Rapid detection and identification of infectious pathogens based on high-throughput sequencing.

BACKGROUND: The dilemma of pathogens identification in patients with unidentified clinical symptoms such as fever of unknown origin exists, which not only poses a challenge to both the diagnostic and therapeutic process by itself, but also to expert physicians.
METHODS: In this report, we have attempted to increase the awareness of unidentified pathogens by developing a method to investigate hitherto unidentified infectious pathogens based on unbiased high-throughput sequencing.
RESULTS: Our observations show that this method supplements current diagnostic technology that predominantly relies on information derived five cases from the intensive care unit. This methodological approach detects viruses and corrects the incidence of false positive detection rates of pathogens in a much shorter period. Through our method is followed by polymerase chain reaction validation, we could identify infection with Epstein-Barr virus, and in another case, we could identify infection with Streptococcus viridians based on the culture, which was false positive.
CONCLUSIONS: This technology is a promising approach to revolutionize rapid diagnosis of infectious pathogens and to guide therapy that might result in the improvement of personalized medicine.

INTRODUCTION

Rapid detection and identification of infectious pathogens in clinical samples is very challenging, but it is essential to guide the therapy and predict the outcome. Traditional clinical microbial diagnostic methods like physiology and biochemical identification, serological tests, and automated detection technology, are mainly based on the conventional culturing of clinical samples.[12] The culture turnaround time is usually 1–2 days for most samples, or longer for samples that require prolonged incubation, like blood cultures.[3] More recently, once an isolate has been detected, species identification and genotyping could be performed following identification of a susceptibility test.[3] In fact, the whole procedure could take several days or even weeks, for example, in terms of the culture time required for tuberculosis, which is 42 days in the clinical laboratory. Therefore, it is not only a time-consuming process, but also is a barrier that impedes the rapid detection and identification of infectious pathogens. Interestingly, these culture-based methods usually fail to identify those pathogens that cannot be cultivated under standard techniques, such as viruses that can be identified through antibody-specific ELISA or sequence-specific molecular methods such as polymerase chain reaction (PCR) assay and array hybridization technologies.

With traditional culture-based diagnostic techniques being partially superseded by several methods, like matrix-assisted laser desorption ionization time-of-flight mass spectrometry, PCR, real-time PCR and molecular hybridization-based technologies, the speed and accuracy of both the diagnosis and prevention of infectious diseases have been greatly improved.[45678] For instance, an internal transcribed spacer-targeted oligonucleotide chip was successfully used to detect and identify important bacterial pathogens associated with sepsis directly from the blood sample.[9] Another study also reported the use of a multiplex real-time PCR assay to identify members of the Mycobacterium tuberculosis complex.[10] Although the accuracy of pathogen identification has been improved, some newly emerging and unknown pathogens still cannot be identified using these current technologies.

In recent years, unbiased high-throughput sequencing (HTS) has been applied in medical microbiology as an emerging and powerful technique due to its low cost and rapid turnaround time. Some notable observations are the detection of a rapid hospital spread of antibiotic-resistant Klebsiella pneumonia, distinguishing Legionella pneumophila outbreak isolates from nonoutbreak isolates and identifying methicillin-resistant Staphylococcus aureus outbreak isolates.[11121314] Recently, the use of metagenomic analysis methods, which applies HTS has been proposed for infectious disease detection.[1516] Although HTS has an important impact on the detection of pathogens, some challenges remain due to a lack of standard operating procedures.[17]

Here, we have developed an efficient, accurate and comprehensive method based on HTS for the rapid detection and identification of infectious pathogens directly from clinical specimens, which can be used in the context of more challenging and troublesome diseases such as a fever of unknown origin (FUO).

METHODS

Clinical sample collection

The blood samples were collected from five patients, when they were transferred immediately to the intensive care unit (Peking Union Medical College Hospital) with the same symptoms as FUO. And then blood samples were processed as the following steps: Centrifuged (4°C, 10 min, 1600 ×g) within 8 h, transfer plasma to another tube and delivered in dry ice type environments. Two blood samples, which were referred to as UPDID2017 and UPDID2011-1, were blood culture positive for Pseudomonas aeruginosa and Streptococcus viridians, respectively. The remaining three samples, which were referred to as UPDID2020, UPDID2026 and UPDID0559-1 were blood culture negative. All samples were collected in sterile tubes and were centrifuged at 1600 ×g, and then the plasma was transferred to another tube for performing the follow-up analyses. Ethical approval for the study was obtained from the Institutional Review Board of BGI-Shenzhen. All patients had signed informed consent, and samples could then be used for research only.

