Molecular epidemiology of Clostridium neonatale and its relationship with the occurrence of necrotizing enterocolitis in preterm neonates.

Clostridia-especially Clostridium butyricum-are among the taxa most frequently identified from stool samples of preterm neonates with necrotizing enterocolitis (NEC). Recently, Clostridium neonatale has also been detected from epidemic cases, but using a culture-based approach we were unable to confirm this discovery in a local cohort. In order to investigate this link by a molecular approach, a specific rpoB-based quantitative real-time PCR was developed to detect C. neonatale directly from patients' stool specimens. Design of this rpoB-based quantitative real-time PCR was based on the genomic analysis of seven clinical isolates of C. neonatale. It was tested on stool samples from 88 preterm neonates with necrotizing enterocolitis and 71 matched controls. C. neonatale was significantly more prevalent in stools from preterm neonates with necrotizing enterocolitis than in controls (respectively 30/88 (34%) versus 9/71 (13%); p 0.003). Whole-genome analysis also allowed the identification of three genomic clusters of C. neonatale. This clustering was associated with a geographical location regardless of isolation from the NEC or control, suggesting asymptomatic carriage. Although less prevalent than C. butyricum in our cohort, C. neonatale is significantly associated with the occurrence of necrotizing enterocolitis.

Introduction

Necrotizing enterocolitis (NEC) is a major cause of morbidity and mortality in neonatal intensive care units (NICUs), especially for very-low-birth-weight infants [1]. Over the past decade, most studies have focused on intestinal immaturity, feeding strategies, and the composition of the gut microbiota, as NEC is recognized as a multifactorial disease [1,2]. Approaches to determining the composition of the gut microbiota were performed using molecular tools such as 16S rRNA pyrosequencing, shotgun metagenomic sequencing, and specific quantitative real-time PCR (qPCR) [1,3,4]. These tools have been used to describe dysbiosis-linked alterations in neonates' gut microbiota. In addition, several bacterial species have been specifically associated with NEC; these include Klebsiella pneumoniae, Enterobacter cloacae, uropathogenic Escherichia coli (UPEC), Clostridium butyricum and Clostridium neonatale [1,[4], [5], [6], [7]]. This latter was first isolated in the context of epidemic NEC cases in a Canadian NICU [8]. Pulsed-field gel electrophoresis identified a similar banding profile from blood and stool culture isolates [8]. Later, Roze et al. reported correlations between nutritional strategies and gut microbiota composition, where C. neonatale was identified from the first stool of preterm neonates with NEC [9], and its genome sequence was determined [10,11]. A causal relationship between the presence of C. neonatale in stool samples and NEC has not yet been clearly demonstrated, but it warrants further study. Based on a cohort of neonates and controls enrolled to study the involvement of C. butyricum in NEC [3,7], we evaluated the association of C. neonatale with the occurrence of NEC. This hypothesis was developed by studying the frequency of C. neonatale in the neonates' stools using a specific rpoB-based qPCR, and then comparing the strains isolated by genomic sequencing.

Methods

Study design and patients

The study was validated by agreements from the ethics committee of the Institut Fédératif de Recherche, IFR48, and the Institut Hospitalo-Universitaire, IHU-2017-007. Written approval was obtained from the parents of all patients [3]. One hundred and fifty-nine stool samples were collected from preterm neonates whose parents had consented to the study; the neonates were treated in five NICUs. Enrolled samples were obtained from our previous cohort studies and included 88 patients with NEC and 71 healthy controls [3,7]. None of the patients were subjected to probiotic therapy. Stool samples were collected, if possible, on the day of symptom onset and stored at –80°C. Patients and controls were grouped by sex, gestational age (under 37 gestational weeks), birth weight, days of life, feeding strategies, mode of delivery, and previous antibiotic therapy (Table 1). Routine microbiological analysis was negative for all samples.

Table 1

Factors associated with the occurrence of necrotizing enterocolitis (NEC) compared with control preterm neonates

Factors	NEC (n = 88)	Controls (n = 71)	Univariate analysis	Multivariate analysis
	n (%)	n (%)	p value	OR (95%CI)	p value
Gestational age ± SD (days)	28.36 (±2.82)	28.24 (±2.93)	0.7864	NA	NA
Days of life ± SD (days)	25.99 (±13.16)	23.99 (±11.61)	0.3166	NA	NA
Male sex	49 (55.7)	43 (60.5)	0.594	NA	NA
Birth weight mean ± SD (g)	1179 (±380)	1174 (±432)	0.950	NA	NA
Very low birth weight (≤1500g)	69 (78.4)	58 (81.7)	0.617	NA	NA
Pasteurized breast milk	59 (67)	49 (69)	0.372	NA	NA
Formula fed	9 (10)	13 (18)	0.139	NA	NA
Vaginal delivery	32 (36)	37 (52)	0.023	2.3 [1.2-4.5]	0.018
Antibiotics before collection	64 (73)	40 (56)	0.092	1.9 [0.9-3.8]	0.07
Clostridium neonatale q-PCR-positive	30 (34)	9 (13)	0.003	3.7 [1.6-8.8]	0.003

SD, standard deviation; NA, not applicable.

