Thymic lymphoma detection in RORγ knockout mice using 5-hydroxymethylcytosine profiling of circulating cell-free DNA.

A high incidence of thymic lymphoma has been noted in mice deficient of retinoid-related orphan receptor γ2 (RORγ2), which is required for differentiation of naïve CD4+ T cells into TH17 cells. Using a RORγ homozygous knockout (KO) mouse model of thymic lymphoma, we characterized this tumor progression and investigated the utility of 5-hydroxymethylcytosine (5hmC) signatures as a non-invasive circulating biomarker for early prediction of malignancy. No evidence for malignancy was noted in the wild-type mice, while primary thymic lymphoma with multi-organ metastasis was observed microscopically in 97% of the homozygous RORγ KO mice. The severity of thymic lymphoma was not age-dependent in the KO mice of 2 to 4 months old. Differential enrichment of 5hmC in thymic DNA and plasma cell-free DNA (cfDNA) was compared across different stages of tumor progression. Random forest modeling of plasma cfDNA achieved good predictivity (AUC = 0.74) in distinguishing early non-metastatic thymic lymphoma compared to cancer-free controls, while perfect predictivity was achieved with advanced multi-organ metastatic disease (AUC = 1.00). Lymphoid-specific genes involved in thymocyte selection during T cell development (Themis, Tox) were differentially enriched in both plasma and thymic tissue. This could help in differentiating thymic lymphoma from other tumors commonly detected in rodent carcinogenicity studies used in pharmaceutical drug development to inform human malignancy risk. Overall, these results provide a proof-of-concept for using circulating cfDNA profiles in rodent carcinogenicity studies for early risk assessment of novel pharmaceutical targets.

Comparing molecular and computational approaches for detecting viral integration of AAV gene therapy constructs.

Many current gene therapy targets use recombinant adeno-associated virus (AAV). The majority of delivered AAV therapeutics persist as episomes, separate from host DNA, yet some viral DNA can integrate into host DNA in different proportions and at genomic locations. The potential for viral integration leading to oncogenic transformation has led regulatory agencies to require investigation into AAV integration events following gene therapy in preclinical species. In the present study, tissues were collected from cynomolgus monkeys and mice 6 and 8 weeks, respectively, following administration of an AAV vector delivering transgene cargo. We compared three different next-generation sequencing approaches (shearing extension primer tag selection ligation-mediated PCR, targeted enrichment sequencing [TES], and whole-genome sequencing) to contrast the specificity, scope, and frequency of integration detected by each method. All three methods detected dose-dependent insertions with a limited number of hotspots and expanded clones. While the functional outcome was similar for all three methods, TES was the most cost-effective and comprehensive method of detecting viral integration. Our findings aim to inform the direction of molecular efforts to ensure a thorough hazard assessment of AAV viral integration in our preclinical gene therapy studies.

Evaluating molecular fingerprint-based models of drug side effects against a statistical control.

There are many machine learning models that use molecular fingerprints of drugs to predict side effects. Characterizing their skill is necessary for understanding their usefulness in pharmaceutical development. Here, we analyze a statistical control of side effect prediction skill, develop a pipeline for benchmarking models, and evaluate how well existing models predict side effects identified in pharmaceutical documentation. We demonstrate that molecular fingerprints are useful for ranking drugs by their likelihood to cause a given side effect. However, the predictions for one or more drugs overall benefit only marginally from molecular fingerprints when ranking the likelihoods of many possible side effects, and display at most modest overall skill at identifying the side effects that do and do not occur.

A Systematic Review of Artificial Intelligence and Machine Learning Applications to Inflammatory Bowel Disease, with Practical Guidelines for Interpretation.

BACKGROUND: Inflammatory bowel disease (IBD) is a gastrointestinal chronic disease with an unpredictable disease course. Computational methods such as machine learning (ML) have the potential to stratify IBD patients for the provision of individualized care. The use of ML methods for IBD was surveyed, with an additional focus on how the field has changed over time. METHODS: On May 6, 2021, a systematic review was conducted through a search of MEDLINE and Embase databases, with the search structure ("machine learning" OR "artificial intelligence") AND ("Crohn* Disease" OR "Ulcerative Colitis" OR "Inflammatory Bowel Disease"). Exclusion criteria included studies not written in English, no human patient data, publication before 2001, studies that were not peer reviewed, nonautoimmune disease comorbidity research, and record types that were not primary research. RESULTS: Seventy-eight (of 409) records met the inclusion criteria. Random forest methods were most prevalent, and there was an increase in neural networks, mainly applied to imaging data sets. The main applications of ML to clinical tasks were diagnosis (18 of 78), disease course (22 of 78), and disease severity (16 of 78). The median sample size was 263. Clinical and microbiome-related data sets were most popular. Five percent of studies used an external data set after training and testing for additional model validation. DISCUSSION: Availability of longitudinal and deep phenotyping data could lead to better modeling. Machine learning pipelines that consider imbalanced data and that feature selection only on training data will generate more generalizable models. Machine learning models are increasingly being applied to more complex clinical tasks for specific phenotypes, indicating progress towards personalized medicine for IBD.

