Infectious Diseases

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/infd20

Diagnosis of pleural empyema/parapneumonic

effusion by next-generation sequencing

Yoshiki Shiraishi, Kirill Kryukov, Katsuyoshi Tomomatsu, Fumio Sakamaki,

Shigeaki Inoue, So Nakagawa, Tadashi Imanishi & Koichiro Asano

To cite this article: Yoshiki Shiraishi, Kirill Kryukov, Katsuyoshi Tomomatsu, Fumio Sakamaki,
Shigeaki Inoue, So Nakagawa, Tadashi Imanishi & Koichiro Asano (2021) Diagnosis of pleural
empyema/parapneumonic effusion by next-generation sequencing, Infectious Diseases, 53:6,
450-459, DOI: 10.1080/23744235.2021.1892178

To link to this article: https://doi.org/10.1080/23744235.2021.1892178

View supplementary material

Published online: 09 Mar 2021.

Submit your article to this journal

Article views: 72

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=infd20
INFECTIOUS DISEASES, https://doi.org/10.1080/23744235.2021.1892178
2021; VOL. 53,
NO. 6, 450–459

ORIGINAL ARTICLE

Diagnosis of pleural empyema/parapneumonic effusion by

next-generation sequencing

Yoshiki Shiraishia , Kirill Kryukovb,c , Katsuyoshi Tomomatsua, Fumio Sakamakid, Shigeaki Inouee,
So Nakagawab , Tadashi Imanishib and Koichiro Asanoa
a
Division of Pulmonary Medicine, Department of Medicine, Tokai University School of Medicine, Isehara, Japan; bDepartment of
Molecular Life Science, Tokai University School of Medicine, Isehara, Japan; cDepartment of Genomics and Evolutionary Biology,
National Institute of Genetics, Mishima, Japan; dDivision of Respiratory Disease, Department of Medicine, Tokai University
Hachioji Hospital, Tokyo, Japan; eDepartment of Emergency and Critical Care Medicine, Tokai University School of Medicine,
Isehara, Japan

ABSTRACT
Background: Although a microbiological diagnosis of pleural infection is clinically important, it is often complicated by
prior antibiotic treatment and/or difficulties with culturing some bacterial species. Therefore, we aimed to identify probable
causative bacteria in pleural empyema/parapneumonic effusions by combining 16S ribosomal RNA (rRNA) gene amplifica-
tion and next-generation sequencing (NGS).

Methods: Pleural fluids were collected from 19 patients with infectious effusions and nine patients with non-infectious
malignant effusions. We analysed DNA extracted from the pleural fluid supernatant by NGS using the Genome Search
Toolkit and GenomeSync database, either directly or after PCR amplification of the 16S rRNA gene. Infectious and non-
infectious effusions were distinguished by semi-quantitative PCR of the 16S rRNA gene.

Results: Only 8 (42%) effusions were culture-positive, however, NGS of the 16S rRNA gene amplicon identified 14 anae-
robes and 7 aerobes/facultative anaerobes in all patients, including Streptococcus sp. (n ¼ 6), Fusobacterium sp. (n ¼ 5),
Porphyromonas sp. (n ¼ 5), and Prevotella sp. (n ¼ 4), accounting for >10% of the total genomes. The culture and NGS
results were discordant for 3 out of 8 patients, all of whom had previously been treated with antibiotics. Total (2DCT value
in semi-quantitative PCR of the 16S rRNA gene) and specific (total bacterial load multiplied by the proportion of primary
bacteria in NGS) bacterial loads could efficiently distinguish empyema/parapneumonic effusion from non-infectious effu-
sion.

Conclusion: Combining NGS with semi-quantitative PCR can facilitate the diagnosis of pleural empyema/parapneumonic
effusion and its causal bacteria.
Abbreviations: A ACP: American College of Chest Physicians; GSTK: Genome Search Toolkit; NGS: next-generation sequencing; ROC:
receiver operating characteristic; rRNA: 16S ribosomal RNA; IQR: Interquartile range

KEYWORDS ARTICLE HISTORY CONTACT

Pleural empyema Received 27 July 2020 Koichiro Asano
parapneumonic effusion Revised 31 January 2021 [email protected]
next-generation sequencing Accepted 10 February 2021 Division of Pulmonary Medicine, Department
malignant pleural effusion of Medicine, Tokai University School of Medicine,
143 Shimokasuya, Isehara, 259-1193,
Kanagawa, Japan

Supplemental data for this article can be accessed here.

ß 2021 Society for Scandinavian Journal of Infectious Diseases
 INFECTIOUS DISEASES 451

