven AI holds great promise in the discovery of new mechanistic and functional insights that explain the complexity of cancer etiology and potentially chart a new roadmap to improve treatment regimens for individual patients.

Endoxifen downregulates AKT phosphorylation through protein kinase C beta 1 inhibition in ERα+ breast cancer.

Endoxifen, a secondary tamoxifen metabolite, is a potent antiestrogen exhibiting estrogen receptor alpha (ERα) binding at nanomolar concentrations. Phase I/II clinical trials identified clinical activity of Z-endoxifen (ENDX), in endocrine-refractory metastatic breast cancer as well as ERα+ solid tumors, raising the possibility that ENDX may have a second, ERα-independent, mechanism of action. An unbiased mass spectrometry approach revealed that ENDX concentrations achieved clinically with direct ENDX administration (5 µM), but not low concentrations observed during tamoxifen treatment (<0.1 µM), profoundly altered the phosphoproteome of the aromatase expressing MCF7AC1 cells with limited impact on the total proteome. Computational analysis revealed protein kinase C beta (PKCß) and protein kinase B alpha or AKT1 as potential kinases responsible for mediating ENDX effects on protein phosphorylation. ENDX more potently inhibited PKCß1 kinase activity compared to other PKC isoforms, and ENDX binding to PKCß1 was confirmed using Surface Plasma Resonance. Under conditions that activated PKC/AKT signaling, ENDX induced PKCß1 degradation, attenuated PKCß1-activated AKTSer473 phosphorylation, diminished AKT substrate phosphorylation, and induced apoptosis. ENDX's effects on AKT were phenocopied by siRNA-mediated PKCß1 knockdown or treatment with the pan-AKT inhibitor, MK-2206, while overexpression of constitutively active AKT diminished ENDX-induced apoptosis. These findings, which identify PKCß1 as an ENDX target, indicate that PKCß1/ENDX interactions suppress AKT signaling and induce apoptosis in breast cancer.

SPIN-AI: A Deep Learning Model That Identifies Spatially Predictive Genes.

Spatially resolved sequencing technologies help us dissect how cells are organized in space. Several available computational approaches focus on the identification of spatially variable genes (SVGs), genes whose expression patterns vary in space. The detection of SVGs is analogous to the identification of differentially expressed genes and permits us to understand how genes and associated molecular processes are spatially distributed within cellular niches. However, the expression activities of SVGs fail to encode all information inherent in the spatial distribution of cells. Here, we devised a deep learning model, Spatially Informed Artificial Intelligence (SPIN-AI), to identify spatially predictive genes (SPGs), whose expression can predict how cells are organized in space. We used SPIN-AI on spatial transcriptomic data from squamous cell carcinoma (SCC) as a proof of concept. Our results demonstrate that SPGs not only recapitulate the biology of SCC but also identify genes distinct from SVGs. Moreover, we found a substantial number of ribosomal genes that were SPGs but not SVGs. Since SPGs possess the capability to predict spatial cellular organization, we reason that SPGs capture more biologically relevant information for a given cellular niche than SVGs. Thus, SPIN-AI has broad applications for detecting SPGs and uncovering which biological processes play important roles in governing cellular organization.

Network Biology-Inspired Machine Learning Features Predict Cancer Gene Targets and Reveal Target Coordinating Mechanisms.

Anticipating and understanding cancers' need for specific gene activities is key for novel therapeutic development. Here we utilized DepMap, a cancer gene dependency screen, to demonstrate that machine learning combined with network biology can produce robust algorithms that both predict what genes a cancer is dependent on and what network features coordinate such gene dependencies. Using network topology and biological annotations, we constructed four groups of novel engineered machine learning features that produced high accuracies when predicting binary gene dependencies. We found that in all examined cancer types, F1 scores were greater than 0.90, and model accuracy remained robust under multiple hyperparameter tests. We then deconstructed these models to identify tumor type-specific coordinators of gene dependency and identified that in certain cancers, such as thyroid and kidney, tumors' dependencies are highly predicted by gene connectivity. In contrast, other histologies relied on pathway-based features such as lung, where gene dependencies were highly predictive by associations with cell death pathway genes. In sum, we show that biologically informed network features can be a valuable and robust addition to predictive pharmacology models while simultaneously providing mechanistic insights.

