Florian P. Breitwieser

Next-generation sequencing in neuropathologic diagnosis of infections of the nervous system

Objective: To determine the feasibility of next-generation sequencing (NGS) microbiome approaches in the diagnosis of infectious disorders in brain or spinal cord biopsies in patients with suspected CNS infections. Methods: In a prospective pilot study, we applied NGS in combination with a new computational analysis pipeline to detect the presence of pathogenic microbes in brain or spinal cord biopsies from 10 patients with neurologic problems indicating possible infection but for whom conventional clinical and microbiology studies yielded negative or inconclusive results. Results: Direct DNA and RNA sequencing of brain tissue biopsies generated 8.3 million to 29.1 million sequence reads per sample, which successfully identified with high confidence the infectious agent in 3 patients for whom validation techniques confirmed the pathogens identified by NGS. Although NGS was unable to identify with precision infectious agents in the remaining cases, it contributed to the understanding of neuropathologic processes in 5 others, demonstrating the power of large-scale unbiased sequencing as a novel diagnostic tool. Clinical outcomes were consistent with the findings yielded by NGS on the presence or absence of an infectious pathogenic process in 8 of 10 cases, and were noncontributory in the remaining 2. Conclusions: NGS-guided metagenomic studies of brain, spinal cord, or meningeal biopsies offer the possibility for dramatic improvements in our ability to detect (or rule out) a wide range of CNS pathogens, with potential benefits in speed, sensitivity, and cost. NGS-based microbiome approaches present a major new opportunity to investigate the potential role of infectious pathogens in the pathogenesis of neuroinflammatory disorders.

A review of methods and databases for metagenomic classification and assembly

Microbiome research has grown rapidly over the past decade, with a proliferation of new methods that seek to make sense of large, complex data sets. Here, we survey two of the primary types of methods for analyzing microbiome data: read classification and metagenomic assembly, and we review some of the challenges facing these methods. All of the methods rely on public genome databases, and we also discuss the content of these databases and how their quality has a direct impact on our ability to interpret a microbiome sample.