Background PCR with NGS to identify pathogenic bacteria in pleural

empyema/parapneumonic effusions.
A parapneumonic effusion is the accumulation of pleural
fluid caused by pneumonia. When microorganisms infect
the pleural space, a massive influx of neutrophils causes Methods
the accumulation of purulent effusion, resulting in Patients
empyema. Among patients with community-acquired
pneumonia, 15%–44% develop parapneumonic effusion Patients with infectious effusions were prospectively
and 6%–10% develop pleural empyema [1,2]. The aspir- recruited from two tertiary referral university hospitals
ation of oral indigenous bacteria in patients with poor between August 2015 and September 2019. Medical his-
dental hygiene is a major risk factor for empyema, which tory, laboratory, microbiological, and radiological data
can be lethal, especially among the elderly [3]. were obtained from their medical records. Patients were
Identification of the pathogenic bacteria causing pleu- categorised into four risk categories for poor outcomes
ral empyema is clinically important for determining suit- according to the American College of Chest Physicians
able antimicrobial therapy. Facultative anaerobic Gram- (AACP) consensus statement [21] as 1 (very low), 2 (low),
positive cocci, including Streptococcus pneumoniae and 3 (moderate), and 4 (high), based on the volume, localisa-
tion, appearance, pH, and Gram staining/bacterial culture
Streptococcus anginosus group, are predominant in cul-
data of the pleural effusions (Categories 1–3, parapneu-
ture-positive patients, whereas Gram-negative rods such
monic effusion; Category 4, empyema). Patients with
as Escherichia coli, Klebsiella pneumoniae, Haemophilus
malignant effusion were recruited as non-infectious con-
influenzae, Pseudomonas aeruginosa, and anaerobes
trols. The institutional review board of Tokai University
including Peptostreptococci, Fusobacterium nucleatum,
Hospital approved this study (#14 R-220), and all patients
and Prevotella sp., etc. are rare [3–6]. In one study,
provided written informed consent to participate.
Gram-positive cocci were isolated from 76% of patients,
anaerobes from 17%, and Gram-negative bacilli from 5%
[2], whereas in another study, aerobic, anaerobic, and DNA extraction from pleural fluids
both aerobic and anaerobic bacteria were isolated from The pleural fluid samples (obtained by thoracentesis)
64%, 13%, and 23% of patients, respectively [4]. were fractionated into cellular components and super-
Low sensitivity and specificity is a recurring problem natant by centrifugation at 50  g at 4  C for 15 min. The
for identifying pathogenic microorganisms by conven- supernatants were divided into 1 mL aliquots and stored
tional means. Up to 40% of the samples can be culture- at 80  C. Ice-cold aliquots of the supernatants (1 mL)
negative [5,7–11] because some bacterial species are were centrifuged at 20,000  g at 4  C for 15 min, and
uncultivable or require specific culture conditions DNA was extracted from the pellets using the DNeasy
[3–6,12], anaerobes are more difficult to culture than Blood & Tissue Kit (Qiagen GmbH, Hilden, Germany). The
aerobic bacteria [13,14], and most patients are treated pellets were resuspended in 200 mL of buffer (10 mM
with antimicrobials before the pleural fluid is Tris-HCl, 0.1 M NaCl, 1 mM EDTA, 5% (v/v) Triton X-100,
sampled [15]. pH 8.0) with 20 mg/mL lysozyme and 4 mL of RNase
To overcome these problems, Kawanami and col- (100 ng/mL) and incubated at 37  C for 30 min.
leagues applied clonal microflora analysis in which the Proteinase K (25 mL, 20 mg/mL) was added and incu-
16S ribosomal RNA (rRNA) gene was amplified, and 100 bated at 56  C from 30 min to overnight. Following alco-
randomly selected clones were sequenced using the hol precipitation, samples were dissolved in 20 mL of
Sanger method [16]. This strategy identified anaerobes nuclease-free water. DNA was quantified using a
more efficiently and comprehensively than in vitro cul- QubitTM fluorometer with a dsDNA HS assay kit (Thermo
ture, but it is time and labour-intensive and thus, Fisher Scientific, Waltham, MA, USA).
impractical for clinical applications. Other researchers
have applied next-generation sequencing (NGS) technol-
16s rRNA gene amplification
ogy for the diagnosis of infectious respiratory diseases
such as pneumonia and pleural empyema [17–20], how- For some experiments, DNA extracted from the pleural
ever, it is often difficult to differentiate pathogenic bac- effusion was PCR amplified before NGS. Variable regions
teria from contaminated species with NGS analysis of the bacterial 16S rRNA gene were amplified using the
alone. Therefore, we combined 16S rRNA gene-specific 16S Primer Sets V2–4–8 that amplify variable regions
452 Y. SHIRAISHI ET AL.

