Deep Sequencing of RNA from Blood and Oral Swab Samples Reveals the Presence of Nucleic Acid from a Number of Pathogens in Patients with Acute Ebola Virus Disease and Is Consistent with Bacterial Translocation across the Gut

Rather than first trying to identify separate pathogens, based on predetermined lists of pathogens that could be potentially present in blood samples taken from individual patients, an unbiased search approach was taken. Sequencing data from all the acute patients with EVD and convalescent patients were pooled to first determine which pathogens were present, using a combination of de novo and reference-based mapping approaches. The identified pathogens were then used to form a specific query database at the individual patient level to determine which pathogens were present in which samples, thus reducing computational time.

Sequence reads were combined from all the patients in these groups and any that aligned to the human genome were removed from the analysis, leaving approximately 1 TB of fastq data. Next, these reads were processed by using the Trinity read normalization software before assembling into ~1.2 million contigs using Trinity. These contigs were searched in BLAST using Diamond BLAST to find the nearest homologues and then they were manually searched to confirm the presence of the EBOV genome. As expected, a near-full-length genome corresponding to EBOV was assembled by use of Trinity, allowing calibration and verification of the approach. In addition, evidence of genetic material belonging to other pathogens was identified (provided as a fully searchable Excel file in Table S1 in the supplemental material and abstracted in Fig. 2).

Examples are also presented in Table 2 (for bacterial, fungal, and parasitic pathogens) and Table 3 (for viral pathogens) to illustrate the mix of microorganisms and gene products identified. The largest number of reads mapped to Acinetobacter spp. (prevalent in intensive care units), Echinococcus granulosus (a tape worm), Bacteroides fragilis (a commensal bacterium associated with the normal flora of the colon), Trichuris trichiura (a whipworm normally found in the intestine), and Burkholderia multivorans (a bacterium associated with chronic infections in cystic fibrosis), followed by EBOV. The data immediately illustrated the potential complexity of microorganisms in acutely ill patients who are positive for EBOV and that these other organisms may potentially confound treatment. Obvious from this effort was the identification of sequence reads mapping to Plasmodium falciparum, the causative agent of malaria tropica. The presence of this parasite was also confirmed by an antigen-based approach (a rapid diagnostic test [RDT]) for many of the samples in the European Mobile Laboratory deployed in Guinea.

The convalescent control group provided a means to test the high-throughput RNA sequencing (RNA-seq) and bioinformatics approaches in terms of sample handling and potential environmental contamination during the stages involved in obtaining samples from patients, sample processing, and transport, all the way through to sequencing. The prediction was that in the convalescent group of patients that had been shown to EBOV negative by qRT-PCR and Plasmodium negative through RDT, these and other organisms should not have been found, as blood is effectively sterile. In this group of 16 patients, no transcripts mapping to Plasmodium falciparum or to EBOV were identified. In only one patient were transcripts identified that mapped to a bacterium, which was identified as Pseudomonas at the species level (Fig. 3). These results thus suggest that any potential environmental contamination was minimal and the informatics pipeline did not identify spurious “hits.”

Sequence analysis of the RNA prepared from swabs taken from community deaths (n = 24) illustrated a range of nucleic acids mapping to bacteria that would be expected in the oral microbiome. This included sequence reads mapping to Neisseria meningitidis, a Gram-negative bacterium that can cause sepsis (Fig. 4). A mass spectrometry-based approach was then used to confirm the identification of individual bacterial species in the oral microbiome of patients in the community death cohort. Oral samples were cultured at high containment facilities, and colonies were recovered from the throat swabs of five separate patients in the community death group; these samples tested positive for bacteria and fungi by mass spectrometry (Table 4). A score for the matrix-assisted laser desorption ionization–time of flight (MALDI-TOF) results between 2.000 and 2.2999 indicated secure genus identification, and scores between 2.300 and 3.000 indicated a high-probability species identification. Identification of bacteria by mass spectrometry matched those identified by the RNA-seq and computational approaches. For example, Proteus mirabilis was identified by RNA-seq and by mass spectrometry from swabs taken from two patients, and Rothia mucilaginosa and Streptococcus salivarius were identified in individual patients.

