An ensemble of features based deep learning neural network for reduction of inappropriate atrial fibrillation detection in implantable cardiac monitors.

Background: Multiple studies have reported on classification of raw electrocardiograms (ECGs) using convolutional neural networks (CNNs). Objective: We investigated an application-specific CNN using a custom ensemble of features designed based on characteristics of the ECG during atrial fibrillation (AF) to reduce inappropriate AF detections in implantable cardiac monitors (ICMs). Methods: An ensemble of features was developed and combined to form an input signal for the CNN. The features were based on the morphological characteristics of AF, incoherence of RR intervals, and the fact that AF begets more AF. A custom CNN model and the RESNET18 model were trained using ICM-detected AF episodes that were adjudicated to be true AF or false detections. The trained models were evaluated using a test dataset from independent patients. Results: The training and validation datasets consisted of 31,757 AF episodes (2516 patients) and 28,506 false episodes (2126 patients). The validation set (20% randomly chosen episodes of each type) had an area under the curve of 0.996 for custom CNN (0.993 for RESNET18). Thresholds were chosen to obtain a relative sensitivity and specificity of 99.2% and 92.8%, respectively (99.2% and 87.9% for RESNET18, respectively). The performance in the independent test set (4546 AF episodes from 418 patients; 5384 false episodes from 605 patients) showed an area under the curve of 0.993 (0.991 for RESNET18) and relative sensitivity and specificity of 98.7% and 91.4%, respectively, at chosen thresholds (98.9% and 88.2% for RESNET18, respectively). Conclusion: An ensemble of features-based CNNs was developed that reduced inappropriate AF detection in ICMs by over 90% while preserving sensitivity.

Detection of Atrial Fibrillation After Central Retinal Artery Occlusion.

Background: Central retinal artery occlusion (CRAO) causes sudden, irreversible blindness and is a form of acute ischemic stroke. In this study, we sought to determine the proportion of patients in whom atrial fibrillation (AF) is detected by extended cardiac monitoring after CRAO. Methods: We performed a retrospective, observational cohort study using data from the Optum deidentified electronic health record of 30.8 million people cross-referenced with the Medtronic CareLink database of 2.7 million people with cardiac monitoring devices in situ. We enrolled patients in 3 groups: (1) CRAO, (2) cerebral ischemic stroke, and (3) age-, sex-, and comorbidity-matched controls. The primary end point was the detection of new AF (defined as ≥2 minutes of AF detected on a cardiac monitoring device). Results: We reviewed 884 431 patient records in common between the two databases to identify 100 patients with CRAO, 6559 with ischemic stroke, and 1000 matched controls. After CRAO, the cumulative incidence of new AF at 2 years was 49.6% (95% CI, 37.4%­61.7%). Patients with CRAO had a higher rate of AF than controls (hazard ratio, 1.64 [95% CI, 1.17­2.31]) and a comparable rate to patients with stroke (hazard ratio, 1.01 [95% CI, 0.75­1.36]). CRAO was associated with a higher incidence of new stroke compared with matched controls (hazard ratio, 2.85 [95% CI, 1.29­6.29]). Conclusions: The rate of AF detection after CRAO is higher than that seen in age-, sex-, and comorbidity-matched controls and comparable to that seen after ischemic cerebral stroke. Paroxysmal AF should be considered as part of the differential etiology of CRAO, and those patients may benefit from long-term cardiac monitoring.

Bioinformatics Data Analysis of Next-Generation Sequencing Data from Heterogeneous Tumor Samples.

Tumor heterogeneity is a major challenge when it comes to treating cancer and also complicates research aimed at determining genetic sources for tumorigenesis. Leveraging high-throughput sequencing technology has been an effective approach for advancing our understanding of genetic diseases, and this type of data can also be used to better understand and make inferences about tumor heterogeneity. Here we describe the basics of genomics data analysis, as well as analysis pipelines for investigating tumor heterogeneity with next-generation sequencing data.

Single-Strand Consensus Sequencing Reveals that HIV Type but not Subtype Significantly Impacts Viral Mutation Frequencies and Spectra.

A long-standing question of human immunodeficiency virus (HIV) genetic variation and evolution has been whether differences exist in mutation rate and/or mutation spectra among HIV types (i.e., HIV-1 versus HIV-2) and among HIV groups (i.e., HIV-1 groups M-P and HIV-2 groups A-H) and HIV-1 Group M subtypes (i.e., subtypes A-D, F-H, and J-K). To address this, we developed a new single-strand consensus sequencing assay for the determination of HIV mutation frequencies and spectra using the Illumina sequencing platform. This assay enables parallel and standardized comparison of HIV mutagenesis among various viral vectors with lower background error than traditional methods of Illumina library preparation. We found significant differences in viral mutagenesis between HIV types but intriguingly no significant differences among HIV-1 Group M subtypes. More specifically, HIV-1 exhibited higher transition frequencies than HIV-2, due mostly to single G-to-A mutations and (to a lesser extent) G-to-A hypermutation. These data suggest that HIV-2 RT exhibits higher fidelity during viral replication, and taken together, these findings demonstrate that HIV type but not subtype significantly affects viral mutation frequencies and spectra. These differences may inform antiviral and vaccine strategies.