Relationship between BCL2 mutations and follicular lymphoma outcome in the chemoimmunotherapy era.

How to identify follicular lymphoma (FL) patients with low disease burden but high risk for early progression is unclear. Building on a prior study demonstrating the early transformation of FLs with high variant allele frequency (VAF) BCL2 mutations at activation-induced cytidine deaminase (AICDA) sites, we examined 11 AICDA mutational targets, including BCL2, BCL6, PAX5, PIM1, RHOH, SOCS, and MYC, in 199 newly diagnosed grade 1 and 2 FLs. BCL2 mutations with VAF ≥20% occurred in 52% of cases. Among 97 FL patients who did not initially receive rituximab-containing therapy, nonsynonymous BCL2 mutations at VAF ≥20% were associated with increased transformation risk (HR 3.01, 95% CI 1.04-8.78, p = 0.043) and a trend toward shorter event-free survival (EFS, median 20 months with mutations versus 54 months without, p = 0.052). Other sequenced genes were less frequently mutated and did not increase the prognostic value of the panel. Across the entire population, nonsynonymous BCL2 mutations at VAF ≥20% were associated with decreased EFS (HR 1.55, 95% CI 1.02-2.35, p = 0.043 after correction for FLIPI and treatment) and decreased overall survival after median 14-year follow-up (HR 1.82, 95% CI 1.05-3.17, p = 0.034). Thus, high VAF nonsynonymous BCL2 mutations remain prognostic even in the chemoimmunotherapy era.

A Knowledge-Based Discovery Approach Couples Artificial Neural Networks With Weight Engineering to Uncover Immune-Related Processes Underpinning Clinical Traits of Breast Cancer.

Immune-related processes are important in underpinning the properties of clinical traits such as prognosis and drug response in cancer. The possibility to extract knowledge learned by artificial neural networks (ANNs) from omics data to explain cancer clinical traits is a very attractive subject for novel discovery. Recent studies using a version of ANNs called autoencoders revealed their capability to store biologically meaningful information indicating that autoencoders can be utilized as knowledge discovery platforms aside from their initial assigned use for dimensionality reduction. Here, we devise an innovative weight engineering approach and ANN platform called artificial neural network encoder (ANNE) using an autoencoder and apply it to a breast cancer dataset to extract knowledge learned by the autoencoder model that explains clinical traits. Intriguingly, the extracted biological knowledge in the form of gene-gene associations from ANNE shows immune-related components such as chemokines, carbonic anhydrase, and iron metabolism that modulate immune-related processes and the tumor microenvironment play important roles in underpinning breast cancer clinical traits. Our work shows that biological "knowledge" learned by an ANN model is indeed encoded as weights throughout its neuronal connections, and it is possible to extract learned knowledge via a novel weight engineering approach to uncover important biological insights.

Gene utility recapitulates chromosomal aberrancies in advanced stage neuroblastoma.

Neuroblastoma (NB) is the most common extracranial solid tumor in children. Although only a few recurrent somatic mutations have been identified, chromosomal abnormalities, including the loss of heterozygosity (LOH) at the chromosome 1p and gains of chromosome 17q, are often seen in the high-risk cases. The biological basis and evolutionary forces that drive such genetic abnormalities remain enigmatic. Here, we conceptualize the Gene Utility Model (GUM) that seeks to identify genes driving biological signaling via their collective gene utilities and apply it to understand the impact of those differentially utilized genes on constraining the evolution of NB karyotypes. By employing a computational process-guided flow algorithm to model gene utility in protein-protein networks that built based on transcriptomic data, we conducted several pairwise comparative analyses to uncover genes with differential utilities in stage 4 NBs with distinct classification. We then constructed a utility karyotype by mapping these differentially utilized genes to their respective chromosomal loci. Intriguingly, hotspots of the utility karyotype, to certain extent, can consistently recapitulate the major chromosomal abnormalities of NBs and also provides clues to yet identified predisposition sites. Hence, our study not only provides a new look, from a gene utility perspective, into the known chromosomal abnormalities detected by integrative genomic sequencing efforts, but also offers new insights into the etiology of NB and provides a framework to facilitate the identification of novel therapeutic targets for this devastating childhood cancer.

Uncovering Pharmacological Opportunities for Cancer Stem Cells-A Systems Biology View.