Nucleic acids from both bacteria (Table 2; Fig. 3) and fungi (Table 2) were identified in the blood from EBOV-positive patients presenting at the Treatment Centre and testing positive for EBOV. Confirmatory blood cultures were attempted at high containment but were not successful, possibly due to multiple freeze-thaw cycles and suboptimal sample transport. Although it was problematic to distinguish what may be commensal microorganisms from these data, nevertheless, several bacterial species were identified, based on sequence reads mapped, that have been associated with disease in humans, particularly sepsis (15). These species included Acinetobacter baumannii, which thrives in individuals with weakened immune systems and is considered a nosocomial agent and has multidrug resistance (16). Streptococcus pneumoniae and Haemophilus influenzae are found asymptomatically in the upper airways of asymptomatic healthy carriers, but if found in the bloodstream, both pathogens can cause pneumonia, meningitis, and sepsis. Haemophilus influenzae was identified in both the blood (and swab) samples. Several patients had evidence of a wide range of bacteria, based on nucleic acid sequences identified, including one patient in the hospitalized survivors category. This patient had evidence of nucleic acids that mapped to bacteria associated with sexual transmission and vaginal infection, including Gardnerella vaginalis, Lactobacillus iners, and Ureaplasma urealyticum. Nucleic acids from Enterobacteriaceae was also identified in many patients. These are the predominant pathogenic bacteria in the bowel, where bacterial translocation would be expected to originate in EVD. Examples included sequence reads mapping to Gram-negative bacteria such as Escherichia coli, Klebsiella spp., Enterobacter spp., Shigella spp., Salmonella spp., Pseudomonas spp., and Gram-positive bacteria such as Enterococcus spp. Therefore, the data indicate that most patients with EVD sequenced in this study had evidence of bacterial translocation across the gut.

Analysis of the host response in peripheral blood samples from patients with acute EVD identified several differences compared to healthy controls (4). There were also differences in the host transcriptomes between patients with EVD who survived versus those who succumbed to EVD (4). These included pathways involved in the innate response, blood coagulation, and the acute-phase response (4). To investigate whether the potential presence of bacteria in the blood of patients with EVD influenced the host response, transcriptome changes in these exemplary pathways were compared between the different groups of patients. This analysis made use of matching host transcriptome data that had been determined for this group of patients (4). The abundance of transcripts mapping to selected genes were chosen to assess the effect of pathogen burden on the different exemplary pathways. The data indicated that certain transcripts associated with the acute-phase response (intercellular adhesion molecule 1 [ICAM-1] and transforming growth factor β1 [TGF-β1]) were present at different abundance levels in EVD patients with versus without sequence reads mapping to bacteria in their blood (Fig. 5). In contrast, there were no significant differences in the abundance levels of selected transcripts associated with blood clotting, interferon (IFN) stimulation, or inflammation, at least in the limited sample sets used in this study.

Matching sequencing reads to the Alveolata superphylum from the blood samples suggested that most transcripts corresponded to P. falciparum (Fig. S1), and on a per-patient basis, the presence of the malaria parasite was confirmed by RDT. The data indicated that 84% of blood and plasma samples contained sequence reads mapping to P. falciparum transcripts. However, these sequence reads varied from an FPKM (a measure of sequence read depth and abundance of RNA transcripts mapping to P. falciparum) of 59, in a sample from a patient with acute EVD with the lowest number of sequences, to an FPKM of 358820, representing the highest number of sequence reads mapping to Plasmodium spp.

Comparison with the RDT results suggested that the presence of P. falciparum may have been underrepresented in the sampled patient population (Fig. 6). The data were then analyzed to investigate whether the P. falciparum load influenced the outcome of EVD (Fig. 6). As the RDT is qualitative, sequence read depth was used as a proxy measurement for the amount of Plasmodium present, and these results are akin to those from qRT-PCR, which is used as a confirmatory technique for the laboratory diagnosis of malaria (17, 18). Here, sequence read depths were divided into arbitrary bins based on approximately the same numbers of patients. Of the patients experiencing the highest reads mapping to P. falciparum, 87% succumbed to a fatal infection (n = 6). However, while the data indicated that in general mortality increased with sequence read depth, there was no absolute correlation, at least at the sequence read depths measured in this analysis.