V2, V4, and V8, and 16S Primer Sets V3–6,7–9 that amp- Enterococcus faecalis, Lactobacillus fermentum, Listeria
lify variable regions V3, V6-7, and V9 from the 16STM monocytogenes, and Staphylococcus aureus) that are diffi-
Metagenomics Kit (Thermo Fisher Scientific). The primer cult to lyse. Fifty mL of the mock microbial community
set V2-4-8 is designed to amplify variable regions V2, V4, standard was added to 1 mL of malignant pleural fluid,
and V8 and the primer set V3-6,7-9 is designed to amp- which was then analysed as above.
lify variable regions V3, V6-7, and V9. Fragments of DNA
were amplified by PCR using the ProFlexTM 3  32-well
Semi-quantitative PCR
PCR System (Thermo Fisher Scientific) according to the
manufacturer’s instructions. The PCR products were puri- We estimated the bacterial loads of the pleural effusions,
fied using an Agencourt AMPure XP system (Beckman by semi-quantitative real-time PCR of the V2 region
Coulter Genomics, Carlsbad, CA, USA) and quantified of the 16S rRNA gene, using the primers, 50 -AGNGGCG
using an Agilent 2100 Bioanalyzer system (Agilent NACGGGTGAGTAAC-30 (forward), and 50 -CTGCCTCCCGTA
Technologies, Santa Clara, CA, USA). GGAGTCTG-30 (reverse). Reactions contained 5 mL of Fast
SYBR Green master mix (Thermo Fisher Scientific), 0.3 mM
of each primer, 1 mL of 10 diluted genomic DNA
NGS and analysis
(equivalent to 2 mL of pleural effusion supernatant), and
For the NGS, barcode adapters were attached using the ultrapure water (to 10 mL final volume). PCR using the
Ion XpressTM Barcode Adapters 1–16 and Ion Plus StepOnePlus Real-Time PCR System (Thermo Fisher
Fragment Library Kit (Thermo Fisher Scientific), each bar- Scientific) was performed as follows: initial denaturation
coded sample diluted to 26 pM and combined as at 95  C for 5 min, 40 amplification cycles at 95  C for
described by the manufacturer. Thereafter, emulsion PCR 20 sec, and 60  C for 30 sec, and then the samples were
was performed using the Ion OneTouchTM 2 System held at 4  C. The reference cycle threshold (CT) value
loaded onto an Ion 316TM Chip. The NGS-based 16S determined by PCR using nuclease-free water was used
rRNA gene amplicon was sequenced using the Ion to calculate the DCT values. We defined the total bacter-
PGMTM with HiQ chemistry (Thermo Fisher Scientific). ial load as 2DCT and the specific bacterial load as the
The Ion Reporter System (Thermo Fisher Scientific) gen- total bacterial load multiplied by the proportion of spe-
erated FASTQ files and the statistics were evaluated cific bacterial species or genera as determined by NGS.
using a SeqKit with a ‘stats’ option [22].
The sequence reads were taxonomically classified by
Statistical analysis
a BLASTN homology search using the Genome Search
Toolkit (GSTK; http://kirill-kryukov.com/study/tools/gstk/) All values are presented as medians and interquartile
against the reference database, the GenomeSync (http:// ranges (IQR). Total or specific bacterial loads were com-
genomesync.org); the GenomeSync contains genomes pared between infectious and non-infectious pleural flu-
from both GenBank [23] and RefSeq [24] and taxonomy ids using Mann-Whitney U tests and receiver operating
annotation from the NCBI Taxonomy Database [25]. The characteristics (ROC) curves. Values with p < .05 were
taxonomic assignment was based on the best-matching considered significantly different. All data were statistic-
genomes between the query sequence of the NGS ally analysed using GraphPad Prism 8 version 9.0.0
results and the reference sequence in the database. We (GraphPad Software, La Jolla, CA, USA).
selected the species with the highest alignment score
computed by BLASTN as the species of interest in the Results
sample. Visual reports were made using KronaTools [26].
Demographic data of the patients
To verify the ability of the 16S rRNA gene amplicon
sequencing to accurately detect bacterial species, we Pleural effusion samples were obtained from 19 patients
used the microbial mixture, ZymoBIOMICS Microbial (male, n ¼ 16; female, n ¼ 3; median age: 69 years) with
Community Standard D6300 (Zymo Research, Irvine CA, infectious effusion, and nine patients with non-infec-
USA), that contains eight bacterial strains (theoretical tious, malignant pleural effusion (male, n ¼ 6; female,
proportion of each genomic DNA: 12%), including three n ¼ 3; median age, 73 years). Table 1 shows the demo-
Gram-negative bacteria (Escherichia coli, Salmonella graphic data. Seven patients with pleural empyema/par-
enterica, and Pseudomonas aeruginosa) that are easy to apneumonic effusion had cancer, five had underlying
lyse and five Gram-positive bacteria (Bacillus subtilis, respiratory diseases such as interstitial pneumonia,
 INFECTIOUS DISEASES 453

chronic obstructive pulmonary disease, or bronchiectasis, antibiotic therapy, 3 (60%) were culture-positive for
two had neuromuscular diseases, and four had a history anaerobes, whereas 5 (36%) of the 14 patients under
of stroke. One patient had an indwelling chest tube to antibiotic therapy were culture-positive, including two
treat malignant effusion at the onset of pleural empy- with anaerobes. Sputum cultures were positive in eight
ema. One patient was being treated with systemic corti- patients and did not match the bacteria isolated from
costeroids for eosinophilic pleurisy. At the time of the pleural effusions in any of the patients. Blood cul-
thoracocentesis, 14 patients had already been treated tures were negative in all patients.
with systemic antibiotics for 1–18 days. According to the
AACP consensus statement regarding the risk for poor
Verification of NGS protocol for
outcomes in patients with parapneumonic effusions [21],
parapneumonic effusion
one, five, and 13 patients were placed in categories 2, 3,
and 4, respectively. Additional file 3: Table S3 summarises the details of the
Tables 2 and 3 and Additional file 1: Table S1 show statistics of our NGS data. We conducted NGS of the first
the bacterial culture results of pleural fluid, sputum, seven samples (Additional file 3: Table S2), both with
and blood. and without 16S rRNA gene amplification. In the
The pleural fluid was culture-positive in eight absence of PCR, bacterial reads accounted for only
patients. Among the five patients without prior 0.03% (IQR: 0.02%  0.04%) of the total reads. In

Table 1. Clinical and laboratory features of 19 infectious and 9 non-infectious effusions.

Infectious effusions
Patient characteristics Non-infectious effusions AACP Categories 1–3 (Parapneumonic) AACP Category 4 (Empyema)
Male/Female 6/3 4/2 12/1
Age (y) 73 (70–75) 71 (66–76) 65 (49–76)
Pleural fluid analysis
Total cell count (cell number/mL) 1023 (539–1582) 3,508 (2,239–4,893) 4,000 (1,034–27,215)
Neutrophil count (cell number/mL) 5 (2–12) 2,500 (1,686–4,122) 2,080 (807–1,7908)
pH 7.44 (7.42–7.46) 7.44 (7.24–7.68) 7.20 (6.98–7.30)
Glucose (mg/dL) 112 (108–135) 120 (90–131) 28 (8–62)
Lactate dehydrogenase (U/mL) 297 (283–686) 557 (475–777) 2,111 (1,312–5,024)
Symptoms/Conditions
Fever 0 6 11
Chest pain 0 6 8
Prior antibiotic therapy 1 4 8
Thoracostomy tube 0 0 1
Immunosuppressive drugs 1 0 1
Underlying diseases
Poor dental hygiene 0 1 9
Cancer 8 4 2
Cardiac failure 2 1 4
Respiratory diseases† 5 2 3
Diabetes mellitus 4 2 2
History of stroke 0 0 4
Renal failure 2 1 1
Neurological disease 0 0 2
Values shown are medians with interquartile ranges.
†
Respiratory diseases included chronic obstructive pulmonary disease, interstitial pneumonia, and bronchiectasis.

