Coinfections identified from metagenomic analysis of cervical lymph nodes from tularemia patients

Underlying coinfections in primary infectious disease are an important variable to consider but are commonly undetected. A growing body of literature points to the high occurrence (10–50%) of coinfections [1–5], and > 75% of coinfections from diverse origins have an overall negative impact on human health [6]. Underlying coinfections complicate disease presentation [7, 8] and the ability to detect their presence is highly relevant to inform medical treatment. The under-diagnosis of coinfections is due, among other things, to a lack of clinical suspicion, overlapping symptoms, and/or the fact that traditional tools have limited ability to identify coinfections in the absence of a priori knowledge. Thus, exploration of new diagnostic strategies is necessary to advance the understanding of the contribution of coinfections to infectious disease manifestations and treatment responses.

Significant advances in next generation sequencing have recently made metagenomics, an unbiased shotgun method of analysis, a widely used tool in just about all areas in biology, including infectious disease diagnostics [9, 10]. Metagenomics is powerful for its ability to diagnose unsuspected microbial agents [11]. It directly analyzes samples in their entirety, which removes the requirement for a priori knowledge to obtain comprehensive information. In this capacity, metagenomics surpasses the limits encountered with traditional diagnostics.

Many infectious disease-causing microbes are considered foreign by the host immune system and, therefore, are actively routed to the lymph nodes. As a consequence, lymph nodes capture and house disparate microbes [12] regardless of their transmission route or ability to persist within the host. In a non-disease state, cervial lymph nodes are normally microbe-free environments [12]. Because of these unique attributes, lymph nodes make an ideal clinical target to detect underlying coinfections.

F. tularensis is the causative agent of the zoonotic disease tularemia and it can infect humans by several routes, including the ingestion of contaminated water or food. Exposure to F. tularensis-contaminated water [13, 14], blood-feeding vectors [15], or, on rare occasion, aerosolized particulates [16] each lead to distinct clinical forms of tularemia: orophyryngeal, ulcergrandular, and respiratory tularemia, respectively. In the rapidly developing nation of Turkey, tularemia has been on the rise since 2009 and oropharyngeal disease is the most common form [17]. This form involves a sore throat and the swelling of cervical lymph nodes. Antibiotic treatment is highly effective at significantly shortening disease duration [17] and very little evidence exists to support the idea that antibiotic resistant F. tularensis strains are prevalent in Turkey [18]. Because swollen cervical nodes are observed from many different types of human infections occurring in Turkey [19], we wanted to use cervical fine-needle aspirates of lymph nodes from eight tularemia patients [20] as a test model to analyze the utility of shotgun metagenomics to assess for the presence of multiple infectious agents.

Metagenomics analysis of fine-needle aspirates of cervical lymph nodes from tularemia patients identified underlying coinfections (HBV and parvovirus B19). The true burden of coinfection may have been underestimated by not accounting for RNA viruses. Metagenomic analysis identified the presence of F. tularensis in all eight clinical samples when analyzed by both MetaGeniE and SUPRI. In addition, both analysis search methods detected other microbes in the same subset of patient samples (Parvovirus B19 in patients 3, 5, and 6, and HBV positive in patient 3, see Table 1). When combining total sequencing reads from all eight clinical samples, we obtained a total of 787,568,687 reads with 99.6% (784,495,044) matching human DNA, 0.31% unknown (2,465,280), and 0.039% (305,738) matching bacteria (Fig. 1). Among 305,738 reads from bacteria, 8848 reads matched F. tularensis, which comprised 2.89% of total bacterial reads (Fig. 1). This composition profile of extremely high levels of human DNA and low-level F. tularensis DNA in these clinical samples is consistent with our real-time PCR data (data not shown). Despite this extreme disproportionate ratio between human vs pathogen DNA species, 1000× sequence coverage provided enough sequences of F. tularensis, HBV, and parvovirus at high sequence match identity to solidly confirm the presence of these pathogens in specific clinical samples (Table 1). The other detected non-Francisella bacterial reads were classified as errors due to poor sequencing match identities with reference bacteria in published databases.

Our metagenomics analysis also detected non-pathogenic commensal skin bacteria, Propionibacterium acnes, which is likely real (Table 1) and is probably an incidential consequence of the fine-needle aspiration process itself, which involves the direct puncture of the skin [30]. We also detected Ralstonia picketti sequence in all eight patient samples (Table 1). Many clinical reagents, including ultra-pure water systems, have been reported to be contaminated with Ralstonia species [31, 32] and we hypothesize that this a likely source for our samples. Since R. picketti and P. acnes are known to have little clinical consequence [33, 34], no molecular confirmation was conducted on these organisms.

Our metagenomics analysis yielded high quality matches for 8848 sequencing reads of F. tularensis generated from all eight samples, which represents a very small fraction of the F. tularensis genome (see Table 1, Additional file 1: Methods Figure S3). Despite limited data for this analysis, more data could have been captured by this metagenomics strategy by increasing the coverage depth for which the sequencing was generated (i.e. > 1000× coverage). We found no evidence for F. tularensis, HBV, human parvovirus B19, and Ralstonia species among other clinical samples prepared and sequenced at the TGen sequencing facility, suggesting that the DNA sequence of these four microbes did not originate from the environment nor commercial reagents used in this facility at the time lymph node aspirates were processed.