Presence of Human Hepegivirus-1 in a Cohort of People Who Inject Drugs

Humans are teeming with microbes that comprise our microbiome and contribute to health and disease. A typical human with 1012 nucleated cells is estimated to have 1015 virions and 1013 bacteria (1, 2). In the past, only the components of the microbiome that could be directly visualized or grown in culture were readily appreciated (3). Even advances in nucleic acid amplification, such as polymerase chain reaction (PCR), are constrained by the technical necessity of having to specify the suspected microbe before testing (4). However, a new frontier of microbial discovery has been opened by the deployment of next-generation metagenomic sequencing (NGMS), in which all of the DNA or RNA in a tissue is sequenced and interpreted with novel bioinformatics tools (58). In clinical practice, NGMS has been used to detect unsuspected pathogens (7, 9, 10), and it is also being used to characterize the composition of complex populations of recognized viruses, such as HIV-1 and hepatitis C virus (HCV) (deep sequencing) (11, 12). However, although NGMS can uncover novel sequences, the limits of detection are not clearly defined. When no microbial nucleic acid is detected by NGMS in a tissue, how confident can we be in excluding a pathogens role? When a drug-resistant virus is not detected in a plasma specimen by deep sequencing, how sure can we be that the virus is not present? Likewise, although NGMS provides a means to estimate the quantity of microbial species in a sample, it is not clear how those measurements compare with well-established clinical laboratory quantitative standards. These questions are critical for clinical applications of NGMS. In this study, we used NGMS to explore the plasma virome of persons heavily exposed to bloodborne infections through long-term injection drug use, in whom we had carefully quantified 2 RNA viruses (HIV and HCV), to determine whether NGMS would reveal additional viruses. To have a reference for the sensitivity of our approach, we studied persons enrolled in a clinical trial of pegylated interferon-2b (IFN) who had a broad, dynamic range of HIV and HCV plasma levels documented by clinical quantitative assays. Methods Participants Characterization of Plasma Nucleic Acids by NGMS Plasma samples were studied from persons co-infected with HIV and HCV who were enrolled in a prospective study of HCV dynamics after IFN administration before and after antiretroviral therapy (ClinicalTrials.gov: NCT01285050) (13). Briefly, participants were enrolled in a study of response to IFN. After a core liver biopsy specimen was obtained, phlebotomy was performed at structured intervals before and after IFN administration. For the present study, pre-IFN samples and the corresponding samples collected at 72 or 168 hours after IFN administration were used. The plasma was centrifuged within 30 minutes of collection and stored at 80C until testing. As detailed in the Sample Preparation for NGMS and Quantitative PCR section, sufficient RNA and DNA was obtained at both time points for NGMS for 8 and 10 of the 20 participants, respectively (Appendix Figure 1); thus, there were 8 pairs of samples for RNA NGMS and 10 pairs for DNA NGMS. Appendix Figure 1. Flow chart showing sample loss during nucleic acid library construction. Characterization of Human Hepegivirus-1 To study human hepegivirus-1 (HHpgV-1) viremia, 177 plasma samples from 156 persons were selected. These persons were participants in ALIVE (AIDS Linked to the Intravenous Experience), a well-characterized cohort study of injection drug users (14, 15). To enhance investigation of co-infection with other bloodborne viruses, participants in this study were selected from more than 2000 participants who had samples that had previously been found to contain GB virus C (GBV-C) RNA (n= 20), SEN virus (SENV) DNA (n= 24), HCV RNA (n= 42), and HCV antibodies but not HCV RNA (n= 50). Because we also identified HHpgV-1 sequences in liver tissue (see Results), we included plasma from persons with unexplained alanine aminotransferase levels more than 10 times the upper limit of normal (n= 20). Persistence of HHpgV-1 viremia and its association with hepatitis C viremia were determined by further testing of plasma samples collected at additional time points from persons with HHpgV-1 viremia (n= 11). Sample Preparation for NGMS and Quantitative PCR Nucleic Acid Extraction The ZR-Duet DNA/RNA MiniPrep kit (Zymo Research) was used to extract DNA and RNA from 200 L of plasma. The kit isolated and purified DNA and RNA separately without the use of carrier RNA. Preextraction steps included spinning the samples at 1600 g for 15 minutes at 4C to remove debris (such as insoluble complexes), followed by filtration using 0.2-m Millex syringe filters (EMD Millipore). The filters were ideal for low sample volumes (<1 mL) because their low holdup volumes (<10 L) resulted in negligible volume loss of the filtrate. NGMS The NGMS library