Table 2. Major bacterial species identified by 16S rRNA metagenomics in non-infectious effusions.
Antibiotics Pleural fluid
Treatment NGS culture
Prior antibiotic Proportion of specific Specific Average nucleotide Bacterial
treatment DCT Bacterial Species bacteria (%) bacterial load identity (%) species
No 0.32 Methylobacterium sp. Leaf89 13.5 0.11 98.72 ND
No 0.96 Candidatus Burkholderia crenata 11.8 0.23 96.84 ND
No 0.28 ND – – – ND
No 0.36 ND – – – ND
No 0.56 Cutibacterium acnes 21.2 0.31 98.72 ND
No 0.08 ND – – – ND
No 0.2 ND – – – ND
No 1.3 ND – – – ND
No 0.8 ND – – – ND
ND: not detected.
454 Y. SHIRAISHI ET AL.

Table 3. Major bacterial species and their proportions identified by 16S rRNA metagenomics in infectious effusions.
Antibiotic Treatment NGS Pleural fluid culture
Specific Average
AACP Prior antibiotic Period Proportion of bacterial nucleotide
Category treatment (Days) DCT Bacterial Species specific bacteria (%) load identity (%) Bacterial species
2 Yes 5 6.6 Streptococcus intermedius 81.1 78.73 99.30 ND
3 No – 7.5 Streptococcus sp. 72.7 130.75 – ND
S. mitis 26.9 48.35 98.69
S. infantis 16.1 29.98 98.64
S. pneumoniae 29.7 53.42 98.54
3 No – 4.2 Lawsonella clevelandensis 14.4 2.71 98.96 ND
Cutibacterium acnes 13.7 2.57 98.09
3 Yes 3 7.7 Cutibacterium acnes 11.9 25.20 98.84 ND
3 Yes 3 6.8 Fusobacterium nucleatum 89.3 101.68 98.69 Pseudomonas aeruginosa
3 Yes 3 0.5 Porphyromonas sp. 12.7 0.09 – ND
P. endodontalis 8.0 0.06 98.71
P. gingivalis 4.7 0.03 98.60
4 No – 11.8 Enterococcus sp. 76.4 2724.46 – Enterococcus faecium
E. faecium 38.1 1358.47 98.79
E. phoeniculicola 19.6 699.05 98.71
E. durans 18.7 666.88 98.81
4 No – 1.6 Fusobacterium nucleatum 55.7 1.72 98.77 Fusobacterium sp.
Prevotella buccae 12.8 0.39 96.79
4 No – 5.7 Fusobacterium nucleatum 62.5 31.68 99.32 Fusobacterium nucleatum
4 Yes 18 6.1 Porphyromonas endodontalis 24.6 17.17 98.61 Peptostreptococcus sp.
Fusobacterium nucleatum 21.2 14.77 98.61 Campylobacter curvus
Prevotella oris 16.2 11.28 98.28
4 Yes 14 13.4 Prevotella oris 76.6 8561.83 98.07 Streptococcus anginosus
4 Yes 6 5.1 Streptococcus intermedius 16.4 5.53 98.78 ND
4 Yes 4 11.5 Porphyromonas gingivalis 84.1 2483.26 98.00 Porphyromonas sp.
Prevotella sp.
4 Yes 2 1.0 Staphylococcus sp. 63.1 1.26 – Staphylococcus aureus
S. aureus 39.1 0.78 98.95
S. simiae 24.0 0.48 98.90
4 Yes 9 0.6 Streptococcus intermedius 34.5 0.51 99.20 ND
Filifactor alocis 18.9 0.28 99.00
4 Yes 6 1.0 Streptococcus intermedius 53.1 1.08 99.11 ND
4 Yes 4 0.1 Streptococcus intermedius 37.9 0.39 99.25 ND
4 Yes 6 4.9 Porphyromonas gingivalis 54.4 16.26 98.62 ND
Fusobacterium sp. 23.4 6.99 –
F. periodonticum 12.2 3.66 98.65
F. nucleatum 11.2 3.36 98.85
4 Yes 4 4.7 Enterococcus sp. 56.9 14.57 – ND
E. durans 29.6 7.56 99.47
E. faecium 16.1 4.13 99.35
E. phoeniculicola 11.2 2.87 99.12
Prevotella oris 29.6 7.56 98.71
Porphyromonas gingivalis 11.5 2.94 98.66
ND: not detected.

contrast, >95% of the total reads from post-PCR samples 10.1%, Lactobacillus fermentum 18.4%, Listeria monocyto-
were annotated as bacterial DNA, regardless of the pleu- genes 14.1%, Pseudomonas aeruginosa 4.2%, Salmonella
ral effusion category (Additional file 2: Table S2). enterica 10.4%, and Staphylococcus aureus 15.5%. The
Comparing NGS with 16S rRNA PCR and NGS without measured proportion of each bacterial species [mean
16S rRNA PCR, only three out of seven cases showed (standard deviation)] were: Bacillus subtilis 12.6 (2.5)%,
identical bacterial species. Therefore, we amplified the Enterococcus faecalis 9.1 (1.6)%, Escherichia coli 14.4
16S rRNA gene by PCR in the subsequent samples (3.6)%, Lactobacillus fermentum 11.4 (1.5)%, Listeria
before NGS. monocytogenes 11.5 (0.4)%, Pseudomonas aeruginosa 3.8
Additional file 4: Figure S1 shows the results of the (1.1)%, Salmonella enterica 23.1 (5.4)%, and
mock standard verification. All bacterial species were Staphylococcus aureus 16.9 (3.0)%.
detectable in the three samples of the malignant pleural
effusion spiked with the mock standard. The theoretical
NGS of non-infectious and infectious effusions
proportion of 16S rRNA genes based on the genome
size and copy number of 16S rRNA gene are: Bacillus We assessed malignant pleural effusions from nine
subtilis 17.4%, Enterococcus faecalis 9.9%, Escherichia coli patients with no signs of active intrapleural infection.
 INFECTIOUS DISEASES 455