Organocatalyst treatment improves variant calling and mutant detection in archival clinical samples.

Formalin fixation of biological specimens damages nucleic acids and limits their use in genomic analyses. Previously, we showed that RNA isolation with an organocatalyst (2-amino-5-methylphenyl phosphonic acid, used to speed up reversal of formalin-induced adducts) and extended heated incubation (ORGΔ) improved RNA-sequencing data from formalin-fixed paraffin-embedded (FFPE) tissue samples. The primary goal of this study was to evaluate whether ORGΔ treatment improves DNA-sequencing data from clinical FFPE samples. We isolated RNA and DNA ± ORGΔ from paired FFPE and frozen human renal and ovarian carcinoma specimens collected as part of the National Cancer Institute Biospecimen Pre-analytical Variables program. Tumor types were microscopically confirmed from adjacent tissue sections. Following extraction, DNA was fragmented and sequenced and differences were compared between frozen and FFPE sample pairs. Treatment with ORGΔ improved concurrent SNP calls in FFPE DNA compared to non-ORGΔ FFPE samples and enhanced confidence in SNP calls for all FFPE DNA samples, beyond that of matched frozen samples. In general, the concordant SNPs identified in paired frozen and FFPE DNA samples agreed for both genotype and homozygosity vs. heterozygosity of calls regardless of ORGΔ treatment. The increased confidence in ORGΔ FFPE DNA variant calls relative to the matched frozen DNA suggests a novel application of this method. With further optimization, this method may improve quality of DNA-sequencing data in FFPE as well as frozen tissue samples.

Author Correction: Exploring use of unsupervised clustering to associate signaling profiles of GPCR ligands to clinical response.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

The MicroRNA-based Liquid Biopsy Improves Early Assessment of Lethal Acetaminophen Poisoning: A Case Report.

BACKGROUND Acetaminophen overdose is the most common cause of acute liver failure. Nevertheless, new biomarker approaches enabling early prediction of the outcome of the acetaminophen overdose are needed. Recently, using next-generation sequencing analysis of serum from human study participants we uncovered injury-specific signatures of circulating microRNAs (miRNAs) that represented underlying molecular mechanisms of toxicity. This case study is first to show the application of miRNA profiling to assess prognosis of acetaminophen poisoning. CASE REPORT The patient was admitted to the hospital following supra therapeutic acetaminophen ingestion. The patient showed elevated levels of biomarkers of hepatocellular injury alanine aminotransferase, aspartate transaminase, and glutamate dehydrogenase. Even though treatment with N-acetyl cysteine was initiated 24 hours post-ingestion, levels of alanine-aminotransferase and aspartate transaminase peaked at about 40 hours post ingestion of acetaminophen. We analyzed global circulating miRNA levels from 24 consecutive serum samples from this study participant covering the period from admission to time of death. CONCLUSIONS The resulting global miRNA profiles were compared with profiles from study participants with non-lethal acetaminophen poisoning and healthy controls. At the admission, the miRNA profiles of both lethal and non-lethal acetaminophen poisoning showed induction of cellular stress and oxidative damage. Later, the miRNA profiles of the lethal poisoning featured fibrosis and coagulation pathways while profiles of non-lethal cases resembled those of healthy study participants. Although additional confirmatory studies are needed, our case study is first to indicate that global miRNA profiles to be used as liquid biopsies have potential to facilitate the assessment of acetaminophen poisoning.

Exploring use of unsupervised clustering to associate signaling profiles of GPCR ligands to clinical response.