Genome sequencing, assembly and annotation of Clostridium neonatale isolates

Five C. neonatale strains—all from Marseille and isolated in a previous work—were sequenced and analysed [7]. Four were NEC-associated isolates (NEC25, NEC26, NEC32, NEC86) and one was from a control neonate (C25). Strains were cultured on 5% Columbia sheep-blood agar (Becton Dickinson®, USA) at 37°C for 48 h under anaerobic conditions before DNA extraction using an EZ1 DNA Tissue Kit (QIAGEN, Germany). Genomic DNA was sequenced with MiSeq Technology (Illumina Inc., San Diego, CA, USA) using the paired-end applications. Strain NEC86 was additionally sequenced by MinION technology (Nanopore, Oxford, UK) [12,13]. SPAdes software was used for NEC86 reads assembly obtained from both sequencing methods [14]. Reads of the other C. neonatale genomes were mapped against the NEC86 dataset using the CLC genomics workbench 7 (Qiagen Inc., Valencia, CA, USA). Coding DNA sequences (CDSs) and annotation were procured using Prokka software [15]. Finally, sequenced genomes were deposited in the European Molecular Biology Laboratory, European Bioinformatics Institute (EMBL-EBI), under accession numbers: NEC25 (UWJD01000001–01000003), NEC26 (UICU01000001–01000003), NEC32 (UICW01000001–01000003), NEC86 (UICR01000001–01000003), and C25 (UICQ01000001–01000003) (Table 2). Further analysis and comparison were conducted by introducing genome sequences from two strains isolated from a Canadian NICU (LCDC99A005: GCA_001458595.1 and LCDC99A006: GCA_002553455.1) [8,11].

Table 2

Characteristics of Clostridium neonatale draft genomes

Isolate	Location, date of isolation	Sequencing technique	Scaffold number	Genome length (base pair)	GC content (%)	ORFs number	Accession number
LCDC 99A005	NICU-1, 2000	Mate-pair	3	4 304 459	28.6	3795	GCA_001458595.1
LCDC 99A006	NICU-1, 2000	NA	58	4 658 596	28.4	4139	GCA_002553455.1
NEC25	NICU-3, December 2011	Paired-end	3	4 352 344	28.7	3883	UWJD01000001–01000003
NEC26	NICU-3, July 2012	Paired-end	3	4 282 644	28.9	3840	UICU01000001–01000003
NEC32	NICU-3, December 2012	Paired-end	3	4 321 416	28.7	3859	UICW01000001–01000003
NEC86	NICU-2, December 2010	Paired-end, Nanopore	3	4 739 641	28.6	4188	UICR01000001–01000003
C25	NICU-3, February 2012	Paired-end	3	4 358 232	28.6	3913	UICQ01000001–01000003

ORF, open-reading frame number; NEC, necrotizing enterocolitis; C25, control; NICU, neonatal intensive care unit; NICU-1, Canada; NICU-2, Marseille, France; NICU-3, Marseille, France; NA: not available.

Design of specific qPCR for Clostridium neonatale detection

As previously performed for the detection of C. butyricum [3], we designed a specific qPCR targeting the rpoB gene of C. neonatale. First, we verified that a unique copy of the rpoB gene was present in the C. neonatale genome (see above). Second, these sequences were aligned with those of four pathogenic species using MEGA7 software [16]. Other Clostridium genomes included in this analysis were C. butyricum E4 str. BoNT E BL5262 (GCA_000182605.1), Clostridium botulinum A str. (GCA_000017025.1), Clostridium perfringens ATCC 13124 (GCA_000013285.1) and Clostridioides (Clostridium) difficile QCD-66c26 (GCA_000003215.1) as an outgroup at the class level (Clostridia). Primers and probe were designed to detect specifically the C. neonatale rpoB gene using Primer3 software [17]. Finally, a 151-nucleotide rpoB sequence was selected; this region is homogenous among C. neonatale strains and highly heterogenic compared to that in other clostridia. This molecular construction was validated by a maximum-likelihood phylogenetic analysis generated by MEGA7 software (Fig. 1) [16]. LightCycler® 480 Probes Master (Roche, Germany) was used to achieve qPCR reactions. The optimized 20 μL qPCR mix contained: master mix (10 μL), primers (0.5 μL, 20 nM), probe (0.5 μL, 5 nM), water (3.5 μL) and DNA (5 μL). The qPCR cycling protocol started with one cycle of 50°C for 2 minutes, followed by one activation cycle (95°C for 5 min) and 40 amplification cycles (95°C for 1 sec; Tm 60°C for 30 sec). Tubes were placed in a thermal cycler CFX96 Touch™ (Bio-rad®, France). Sequences of the qPCR system were designed as follows: forward primer (AATAGTTGATAAAGAACACGGTAGAGC), reverse primer (TAGCGGTTCTTTTGCTTGAG), and probe (FAM-TGCAGATGAAGAAGATCAGTTTTT-TAMRA) (supplementary material Fig. S1). First, the qPCR system was tested on a collection of 79 bacterial species associated with the gut microbiota, including five locally isolated C. neonatale strains and 24 Clostridium species (supplementary material Table S1). Total genomic DNA was extracted directly from stool samples using the NucleoSpin Tissue Kit (Macherey-Nagel, Hoerdt, France) as previously described [7].