Bacterial species accounting for more than 10% of the Discrimination between infectious and non-infectious
total bacterial genomes by 16S rRNA gene sequence pleural effusion
analysis were identified in three patients (33%); NGS can only determine the proportions of bacterial
Cutibacterium acnes, Methylobacterium sp. Leaf89, genomes in a given sample. To discriminate between
Candidatus burkholderia crenata in one patient each infectious and non-infectious pleural effusions, we ana-
(Tables 2 and 3). lysed the largest proportions of the bacterial species,
Major bacterial species accounting for >10% of the the bacterial genome load by real-time PCR of the V2
total genomes including fourteen anaerobes and five region of the 16S rRNA gene, and the specific bacterial
aerobes/facultative anaerobes were identified in all the load, which is the total bacterial load multiplied by the
pleural empyema/parapneumonic effusions. proportion of specific bacteria determined by NGS
Streptococcus sp. (n ¼ 6), Fusobacterium sp. (n ¼ 5), (Table 2, Figure 2). The median value of the largest pro-
Porphyromonas sp. (n ¼ 5), Prevotella sp. (n ¼ 4), portions of bacterial species from infectious effusions
Cutibacterium acnes (n ¼ 2), and Enterococcus sp. (n ¼ 2) was 0.56 (IQR, 0.30  0.75), which was significantly
were identified in two or more effusions. Detection rates higher than that observed in non-infectious effusions
and bacterial species did not differ substantially based (0.07; IQR, 0.05  0.12, p < .0001). The area under the
on the AACP risk category or prior antibiotic treatment ROC curve (AUC) was 0.96 and the cut-off value to dis-
(Figure 1). The predominant bacteria comprising >50% criminate between infectious and non-infectious samples
total bacterial genomes identified in 12 samples were: was 0.125 with a Youden index of 0.73 (Figure 2(A,B)).
Streptococcus sp. (S. intermedius or S. mitis/S. infantis/S. The median total bacterial load in infectious pleural effu-
pneumoniae) and F. nucleatum in three effusions each, sions was 33.7 (IQR, 5.8–146.9), which was also signifi-
Enterococcus sp. (E. durans/E. faecium/E. phoeniculicola) cantly higher than that observed non-infectious
and Porphyromonas gingivalis in two effusions each, and malignant effusions (1.3; IQR, 1.1–1.7; p < .01). The AUC
Prevotella oris, and Staphylococcus sp. (S. aureus/S. sim- of the total bacterial load was 0.88 and the cut-off to
iae/S. haemolyticus) in one effusion each. Four effusions discriminate between infectious and non-infectious sam-
were probably infected with two or more bacteria, three ples was 2.73 with a Youden index of 0.79 (Figure
2(C,D)). The median specific bacterial loads were 16.3
had multiple anaerobes, one had Enterococcus sp. and
(IQR, 2.2–90.2) and 0.08 (IQR, 0.07–0.13) for infectious
anaerobes, and two had mixed infections (Figure 1(A)).
and non-infectious effusions, respectively (p < .001). The
AUC of the specific bacterial load was 0.98 and the cut-
off to discriminate infectious from non-infectious sam-
ples was 0.35 with a Youden index of 0.95
Comparison of bacterial species identified by culture (Figure 2(E,F)).
and NGS
The predominant bacteria identified by NGS matched Discussion
the culture results in five effusions (Table 2). In contrast,
We show here that combining 16S rRNA gene amplicon
the predominant bacteria identified in three effusions by
sequencing with semi-quantitative PCR can reliably
NGS did not match those identified by culture. In an
detect bacterial infection and identify the causative
effusion that was positive for Peptostreptococcus sp. and
microorganism(s) of pleural empyema and parapneu-
Campylobacter curvus in culture, these bacteria repre- monic effusions more efficiently and accurately than
sented only 0.01% and 1.6% of the total bacterial conventional culture methods, even when patients have
genomes, respectively, whereas NGS identified the anae- previously been treated with antibiotics.
robes, Porphyromonas endodontalis (25%), Fusobacterium Prokaryote-specific 16S rRNA gene amplicon sequencing
nucleatum (21%), and Prevotella oris (16%). In the other has been widely applied to microbiome analyses, but few
two effusions, culture-positive Streptococcus anginosus studies have highlighted the value of NGS for diagnosing
and Pseudomonas aeruginosa, accounted for only 0.9% infectious diseases such as pneumonia and pleural empy-
and 0.2% of the total bacterial genomes, respectively. ema [17–20]. One reason could be the difficulty of identify-
Prevotella oris (77%) and F. nucleatum (89%) were the ing pathogenic species due to the similarity of the 16S
predominant bacteria according to NGS in rRNA gene among species. We transcended this issue using
these samples. GenomeSync and GSTK, our in-house database and
456 Y. SHIRAISHI ET AL.