constructions were done using the Ovation Single Cell RNA-Seq System (NuGEN) and the Ovation Ultralow System V2 (NuGEN) for RNA and DNA, respectively. The DNA was sheared using a Bioruptor (Diagenode) with the following settings: 30 seconds on and 30 seconds off for 13 cycles. The size distribution of the sheared DNA samples was analyzed using a 2100 Bioanalyzer (Agilent Technologies). The kits were selected on the basis of their ability to generate libraries from low-concentration inputs. In addition, during library preparation, 4 extra amplification cycles were incorporated in each of the 2 amplification steps to increase the final concentration of the libraries. We were able to generate good-quality sequencing libraries for 8 of 20 participants for RNA and 10 of 20 participants for DNA. The libraries were bar-coded, pooled (10 samples per lane), and sequenced at the Johns Hopkins Genetic Resources Core Facility using a HiSeq 2500 System (Illumina) in high-output mode with a read length of 2100 bp reads (approximately 500 million reads per lane). Quantitative PCR Plasma HCV and HIV RNA testing were done using commercial kits from Abbott, as previously described (13). Quantitative PCR for HHpgV-1 was performed by using primers targeting the NS2-3 region, as described by Berg and colleagues (6). Quantitation standards for the HHpgV-1 PCR were developed using gBlocks Gene Fragments (Integrated DNA Technologies). Analysis Kraken, version 0.10.5-beta (16), was used for metagenomics read classification, with a custom database built from 1) contaminant sequences from the EMVec and UniVec databases as well as other low-complexity sequences (to discard possible laboratory contaminants and nonmicrobial sequences); 2) the human genome build GRCh38.p2 and the mouse genome build GRCm38.p4 (the latter was done to discard less common cases of contamination); 3) all complete genomes in the RefSeq database as of 13 January 2016 in the bacterial (4111 genomes), archaeal (202 genomes), and viral (5412 genomes) domains; 4) all viral genomes listed on the National Center for Biotechnology Information Viral Genomes Resource (17) as of 13 January 2016 (84272 genomes); and 5) 14 fungal pathogen genomes and 11 protist pathogen genomes. The Kraken index had a total size of 154 GB. The nonhuman and noncontaminant reads were extracted from the Kraken results and aligned to HHpgV-1 sequences collected from GenBank (accessions NC_027998.2, KT427413.1, KT427408.1, KT427407.1, KU159665.1, KU159664.1, KT427414.1, KT427412.1, KT427411.1, KT427410.1, KT427409.1, and KT439329.1) using the very-sensitive-local option in Bowtie 2, version 2.2.6 (18). Polymerase chain reaction duplicates were removed from the aligned reads using Picard (http://broadinstitute.github.io/picard), and the reads were quality-trimmed using seqtk trimfq with the q 0.01 option. The assembly was done using the -meta option in SPAdes, version 3.6.0 (19). The alignments were visualized using Pavian (20). All available HHpgV-1 genome sequences were compared in SeaView (21); the multiple-sequence analysis was done with Clustal Omega (22), and phylogeny reconstruction was done for NS5B sequences in MEGA7 using the JukesCantor model and the maximum-likelihood algorithm, with 1000 bootstrap replicates used to calculate branch strength (23). Hepatitis C virus (KX621472) was used as the out-group sequence for the analysis. To compare the quantity of viral reads by NGMS versus our clinical standards, viral reads were expressed per million mapped reads to normalize the sequence data. Institutional Review Board Approval Samples were obtained from persons who provided consent using forms and a protocol approved by the Institutional Review Board of the Johns Hopkins University School of Medicine and the Johns Hopkins Bloomberg School of Public Health. Role of the Funding Source The study was funded by the National Institutes of Health (NIH), which had no role in study design, data collection and interpretation, or the decision to submit the manuscript for publication. Results Characterization of Plasma Nucleic Acids From the 18 plasma samples, approximately 600 million nucleic acid sequences (paired reads) were identified (an average of 16 million paired reads per sample). The reads were classified using the Kraken program, which compares each read with a large database of viruses and other species (16). In addition to the expected HCV- and HIV-derived RNA reads, sequences were detected that aligned with the novel RNA virus HHpgV-1 (5, 6). There also were reads that assigned to murine leukemia virus and orthopoxviruses, such as vaccinia and ectromelia virus (Appendix Figure 2). However, these seemed to be artifactual because the reads were nearly identical and mapped to the same region of the reference sequence. Most DNA reads (99%) mapped as expected to the human genome. The predominant DNA viral reads mapped to the EpsteinBarr virus (6 of 10 participants), members of the human endogenous retrovirus K (6 of 10 participant