Signaling diversity of G protein-coupled (GPCR) ligands provides novel opportunities to develop more effective, better-tolerated therapeutics. Taking advantage of these opportunities requires identifying which effectors should be specifically activated or avoided so as to promote desired clinical responses and avoid side effects. However, identifying signaling profiles that support desired clinical outcomes remains challenging. This study describes signaling diversity of mu opioid receptor (MOR) ligands in terms of logistic and operational parameters for ten different in vitro readouts. It then uses unsupervised clustering of curve parameters to: classify MOR ligands according to similarities in type and magnitude of response, associate resulting ligand categories with frequency of undesired events reported to the pharmacovigilance program of the Food and Drug Administration and associate signals to side effects. The ability of the classification method to associate specific in vitro signaling profiles to clinically relevant responses was corroborated using ß2-adrenergic receptor ligands.

Evaluation of in Vitro Mitochondrial Toxicity Assays and Physicochemical Properties for Prediction of Organ Toxicity Using 228 Pharmaceutical Drugs.

Mitochondrial toxicity has been shown to contribute to a variety of organ toxicities such as liver, cardiac, and kidney. In the past decades, two high-throughput applicable screening assays (isolated rat liver mitochondria; glucose-galactose grown HepG2 cells) to assess mitochondrial toxicity have been deployed in many pharmaceutical companies, and numerous publications have demonstrated its usefulness for mechanistic investigations. However, only two publications have demonstrated the utility of these screens as a predictor of human drug-induced liver injury. In the present study, we screened 73 hepatotoxicants, 46 cardiotoxicants, 49 nephrotoxicants, and 60 compounds not known to cause human organ toxicity for their effects on mitochondrial function(s) in the assays mentioned above. Predictive performance was evaluated using specificity and sensitivity of the assays for predicting organ toxicity. Our results show that the predictive performance of the mitochondrial assays are superior for hepatotoxicity as compared to cardiotoxicity and nephrotoxicity (sensitivity 63% vs 33% and 28% with similar specificity of 93%), when the analysis was done at 100* Cmax (drug concentration in human plasma level). We further explored the association of mitochondrial toxicity with physicochemical properties such as calculated log partition coefficient (cLogP), topological polar surface area, ionization status, and molecular weight of the drugs and found that cLogP was most significantly associated mitochondrial toxicity. Since these assays are amenable to higher throughput, we recommend that chemists use these assays to perform structure activity relationship early in the drug discovery process, when chemical matter is abundant. This assures that compounds that lack the propensity to cause mitochondrial dysfunction (and associated organ toxicity) will move forward into animals and humans.

Demodifying RNA for Transcriptomic Analyses of Archival Formalin-Fixed Paraffin-Embedded Samples.

Archival formalin-fixed paraffin-embedded (FFPE) tissue samples offer a vast but largely untapped resource for genomic research. The primary technical issues limiting use of FFPE samples are RNA yield and quality. In this study, we evaluated methods to demodify RNA highly fragmented and crosslinked by formalin fixation. Primary endpoints were RNA recovery, RNA-sequencing quality metrics, and transcriptional responses to a reference chemical (phenobarbital, PB). Frozen mouse liver samples from control and PB groups (n = 6/group) were divided and preserved for 3 months as follows: frozen (FR); 70% ethanol (OH); 10% buffered formalin for 18 h followed by ethanol (18F); or 10% buffered formalin (3F). Samples from OH, 18F, and 3F groups were processed to FFPE blocks and sectioned for RNA isolation. Additional sections from 3F received the following demodification protocols to mitigate RNA damage: short heated incubation with Tris-Acetate-EDTA buffer; overnight heated incubation with an organocatalyst using 2 different isolation kits; or overnight heated incubation without organocatalyst. Ribo-depleted, stranded, total RNA libraries were built and sequenced using the Illumina HiSeq 2500 platform. Overnight incubation (± organocatalyst) increased RNA yield >3-fold and RNA integrity numbers and fragment analysis values by > 1.5- and >3.0-fold, respectively, versus 3F. Postsequencing metrics also showed reduced bias in gene coverage and deletion rates for overnight incubation groups. All demodification groups had increased overlap for differentially expressed genes (77%-84%) and enriched pathways (91%-97%) with FR, with the highest overlap in the organocatalyst groups. These results demonstrate simple changes in RNA isolation methods that can enhance genomic analyses of FFPE samples.

Use of Rat Primary Mesenteric Cells for the Prediction of PDE4 Inhibitor Drug-Induced Vascular Injury.