(A) 1.0
, Cutibacterium

Proportion of Major
Bacterial Genus
0.8 Enterococcus
Fusobacterium
0.6 Porphyromonas
0.4 Prevotella
Pseudomonas
0.2 Staphylococcus
Streptococcus
0
(B) Other genus of Bacteria
104
Specific Bacterial

103
Genus Load

102
10
1
10-1
10-2
AACP Category 2 3 3 3 3 3 4 4 4 4 4 4 4 4 4 4 4 4 4
Antibiotics Treatment + + + + + + + + + + + + + + + +
Non-infectious Infectious

Figure 1. Distribution of bacterial genera in non-infectious and infectious pleural effusions. The genera of major bacteria in non-infectious
(n ¼ 9) and infectious (n ¼ 19) samples are presented with the AACP risk category and prior antibiotic treatment in the order shown in
Table 2. (A) Proportion of major bacterial genera in each sample. (B) Specific bacterial load defined as total bacterial load (2DCT in PCR of
the 16S rRNA gene V2 region) multiplied by the proportion of the most dominant bacterial genus determined by NGS.

software. Our sequence search comprised all-to-all non-infectious patients, suggesting that this bacterium,
sequence similarity searches without clustering in GSTK which is also found in the oral cavity, might have caused
[27–31]. Besides, the GenomeSync database is continuously the pyogenic pleural infections as previously
updated by peer-to-peer networking with the NCBI reported [34–37].
GenBank [23] and RefSeq [24] databases and is now one of Several technical limitations remain in our method-
the largest genome databases worldwide. ology. The efficiency of DNA extraction and PCR amplifi-
Possible causative bacteria were identified by NGS in cation is inconsistent among bacterial species. We
almost all effusion samples, whereas the culture-positive examined this variability using a mock standard that
rate was only 42% as previously reported [5,7–11]. We contained eight bacterial species with theoretical
found that over 50% of our patients had empyema/para- amounts of 16S rRNA genes (4.2–18.4%). The proportion
pneumonic effusions caused by mono- or poly-infection determined by 16S rRNA gene amplicon sequencing
with anaerobes, a result compatible with the NGS find- showed some deviation from the theoretical proportion,
ings of Dyrhovden et al. [17], who also found discordant ranging between 3.8% and 23.1%. Direct sequencing
results with bacterial cultures in 63% of their samples, without 16S rRNA gene amplification would avoid vari-
similar to our results (73%). This discordance might ability during PCR amplification, but the results were not
result from laboratory culture conditions or antibiotic reliable in 4 (57%) of our samples due to a low bacterial
therapy before pleural effusion sampling. Streptococcus load (Additional file 2: Table S2). Another problem was
pneumoniae, S. anginosus group, and other aerobes/fac- that multiple species in the same genus such as S. pneu-
ultative anaerobes are more likely to be detected in cul- moniae/mitis/infantis, E. faecium/phoeniculicola/durans, or
ture than anaerobic microbes [4,14,32]. However, one S. aureus/simiae were annotated in some samples. The
study found that anaerobes such as Gram-positive alignment scores among these species were so close
anaerobic cocci, Fusobacterium, microaerophilic strepto- that we could not determine whether the results were
cocci, Prevotella, and Bacteroides species could be identi- due to mixed infections or mis-annotation. Sequencing
fied in 74% of empyema cases under culture conditions other targets, such as the atpA, groEL, RecN, or kat
more suitable for anaerobes [33]. Cutibacterium acnes, genes, groES locus, or 16S-23S rRNA gene internal tran-
identified in two of our patients with AACP category 3 scribed spacer regions might help overcome these prob-
effusions, might have originated from the skin during lems [38–45], although it is necessary to design primers
thoracentesis. Nevertheless, the specific bacterial load of that target specific groups of bacteria and perform
C. acnes was higher in these samples than in those from another round of NGS analysis. We previously combined
 INFECTIOUS DISEASES 457

(A) (B)
1.0
periodontopathic microorganisms are more prone to
1.0
Largest Proportion of

phagocytosis by neutrophils than periodontopathic

Bacterial Species
0.8 0.8
pathogens such as Porphyromonas and Prevotella sp.

Sensitivity
0.6 0.6
[47]. This might explain why bacterial loads were smaller
0.4 0.4
in patients with S. intermedius-induced pleural empyema
0.2 0.2
(Table 3). Further analysis of more patients is needed to
0.0 0.0
0.0 0.5 1.0 establish appropriate cut-off values for specific genome
us

us
io

io

1 - Specificity
ct

ct

loads to discriminate infectious from non-infectious

fe

fe
-in

In
on

pleural effusions.
N

(C) (D)

Conclusions
NGS analysis combined with semi-quantitative PCR of
the 16S rRNA gene can identify causative micro-organ-
isms more accurately than conventional culture meth-
ods, and thus serve as a powerful tool that could
facilitate appropriate and timely treatment of pleural
empyema and parapneumonic effusions.
(E) (F)

Acknowledgments
We acknowledge the valuable technical support and comments
regarding the NGS analysis provided by Keiko Yokoyama, Takuma
Araki, Masayuki Tanaka, Tadayuki, Wang Ting, Tadayuki Satou,
Hideki Hayashi, and Nobuo Watanabe. We are grateful to Naoki
Hayama, Tsuyoshi Oguma, Takuya Aoki, Tomoe Takeuchi, Keito
Enokida, and Shohei Obayashi for recruiting participants and pro-
viding clinical support.
Figure 2. Comparison of non–infectious and infectious pleural effu-
sions. Proportions of the largest bacterial species (A), total (C), and
specific (E) bacterial loads of non-infectious and infectious pleural Ethics approval and consent to participate
effusions are shown as medians with IQR. ROC curve of proportions The Institutional Review Board for Clinical Research at Tokai
of bacterial genome (B), total (D), and specific (F) bacterial loads
University approved the study (Approval No: 14 R-220), which was
were analysed to discriminate non-infectious from parapneu-
monic samples. conducted according to the Declaration of Helsinki (2013 amend-
ment). The study was initiated after patients received written
explanations of the study and its protocols. After ensuring all the
atpA and 16S rRNA gene amplification to identify a bac-
details were fully understood, the patients provided written
terium that was annotated as a subspecies of
informed consent to participate.
Campylobacter curvus [27]. Alternatively, sequencing lon-
ger fragments of 16S rRNA gene such as V1-V2 region
might improve annotation rates [46]. Author contributions
The NGS data provided information about the bacter- YS and KA contributed to the conceptualisation and the design of
ial proportion, but not the number of specific bacteria. the study and data interpretation.
Therefore, NGS alone cannot differentiate infectious KK analysed the NGS data. SN and TI supervised the bioinfor-
from non-infectious pleural effusions. Therefore, we per- matic analysis; KT, FS, and SI acquired the clinical data and the
pleural effusion samples. YS drafted the manuscript and con-
formed semi-quantitative PCR of the 16S rRNA gene to
ducted statistical analyses. All authors contributed to the critical
calculate total and specific bacterial loads. The AUC indi-
revision of the manuscript and approved its submission for
cated that the specific bacterial loads are more suitable publication.
for identifying infectious pleural effusions. Three of our
patients with AACP category 4 effusions caused by
Streptococcus intermedius had specific bacterial loads Disclosure statement
close to the cut-off value. Streptococci and other non- The authors declare that they have no competing interests.
458 Y. SHIRAISHI ET AL.

