Pathogen identification in culture-negative cerebrospinal fluid specimens of patients with purulent meningitis using next-generation sequencing technology

Abstract

Antimicrobial therapy necessitates fast and accurate detection of pathogens in purulent meningitis (PM) diagnosis. Although considered a gold standard, the traditional bacterial detection and identification methods based on routine bacterial culture are time-consuming and show low sensitivity and specificity. In order to reduce PM-related deaths and improve the treatment efficiency, we employed trace DNA extraction technology and combined it with Next-Generation Sequencing technology for bacteria identification. We extracted DNA from 31 cerebrospinal fluid (CSF) samples from people with PM. The samples were collected by lumbar puncture at Xi’an Children Hospital under a stringent protocol. Ion PGM^TM System Next-Generation Sequencing was used to screen for pathogens in the CSF samples based on their 16S ribosomal RNA (rRNA) genes. Of the 31 CSF samples, 29 were culture negative. Streptococcus was detected in 2 samples by culture-based methods. The species identified by Next-Generation Sequencing showed a diverse bacterial population that included Streptococcus (22.6%), Escherichia (15.9%), Peptostreptococcus (10.7%), Pseudomonas (10.5%), Rothia (8.8%), Acinetobacter (4.9%), Prevotella (4.1%), Bacillus (3.3%), Neisseria (2.5%), Catonella (2.4%), Acitinomyces (2.0%), and Citrobacter (2.0%). Clustering analysis revealed three major bacterial groups, namely Streptococcus, Escherichia and Acinetobacter, each group with a different proportion of bacterial species. The results indicated that PM may result from infections by multiple bacterial species as opposed to a single species, with infectivity determined by abundance of the dominant species. Furthermore, Next-Generation Sequencing allowed a sensitive and specific etiological basis for identification of pathogens not only in the cerebrospinal fluid specimens, but also in other clinical samples.

Introduction

Meningitis caused by bacteria is a clinical emergency that leads to high mortality and morbidity. Over the past several decades, the incidence of bacterial meningitis in children has been reported to be 0.2-1% in developed countries and higher in developing countries with the incidence of 0.8-6.1%. In some areas of China, the mortality rate associated with meningitis has been reported to be as high as 21.2%, with an incidence of 10%-30% [1,2].

To date, hundreds of pathogenic microorganisms have been proven to cause meningitis. However, in most cases (40-60%), the etiology remains unknown, which could be due to a wide range of factors such as suboptimal time of sample collection, inability to identify the causative agent due to large number of causative microorganisms, or the detected pathogen being unknown. At present, routine diagnostics, which typically includes serologic and molecular tests, enables the detection of only the common pathogens [3,4]. The traditional microbiological work-up depends on routine bacterial culture [5], Gram staining and phenotypic identification schemes [4], which are all extremely useful in providing treatment guidance. However, the fact remains that these traditional detection methods are limited by their low positive rate of clinical diagnosis, and apparent drawbacks. First, CSF cultures are time-consuming. Second, detection is limited to bacteria that can be grown on suitable media. Third, the results from the CSF bacterial culture do not reflect the results from phenotypic identification methodologies such as microscopy. Regardless of the type of organism identified by the CSF culture, only 25% show positive microscopy results with bacterial concentration <10³ CFU/ml and 60% show positive microscopy results with bacterial concentration in the range of 10³ to 10⁵ CFU/ml [6]. Fourth, in an extensive study over a period of 27 years, it appeared that culture might miss the diagnosis of bacterial meningitis in at least 13% of cases [5].

In order to reduce PM-associated deaths and improve treatment efficiency, sensitive detection and accurate identification of bacterial pathogens is critical [7]. This is where Next-Generation Sequencing (NGS) has its best utility. Next-Generation Sequencing with its high throughput capacity enables the detection and identification of known as well as novel pathogens [8], and therefore could be useful in the diagnosis of meningitis caused by a variety of microorganisms [9].