Truncation and constitutive activation of the androgen receptor by diverse genomic rearrangements in prostate cancer.

Molecularly targeted therapies for advanced prostate cancer include castration modalities that suppress ligand-dependent transcriptional activity of the androgen receptor (AR). However, persistent AR signalling undermines therapeutic efficacy and promotes progression to lethal castration-resistant prostate cancer (CRPC), even when patients are treated with potent second-generation AR-targeted therapies abiraterone and enzalutamide. Here we define diverse AR genomic structural rearrangements (AR-GSRs) as a class of molecular alterations occurring in one third of CRPC-stage tumours. AR-GSRs occur in the context of copy-neutral and amplified AR and display heterogeneity in breakpoint location, rearrangement class and sub-clonal enrichment in tumours within and between patients. Despite this heterogeneity, one common outcome in tumours with high sub-clonal enrichment of AR-GSRs is outlier expression of diverse AR variant species lacking the ligand-binding domain and possessing ligand-independent transcriptional activity. Collectively, these findings reveal AR-GSRs as important drivers of persistent AR signalling in CRPC.

Synergistic reduction of HIV-1 infectivity by 5-azacytidine and inhibitors of ribonucleotide reductase.

Although many compounds have been approved for the treatment of human immunodeficiency type-1 (HIV-1) infection, additional anti-HIV-1 drugs (particularly those belonging to new drug classes) are still needed due to issues such as long-term drug-associated toxicities, transmission of drug-resistant variants, and development of multi-class resistance. Lethal mutagenesis represents an antiviral strategy that has not yet been clinically translated for HIV-1 and is based on the use of small molecules to induce excessive levels of deleterious mutations within the viral genome. Here, we show that 5-azacytidine (5-aza-C), a ribonucleoside analog that induces the lethal mutagenesis of HIV-1, and multiple inhibitors of the enzyme ribonucleotide reductase (RNR) interact in a synergistic fashion to more effectively reduce the infectivity of HIV-1. In these drug combinations, RNR inhibitors failed to significantly inhibit the conversion of 5-aza-C to 5-aza-2'-deoxycytidine, suggesting that 5-aza-C acts primarily as a deoxyribonucleoside even in the presence of RNR inhibitors. The mechanism of antiviral synergy was further investigated for the combination of 5-aza-C and one specific RNR inhibitor, resveratrol, as this combination improved the selectivity index of 5-aza-C to the greatest extent. Antiviral synergy was found to be primarily due to the reduced accumulation of reverse transcription products rather than the enhancement of viral mutagenesis. To our knowledge, these observations represent the first demonstration of antiretroviral synergy between a ribonucleoside analog and RNR inhibitors, and encourage the development of additional ribonucleoside analogs and RNR inhibitors with improved antiretroviral activity.

5-Azacytidine Enhances the Mutagenesis of HIV-1 by Reduction to 5-Aza-2'-Deoxycytidine.

5-Azacytidine (5-aza-C) is a ribonucleoside analog that induces the lethal mutagenesis of human immunodeficiency virus type 1 (HIV-1) by causing predominantly G-to-C transversions during reverse transcription. 5-Aza-C could potentially act primarily as a ribonucleotide (5-aza-CTP) or as a deoxyribonucleotide (5-aza-2'-deoxycytidine triphosphate [5-aza-dCTP]) during reverse transcription. In order to determine the primary form of 5-aza-C that is active against HIV-1, Illumina sequencing was performed using proviral DNA from cells treated with 5-aza-C or 5-aza-dC. 5-Aza-C and 5-aza-dC were found to induce highly similar patterns of mutation in HIV-1 in terms of the types of mutations observed, the magnitudes of effects, and the distributions of mutations at individual sequence positions. Further, 5-aza-dCTP was detected by liquid chromatography-tandem mass spectrometry in cells treated with 5-aza-C, demonstrating that 5-aza-C was a substrate for ribonucleotide reductase. Notably, levels of 5-aza-dCTP were similar in cells treated with equivalent effective concentrations of 5-aza-C or 5-aza-dC. Lastly, HIV-1 reverse transcriptase was found to incorporate 5-aza-CTPin vitroat least 10,000-fold less efficiently than 5-aza-dCTP. Taken together, these data support the model that 5-aza-C enhances the mutagenesis of HIV-1 primarily after reduction to 5-aza-dC, which can then be incorporated during reverse transcription and lead to G-to-C hypermutation. These findings may have important implications for the design of new ribonucleoside analogs directed against retroviruses.