Funding [12] Maskell NA, Davies CWH, Nunn AJ, et al. U.K. Controlled
trial of intrapleural streptokinase for pleural infection. N
This study was supported by the Japan Initiative for Global
Engl J Med. 2005;352(9):865–874.
Research Network on Infectious Diseases (J-GRID) of Japan Agency
[13] Bartlett JG. Anaerobic bacterial infections of the lung.
for Medical Research and Development (AMED) under Grant
Chest. 1987;91(6):901–909.
Number JP17fm0108023, and the Takeda Science Foundation. [14] Menzies SM, Rahman NM, Wrightson JM, et al. Blood cul-
ture bottle culture of pleural fluid in pleural infection.
Thorax. 2011;66(8):658–662.
ORCID [15] Vincent JL, De Backer D. Circulatory shock. N Engl J Med.
2013;369(18):1726–1734.
Yoshiki Shiraishi http://orcid.org/0000-0001-5275-6318 [16] Kawanami T, Fukuda K, Yatera K, et al. A higher significance
Kirill Kryukov http://orcid.org/0000-0002-0286-0288 of anaerobes: the clone library analysis of bacterial pleurisy.
So Nakagawa http://orcid.org/0000-0003-1760-3839 Chest. 2011;139(3):600–608.
Tadashi Imanishi http://orcid.org/0000-0002-1182-9127 [17] Dyrhovden R, Nygaard RM, Patel R, et al. The bacterial aeti-
Koichiro Asano http://orcid.org/0000-0002-9044-3061
ology of pleural empyema. A descriptive and comparative
metagenomic study. Clin Microbiol Infect. 2019;25(8):
981–986.
Data availability statement
[18] Sung JY, Hwang Y, Shin MH, et al. Utility of conventional
The dataset used and/or analysed during the current study is culture and MALDI-TOF MS for identification of microbial
available from the corresponding author on reasonable request. communities in bronchoalveolar lavage fluid in comparison
with the GS junior next generation sequencing system.
Ann Lab Med. 2018;38(2):110–118.
References [19] Kitsios GD, Fitch A, Manatakis DV, et al. Respiratory micro-
[1] Light RW. Parapneumonic effusions and empyema. Proc biome profiling for etiologic diagnosis of pneumonia in
Am Thorac Soc. 2006;3(1):75–80. mechanically ventilated patients. Front Microbiol. 2018;9:
[2] Falguera M, Carratala J, Bielsa S, et al. Predictive factors, 1413.
microbiology and outcome of patients with parapneu- [20] Yang L, Haidar G, Zia H, et al. Metagenomic identification
monic effusion. Eur Respir J. 2011;38(5):1173–1179. of severe pneumonia pathogens in mechanically-ventilated
[3] Davies CW, Kearney SE, Gleeson FV, et al. Predictors of out- patients: a feasibility and clinical validity study. Respir Res.
come and long-term survival in patients with pleural infec- 2019;20(1):265.
tion. Am J Respir Crit Care Med. 1999;160(5):1682–1687. [21] Colice GL, Curtis A, Deslauriers J, et al. Medical and surgical
[4] Brook I, Frazier EH. Aerobic and anaerobic microbiology of treatment of parapneumonic effusions: an evidence-based
empyema. A retrospective review in two military hospitals. guideline. Chest. 2000;118(4):1158–1171.
[22] Shen W, Le S, Li Y, et al. SeqKit: a cross-platform and ultra-
Chest. 1993;103(5):1502–1507.
[5] Barnes TW, Olson EJ, Morgenthaler TI, et al. Low yield of fast toolkit for FASTA/Q file manipulation. PLoS One. 2016;
11(10):e0163962.
microbiologic studies on pleural fluid specimens. Chest.
[23] Benson DA, Cavanaugh M, Clark K, et al. GenBank. Nucleic
2005;127(3):916–921.
Acids Res. 2017;45(D1):D37–D42.
[6] Ozol D, Oktem S, Erdinc E. Complicated parapneumonic
[24] O’Leary NA, Wright MW, Brister JR, et al. Reference
effusion and empyema thoracis: microbiologic and thera-
sequence (RefSeq) database at NCBI: current status, taxo-
peutic aspects. Respir Med. 2006;100(2):286–291.
nomic expansion, and functional annotation. Nucleic Acids
[7] Byington CL, Spencer LY, Johnson TA, et al. An epidemio-
Res. 2016;44:D733–45.
logical investigation of a sustained high rate of pediatric
[25] Federhen S. Type material in the NCBI Taxonomy Database.
parapneumonic empyema: risk factors and microbiological
Nucleic Acids Res. 2015;43(Database issue):D1086–D1098.
associations. Clin Infect Dis. 2002;34(4):434–440. [26] Ondov BD, Bergman NH, Phillippy AM. Interactive metage-
[8] Jimenez D, Diaz G, Garcia-Rull S, et al. Routine use of pleu- nomic visualization in a Web browser. BMC Bioinformatics.
ral fluid cultures. Are they indicated? Limited yield, minimal 2011;12:385.
impact on treatment decisions. Respir Med. 2006;100(11): [27] Horio Y, Shiraishi Y, Watanabe N, et al. Empyema associ-
2048–2052. ated with Campylobacter curvus infection. Respirol Case
[9] Labelle AJ, Arnold H, Reichley RM, et al. A comparison of Rep. 2017;5(4):e00234.
culture-positive and culture-negative health-care-associated [28] Kryukov K, Ueda MT, Imanishi T, et al. Systematic survey of
pneumonia. Chest. 2010;137(5):1130–1137. non-retroviral virus-like elements in eukaryotic genomes.
[10] Medford AR, Medford A, Maskell N. Pleural effusion. Virus Res. 2019;262:30–36.
Postgrad Med J. 2005;81(961):702–710. [29] Mitsuhashi S, Kryukov K, Nakagawa S, et al. A portable sys-
[11] Thomson AH, Hull J, Kumar MR, et al. Randomised trial of tem for rapid bacterial composition analysis using a nano-
intrapleural urokinase in the treatment of childhood empy- pore-based sequencer and laptop computer. Sci Rep. 2017;
ema. Thorax. 2002;57(4):343–347. 7(1):5657.
 INFECTIOUS DISEASES 459