Bacterial species identification and taxonomic studies have been performed by 16S rRNA gene analysis [10,11]. The nine “hypervariable regions” within the bacterial 16S rRNA genes demonstrate considerable sequence diversity among different bacterial species and can be used for species identification. Amplification of the hypervariable regions for sequencing purposes is achieved by designing universal primers that target the conserved sequences stretches that flank the hypervariable regions in most bacteria [12]. The use of 16S rRNA gene universal PCR primer can lead to greater recovery of genomic DNA extracted from a bacterial population [13]. Therefore, instead of the conventional microbiological workup, NGS of the 16S rRNA gene can be applied for taxonomic identification of all bacteria presented in the sample.

In this study, we established an efficient method to extract genomic DNA from a small amount of CSF samples. We used the Ion Personal Genome Machine (Ion PGM) Sequencing System, which allowed sensitive and specific identification of pathogens causing rapid infection, and provided a basis for accurate etiologic diagnosis. The NGS tools such as Ion PGM with advantages such as a short analysis time, high specificity, and high resolution are increasingly being applied as alternative molecular technique in clinical microbiology [8,14]. More importantly, the sensitivity of NGS allows diagnosis of encephalitis even when routine tests fail in identifying the causative pathogens.

Materials and methods

Patients and CSF sampling

Thirty-one CSF specimens were collected from patients confirmed to have PM. The CSF samples were collected by a lumbar puncture in Xi’an Children Hospital under a stringent protocol and stored at -80°C immediately. All the samples were examined according to the contemporary standard that includes microscopic evaluation and subsequent culture. Clinical data from the 31 cerebrospinal fluid culture samples are shown in Table 1.

Table 1.

Clinical data of 31 culture cerebrospinal fluid samples

Sample ID	Gender	Age (Months)	Main symptom	Cerebral spinal fluid		Routine culture	Resident place
				WBC (10^6/L)	N (%)
LZI160001	Male	2	Fever, headache	145	83	Negative	Yulin, Shaanxi
LZI160002	Female	8	Fever, cough	187	77	Negative	Yanan, Shaanxi
LZI160003	Female	3	Fever, convulsion	174	69	Negative	Yulin, Shaanxi
LZI160004	Female	1	Fever	132	69	Negative	Yulin, Shaanxi
LZI160005	Male	8	Fever, headache	209	78	Negative	Yulin, Shaanxi
LZI160006	Male	5	Fever	147	83	Streptococcus	Tianshui, Gansu
LZI160007	Male	2	Fever	167	82	Negative	Yulin, Shaanxi
LZI160008	Male	1	Fever, nausea	216	83	Negative	Baoji, Shaanxi
LZI160009	Male	3	Fever, headache	279	88	Negative	Ankang, Shaanxi
LZI160010	Male	3	Fever, headache	159	66	Negative	Hanzhong, Shaanxi
LZI160011	Male	2	Nausea	178	81	Negative	Baoji, Shaanxi
LZI160012	Female	7	Fever, convulsion	167	78	Negative	Xi’an, Shaanxi
LZI160013	Male	4	Ferver, nausea	253	83	Negative	Baoji, Shaanxi
LZI160014	Female	11	Ferver, nausea	294	85	Negative	Shangluo, Shaanxi
LZI160015	Male	12	Ferver, nausea	261	80	Negative	Shangluo, Shaanxi
LZI160016	Male	7	Fever	157	79	Negative	Ankang, Shaanxi
LZI170001	Male	9	Fever, headache	154	81	Negative	Baoji, Shaanxi
LZI170002	Male	5	Fever, headache	182	74	Negative	Xi’an, Shaanxi
LZI170003	Male	6	Ferver, nausea	203	71	Negative	Yanan, Shaanxi
LZI170004	Male	19	Ferver, headache, nausea	257	73	Negative	Tianshui, Gansu
LZI170005	Female	12	Nausea	149	79	Negative	Xi’an, Shaanxi
LZI170006	Male	7	Fever, cough	139	84	Negative	Xi’an, Shaanxi
LZI170007	Male	3	Fever, cough	177	85	Negative	Xi’an, Shaanxi
LZI170008	Male	3	Fever	242	63	Negative	Xianyang, Shaanxi
LZI170009	Female	8	Ferver, nausea	283	69	Negative	Baoji, Shaanxi
LZI170010	Female	53	Ferver, nausea	257	75	Negative	Baoji, Shaanxi
LZI170011	Male	4	Ferver, nausea	169	79	Negative	Hanzhong, Shaanxi
LZI170012	Male	5	Ferver, headache, nausea	174	81	Negative	Weinan, Shaanxi
LZI170013	Female	2	Fever, convulsion	183	73	Negative	Hanzhong, Shaanxi
LZI170014	Male	6	Fever, headache	164	75	Negative	Weinan, Shaanxi
LZI170015	Male	1	Nausea	147	63	Negative	Xi’an, Shaanxi