Fig. 1

In silico analysis of Clostridium neonatale rpoB-specific region. This analysis was performed on the selected region of 151 nucleotides. Phylogenetic analysis showed that the rpoB sequence is homogenous among C. neonatale strains and classified it among Clostridium species. C. difficile was out-grouped. NEC, necrotizing enterocolitis; C25, control.

Statistical analysis

Statistical analysis was performed using SPSS® statistics 2016 (IBM, NY, USA). Mean and standard deviation were used to describe continuous variables. Percentage and number of events were used for quantitative variables. The Student t-test or Mann–Whitney U test were used to perform two-group comparisons for quantitative variables. The χ-square (Mantel–Haenszel) test was used to perform two-group comparisons for qualitative variables, or the Fisher exact test was used when the expected count was <5. A multivariate analysis using logistic regression was performed to identify independent risk factors for NEC. Variables clinically relevant and associated (p < 0.10) with NEC in univariate analysis were used in the model. A p value < 0.05 was considered statistically significant.

Draft genome-based and core-genome phylogeny

For draft genome-based phylogeny, only mapped genomes were included. Scaffolds were concatenated, and aligned using the MAFFT software [18]. For core genome analysis, a dataset comprising seven genomes was generated. Orthologous proteins were obtained from ProteinOrtho software using the following parameters: identity 50%, coverage 60% and E-value 1.10−10. CDSs of core genes were inferred from the pan-genome, then concatenated and aligned using a Python script. The phylogenetic trees were generated using the maximum-likelihood method within PhyML [19] and edited by TreeGraph 2 software [20]. BLAST was used to identify haemolysin (A, B, C), β-haemolysin sequences, dlt operon (dlt A, B, C, D) and the clusters of orthologous groups (COG) identification (E-value 1e−03, coverage 0,7 and identity percentage 30%).

Results

Evaluation of patient features and Clostridium neonatale-specific rpoB-based qPCR system

There was no statistically significant difference between NEC and control groups regarding gestational age, days of life, sex ratio, birth weight and feeding strategies. Only vaginal delivery was significantly more prevalent in controls than in NEC: 37/71 (52%) versus 32/88 (36%), p < 0.05. This statistical analysis improved the rationality of clinical features between NEC and healthy controls. Specificity of primers and probe were validated due to positive amplification for all C. neonatale strains (5/5) and the absence of amplification for other tested bacterial strains isolated from gut microbiota (0/74). This result excluded cross-reaction between the C. neonatale qPCR system and bacterial species inhabiting the gut microbiota. C. neonatale was more frequently detected in stools from preterm neonates with NEC than in controls: respectively 30/88 (34%) versus 9/71 (13%), p 0.003. The presence of C. neonatale in stool samples was significantly associated with the occurrence of NEC. Statistical results are summarized in Table 1. Means of C. neonatale qPCR cycle thresholds (Ct) were statistically non-significant between NEC and controls (26.9 and 25.4 respectively, p 0.5292, unpaired student t test), suggesting that the density of C. neonatale is similar in both cohorts. The prevalence of C. neonatale was also compared with that of C. butyricum reported in Hosny et al. [7]. In brief, the prevalence of C. butyricum, evaluated by both culture and qPCR, was assessed from each cohort and compared with the frequency of detection of C. neonatale (present work). In the majority of NEC cases, C. neonatale was frequently detected in association with C. butyricum when compared to controls: respectively 20/30, (66.7%) versus 1/9 (11.1%), p 0.003 (supplementary material Table S2).