Pavian: Interactive analysis of metagenomics data for microbiomics and pathogen identification

Summary Pavian is a web application for exploring metagenomics classification results, with a special focus on infectious disease diagnosis. Pinpointing pathogens in metagenomics classification results is often complicated by host and laboratory contaminants as well as many non-pathogenic microbiota. With Pavian, researchers can analyze, display and transform results from the Kraken and Centrifuge classifiers using interactive tables, heatmaps and flow diagrams. Pavian also provides an alignment viewer for validation of matches to a particular genome. Availability and implementation Pavian is implemented in the R language and based on the Shiny framework. It can be hosted on Windows, Mac OS X and Linux systems, and used with any contemporary web browser. It is freely available under a GPL-3 license from http://github.com/fbreitwieser/pavian. Furthermore a Docker image is provided at https://hub.docker.com/r/florianbw/pavian. Contact fbreitw1@jhu.edu Supplementary information Supplementary data is available at Bioinformatics online.

Identifying Corneal Infections in Formalin-Fixed Specimens Using Next Generation Sequencing

Purpose We test the ability of next-generation sequencing, combined with computational analysis, to identify a range of organisms causing infectious keratitis. Methods This retrospective study evaluated 16 cases of infectious keratitis and four control corneas in formalin-fixed tissues from the pathology laboratory. Infectious cases also were analyzed in the microbiology laboratory using culture, polymerase chain reaction, and direct staining. Classified sequence reads were analyzed with two different metagenomics classification engines, Kraken and Centrifuge, and visualized using the Pavian software tool. Results Sequencing generated 20 to 46 million reads per sample. On average, 96% of the reads were classified as human, 0.3% corresponded to known vectors or contaminant sequences, 1.7% represented microbial sequences, and 2.4% could not be classified. The two computational strategies successfully identified the fungal, bacterial, and amoebal pathogens in most patients, including all four bacterial and mycobacterial cases, five of six fungal cases, three of three Acanthamoeba cases, and one of three herpetic keratitis cases. In several cases, additional potential pathogens also were identified. In one case with cytomegalovirus identified by Kraken and Centrifuge, the virus was confirmed by direct testing, while two where Staphylococcus aureus or cytomegalovirus were identified by Centrifuge but not Kraken could not be confirmed. Confirmation was not attempted for an additional three potential pathogens identified by Kraken and 11 identified by Centrifuge. Conclusions Next generation sequencing combined with computational analysis can identify a wide range of pathogens in formalin-fixed corneal specimens, with potential applications in clinical diagnostics and research.

KrakenHLL: Confident and fast metagenomics classification using unique k-mer counts

Motivation False positive identifications are a significant problem in metagenomics. When matching sequencing reads to a database of genomes, low-complexity regions and contaminants in genome assemblies can attract many reads to the wrong genomes. Alignment information may be used to recognize that many reads are piling up in a small number of locations, but fast k-mer based metagenomic classifiers do not provide alignments, and re-alignment is expensive. We propose using k-mer coverage, which can be computed during classification, as a proxy for genome base coverage. Results We present KrakenHLL, a sequence classifier that records the number of unique k-mers observed in each taxon in a metagenomics data set. KrakenHLL is based on the ultra-fast classification engine Kraken, which it combines with a cardinality estimator known as HyperLogLog (HLL). We demonstrate that many false-positive identifications can be identified using unique k-mer counts, especially when looking at species present in low abundance. KrakenHLL’s database includes over 100,000 additional viral strain sequences not used in the standard Kraken database, and it also has the ability to map against multiple databases and identify plasmids and strains using an extended taxonomy. KrakenHLL’s run time is similar to that of Kraken, and it requires only a very modest increase in memory. Availability and Implementation KrakenHLL is implemented in C++ and Perl, and available under the GPL v3 license at https://github.com/fbreitwieser/krakenhll. Contact florian.bw@gmail.com.

Re-analysis of metagenomic sequences from acute flaccid myelitis patients reveals alternatives to enterovirus D68 infection

Metagenomic sequence data can be used to detect the presence of infectious viruses and bacteria, but normal microbial flora make this process challenging. We re-analyzed metagenomic RNA sequence data collected during a recent outbreak of acute flaccid myelitis (AFM), caused in some cases by infection with enterovirus D68. We found that among the patients whose symptoms were previously attributed to enterovirus D68, one patient had clear evidence of infection with Haemophilus influenzae, and a second patient had a severe Staphylococcus aureus infection caused by a methicillin-resistant strain. Neither of these bacteria were identified in the original study. These observations may have relevance in cases that present with flaccid paralysis because bacterial infections, co-infections or post-infection immune responses may trigger pathogenic processes that may present as poliomyelitis-like syndromes and may mimic AFM. A separate finding was that large numbers of human sequences were present in each of the publicly released samples, although the original study reported that human sequences had been removed before deposition.