Genomic DNA extraction

CSF samples were processed, and DNA recovered by following methodologies. First, 500 µl of each CSF sample was ultra-filtrated with Amicon ultra-0.5 Centrifugal Filter Devices, and then concentrated recovery of the samples was performed by using QIAamp mini kit (Cat. no. 51306) according to the standard protocol. For extraction of small amount of nucleic acids, Ethachinmate (Cat. no. 312-01791) was used, which is a specially prepared neutral polyacrylamide polymer solution and a useful reagent for recovery of an extremely small quantity of nucleic acids. The yield and purity of the extracted DNA was quantified by Qubit 2.0 Fluorometerand (Life Technologies, USA) with DNA HS Assay Kit (Thermo Scientific, Waltham, MA). The DNA samples were then stored at -20°C until further use.

16S rDNA library construction, sequencing, and bacterial identification

A library was prepared from the 31 cerebrospinal fluid samples using Ion 16s^TM Metagenomics kit (Cat. no. A26216). 12.5 ul of the extracted genomic DNA was used to prepare amplicons. Two primer sets within the kit were used to amplify the corresponding hypervariable regions of the 16S region of bacteria. The polymerase chain reaction (PCR) was performed under the following conditions: initial denaturation for 10 minutes at 95°C, followed by 25 cycles consisting of denaturation for 30 seconds at 95°C, annealing for 30 seconds at 58°C, and extension for 20 seconds at 72°C, followed by extension for 7 minutes at 72°C. Samples were stored at -20°C until ready for use. E. coli genomic DNA and H₂O were set as positive and negative control, respectively. The amplification products were allowed to come to room temperature for 30 minutes and purified with the Agencourt® AMPure® XP beads (Cat. no. A63881). The purified PCR products were analyzed with the Agilent® 2100 Bioanalyzer® instrument (Agilent Technologies, USA) and an Agilent® High Sensitivity DNA Kit (Cat. no. 5067-4626). The amount of target amplicons was determined by Agilent® software and the volume of two sets of purified amplification products for every sample was combined equally. Amplification end repair was conducted with the Ion Plus Fragment Library Kit (Cat. no. 4471252), and then purified with the Agencourt® AMPure® XP beads (Cat. no. A63881). Barcoded libraries were prepared by ligating the adapters from the Ion Xpress^TM Barcode Adapters 1-16 Kit (Cat. no. 4471250). The libraries were then purified with the Agencourt® AMPure® XP beads (Cat. no. A63881). The Ion Universal Library Quantitation Kit (Cat. no. A26217) was used to determine library concentration using qPCR.

The amplified fragments were sequenced on the Ion PGM^TM System (Life Technologies, USA) using the Ion PGM^TM Hi-Q (TM) Sequencing Kit (Cat. no. A25592).

High-throughput sequencing reads were reassigned to samples based on barcodes. After trimming, the quality-controlled clean reads were de-chymed and clustered using the Usearch software to obtain the OTUs according to a standard of 97% similarity. These OTUs are aggregated into OTU abundance tables. Subsequent analysis was performed based on OTU abundance table by using Qiime software (http://qiime.org/home_static/DataFiles.html)[15]. The three 16S rRNA databases RDP Classifier tool (http://rdp.cme.msu.edu/; East Lansing, MI, USA), EzBioCloud 16S Database (https://www.ezbiocloud.net/tools) [16] and Greengenes Database were used to align the representative sequences extracted from the obtained OTUs with the 16S rRNA [17].

Diversity estimation of bacteria identified in CSF samples

The species and diversity of bacteria identified in CSF samples were evaluated using bioinformatics analysis tools.