Cancer stem cells (CSCs) represent a small fraction of the total cancer cell population, yet they are thought to drive disease propagation, therapy resistance and relapse. Like healthy stem cells, CSCs possess the ability to self-renew and differentiate. These stemness phenotypes of CSCs rely on multiple molecular cues, including signaling pathways (for example, WNT, Notch and Hedgehog), cell surface molecules that interact with cellular niche components, and microenvironmental interactions with immune cells. Despite the importance of understanding CSC biology, our knowledge of how neighboring immune and tumor cell populations collectively shape CSC stemness is incomplete. Here, we provide a systems biology perspective on the crucial roles of cellular population identification and dissection of cell regulatory states. By reviewing state-of-the-art single-cell technologies, we show how innovative systems-based analysis enables a deeper understanding of the stemness of the tumor niche and the influence of intratumoral cancer cell and immune cell compositions. We also summarize strategies for refining CSC systems biology, and the potential role of this approach in the development of improved anticancer treatments. Because CSCs are amenable to cellular transitions, we envision how systems pharmacology can become a major engine for discovery of novel targets and drug candidates that can modulate state transitions for tumor cell reprogramming. Our aim is to provide deeper insights into cancer stemness from a systems perspective. We believe this approach has great potential to guide the development of more effective personalized cancer therapies that can prevent CSC-mediated relapse.

LIN28 coordinately promotes nucleolar/ribosomal functions and represses the 2C-like transcriptional program in pluripotent stem cells.

LIN28 is an RNA binding protein with important roles in early embryo development, stem cell differentiation/reprogramming, tumorigenesis and metabolism. Previous studies have focused mainly on its role in the cytosol where it interacts with Let-7 microRNA precursors or mRNAs, and few have addressed LIN28's role within the nucleus. Here, we show that LIN28 displays dynamic temporal and spatial expression during murine embryo development. Maternal LIN28 expression drops upon exit from the 2-cell stage, and zygotic LIN28 protein is induced at the forming nucleolus during 4-cell to blastocyst stage development, to become dominantly expressed in the cytosol after implantation. In cultured pluripotent stem cells (PSCs), loss of LIN28 led to nucleolar stress and activation of a 2-cell/4-cell-like transcriptional program characterized by the expression of endogenous retrovirus genes. Mechanistically, LIN28 binds to small nucleolar RNAs and rRNA to maintain nucleolar integrity, and its loss leads to nucleolar phase separation defects, ribosomal stress and activation of P53 which in turn binds to and activates 2C transcription factor Dux. LIN28 also resides in a complex containing the nucleolar factor Nucleolin (NCL) and the transcriptional repressor TRIM28, and LIN28 loss leads to reduced occupancy of the NCL/TRIM28 complex on the Dux and rDNA loci, and thus de-repressed Dux and reduced rRNA expression. Lin28 knockout cells with nucleolar stress are more likely to assume a slowly cycling, translationally inert and anabolically inactive state, which is a part of previously unappreciated 2C-like transcriptional program. These findings elucidate novel roles for nucleolar LIN28 in PSCs, and a new mechanism linking 2C program and nucleolar functions in PSCs and early embryo development.

De novo individualized disease modules reveal the synthetic penetrance of genes and inform personalized treatment regimens.

Current understandings of individual disease etiology and therapeutics are limited despite great need. To fill the gap, we propose a novel computational pipeline that collects potent disease gene cooperative pathways to envision individualized disease etiology and therapies. Our algorithm constructs individualized disease modules de novo, which enables us to elucidate the importance of mutated genes in specific patients and to understand the synthetic penetrance of these genes across patients. We reveal that importance of the notorious cancer drivers TP53 and PIK3CA fluctuate widely across breast cancers and peak in tumors with distinct numbers of mutations and that rarely mutated genes such as XPO1 and PLEKHA1 have high disease module importance in specific individuals. Furthermore, individualized module disruption enables us to devise customized singular and combinatorial target therapies that were highly varied across patients, showing the need for precision therapeutics pipelines. As the first analysis of de novo individualized disease modules, we illustrate the power of individualized disease modules for precision medicine by providing deep novel insights on the activity of diseased genes in individuals.

NKG7 Is a T-cell-Intrinsic Therapeutic Target for Improving Antitumor Cytotoxicity and Cancer Immunotherapy.