Lack of mutational hot spots during decitabine-mediated HIV-1 mutagenesis.

Decitabine has previously been shown to induce lethal mutagenesis of human immunodeficiency virus type 1 (HIV-1). However, the factors that determine the susceptibilities of individual sequence positions in HIV-1 to decitabine have not yet been defined. To investigate this, we performed Illumina high-throughput sequencing of multiple amplicons prepared from proviral DNA that was recovered from decitabine-treated cells infected with HIV-1. We found that decitabine induced an ≈4.1-fold increase in the total mutation frequency of HIV-1, primarily due to a striking ≈155-fold increase in the G-to-C transversion frequency. Intriguingly, decitabine also led to an ≈29-fold increase in the C-to-G transversion frequency. G-to-C frequencies varied substantially (up to ≈80-fold) depending upon sequence position, but surprisingly, mutational hot spots (defined as upper outliers within the mutation frequency distribution) were not observed. We further found that every single guanine position examined was significantly susceptible to the mutagenic effects of decitabine. Taken together, these observations demonstrate for the first time that decitabine-mediated HIV-1 mutagenesis is promiscuous and occurs in the absence of a clear bias for mutational hot spots. These data imply that decitabine-mediated G-to-C mutagenesis is a highly effective antiviral mechanism for extinguishing HIV-1 infectivity.

HIV-1 and HIV-2 exhibit similar mutation frequencies and spectra in the absence of G-to-A hypermutation.

BACKGROUND: Human immunodeficiency virus type 2 (HIV-2) is often distinguished clinically by lower viral loads, reduced transmissibility, and longer asymptomatic periods than for human immunodeficiency virus type 1 (HIV-1). Differences in the mutation frequencies of HIV-1 and HIV-2 have been hypothesized to contribute to the attenuated progression of HIV-2 observed clinically. RESULTS: To address this hypothesis, we performed Illumina sequencing of multiple amplicons prepared from cells infected with HIV-1 or HIV-2, resulting in ~4.7 million read pairs and the identification of ~200,000 mutations after data processing. We observed that: (1) HIV-2 displayed significantly lower total mutation, substitution, and transition mutation frequencies than that of HIV-1, along with a mutation spectrum markedly less biased toward G-to-A transitions, (2) G-to-A hypermutation consistent with the activity of APOBEC3 proteins was observed for both HIV-1 and HIV-2 despite the presence of Vif, (3) G-to-A hypermutation was significantly higher for HIV-1 than for HIV-2, and (4) HIV-1 and HIV-2 total mutation frequencies were not significantly different in the absence of G-to-A hypermutants. CONCLUSIONS: Taken together, these data demonstrate that HIV-2 exhibits a distinct mutational spectrum and a lower mutation frequency relative to HIV-1. However, the observed differences were primarily due to reduced levels of G-to-A hypermutation for HIV-2. These findings suggest that HIV-2 may be less susceptible than HIV-1 to APOBEC3-mediated hypermutation, but that the fidelities of other mutational sources (such as reverse transcriptase) are relatively similar for HIV-1 and HIV-2. Overall, these data imply that differences in replication fidelity are likely not a major contributing factor to the unique clinical features of HIV-2 infection.

Transposon Mutagenesis Screen Identifies Potential Lung Cancer Drivers and CUL3 as a Tumor Suppressor.

UNLABELLED: Non-small cell lung cancers (NSCLC) harbor thousands of passenger events that hide genetic drivers. Even highly recurrent events in NSCLC, such as mutations in PTEN, EGFR, KRAS, and ALK, are detected, at most, in only 30% of patients. Thus, many unidentified low-penetrant events are causing a significant portion of lung cancers. To detect low-penetrance drivers of NSCLC, a forward genetic screen was performed in mice using the Sleeping Beauty (SB) DNA transposon as a random mutagen to generate lung tumors in a Pten-deficient background. SB mutations coupled with Pten deficiency were sufficient to produce lung tumors in 29% of mice. Pten deficiency alone, without SB mutations, resulted in lung tumors in 11% of mice, whereas the rate in control mice was approximately 3%. In addition, thyroid cancer and other carcinomas, as well as the presence of bronchiolar and alveolar epithelialization, in mice deficient for Pten were also identified. Analysis of common transposon insertion sites identified 76 candidate cancer driver genes. These genes are frequently dysregulated in human lung cancers and implicate several signaling pathways. Cullin3 (Cul3), a member of a ubiquitin ligase complex that plays a role in the oxidative stress response pathway, was identified in the screen and evidence demonstrates that Cul3 functions as a tumor suppressor. IMPLICATIONS: This study identifies many novel candidate genetic drivers of lung cancer and demonstrates that CUL3 acts as a tumor suppressor by regulating oxidative stress.