Drug-induced vascular injury (DIVI) in preclinical studies can delay, if not terminate, a drug development program. Clinical detection of DIVI can be very difficult as there are no definitive biomarkers known to reliably detect this disorder in all instances. The preclinical identification of DIVI requires detailed microscopic examination of a wide range of tissues although one of the most commonly affected areas in rats is the mesenteric vasculature. The reason for this predisposition of mesenteric arteries in rats as well as the exact mechanism and cell types involved in the initial development of these lesions have not been fully elucidated. We hypothesized that by using a mixed culture of cells from rat mesenteric tissue, we would be able to identify an RNA expression signature that could predict the invivo development of DIVI. Five compounds designed to inhibit Phosphodiesterase 4 activity (PDE4i) were chosen as positive controls. PDE4i's are well known to induce DIVI in the mesenteric vasculature of rats and there is microscopic evidence that this is associated, at least in part, with a proinflammatory mechanism. We surveyed, by qRT-PCR, the expression of 96 genes known to be involved in inflammation and using a Random-Forest model, identified 12 genes predictive of invivo DIVI outcomes in rats. Using these genes, we were able to cross-validate the ability of the Random-Forest modeling to predict the concentration at which PDE4i caused DIVI invivo.

Induction of endogenous retroelements as a potential mechanism for mouse-specific drug-induced carcinogenicity.

A number of chemical compounds have been shown to induce liver tumors in mice but not in other species. While several mechanisms for this species-specific tumorigenicity have been proposed, no definitive mechanism has been established. We examined the effects of the nongenotoxic rodent hepatic carcinogen, WY-14,643, in male mice from a high liver tumor susceptible strain (C3H/HeJ), and from a low tumor susceptible strain (C57BL/6). WY-14,643, a PPARα activator induced widespread increases in the expression of some endogenous retroelements, namely members of LTR and LINE elements in both strains. The expression of a number of known retroviral defense genes was also elevated. We also demonstrated that basal immune-mediated viral defense was elevated in C57BL/6 mice (the resistant strain) and that WY-14,643 further activated those immuno-defense processes. We propose that the previously reported >100X activity of retroelements in mice drives mouse-specific tumorigenicity. We also propose that C57BL/6's competent immune to retroviral activation allows it to remove cells before the activation of these elements can result in significant chromosomal insertions and mutation. Finally, we showed that WY-14,643 treatment induced gene signatures of DNA recombination in the sensitive C3H/HeJ strain.

Serum microRNA signatures as "liquid biopsies" for interrogating hepatotoxic mechanisms and liver pathogenesis in human.

MicroRNAs (miRNAs) released into the peripheral circulation upon cellular injury have shown a promise as a new class of tissue-specific biomarkers. We were first to demonstrate that next-generation sequencing analysis of serum from human subjects with acetaminophen-induced liver injury revealed a specific signature of circulating miRNAs. We consequently hypothesized that different types of hepatic liver impairments might feature distinct signatures of circulating miRNAs and that this approach might be useful as minimally invasive diagnostic "liquid biopsies" enabling the interrogation of underlying molecular mechanisms of injury in distant tissues. Therefore we examined serum circulating miRNAs in a total of 72 serum samples from a group of 53 subjects that included patients with accidental acetaminophen overdose, hepatitis B infection, liver cirrhosis and type 2 diabetes as well as gender- and age-matched healthy subjects with no evidence of liver disease. The miRNA signatures were identified using next-generation sequencing that provided analysis for the whole miRNome, including miRNA isoforms. Compared to the healthy subjects, a total of 179 miRNAs showed altered serum levels across the diseased subjects. Although many subjects have elevated alanine aminotransferase suggesting liver impairments, we identified distinct miRNA signatures for different impairments with minimum overlap. Furthermore, the bioinformatics analysis of miRNA signatures revealed relevant molecular pathways associated with the mechanisms of toxicity and or pathogenesis of disease. Interestingly, the high proportion of miRNA isoforms present in the respective signatures indicated a new level of complexity in cellular response to stress or disease. Our study demonstrates for the first time that signatures of circulating miRNAs show specificity for liver injury phenotypes and, once validated, might become useful for diagnosis of organ pathologies as "liquid biopsies".

ToxReporter: viewing the genome through the eyes of a toxicologist.