Taxonomic alpha-diversity was estimated as the number of observed OTUs and by the calculated Chao1, Simpson and Shannon indices. The taxonomic alpha-diversity estimates were then compared between gene regions (16S_v4 versus full 16S) using non-parametric versions of the t-test. To validate the beta-diversity conclusions, regardless of which gene region was used to compare the samples, we used Procrustes analysis [18]. We then compared ordinations generated by principal coordinates analysis (PCoA) of Bray-Curtis and weighted Unifrac distances between 16S_V4 and full 16S groups [18]. Significance was estimated by Monte Carlo simulation (10,000 iterations). After classification of the species, the sequences in each OTU were summarized into OTU abundance tables. Alpha diversity was analyzed based on OTU abundance table [19]. In this study, Qiime software was used to calculate the Alpha diversity indices. The Alpha Diversity Index mainly includes observed species indices such as the Chao1, Shannon and Simpson indices, and the PD_whole_tree index. We made a corresponding dilution curve for each index to describe the species diversity in a single sample. Furthermore, the beta diversity analysis was performed to assess the distribution and content of the bacteria and evaluate the total diversity in different samples based on the bacterial profile.

Results

Consequence of sequencing and bacteria identified by 16S rDNA analysis

After quality control analysis and OTU filtering, a total of 3,478,834 sequences ranging from 269,392 to 30,514 per sample (mean =112,220) sequences were obtained. Table 2 presents the number of original sequences, valid sequences, and OTUs in all the 31 samples. From these sequences, 14,593 OTUs were identified. About 98.5% OTUs could be classified to the phylum level and 95.2% could be classified to the genus level.

Table 2.

Summary of sequences and OTUs in all samples

Sample ID	Original sequence	Valid sequence	Ratio	OTUs
LZI160001	114283	48106	42.1%	460
LZI160002	82529	43429	52.6%	587
LZI160003	66264	35676	53.8%	584
LZI160004	104021	46660	44.9%	458
LZI160005	93635	39626	42.3%	447
LZI160006	377566	149981	39.7%	305
LZI160007	335835	173460	51.7%	505
LZI160008	600359	257243	42.8%	442
LZI160009	496821	195235	39.3%	370
LZI160010	79043	33262	42.1%	464
LZI160011	160104	50191	31.3%	393
LZI160012	150225	49323	32.8%	423
LZI160013	142437	48976	34.4%	493
LZI160014	65968	30514	46.3%	536
LZI160015	601349	238131	39.6%	399
LZI160016	93005	33417	35.9%	507
LZI170001	472353	193029	40.9%	548
LZI170002	240067	101323	42.2%	324
LZI170003	383500	185012	48.2%	425
LZI170004	407830	153176	37.6%	327
LZI170005	449354	195261	43.5%	406
LZI170006	387461	192618	49.7%	438
LZI170007	115911	51977	44.8%	502
LZI170008	199179	143335	71.9%	803
LZI170009	229469	167239	72.8%	657
LZI170010	485191	203191	41.9%	369
LZI170011	561106	269392	48.0%	551
LZI170012	149508	96438	64.5%	458
LZI170013	133263	87324	65.5%	421
LZI170014	148364	68492	46.2%	502
LZI170015	192715	93058	48.3%	489
ddH2O	956	416	43.5%	/
Total	7669361	3478834	45.9%	14593

Figure 1 shows an overview of pathogens identified in the 31 CSF samples. At the phylum level, pathogens were classified into 7 phyla with the proportion >1.0% in the total bacteria, including Firmicutes (47.5%), Proteobacteria (25.8%), Actinobacteria (14.6%), Bacteroidetes (6.8%), Fusobavteria (1.4%), Cyanobacteria (1.2%), Streptophyta (1.0%). At the genus level, pathogens were classified into 16 genus with the proportion greater than 1.0% in the total bacteria, including Streptococcus (22.6%), Escherichia (15.9%), Peptostreptococcus (10.7%), Pseudomonas (10.5%), Rothia (8.8%), Acinetobacter (4.9%), Prevotella (4.1%), Bacillus (3.3%), Neisseria (2.5%), Catonella (2.4%), Acitinomyces (2.0%), Citrobacter (2.0%), Staphylococcus (1.1%), Haemophilus (1.0%), Anaerococcus (1.0%), Listeria (1.0%). Then the frequency of these dominant bacteria in each sample were analyzed. The results are presented in Table 4 and Figure 2. Streptococcus, Escherichia, Pseudomonas, Rothia, Acinetobacter, Prevotella and Citrobacter had the highest detection rate and were detected in all the 31 samples. No correlation was found between the number of reads and the frequency of bacteria detected in the samples. Moreover, the dominant bacteria detected may not be pathogenic bacteria.