For EVD patients with the highest depth of sequence mapping to P. falciparum, the host response may have been different for patients infected only with EBOV. To investigate this, the abundance of transcripts corresponding to markers of innate immunity and associated with malaria tropica were compared. The two groups of patients selected were patients with EBOV that were positive for P. falciparum, determined by both RNA-seq and RDT and using data from patients with the highest mapped reads to P. falciparum (n = 10), versus patients with EBOV but for whom both RDT and sequence read results were negative for P. falciparum (n = 13). This analysis made use of matching host transcriptome data that had been determined for this group of patients (4). Selected genes were chosen whose abundance levels had previously been associated with severe malaria tropica in the absence of EBOV infection (IFIT2, IFIT3, IFITM3, ISG15, STAT1, MX1, TGF-β, PRF1, PGRMC1, CTSW, ICAM-1, CD36, IFN-γ, interleukin-10, and IFN regulatory factor 9) (19–22). The abundance levels of the gene transcripts were compared, and these data are presented in two ways (Fig. 7). The first way shows the mean value plus the standard deviation for each transcript. The second way shows the median value plus the total range; the latter analysis is additionally presented because the transcript abundance in individual patients did not follow a normal distribution. The data indicated no significant differences in the abundance of transcripts associated with the innate response between patients who were EBOV positive and P. falciparum negative compared to patients who were positive for both EBOV and P. falciparum. This suggested that P. falciparum has no additional effect on the innate response. A similar analysis was performed on the abundance of transcripts associated with the coagulation pathway (FGA, FGB, FGG, FGL1, and ALB) (Fig. 8). These were the most differentially regulated transcripts between the EVD and control groups (4). The data indicated that transcripts associated with the coagulation pathway were increased in abundance in several of the EVD patients who tested negative for P. falciparum, compared to EVD patients who tested positive for P. falciparum. Note that this analysis was based on a limited patient sample size (23 in total) due to the stringent inclusion criteria used.

Nucleic acid mapping to several different viruses were identified in the blood of EVD patients. These included GB virus C, human herpesvirus 4 (HHV4, or Epstein-Barr virus), human herpesvirus 5 (cytomegalovirus), and rotavirus. The latter virus is associated with gastroenteritis and may be present in the blood through gastrointestinal bleeding (23) and has been associated with extraintestinal disease (24). No evidence was found in the samples for common circulating respiratory viruses. Data indicated that in approximately 30% of the samples, nucleic acid mapping to HHV4 was identified. The identification of BALF2 mRNA transcript from HHV4 (Table 3) implies there was HHV4 activation in EVD patients (25). In contrast to a previously report (9), in our patient sample set we found no correlation between the presence of nucleic acid from GBV-C virus and survival. However, our sample population included only 5 patients identified to have nucleic acid from GBV-C, with only 3 of these samples have been obtained during acute infection with EBOV. Nucleic acid mapping to HHV4 was most prevalent in the 1- to 4-years age group category (as was P. falciparum), whereas nucleic acid mapping to GBV-C virus appeared more prevalent in older age groups (Fig. 9). Somewhat surprisingly, several plant virus genomes/transcripts were also identified, with the highest and most confident hit being tobacco mosaic virus (TMV), for which the full genome was recovered. Our review of the literature suggested that serum antibodies to TMV can be identified in humans (26). Taken at face value, our data set suggest that plant virus genomes (and, by implication, plant viruses) can be found in blood, and this finding again may be indicative of a breakdown in the integrity of the gut wall. However, environmental contamination cannot be ruled out, especially during sample gathering. However, we note that at least for sequencing, none of the staff was cigarette or cigar smokers, and no plants were grown in the laboratory area.

Four main categories of patients were selected for analysis (Table 1). In the first two groups, RNA was extracted from (mainly) blood or plasma samples, taken from acutely ill patients during assessment of potential EVD cases, by the European Mobile Laboratory, located in the Médecins Sans Frontières (MSF) Ebola Treatment Centre Guéckédou in Guinea. These patients either went on to survive (hospitalized survivors, n = 44) or had a fatal infection (hospitalized fatalities, n = 118). These patients were also tested for Plasmodium. For the latter analysis, noninactivated EDTA blood was applied to an immunochromatographic assay (BinaxNOW Malaria; Alere GmbH, Cologne, Germany). These acute patients were treated according to MSF guidelines. In the third category, RNA was prepared from blood samples taken from patients who were convalescent for EVD and were qRT-PCR negative for EBOV (convalescent controls, n = 16). In a fourth group, RNA was extracted from oral and throat swabs taken from people who had already died of EVD (community deaths, n = 24). This last group was selected to evaluate microbiome sampling using swabs, as these are a suggested alternative to peripheral blood for determining EBOV status or an alternative to cardiac puncture blood for deceased patients. All RNA was stored at −20°C until shipment.