Cytotoxic CD8+ T cells (CTL) are a crucial component of the immune system notable for their ability to eliminate rapidly proliferating malignant cells. However, the T-cell intrinsic factors required for human CTLs to accomplish highly efficient antitumor cytotoxicity are not well defined. By evaluating human CD8+ T cells from responders versus nonresponders to treatment with immune checkpoint inhibitors, we sought to identify key factors associated with effective CTL function. Single-cell RNA-sequencing analysis of peripheral CD8+ T cells from patients treated with anti-PD-1 therapy showed that cells from nonresponders exhibited decreased expression of the cytolytic granule-associated molecule natural killer cell granule protein-7 (NKG7). Functional assays revealed that reduced NKG7 expression altered cytolytic granule number, trafficking, and calcium release, resulting in decreased CD8+ T-cell-mediated killing of tumor cells. Transfection of T cells with NKG7 mRNA was sufficient to improve the tumor-cell killing ability of human T cells isolated from nonresponders and increase their response to anti-PD-1 or anti-PD-L1 therapy in vitro. NKG7 mRNA therapy also improved the antitumor activity of murine tumor antigen-specific CD8+ T cells in an in vivo model of adoptive cell therapy. Finally, we showed that the transcription factor ETS1 played a role in regulating NKG7 expression. Together, our results identify NKG7 as a necessary component for the cytotoxic function of CD8+ T cells and establish NKG7 as a T-cell-intrinsic therapeutic target for enhancing cancer immunotherapy.See related article by Li et al., p. 154.

Constitutive BAK/MCL1 complexes predict paclitaxel and S63845 sensitivity of ovarian cancer.

We previously found that preformed complexes of BAK with antiapoptotic BCL2 proteins predict BH3 mimetic sensitivities in lymphohematopoietic cells. These complexes have not previously been examined in solid tumors or in the context of conventional anticancer drugs. Here we show the relative amount of BAK found in preformed complexes with MCL1 or BCLXL varies across ovarian cancer cell lines and patient-derived xenografts (PDXs). Cells bearing BAK/MCL1 complexes were more sensitive to paclitaxel and the MCL1 antagonist S63845. Likewise, PDX models with BAK/MCL1 complexes were more likely to respond to paclitaxel. Mechanistically, BIM induced by low paclitaxel concentrations interacted preferentially with MCL1 and displaced MCL1-bound BAK. Further studies indicated that cells with preformed BAK/MCL1 complexes were sensitive to the paclitaxel/S63845 combination, while cells without BAK/MCL1 complexes were not. Our study suggested that the assessment of BAK/MCL1 complexes might be useful for predicting response to paclitaxel alone or in combination with BH3 mimetics.

Characterization of a RAD51C-silenced high-grade serous ovarian cancer model during development of PARP inhibitor resistance.

Acquired PARP inhibitor (PARPi) resistance in BRCA1- or BRCA2-mutant ovarian cancer often results from secondary mutations that restore expression of functional protein. RAD51C is a less commonly studied ovarian cancer susceptibility gene whose promoter is sometimes methylated, leading to homologous recombination (HR) deficiency and PARPi sensitivity. For this study, the PARPi-sensitive patient-derived ovarian cancer xenograft PH039, which lacks HR gene mutations but harbors RAD51C promoter methylation, was selected for PARPi resistance by cyclical niraparib treatment in vivo. PH039 acquired PARPi resistance by the third treatment cycle and grew through subsequent treatment with either niraparib or rucaparib. Transcriptional profiling throughout the course of resistance development showed widespread pathway level changes along with a marked increase in RAD51C mRNA, which reflected loss of RAD51C promoter methylation. Analysis of ovarian cancer samples from the ARIEL2 Part 1 clinical trial of rucaparib monotherapy likewise indicated an association between loss of RAD51C methylation prior to on-study biopsy and limited response. Interestingly, the PARPi resistant PH039 model remained platinum sensitive. Collectively, these results not only indicate that PARPi treatment pressure can reverse RAD51C methylation and restore RAD51C expression, but also provide a model for studying the clinical observation that PARPi and platinum sensitivity are sometimes dissociated.

Modular network mechanism of CCN1-associated resistance to HSV-1-derived oncolytic immunovirotherapies for glioblastomas.