Figure 1.

Distribution of bacteria in each sample. Notes: A shows the composition of bacterial community of each sample at the phylum level. B shows the composition of bacterial community composition of each sample at the genus level. The vertical axis represents relative bacterial abundance of corresponding genus. The horizontal axis represents 31 culture-negative CSF samples. Each bar in the histogram represents an individual sample. Different colors identify different bacteria. The size of the different colored bars characterizes the number of reads.

Table 4.

Distribution of predominant microflora in 31 CSF samples at taxonomic level

Taxonomy	Proportion in total (%)	Detection samples	Detection ratio (%)
Streptococcus	22.6	31	100
Escherichia	15.9	31	100
Peptostreptococcus	10.7	29	93.6
Pseudomonas	10.5	31	100
Rothia	8.8	31	100
Acinetobacter	4.9	31	100
Prevotella	4.1	31	100
Bacillus	3.3	23	74.2
Neisseria	2.5	27	87.1
Catonella	2.4	28	90.3
Actinomyces	2.0	26	83.9
Citrobacter	2.0	31	100
Sphingomonas	1.8	30	96.8
Staphylococcus	1.1	15	48.4
Haemophilus	1.0	29	93.5
Anaerococcus	1.0	11	35.5

Figure 2.

Distribution of predominant microflora in 31 CSF samples. Note: A and B show predominant bacteria at the phylum and genus level, respectively. B shows the composition of the bacterial community within each sample at the genus level. Different colors identify different bacteria. The sector area in the pie chart of different colors characterizes the number of reads.

Alpha and Beta diversity estimation of bacteria identified in CSF samples

Alpha diversity was calculated using bioinformatics analysis based on OTU abundance table. Sequences and OTUs in all samples are summarized in Table 2. To evaluate the species richness and the diversity of bacteria in all samples, the alpha diversity indices including Chao1, Shannon and Simpson were calculated and are presented in Table 3. As shown in Figure 3, the sequencing depth of 31 culture-negative CSF samples was represented by rarefaction curves, which provided estimates of whether the sequencing depth was sufficient to cover all species. Shannon-Wiener curves were drawn by using the microbial diversity index (the Shannon index) at different sequencing depths for each sample to reflect the microbial diversity of each sample at different sequencing numbers. Simpson index was used to estimate the microbial diversity within a sample. The larger the Simpson index, the higher the community diversity. The Simpson curves become more flat with increasing number of sequencing reads, reflecting the sequencing depth needed to cover the bacterial species in each sample. As a result, the curve displays whether the amount of sequencing data is sufficient to reflect the information of most microbial species in each sample.

Table 3.