The National Committee of Ethics in Medical Research of Guinea approved the use of diagnostic leftover samples and corresponding patient data for this study (permit number 11/CNERS/14). Ethical permission for the sequencing work conducted at the University of Liverpool on RNA from patient samples was reviewed and approved by the institution under reference number RETH000784. Permission to sequence biological samples (made safe) containing genetic material from human pathogen hazard group 4 viruses was granted by the UK Home Office and the United Kingdom National Counter Terrorism Security Office. Ethical permission was also obtained from the Ethik-Kommission Der Arztekammer Hamburg (PV4910). Experimental methods complied with the Helsinki Declaration. For samples obtained from community deaths, hospitalized survivors, and hospitalized fatalities, these samples were collected as part of the public health response to contain the outbreak in Guinea, and informed consent was not obtained from these patients. For samples from convalescent patients, informed consent was obtained.

There was no significant difference between onset of symptoms (as reported by the patient) and sample procurement for admittance into the Ebola Treatment Centre (Fig. 1).

Extracted RNA was DNase treated and sequenced using a HiSeq2500 system, with no preamplification of any sequence, as described elsewhere (2, 35) and reproduced herein. Samples from patients were sequenced on a HiSeq2500 system, and several criteria were applied to the selection of data postsequencing. The RNA was DNase treated using Ambion Turbo DNase. RNA-seq libraries were prepared from the DNase-treated RNA by using the Epicentre ScriptSeq v2 RNA-seq library preparation kit and performing 10 to 15 cycles of amplification. Libraries were purified using AMPure XP beads. Each library was quantified using Qubit, the size distribution was assessed using the Agilent 2100 Bioanalyser, and the final libraries were pooled in equimolar ratios. The quantity and quality of each pool were assessed with the Bioanalyzer and subsequently by qPCR using the Illumina library quantification kit from Kapa on a Roche Light Cycler LC480II system according to the manufacturer’s instructions. The template DNA was denatured according to the protocol described in the Illumina User Guide and loaded at 12 pM. To improve sequencing quality control, samples were spiked with 1% PhiX. The sequencing was undertaken on the Illumina HiSeq 2500 with version 4 chemistry, generating 2- by 125-bp paired-end reads. Base calling and demultiplexing of indexed reads were performed by using Casava version 1.8.2 (Illumina) to produce all the sequence data in fastq format. The raw fastq files were trimmed to remove Illumina adapter sequences by using Cutadapt version 1.2.1. The option -O 3′ was set, so that the 3′ ends of any reads which matched the adapter sequence over at least 3 bp were trimmed off. The reads were further trimmed to remove low-quality bases, using Sickle version 1.200 with a minimum window quality score of 20. After trimming, reads shorter than 10 bp were removed. Resulting sequence reads were initially mapped to the human genome using the short-read mapper Bowtie2 as previously described (2). Those that did not map to the human genome were then pooled and mapped again to a list of known human transcripts again using Bowtie2. Sequences that did not align to either the human genome or human transcriptome were pooled, but the size of the data set (~1 TB of fastq reads) required them to be preprocessed first by the Trinity read normalization software to reduce the memory requirements and runtimes prior to de novo assembly using Trinity (36). The assembled data from Trinity were first manually checked to determine that the EBOV genome had been assembled before all the reads were checked for matches using Diamond BLASTX analysis against the nonredundant protein database (37). Vertebrate BLASTX hits, including some for several nonhuman primates, were excluded, as these were deemed likely high-copy human sequences which had passed through the previous filters. Thirty-two contigs with a top hit to Medicago sativa (alfalfa) were also excluded, as these had a taxonomically diverse array of matches with similar E values, indicative of spurious annotation. For each transcript, the best hit was retained and the protein name and organism for that hit was associated with the transcript using in-house scripts. Quality and primer trimmed reads were used as input for MetaPhlAn 2 for estimating genome coverage of bacterial species for each sample. In general, for viral species a 20% mapping to a transcriptome was adopted as a threshold for positive identification.

Selected EBOV-positive samples underwent bacterial cultivation under biosafety level 4 conditions, including samples of whole blood and throat swabs. The samples were inoculated into bouillon broth enriched with 10% fetal bovine serum, incubated at 37°C in the presence of 5% CO₂, and after 24 h, 48 h, and 72 h the samples were plated onto blood agar. The mixed cultures were isolated and underwent three subculture passages on blood agar. Colonies were identified from throat swabs, but not from peripheral blood samples, and the colony samples were analyzed by MALDI-TOF (Bruker MALDI Biotyper).

Raw sequence read data for all the samples used in this analysis were deposited with NCBI, BioProject ID PRJNA352396 . Note that the data contain no patient-identifiable information, only the outcome of infection.