One of the many roles of a toxicologist is to determine if an observed adverse event (AE) is related to a previously unrecognized function of a given gene/protein. Towards that end, he or she will search a variety of public and propriety databases for information linking that protein to the observed AE. However, these databases tend to present all available information about a protein, which can be overwhelming, limiting the ability to find information about the specific toxicity being investigated. ToxReporter compiles information from a broad selection of resources and limits display of the information to user-selected areas of interest. ToxReporter is a PERL-based web-application which utilizes a MySQL database to streamline this process by categorizing public and proprietary domain-derived information into predefined safety categories according to a customizable lexicon. Users can view gene information that is 'red-flagged' according to the safety issue under investigation. ToxReporter also uses a scoring system based on relative counts of the red-flags to rank all genes for the amount of information pertaining to each safety issue and to display their scored ranking as an easily interpretable 'Tox-At-A-Glance' chart. Although ToxReporter was originally developed to display safety information, its flexible design could easily be adapted to display disease information as well.Database URL: ToxReporter is freely available at https://github.com/mgosink/ToxReporter.

The genomic sequence of lymphocryptovirus from cynomolgus macaque.

Lymphocryptoviruses such as Epstein-Barr virus (EBV) cause persistent infections in human and non-human primates, and suppression of the immune system can increase the risk of lymphocryptovirus (LCV)-associated tumor development in both human and non-human primates. To enable LCV infection as a non-clinical model to study effects of therapeutics on EBV immunity, we determined the genomic DNA sequence of the LCV from cynomolgus macaque, a species commonly used for non-clinical testing. Comparison to rhesus macaque LCV and human EBV sequences indicates that LCV from the cynomolgus macaque has the same genomic arrangement and a high degree of similarity in most genes, especially with rhesus macaque LCV. Genes showing lower similarity were those encoding proteins involved in latency and/or tumor promotion or immune evasion. The genomic sequence of LCV from cynomolgus macaque should aid the development of non-clinical tools for identifying therapeutics that impact LCV immunity and carry potential lymphoma risk.

Identification of tubular injury microRNA biomarkers in urine: comparison of next-generation sequencing and qPCR-based profiling platforms.

BACKGROUND: MicroRNAs (miRNAs) are small, non-coding RNAs that regulate protein levels post-transcriptionally. miRNAs play important regulatory roles in many cellular processes and have been implicated in several diseases. Recent studies have reported significant levels of miRNAs in a variety of body fluids, raising the possibility that miRNAs could serve as useful biomarkers. Next-generation sequencing (NGS) is increasingly employed in biomedical investigations. Although concordance between this platform and qRT-PCR based assays has been reported in high quality specimens, information is lacking on comparisons in biofluids especially urine. Here we describe the changes in miRNA expression patterns in a rodent model of renal tubular injury (gentamicin). Our aim is to compare RNA sequencing and qPCR based miRNA profiling in urine specimen from control and rats with confirmed tubular injury. RESULTS: Our preliminary examination of the concordance between miRNA-seq and qRT-PCR in urine specimen suggests minimal agreement between platforms probably due to the differences in sensitivity. Our results suggest that although miRNA-seq has superior specificity, it may not detect low abundant miRNAs in urine samples. Specifically, miRNA-seq did not detect some sequences which were identified by qRT-PCR. On the other hand, the qRT-PCR analysis was not able to detect the miRNA isoforms, which made up the majority of miRNA changes detected by NGS. CONCLUSIONS: To our knowledge, this is the first time that miRNA profiling platforms including NGS have been compared in urine specimen. miRNAs identified by both platforms, let-7d, miR-203, and miR-320, may potentially serve as promising novel urinary biomarkers for drug induced renal tubular epithelial injury.

Transcriptomic analysis of genetically defined autism candidate genes reveals common mechanisms of action.

BACKGROUND: Austism spectrum disorder (ASD) is a heterogeneous behavioral disorder or condition characterized by severe impairment of social engagement and the presence of repetitive activities. The molecular etiology of ASD is still largely unknown despite a strong genetic component. Part of the difficulty in turning genetics into disease mechanisms and potentially new therapeutics is the sheer number and diversity of the genes that have been associated with ASD and ASD symptoms. The goal of this work is to use shRNA-generated models of genetic defects proposed as causative for ASD to identify the common pathways that might explain how they produce a core clinical disability. METHODS: Transcript levels of Mecp2, Mef2a, Mef2d, Fmr1, Nlgn1, Nlgn3, Pten, and Shank3 were knocked-down in mouse primary neuron cultures using shRNA constructs. Whole genome expression analysis was conducted for each of the knockdown cultures as well as a mock-transduced culture and a culture exposed to a lentivirus expressing an anti-luciferase shRNA. Gene set enrichment and a causal reasoning engine was employed to identify pathway level perturbations generated by the transcript knockdown. RESULTS: Quantification of the shRNA targets confirmed the successful knockdown at the transcript and protein levels of at least 75% for each of the genes. After subtracting out potential artifacts caused by viral infection, gene set enrichment and causal reasoning engine analysis showed that a significant number of gene expression changes mapped to pathways associated with neurogenesis, long-term potentiation, and synaptic activity. CONCLUSIONS: This work demonstrates that despite the complex genetic nature of ASD, there are common molecular mechanisms that connect many of the best established autism candidate genes. By identifying the key regulatory checkpoints in the interlinking transcriptional networks underlying autism, we are better able to discover the ideal points of intervention that provide the broadest efficacy across the diverse population of autism patients.