Nucleic acid extraction, library preparation and sequencing

Nucleic acid (including DNA and RNA) was extracted directly from the clinical samples with QIAamp Viral RNA Mini Kit (QIAGEN). After extraction, the reverse transcription reaction was performed with PrimeScript RT-PCR Kit to generate single strand cDNA. Next, the double strand cDNA was obtained according to the Second Strand cDNA Synthesis Kit manufacture's instruction. Finally, the reaction product was purified using the QIAquick PCR Purification Kit (QIAGEN). The double stranded cDNA was disrupted into 200–300 bp fragments. The cDNA library was constructed after end repairing, A-Tailing, Adapter ligation and PCR. The library was sequenced using the Proton platform after quality control.

Data processed and analysis

The high-quality sequence data were generated through removing low quality reads, adapter contamination, duplication reads and discarding reads shorter than 35 bp. To eliminate the effect of the human sequence, the data were removed that was mapped to a human reference sequence hg19 using Burrows–Wheeler Alignment.[18] Subsequently, we mapped the remaining sequence to the bacterial, virus, and fungi database.

For a nonredundant bacterial database, we downloaded the complete genome set of version 20140829 from NCBI (). For the nonredundant viral database, we chose 100 different viruses that are associated with human health based on version R194 from GenBank (). For the nonredundant fungal database, we downloaded the fungal genomic sequences associated with human disease from NCBI. At present, our databases contain 580 genus of bacteria, and 110 species of virus that are related to human diseases, and 14 species of fungi that can cause infectious in humans.

The time cost of our projected pipeline is almost 3 days (2 days and 10 h) based on the current technology of library construction and sequencing [Figure 1a]. The complete process of data analysis is shown in detail [Figure 1b]. This could still be extended according to one's requirements.

Figure 1

The pipeline of pathogen detection that contains detailed information. (a) The framework of rapid pathogen detection, which includes DNA/RNA extraction, sample preparation, DNA library construction, sequencing, and data analysis. The analysis took 2 days and 10 h. H stands for hour and D stands for day; (b) The complete data analysis process, followed by QC, trim host sequences, and database alignment.

Statistical analysis

We used the software named SoapCoverage from the SOAP website () to calculate the depth and coverage of each position. For bacterium, the result with coverage more than 1% and the number of stringent reads more than 10 was adopted. For fungi, the result with coverage of more than 0.1% and the number of stringent reads of more than 50 was reserved. Because of the high virus mutation rate, we chose the result with coverage of more than 10% and sequence length more than 1000 bp. The relative abundance of each microbe was made according to the following definition that modified the RPKM definition, because the pathogen content is ultimately low:[19]

RPMM = (1012 × C)/(N × L)

The RPMM represents the relative abundance of 1 million base-pairs in length, C is the number of reads that aligned to this microbe reference, N is the total number of reads, and L is the length of this microbe.

Polymerase chain reaction and Sanger sequencing validation

We carried out 16s rRNA identification for P. aeruginosa and S. viridians, and sequence-specific PCR identification for Epstein–Barr virus (EBV) with a target fragment of 250 bp to validate the result, respectively. The primers for 16S are (forward and reverse):

5’-CAACGCGAAGAACCTTACC-3’, and 5’-GACGGGCRGTGWGTRCA-3’

The EBV specific primers we used were:

5’-TCTGGCAGCTTTTGGCCTT-3’, and 5’-GTTGGAGTTAGAGTCAGAT-3’

Subsequently, we sequenced these target products by Sanger sequencing.

RESULTS

All these samples were cultured in the laboratory and performed sequencing with proton platform. Culture results and data statistics are listed in Tables 1 and 2, respectively. Species of P. aeruginosa and S. viridians in sample UPDID2017 and UPDID2011-1 were found to be positively grown. But in the other three samples, no microbes had been detected. For each sample, we adopted pooling sequencing with an 8 bp index for all of the samples, which improved the utilization of proton. The raw-data output was 5.1M, 46.2M, 4.9M, 6.1M and 55.7M reads for UPDID2017, UPDID2011-1, UPDID2020, UPDID2026 and UPDID0559-1 respectively, which fulfilled the analytical standard quantity.