Statistical analysis of co-occurrence patterns in microbial presence-absence datasets

Drawing on a long history in macroecology, correlation analysis of microbiome datasets is becoming a common practice for identifying relationships or shared ecological niches among bacterial taxa. However, many of the statistical issues that plague such analyses in macroscale communities remain unresolved for microbial communities. Here, we discuss problems in the analysis of microbial species correlations based on presence-absence data. We focus on presence-absence data because this information is more readily obtainable from sequencing studies, especially for whole-genome sequencing, where abundance estimation is still in its infancy. First, we show how Pearson’s correlation coefficient (r) and Jaccard’s index (J)–two of the most common metrics for correlation analysis of presence-absence data–can contradict each other when applied to a typical microbiome dataset. In our dataset, for example, 14% of species-pairs predicted to be significantly correlated by r were not predicted to be significantly correlated using J, while 37.4% of species-pairs predicted to be significantly correlated by J were not predicted to be significantly correlated using r. Mismatch was particularly common among species-pairs with at least one rare species (<10% prevalence), explaining why r and J might differ more strongly in microbiome datasets, where there are large numbers of rare taxa. Indeed 74% of all species-pairs in our study had at least one rare species. Next, we show how Pearson’s correlation coefficient can result in artificial inflation of positive taxon relationships and how this is a particular problem for microbiome studies. We then illustrate how Jaccard’s index of similarity (J) can yield improvements over Pearson’s correlation coefficient. However, the standard null model for Jaccard’s index is flawed, and thus introduces its own set of spurious conclusions. We thus identify a better null model based on a hypergeometric distribution, which appropriately corrects for species prevalence. This model is available from recent statistics literature, and can be used for evaluating the significance of any value of an empirically observed Jaccard’s index. The resulting simple, yet effective method for handling correlation analysis of microbial presence-absence datasets provides a robust means of testing and finding relationships and/or shared environmental responses among microbial taxa.

Deconvoluting Virome-Wide Antiviral Antibody Profiling Data

The ability to comprehensively characterize exposures and immune responses to viral infections will be critical to better understanding human health and disease. We previously described the VirScan system, a phage-display based technology for profiling antibody binding to a comprehensive library of peptides designed to represent the human virome. The previous VirScan analytical approach did not fully account for disproportionate representation of viruses in the library or for antibody cross-reactivity among sequences shared by related viruses. Here we present the ‘AntiViral Antibody Response Deconvolution Algorithm’ (‘AVARDA’), a multi-module software package for analyzing VirScan datasets. AVARDA provides a probabilistic assessment of infection at species-level resolution by considering alignment of all library peptides to each other and to all human viruses. We employed AVARDA to analyze VirScan data from a cohort of encephalitis patients with either known viral infections or undiagnosed etiologies. By comparing acute and convalescent sera, AVARDA successfully confirmed or detected antibody responses to human herpesviruses 1, 3, 4, 5, and 6, thereby improving the rate of diagnosing viral encephalitis in this cohort by 62.5%. We further assessed AVARDA’s utility in the setting of an epidemiological study, demonstrating its ability to determine infections acquired in a child followed prospectively from infancy. We consider ways in which AVARDA’s conceptual framework may be further developed in the future and describe how its analyses may be extended beyond investigations of viral infection. AVARDA, in combination with VirScan and other pan-pathogen serological techniques, is likely to find broad utility in the epidemiology and diagnosis of infectious diseases.

CMPK2 and BCL-G are associated with type 1 interferon–induced HIV restriction in humans

We identified two genes induced by type 1 interferon in activated CD4+ T cells that are associated with HIV restriction in humans. Type 1 interferons (IFN) are critical for host control of HIV and simian immunodeficiency virus. However, it is unknown which of the hundreds of interferon-stimulated genes (ISGs) restrict HIV in vivo. We sequenced RNA from cells that support HIV replication (activated CD4+ T cells) in 19 HIV-infected people before and after interferon-α2b (IFN-α2b) injection. IFN-α2b administration reduced plasma HIV RNA and induced mRNA expression in activated CD4+ T cells: The IFN-α2b–induced change of each mRNA was compared to the change in plasma HIV RNA. Of 99 ISGs, 13 were associated in magnitude with plasma HIV RNA decline. In addition to well-known restriction factors among the 13 ISGs, two novel genes, CMPK2 and BCL-G, were identified and confirmed for their ability to restrict HIV in vitro: The effect of IFN on HIV restriction in culture was attenuated with RNA interference to CMPK2, and overexpression of BCL-G diminished HIV replication. These studies reveal novel antiviral molecules that are linked with IFN-mediated restriction of HIV in humans.