[30] Watanabe N, Kryukov K, Nakagawa S, et al. Detection of [40] Miller WG, Yee E, Jolley KA, et al. Use of an improved atpA
pathogenic bacteria in the blood from sepsis patients using amplification and sequencing method to identify members
16S rRNA gene amplicon sequencing analysis. PLoS One. of the Campylobacteraceae and Helicobacteraceae. Lett
2018;13(8):e0202049. Appl Microbiol. 2014;58(6):582–590.
[31] Nakagawa S, Inoue S, Kryukov K, et al. Rapid sequencing- [41] Tang Y, Underwood A, Gielbert A, et al. Metaproteomics
based diagnosis of infectious bacterial species from menin- analysis reveals the adaptation process for the chicken gut
gitis patients in Zambia. Clin Transl Immunology. 2019; microbiota. Appl Environ Microbiol. 2014;80(2):478–485.
8(11):e01087. [42] Glazunova OO, Raoult D, Roux V. Partial recN gene
[32] Ferguson AD, Prescott RJ, Selkon JB, et al. The clinical
sequencing: a new tool for identification and phylogeny
course and management of thoracic empyema. QJM. 1996;
within the genus Streptococcus. Int J Syst Evol Microbiol.
89(4):285–289.
2010;60(Pt 9):2140–2148.
[33] Boyanova L, Vladimir D, Gergova G, et al. Anaerobic micro-
[43] Kim W, Park HK, Hwang WJ, et al. Simultaneous detection
biology in 198 cases of pleural empyema: a Bulgarian
study. Anaerobe. 2004;10:261–267. of Streptococcus pneumoniae, S. mitis, and S. oralis by a
[34] Pan SC, Wang JT, Hsueh PR, et al. Endocarditis caused by novel multiplex PCR assay targeting the gyrB gene. J Clin
Propionibacterium acnes: an easily ignored pathogen. J Microbiol. 2013;51(3):835–840.
Infect. 2005;51(4):e229–e231. [44] Blaiotta G, Fusco V, Ercolini D, et al. Diversity of
[35] Odunukan OW, Masannat F, Baka JJ. Propionibacterium Staphylococcus species strains based on partial kat (cata-
acnes brain abscess in an immunocompetent man in the lase) gene sequences and design of a PCR-restriction frag-
absence of prior neurosurgery. S D Med. 2016;69:71–73. ment length polymorphism assay for identification and
[36] Veitch D, Abioye A, Morris-Jones S, et al. Propionibacterium differentiation of coagulase-positive species (S. aureus, S.
acnes as a cause of lung abscess in a cardiac transplant delphini, S. hyicus, S. intermedius, S. pseudintermedius, and S.
recipient. BMJ Case Rep. 2015;2015:bcr2015212431. schleiferi subsp. coagulans. J Clin Microbiol. 2010;48(1):
[37] Claeys G, Verschraegen G, De Potter C, et al. 192–201.
Bronchopneumonia caused by Propionibacterium acnes. [45] Conrads G, Citron DM, Tyrrell KL, et al. 16S-23S rRNA gene
European journal of clinical microbiology & infectious. Eur J
internal transcribed spacer sequences for analysis of the
Clin Microbiol Infect Dis. 1994;13(9):747–749.
phylogenetic relationships among species of the genus
[38] Ward NL, Rainey FA, Hedlund BP, et al. Comparative phylo-
Porphyromonas. Int J Syst Evol Microbiol. 2005;55(Pt 2):
genetic analyses of members of the order
Planctomycetales and the division Verrucomicrobia: 23S 607–613.
rRNA gene sequence analysis supports the 16S rRNA gene [46] Watts GS, Youens-Clark K, Slepian MJ, et al. 16S rRNA gene
sequence-derived phylogeny. Int J Syst Evol Microbiol. sequencing on a benchtop sequencer: accuracy for identifi-
2000;50 Pt 6:1965–1972. cation of clinically important bacteria. J Appl Microbiol.
[39] Wang Y, Qian PY. Conservative fragments in bacterial 16S 2017;123(6):1584–1596.
rRNA genes and primer design for 16S ribosomal DNA [47] Ji S, Hyun J, Park E, et al. Susceptibility of various oral bac-
amplicons in metagenomic studies. PLoS One. 2009;4(10): teria to antimicrobial peptides and to phagocytosis by neu-
e7401. trophils. J Periodontal Res. 2007;42(5):410–419.