Genetic signature of histiocytic sarcoma revealed by a sleeping beauty transposon genetic screen in mice.

Histiocytic sarcoma is a rare, aggressive neoplasm that responds poorly to therapy. Histiocytic sarcoma is thought to arise from macrophage precursor cells via genetic changes that are largely undefined. To improve our understanding of the etiology of histiocytic sarcoma we conducted a forward genetic screen in mice using the Sleeping Beauty transposon as a mutagen to identify genetic drivers of histiocytic sarcoma. Sleeping Beauty mutagenesis was targeted to myeloid lineage cells using the Lysozyme2 promoter. Mice with activated Sleeping Beauty mutagenesis had significantly shortened lifespan and the majority of these mice developed tumors resembling human histiocytic sarcoma. Analysis of transposon insertions identified 27 common insertion sites containing 28 candidate cancer genes. Several of these genes are known drivers of hematological neoplasms, like Raf1, Fli1, and Mitf, while others are well-known cancer genes, including Nf1, Myc, Jak2, and Pten. Importantly, several new potential drivers of histiocytic sarcoma were identified and could serve as targets for therapy for histiocytic sarcoma patients.

SHEAR: sample heterogeneity estimation and assembly by reference.

BACKGROUND: Personal genome assembly is a critical process when studying tumor genomes and other highly divergent sequences. The accuracy of downstream analyses, such as RNA-seq and ChIP-seq, can be greatly enhanced by using personal genomic sequences rather than standard references. Unfortunately, reads sequenced from these types of samples often have a heterogeneous mix of various subpopulations with different variants, making assembly extremely difficult using existing assembly tools. To address these challenges, we developed SHEAR (Sample Heterogeneity Estimation and Assembly by Reference; http://vk.cs.umn.edu/SHEAR), a tool that predicts SVs, accounts for heterogeneous variants by estimating their representative percentages, and generates personal genomic sequences to be used for downstream analysis. RESULTS: By making use of structural variant detection algorithms, SHEAR offers improved performance in the form of a stronger ability to handle difficult structural variant types and better computational efficiency. We compare against the lead competing approach using a variety of simulated scenarios as well as real tumor cell line data with known heterogeneous variants. SHEAR is shown to successfully estimate heterogeneity percentages in both cases, and demonstrates an improved efficiency and better ability to handle tandem duplications. CONCLUSION: SHEAR allows for accurate and efficient SV detection and personal genomic sequence generation. It is also able to account for heterogeneous sequencing samples, such as from tumor tissue, by estimating the subpopulation percentage for each heterogeneous variant.

Reasons for loss of cardiac resynchronization therapy pacing: insights from 32 844 patients.

BACKGROUND: The efficacy of cardiac resynchronization therapy (CRT) is associated with the amount of CRT pacing delivered. The specific causes of CRT pacing loss and their relative frequencies remain poorly defined. METHODS AND RESULTS: CRT patients who transmitted device data from 2006 to 2011 were screened for inclusion. Device diagnostics were analyzed using an automated algorithm to categorize CRT loss into 10 different causes. The algorithm was validated against manual adjudications using a portion of the entire cohort. There were 80 768 patients analyzed with a median time of 594 (interquartile range, 294-1003) days from implant to time of analysis. In this cohort, 40.7% of patients had <98% pacing, and 11.5% of patients had <90% pacing. For patients with <98% pacing, device diagnostics explained 55.8% of pacing loss: 30.6% atrial tachycardia/atrial fibrillation; 16.6% premature ventricular contractions; and 8.6% captured as episodes with at least 10 consecutive beats of CRT loss (ventricular sensing episodes). Inappropriately programmed sensed and paced atrioventricular (AV) intervals (SAV/PAV) accounted for 34.5% of all ventricular sensing episodes. As the severity of CRT loss increased, the contribution of atrial tachycardia/atrial fibrillation and SAV/PAV to the loss increased. Atrial tachycardia/atrial fibrillation accounted for >50% and premature ventricular contractions accounted for <10% of CRT loss in those with <90% CRT pacing. CONCLUSIONS: CRT pacing <98% was observed in 40.7% of patients. Among those with suboptimal pacing, atrial tachycardia/atrial fibrillation was the most common reason for CRT pacing loss. Inappropriately programmed SAV/PAV intervals was the most common reason for episodes of sustained loss of CRT pacing. This information can help in defining more effective treatments to improve CRT delivery.