Alpha diversity degree of 31 CSF samples in the 0.03 distances

Sample_ID	Observed_species	Chao1	ACE	Shannon	Simpson	Goods_coverage
LZI160001	3600	5124.389	5268.316	8.218	0.974	0.985
LZI160002	2587	4840.030	5623.938	5.112	0.902	0.973
LZI160003	2684	4206.299	4346.463	6.079	0.905	0.976
LZI160004	3758	5249.524	5624.502	6.426	0.937	0.980
LZI160005	3847	5283.102	5363.384	8.474	0.987	0.980
LZI160006	5305	8795.696	9371.004	3.633	0.707	0.990
LZI160007	8105	11502.357	12271.304	7.103	0.948	0.982
LZI160008	14142	23784.937	26425.319	7.466	0.966	0.982
LZI160009	11270	20956.982	23540.904	7.209	0.962	0.980
LZI160010	3364	9184.377	10499.347	6.966	0.959	0.947
LZI160011	7793	14144.417	16007.886	7.592	0.958	0.947
LZI160012	7423	12155.603	13721.923	7.904	0.969	0.953
LZI160013	7893	14584.668	16327.395	8.212	0.974	0.942
LZI160014	3136	8708.450	9781.548	7.465	0.969	0.934
LZI160015	10099	19192.493	21374.556	6.706	0.947	0.986
LZI160016	3507	8574.558	9287.817	7.207	0.963	0.954
LZI170001	1348	3010.004	3335.181	6.132	0.918	0.928
LZI170002	5124	9974.559	10593.606	6.550	0.949	0.983
LZI170003	4425	4850.566	5140.253	5.599	0.932	0.995
LZI170004	6327	9522.389	10115.567	4.398	0.762	0.990
LZI170005	11206	15255.847	16603.354	7.635	0.967	0.983
LZI170006	9038	15747.840	17263.887	7.347	0.966	0.980
LZI170007	3502	8830.681	9411.126	6.281	0.945	0.971
LZI170008	6803	13383.562	15475.655	7.279	0.965	0.969
LZI170009	6357	13508.296	14891.596	6.523	0.946	0.976
LZI170010	11069	17290.655	19044.638	7.048	0.952	0.983
LZI170011	9651	16973.091	18771.508	7.325	0.971	0.987
LZI170012	4058	8691.450	9134.672	6.915	0.960	0.979
LZI170013	4121	9103.084	9330.376	7.141	0.970	0.974
LZI170014	5020	11868.621	13029.676	6.888	0.963	0.966
LZI170015	4893	9205.290	10021.252	6.646	0.957	0.981

Chao and ACE were used to estimate species richness; Shannon index and Simpson index are diversity indexes. If Shannon index is larger and Ssimpson index closer to 1.0 it means a more abundant species in the samples.

Figure 3.

Rarefaction curves of each sample. Notes: Rarefaction curves of index including observed_species, PD whole tree, and Shannon and Simpson are plotted in (A-D). The horizontal axis represents random number of sequences. The vertical axis represents each index. Curves of different colors indicate different samples. (A) The saturation point of each rarefaction curve characterizes the actual number of OTUs for each sample. (B) The shape of the rarefaction curve characterizes the depth of sequencing. (C) The saturation point of each rarefaction curve characterizes the value of Shannon index for each sample. The Shannon index is used to estimate the species diversity. (D) The saturation point of each rarefaction curve characterizes the value of Simpson index for each sample. The Simpson index characterizes the degree of species uniformity and the number of species in the CSF sample.

Furthermore, beta diversity analysis was performed to evaluate the total diversity and assess the distribution and content of bacteria in the 31 CSF samples. As shown in Figure 4A the PCoA Weighted-UniFrac analysis found several regions of variability, and categorized the samples them into three groups (A, B, C). Group A containing 22 cases of CSF samples were derived from south of the Shaanxi and Guanzhong area. 7 cases belonging to Group B were derived from north of Shaanxi. 2 cases belonging to Group C were derived from Tianshui of Gansu province. Figure 4A showed that as a result of regional variability, the species distribution was similar between the samples within the same group, while there were substantial differences between the three groups. Figure 4B and 4C showed the analysis of species differences among samples of these three groups. The results showed that the bacterial species and composition of Group A and B were similar because of the close proximity of the regions. Streptococcus, Escherichia and Acinetobacter were the common species in the 3 groups, while the proportion of each species was different. The variable regions led to the differences of the overall prevailing species in species and abundance.

Figure 4.

Beta diversity analysis of Species Differences among Samples of Different Groups. Note: beta diversity analysis was performed to evaluate the total diversity and assess the distribution and content of bacteria in the 31 CSF samples. A shows the result of PCoA Weighted-UniFrac analysis; based on distances between all the samples, they could be divided into three groups (A, B, C). B and C show the analysis of species differences among samples of these three groups.

Discussion

Although set as a gold standard, conventional bacterial culture and identification techniques are limited by the type of culture medium and the sample sizes. Most of the bacteria are undetectable and result in false negative culture. Since 2005, advances in Next-Generation Sequencing (NGS) technologies have revolutionized biological science. NGS is an unbiased sequencing platform and therefore has the potential to enhance our ability to discover trace amounts of pathogens as well as emerging pathogens and also enable us to get a more comprehensive analysis of bacterial diversity from complex environmental samples.