Standardized traditional diagnostics independently confirmed the HBV coinfection in patient 3 that was initially detected through metagenomics. Active infection with HBV was confirmed in patient 3 via a serological diagnostic test, despite being missed by prior clinical examination. This confirmation was communicated using a method that retained the integrity of the patient de-identification system. No further information in respect to the stage of disease for this patient was obtained.

Molecular methods confirmed the presence of DNA sequence from multiple pathogens in three of the eight clinical samples, consistent with coinfection in these patients. We tested all eight samples that were PCR-positive for F. tularensis for the presence of parvovirus B19 and HBV. Through a combination of nested PCR followed by Sanger sequencing using parvovirus B19-specific primers [29], we confirmed detection from patients 3 and 6 but not 5 (Table 1). The 251 bp B19-specific amplicon from patients 3 had 100% sequence identity with published strains of human parvovirus B19 encoding a VP1 gene (EU478584), and the B19-specific amplicon from patient 6 had 99% sequence identity to published strains. This comparison identified a single base mutation that did not match any of the six deliberate mutations engineered in the synthetic positive control. Thus, this mutation either reflects the sequence of the original template or arose as an artifact introduced during PCR and sequencing process. An HBV-specific amplicon was generated from patient 3 and not from the other seven samples (Table 1). HBV-specific primers amplified a 37 bp fragment in patient 3 that perfectly matched published strains for C12 X protein (X) and core protein (C) genes (KP309751).

Not all pathogens initially detected by bioinformatics were confirmed through molecular methods. Parvovirus B19 was detected in patient 5 by metagenomics but not by our nested PCR Sanger sequencing molecular techniques. Although this suggests that the PCR assay used in our study is less sensitive than deep sequencing technology, it is thought that deep sequence Illumina output is nearly comparable to well optimized real-time PCR assay [35]. Thus, the sensitivity difference more likely stems from the technical differences between the two detection strategies. Unlike real-time PCR, which used 1 μL of DNA extract per reaction, metagenomics sequencing captured information from 100 μL of DNA extract. The results suggest that the larger volume of template enabled the capture of enough low-level parvovirus DNA in patient 5 for successful sequencing that was missed using the PCR strategy.

The importance of identifying underlying coinfection(s) is gaining greater appreciation [5, 6] but obtaining such information still remains challenging. We demonstrate an effective strategy to capture existing coinfections by using fine-needle aspirates obtained from cervical lymph nodes from tularemia patients. Other clinical sample types may be inferior at detecting coinfections as suggested by our finding that F. tularensis was PCR negative in blood samples of all eight patients (data not shown) but positive from the lymph node aspirates [20]. Using the metagenomics approach, we were able to detect diverse organisms (bacterial and viruses) that greatly differed in transmission routes and host persistence, indicating a lack of bias based on these differing biological parameters.

Surveying for pathogens from a clean microbial environment, such as lymph nodes, may be a good approach to diagnose clinically relevant microbes. However, not all diagnoses necessarily reflect active disease or an infection that has clinical relevance. There are pathogens, including parvovirus B19 DNA, that are never cleared but, rather, continue to persist in a dormant state in the host [36–38]. Studies have documented that parvovirus B19 DNA sequence is detectable, albeit at very low levels, from a wide range of human clinical samples (skin, synovium, tonsil, heart or liver [36] and bone marrow [39]) years after seroconversion. Although such surveys have not been conducted in lymph nodes, fine-needle lymph node aspirates contain cellular material, including immune cells continuously migrating between these nodes, the circulation system, and the bone marrow [40]. For this reason, we cannot conclude that the parvovirus B19 detected in three pediatric patient samples in this study was the result of acute infections. Initial patient medical examinations did not note signs of active skin rashes (data not shown). However, missed symptoms could be explained by the examination occurring during the early or late phase of this acute disease. In short, metagenomic diagnostics is highly informative for detecting unsuspected pathogens, but clinicians must continue to apply judgement to determine if detected pathogens have clinical relevance and/or warrant treatment.

Although fine-needle aspirates of lymph nodes are highly informative clinical samples, their availability varies. In Turkey, fine needle aspiration is considered routine as part of diagnosis and treatment for oropharyngeal tularemia and other diseases affecting lymph nodes [17, 41]. However, due to the clinical invasiveness of lymph node aspiration, in other countries this approach may be reserved for only those patient cases with lymphadenopathy of uncertain etiology.

Coinfections are not commonly considered when diagnosing and treating tularemia and, therefore, the clinical significance of coinfections is uncertain. Our results, however, indicate that coinfections are not rare in tularemia patients in Turkey. In fact, the rates of HBV and human parvovirus in our pediatric tularemia patients coincide well within the overall prevalence rates of these two diseases in the general Turkish population (10 and 21%, respectively) providing some indirect evidence that the detection could be unrelated to the acute F. tularensis infection [42, 43]. Very little is known regarding the effects of coinfection on clinical manifestation of tularemia and it is beyond the scope of this study to glean insight as to the clinical significance of tularemia patients with the identified coinfections.

Our study reveals that shotgun metagenomics targeting fine-needle lymph node aspirate samples is a promising clinical diagnostic strategy to identify underlying coinfection in primary disease as demonstrated by our ability to simultaneously detect F. tularensis and possible coinfections. Other clinical specimens such as blood may not be as informative for this purpose. In-depth exploration of new broad diagnostic methods that identify multiple microbes and possible coinfections is an important first step to advance the understanding of disease manifestations and treatment responses, and to possibly promote this capability into mainstream clinical practice.