Multinomial modeling and an evaluation of common data-mining algorithms for identifying signals of disproportionate reporting in pharmacovigilance databases.

MOTIVATION: A principal objective of pharmacovigilance is to detect adverse drug reactions that are unknown or novel in terms of their clinical severity or frequency. One method is through inspection of spontaneous reporting system databases, which consist of millions of reports of patients experiencing adverse effects while taking one or more drugs. For such large databases, there is an increasing need for quantitative and automated screening tools to assist drug safety professionals in identifying drug-event combinations (DECs) worthy of further investigation. Existing algorithms can effectively identify problematic DECs when the frequencies are high. However these algorithms perform differently for low-frequency DECs. RESULTS: In this work, we provide a method based on the multinomial distribution that identifies signals of disproportionate reporting, especially for low-frequency combinations. In addition, we comprehensively compare the performance of commonly used algorithms with the new approach. Simulation results demonstrate the advantages of the proposed method, and analysis of the Adverse Event Reporting System data shows that the proposed method can help detect interesting signals. Furthermore, we suggest that these methods be used to identify DECs that occur significantly less frequently than expected, thus identifying potential alternative indications for these drugs. We provide an empirical example that demonstrates the importance of exploring underexpected DECs. AVAILABILITY: Code to implement the proposed method is available in R on request from the corresponding authors. CONTACT: kjell@arboranalytics.com or Mark.M.Gosink@Pfizer.com SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

GenSensor Suite: A Web-Based Tool for the Analysis of Gene and Protein Interactions, Pathways, and Regulation.

The GenSensor Suite consists of four web tools for elucidating relationships among genes and proteins. GenPath results show which biochemical, regulatory, or other gene set categories are over- or under-represented in an input list compared to a background list. All common gene sets are available for searching in GenPath, plus some specialized sets. Users can add custom background lists. GenInteract builds an interaction gene list from a single gene input and then analyzes this in GenPath. GenPubMed uses a PubMed query to identify a list of PubMed IDs, from which a gene list is extracted and queried in GenPath. GenViewer allows the user to query one gene set against another in GenPath. GenPath results are presented with relevant P- and q-values in an uncluttered, fully linked, and integrated table. Users can easily copy this table and paste it directly into a spreadsheet or document.

The role of hypoxia in 2-butoxyethanol-induced hemangiosarcoma.

To understand the molecular mechanisms underlying compound-induced hemangiosarcomas in mice, and therefore, their human relevance, a systems biology approach was undertaken using transcriptomics and Causal Network Modeling from mice treated with 2-butoxyethanol (2-BE). 2-BE is a hemolytic agent that induces hemangiosarcomas in mice. We hypothesized that the hemolysis induced by 2-BE would result in local tissue hypoxia, a well-documented trigger for endothelial cell proliferation leading to hemangiosarcoma. Gene expression data from bone marrow (BM), liver, and spleen of mice exposed to a single dose (4 h) or seven daily doses of 2-BE were used to develop a mechanistic model of hemangiosarcoma. The resulting mechanistic model confirms previous work proposing that 2-BE induces macrophage activation and inflammation in the liver. In addition, the model supports local tissue hypoxia in the liver and spleen, coupled with increased erythropoeitin signaling and erythropoiesis in the spleen and BM, and suppression of mechanisms that contribute to genomic stability, events that could be contributing factors to hemangiosarcoma formation. Finally, an immunohistochemistry method (Hypoxyprobe) demonstrated that tissue hypoxia was present in the spleen and BM. Together, the results of this study identify molecular mechanisms that initiate hemangiosarcoma, a key step in understanding safety concerns that can impact drug decision processes, and identified hypoxia as a possible contributing factor for 2-BE-induced hemangiosarcoma in mice.