In this study, we established an efficient method to extract genomic DNA from a small amount of CSF samples. Trace detection and contamination were crucial to our study. The first task was to be able to detect trace amounts of DNA to a great extent. This method was designed to be able to detect and identify pathogens in the CSF, available in trace amounts. Genomic DNA from CSF samples were first collected using a sensitive DNA extraction kit (QIAamp mini kit). Upon optimization of the protocols, including ultrafiltration, precipitation by Ethachinmate (for DNA sedimentation) and cracking method, the overall low biomass of bacteria was successfully extracted from the CSF samples.

Contamination is a common and critical problem in bacteria detection. Contamination could be due to environmental contamination or a cross-contamination between samples during the DNA extraction and NGS library preparation steps. Several steps were taken to avoid contaminants. To exclude the risk of environmental contamination, the CSF samples were collected under a strict protocol (under clean and sterile conditions). To prevent contamination during genomic DNA extraction and library preparation, we followed the protocols in a strictly controlled, decontaminated, and sterile environment. To prevent reagent contamination, appropriate negative controls were used and repeat analysis of independent experiments were done. No 16S sequence amplification were found in the negative control sample, validating that our samples were free of contaminants and confounding factors.

Differences in the experimental protocols used may produce variations that outweighs biologic differences. The composition of bacteria within samples is typically evaluated by targeting the bacterial 16S rRNA gene as a phylogenetic marker. We developed a combination of the two primer pools for sequence-based identification of a broad range of bacteria within a mixed population. The results showed that 16S rRNA genes from the CSF of patients with purulent meningitis could be amplified by two sets of primer pools V2-4-8 and V3-6, 7-9 regions. Both sets of primer pools essentially covered the entire hypervariable region of the bacterial 16S rDNA gene, thereby not only increasing the resolution of the assay but also increasing the likelihood of detecting more bacterial species. Among the 31 samples, the only culture positive sample was LZI160006, which was detected and identified to be Streptococcus. This was consistent with the sequence analysis result. However, the difference is that besides Streptococcus, there were other bacteria species annotated and identified by 16S sequencing. As mentioned above, most samples contained multiple bacteria. Increasing evidence shows that multiple pathogens are involved in the pathogenesis of infectious diseases, including PM. It has been reported that meningitis is caused by multiple microorganisms [20,21]. Based on our study and that of others, purulent meningitis is not just caused by a single pathogen, but instead by a variety of pathogens. We detected a variety of pathogens by NGS in culture-negative CSF, and that might be associated with the pathogenesis of PM. The study results indicate that purulent meningitis is not caused by one species of bacteria. Based on our results of the species diversity, the pathogens causing purulent meningitis are basically mixed pathogens.

Not all the bacteria identified were pathogenic, such as Streptococcus. But the bacteria were distributed in most of the samples. It is possible that normal human CSF, which has been long believed to be completely sterile, may harbor microorganisms, similar to the flora lining the oral cavity, the intestines, and the placenta reported in 2014 by Aagaard et al. [22]. The disorder of normal flora can also cause disease. This is part of the experimental design that we would like to incorporate in the upcoming experiments.

In conclusion, high-throughput sequencing methods have demonstrated the ability to identify bacterial species from culture-negative CSF in patients with PM. High-throughput sequencing simultaneously greatly remedies the shortcomings of current clinical detection methods and provides a scientific basis for the diagnosis of clinically relevant bacterial infectious diseases quickly and accurately.

Disclosure of conflict of interest

The authors declare no conflicts of interest. Author Zhang J.F., Zhao W. and Yang Y. was employed by the company Lifegen Co. Ltd. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Abbreviations

PM

Purulent Meningitis

CSF

Cerebrospinal Fluid

CNS

Central Nervous System

PCR

Polymerase Chain Reaction

QPCR

Real-time Quantitative PCR Detecting System

rDNA

Ribosomal DNA

NGS

Next-Generation Sequencing