Table 1

The result of blood culture in clinical laboratory

Sample	Pseudomonas aeruginosa	Streptococcus viridians	Other
UPDID2017	+	−	−
UPDID2011-1	−	+	−
UPDID2020	−	−	−
UPDID2026	−	−	−
UPDID0559-1	−	−	−

“+” represents positive, and “−” represents negative outcomes.

Table 2

Data statistics

Sample	Raw data	Clean data	Nonhuman data	Bacterial	Virus	Fungi
UPDID2017	5.1M	4.9M	277.5K	4334	17	538
UPDID2011-1	46.2M	40.7M	2.9M	115	2	1217
UPDID2020	4.9M	4.7M	257.6K	4467	55	545
UPDID2026	6.1M	5.8M	295K	2873	51	478
UPDID0559-1	55.7M	52.9M	2.2M	320	5.3K	1138

After removing the human sequence, the remaining data were mapped to the bacterial database with the mapping rate from 0.08% to 15.22%. However, for viruses whose genome is short, the proportions were all lower than the bacterial data. Only UPDID2026 and UPDID0559-1 samples had a higher proportion of 2.27% and 2%.

Using our method based on proton sequencing, we detected the presence of P. aeruginosa in sample UPDID2017 with the coverage of 6.7%. This was consistent with the blood culture result. P. aeruginosa is a common opportunistic pathogen, and is capable of causing serious infections that can cause wound infection, otitis media, meningitis, respiratory tract infection, urinary tract infection and septicemia, with varying degrees of fever.[20] In addition, we also found traces of other bacteria with low coverage and depth, such as Azotobacter vinelandii (a nitrogen fixing bacteria), Escherichia coli and Shigella sonnei. E. coli is an opportunistic pathogen that mainly colonizes the human gut and sometimes causes intestinal diseases and genitourinary disorders. By contrast, Shigella is a pathogen that can cause bacillary dysentery. Given their low coverage and depth, we thought they may be the homologous sequences that cannot cause such infectious symptoms. According to our parameters and clinical symptoms, we confirmed that P. aeruginosa was the cause of fever in sample UPDID2017.

In sample UPDID2011-1 that was positive for S. viridians after blood culture, we did not find any trace of S. viridians or other bacteria or viral species, which suggested the possibility of a false positive in the blood culture. In fact, S. viridians is an important bacteria among normal human flora that mainly colonizes the oral microbiome, followed by the respiratory tract, gastrointestinal tract, and female genital tract. So, we suspected contaminants during blood culture.

Interestingly, we found EBV, namely human herpesvirus 4, with the coverage of 86% in the sample UPDID0559-1, which was negative when we performed blood culture. EBV is the cause of nasopharyngeal carcinoma, which is also associated with chronic fatigue syndrome, Infectious mononucleosis, and lymphoproliferative diseases.[21] In addition to EBV, we did not detect any other pathogens. Due to the uncultured nature of EBV using the conventional blood culture approach, we concluded that EBV was closely related to the fever in this patient. For the two remaining blood culture negative samples, in addition to E. coli and Shigella that were detected in other samples as fragmented and homologue sequences, we did not find any closely or relevant pathogens.

To profile for positive results whatever based on culture and sequencing, detailed information on data statistics is listed in Table 3, and the coverage and depth of P. aeruginosa, S. viridians and EBV along the genomic sequence in sample UPDID2017, UPDID2011-1 and UPDID0559-1 were figured out. We found that the coverage distribution of P. aeruginosa and EBV genomes were dispersed and uniform, although S. viridians contained reads in only some special regions [Figure 2].

Table 3

The annotation result for each sample

Sample no.	Species	Coverage	Abundance (RPMM)	Unique reads	GI accession
UPDID2017	Pseudomonas aeruginosa	430352/6764661	117.42	3332	NC_017549.1
UPDID0559	Human herpesvirus 4	144909/171823	583.61	4884	NC_007605

RPMM: Represents the relative abundance of 1 million base pairs in length, which was similar with the definition of RPKM value.