Glioblastomas (GBMs) are the most common and lethal primary brain malignancy in adults. Oncolytic virus (OV) immunotherapies selectively kill GBM cells in a manner that elicits antitumor immunity. Cellular communication network factor 1 (CCN1), a protein found in most GBM microenvironments, expression predicts resistance to OVs, particularly herpes simplex virus type 1 (HSV-1). This study aims to understand how extracellular CCN1 alters the GBM intracellular state to confer OV resistance. Protein-protein interaction network information flow analyses of LN229 human GBM transcriptomes identified 39 novel nodes and 12 binary edges dominating flow in CCN1high cells versus controls. Virus response programs, notably against HSV-1, and cytokine-mediated signaling pathways are highly enriched. Our results suggest that CCN1high states exploit IDH1 and TP53, and increase dependency on RPL6, HUWE1, and COPS5. To validate, we reproduce our findings in 65 other GBM cell line (CCLE) and 174 clinical GBM patient sample (TCGA) datasets. We conclude through our generalized network modeling and system level analysis that CCN1 signals via several innate immune pathways in GBM to inhibit HSV-1 OVs before transduction. Interventions disrupting this network may overcome immunovirotherapy resistance.

Machine Learning and Systems Biology Approaches to Characterize Dosage-Based Gene Dependencies in Cancer Cells.

Mapping of cancer survivability factors allows for the identification of novel biological insights for drug targeting. Using genomic editing techniques, gene dependencies can be extracted in a high-throughput and quantitative manner. Dependencies have been predicted using machine learning techniques on -omics data, but the biological consequences of dependency predictor pairs has not been explored. In this work we devised a framework to explore gene dependency using an ensemble of machine learning methods, and our learned models captured meaningful biological information beyond just gene dependency prediction. We show that dosage-based dependent predictors (DDPs) primarily belonged to transcriptional regulation ontologies. We also found that anti-sense RNAs and long- noncoding RNA transcripts display DDPs. Network analyses revealed that SOX10, HLA-J, and ZEB2 act as a triad of network hubs in the dependent-predictor network. Collectively, we demonstrate the powerful combination of machine learning and systems biology approach can illuminate new insights in understanding gene dependency and guide novel targeting avenues.

Synthetic and systems biology principles in the design of programmable oncolytic virus immunotherapies for glioblastoma.

Oncolytic viruses (OVs) are a class of immunotherapeutic agents with promising preclinical results for the treatment of glioblastoma (GBM) but have shown limited success in recent clinical trials. Advanced bioengineering principles from disciplines such as synthetic and systems biology are needed to overcome the current challenges faced in developing effective OV-based immunotherapies for GBMs, including off-target effects and poor clinical responses. Synthetic biology is an emerging field that focuses on the development of synthetic DNA constructs that encode networks of genes and proteins (synthetic genetic circuits) to perform novel functions, whereas systems biology is an analytical framework that enables the study of complex interactions between host pathways and these synthetic genetic circuits. In this review, the authors summarize synthetic and systems biology concepts for developing programmable, logic-based OVs to treat GBMs. Programmable OVs can increase selectivity for tumor cells and enhance the local immunological response using synthetic genetic circuits. The authors discuss key principles for developing programmable OV-based immunotherapies, including how to 1) select an appropriate chassis, a vector that carries a synthetic genetic circuit, and 2) design a synthetic genetic circuit that can be programmed to sense key signals in the GBM microenvironment and trigger release of a therapeutic payload. To illustrate these principles, some original laboratory data are included, highlighting the need for systems biology studies, as well as some preliminary network analyses in preparation for synthetic biology applications. Examples from the literature of state-of-the-art synthetic genetic circuits that can be packaged into leading candidate OV chassis are also surveyed and discussed.

CDK2-Mediated Upregulation of TNFα as a Mechanism of Selective Cytotoxicity in Acute Leukemia.