Figure 2

(a) The coverage profile of Streptococcus viridians we detected. The horizontal axis represents the position of each species while the vertical axis is the corresponding bin depth; (b) The coverage profile of Pseudomonas aeruginosa we detected. The horizontal axis represents the position of each species while the vertical axis is the corresponding bin depth; (c) The coverage profile of Epstein–Barr virus we detected. The horizontal axis represents the position of each species while the vertical axis is the corresponding bin depth.

Sanger sequencing was carried out on the original sample with 16s rRNA identification for UPDID2017 and UPDID2011-1, as well as sequence-specific PCR identification for UPDID0559-1. The results showed that P. aeruginosa and EBV were present in the original samples UPDID2017 and UPDID0559-1, and in RTI2011-1 there was no trace of S. viridians, respectively [Figure 3].

Figure 3

Polymerase chain reaction (PCR) result and Sanger sequencing result of pathogen validation. (a) The result of 16s PCR validation (UPDID2011-1, UPDID2020, UPDID2026 and UPDID2017). The marker we used was D2000; (b) The Epstein–Barr virus sequence-specific PCR identification of UPDID0559-1. The marker we used is Trans 2K Plus DNA Maker; (c and d) Is the Sanger sequence of target fragments. “M” and “N” represent marker and negative control, respectively.

DISCUSSION

The current diagnostic paradigm consisted of four main stages, starting with the detection of a pathogen in the sample.[3] If a clinically relevant pathogen was detected, then this might be further tested for identification of drug susceptibility and epidemiological typing.[3] However, for most viruses, drug susceptibility, and epidemiological typing are not performed. The traditional method of detection and identification always takes several days or even weeks, which might delay the most optimal treatment times. Besides this, due to the limited resolution of the method itself, the accuracy still needs to be improved.

Therefore, in order to detect the pathogen directly in the clinical sample, we developed a rapid, efficient and accurate method containing experimental and bioinformatics analysis processes based on proton sequencing. The whole process including sample preparation, library construction, sequencing, data analysis and issuing reports, could take almost 3 days (2 days and 10 h). Compared with the traditional clinical microbial diagnostic methods based on conventional culturing, it shortens the whole period greatly. In addition, its ability to detect pathogens that are not cultivable or unknown has improved dramatically.

Despite the advantage of our method shows, there are still several problems to be improved upon. In regard to the experiment process, more technologies are essential to be developed such as with low DNA input based on proton sequencing. For most of the clinical samples, like blood samples, the nucleic acid content that is used for library construction after removing the majority of human cells is very low, and even at the nanogram (ng) scale, which will cause false negative result. This process is not only applied to blood samples, but it is also necessary to develop various methods of sample preparation for different types of samples. To optimize this process further, sample preparation must be simplified, which would reduce the turnaround time to a few hours and the need for highly skilled and trained technical staff.[3]

In the context of bioinformatics analysis, considering the present opportunistic pathogens belong to the human normal microbial flora, the threshold to determine whether a pathogen is clinically relevant should be more stringent. Thus, the research aiming at the distribution, coverage and abundance of the normal microbial flora should be carried out. Besides this, the database we used will be updated continuously. It would not only encompass bacterial and viral genomes, but will also incorporate fungal and protist genomes. It will also represent a catalogue of point mutations or genes that account for drug resistance, which would also be added.[3] In the present study, we only employed our method to perform detection and identification of clinically relevant pathogens. However, both drug susceptibility and genotyping will be carried out in the future.

For those sequences that are both unmapped to human sequences and microbial databases, we should determine whether they are contaminated or whether this is due to homologous sequence of other eukaryotic genomes, or novel sequences of the human genome, or even perhaps novel microbial genomic sequences that we had not discovered thus far. However, unknown pathogen identification should be considered such as mapping to the NR database according to protein sequence homology. The sensitivity of HTS will cause false positive result.

Finally, the advancement of HTS sequencing technology will reduce the time and bring down the cost of HTS. With the increasing demand for rapid detection and identification of infectious pathogens in clinical samples, the method we developed would play an important role in the clinical laboratory and potentially revolutionize the capability of identifying infectious and novel pathogens.