Although inhibitors of the kinases CHK1, ATR, and WEE1 are undergoing clinical testing, it remains unclear how these three classes of agents kill susceptible cells and whether they utilize the same cytotoxic mechanism. Here we observed that CHK1 inhibition induces apoptosis in a subset of acute leukemia cell lines in vitro, including TP53-null acute myeloid leukemia (AML) and BCR/ABL-positive acute lymphoid leukemia (ALL), and inhibits leukemic colony formation in clinical AML samples ex vivo. In further studies, downregulation or inhibition of CHK1 triggered signaling in sensitive human acute leukemia cell lines that involved CDK2 activation followed by AP1-dependent TNF transactivation, TNFα production, and engagement of a TNFR1- and BID-dependent apoptotic pathway. AML lines that were intrinsically resistant to CHK1 inhibition exhibited high CHK1 expression and were sensitized by CHK1 downregulation. Signaling through this same CDK2-AP1-TNF cytotoxic pathway was also initiated by ATR or WEE1 inhibitors in vitro and during CHK1 inhibitor treatment of AML xenografts in vivo. Collectively, these observations not only identify new contributors to the antileukemic cell action of CHK1, ATR, and WEE1 inhibitors, but also delineate a previously undescribed pathway leading from aberrant CDK2 activation to death ligand-induced killing that can potentially be exploited for acute leukemia treatment. SIGNIFICANCE: This study demonstrates that replication checkpoint inhibitors can kill AML cells through a pathway involving AP1-mediated TNF gene activation and subsequent TP53-independent, TNFα-induced apoptosis, which can potentially be exploited clinically.

Reactivating latent HIV with PKC agonists induces resistance to apoptosis and is associated with phosphorylation and activation of BCL2.

Eradication of HIV-1 by the "kick and kill" strategy requires reactivation of latent virus to cause death of infected cells by either HIV-induced or immune-mediated apoptosis. To date this strategy has been unsuccessful, possibly due to insufficient cell death in reactivated cells to effectively reduce HIV-1 reservoir size. As a possible cause for this cell death resistance, we examined whether leading latency reversal agents (LRAs) affected apoptosis sensitivity of CD4 T cells. Multiple LRAs of different classes inhibited apoptosis in CD4 T cells. Protein kinase C (PKC) agonists bryostatin-1 and prostratin induced phosphorylation and enhanced neutralizing capability of the anti-apoptotic protein BCL2 in a PKC-dependent manner, leading to resistance to apoptosis induced by both intrinsic and extrinsic death stimuli. Furthermore, HIV-1 producing CD4 T cells expressed more BCL2 than uninfected cells, both in vivo and after ex vivo reactivation. Therefore, activation of BCL2 likely contributes to HIV-1 persistence after latency reversal with PKC agonists. The effects of LRAs on apoptosis sensitivity should be considered in designing HIV cure strategies predicated upon the "kick and kill" paradigm.

Anastrozole has an Association between Degree of Estrogen Suppression and Outcomes in Early Breast Cancer and is a Ligand for Estrogen Receptor α.

PURPOSE: To determine if the degree of estrogen suppression with aromatase inhibitors (AI: anastrozole, exemestane, letrozole) is associated with efficacy in early-stage breast cancer, and to examine for differences in the mechanism of action between the three AIs. EXPERIMENTAL DESIGN: Matched case-control studies [247 matched sets from MA.27 (anastrozole vs. exemestane) and PreFace (letrozole) trials] were undertaken to assess whether estrone (E1) or estradiol (E2) concentrations after 6 months of adjuvant therapy were associated with risk of an early breast cancer event (EBCE). Preclinical laboratory studies included luciferase activity, cell proliferation, radio-labeled ligand estrogen receptor binding, surface plasmon resonance ligand receptor binding, and nuclear magnetic resonance assays. RESULTS: Women with E1 ≥1.3 pg/mL and E2 ≥0.5 pg/mL after 6 months of AI treatment had a 2.2-fold increase in risk (P = 0.0005) of an EBCE, and in the anastrozole subgroup, the increase in risk of an EBCE was 3.0-fold (P = 0.001). Preclinical laboratory studies examined mechanisms of action in addition to aromatase inhibition and showed that only anastrozole could directly bind to estrogen receptor α (ERα), activate estrogen response element-dependent transcription, and stimulate growth of an aromatase-deficient CYP19A1-/- T47D breast cancer cell line. CONCLUSIONS: This matched case-control clinical study revealed that levels of estrone and estradiol above identified thresholds after 6 months of adjuvant anastrozole treatment were associated with increased risk of an EBCE. Preclinical laboratory studies revealed that anastrozole, but not exemestane or letrozole, is a ligand for ERα. These findings represent potential steps towards individualized anastrozole therapy.