Genomic analysis of Clostridium neonatale

Final draft genomes of C. neonatale isolates sequenced herein consist of three scaffolds. The average length of included genomes was 4 431 047 bp; NEC86 had the longest genome (4 739 641 bp) and NEC26 the shortest (4 282 644 bp). GC content varied between 28.4% (LCDC99A006) and 28.9% (NEC26). Predicted CDSs ranged between 3795 genes (LCDC99A005) and 4188 genes (NEC86) with an average of 3945. General features of these genomes are summarized in Table 2. The draft genome-based phylogeny of C. neonatale isolates allowed the identification of three clusters, revealing clonality among isolates from the same NICU by the following: (a) a strain isolated from NICU-1 (LCDC99A005), (b) a strain isolated from NICU-2 (NEC86) and (c) strains isolated from NICU-3 (NEC25, NEC26, NEC32 and C25) (Fig. 2A). Furthermore, core genome analysis distinguished the same clustering where the strict pangenome consisted of 5150 genes, including 3157 core genes (Fig. 2B). Predicted COG categories were homogenous in all C. neonatale genomes. The COGs encoding carbohydrate metabolism and transport, and therefore general functional prediction (G and R), were over-represented; in contrast, no treatment or modification of RNA, and therefore the structure and dynamics of chromatin, were identified (Fig. 2A and B) (supplementary material Fig. S2 and Table S3).

Fig. 2

Relationship of Clostridium neonatale strains based on genomic analysis. (A) Phylogenetic tree based on core genome. (B) Phylogenetic tree based on whole-genome alignment. NEC, necrotizing enterocolitis; C25, control.

Identification of virulent and unique genes

Comparative analyses identified haemolysin-encoding genes in all strains as follows: haemolysin A (coverage: 100%, identity: 84–92%), two protein sequences encoded for haemolysin B (coverage: 100%, identity: 44–99%) for the first protein and (coverage: 99%, identity: 76–77%) for the second one, haemolysin C (coverage: 96%, identity: 81%) and β-haemolysin (coverage: 100%, identity: 92%). We also identified genes coding for the secretion of C. difficile toxins A and B. Moreover, sequences of the dlt operon (dlt A, B, C, D) were not detected in all draft genomes of C. neonatale (supplementary material Table S4). Unique genes also existed and mostly represented hypothetical proteins, as follows: NEC25 (1/4, 25%), NEC26 (77/113, 68%), NEC32 (28/39, 72%), C25 (2/3, 67%), NEC86 (139/212, 65.5%) and LCDC99A005 (31/59, 52.5%) LCDC99A006 (supplementary material Table S5).

Discussion

Dysbiosis remains a major risk factor inducing the establishment of NEC by promoting the translocation of pathogenic bacteria. Specific microorganisms were involved: the predominance of γ-proteobacteria, generating an excessive inflammatory response, and irregular colonization by strictly anaerobic bacteria, including clostridia [1]. Multidisciplinary approaches have indicated a link between this class—especially C. butyricum, C. neonatale and C. perfringens—and the occurrence of NEC [1,21]. To date, C. butyricum is the only Clostridium species among the cases of clostridia-associated NEC which has been clearly described as correlating with NEC [3,7,22]. However, several studies have reported the involvement of C. neonatale in NEC [1,8]. Using a culture-based strategy, we observed that the prevalence of C. neonatale in stool samples from patients with NEC was not significantly different from that of controls: respectively 4/88 (4.5%) versus 1/71 (1.14%), p 0.26 [7]. In a review of the previous work it was proposed that this discrepancy could be due to the use of a heat-shock-based protocol. If this is supposed to kill all bacteria except spore-formers such as C. neonatale, we could not exclude that the method avoids the isolation of in vivo non-spore-forming Clostridium species. Therefore, we decided to investigate our cohort using C. neonatale-specific rpoB-based qPCR. With a strong phylogeny and taxonomy index, rpoB is a conserved gene whose efficacy for the identification of fastidious microorganisms directly from samples has been previously demonstrated [23,24] and which was used for the heterogeneous detection of BoNT-producing clostridia [25] and C. butyricum directly from stool samples [3]. A significant frequency of C. neonatale in stools from NEC was detected compared to controls (respectively 34% versus 13%, p 0.003). In contrast with the study conducted by Rozé et al., in which the authors reported links between feeding strategies and NEC and the abundance of C. neonatale in NEC patients [9], the feeding strategy was herein not associated with the development of NEC. The only correlation identified was vaginal delivery, a feature already observed in NEC cases [2]. Beside C. neonatale and C. butyricum, several NEC outbreaks have been reported to be associated with other bacterial species [1,2]. A case report described toxin-producing C. perfringens where NEC severity was independent of the α-toxin concentration [26]. Also, nosocomial colonization by Klebsiella pneumoniae type 26 has been reported and antibiotic typing was used to specify strain characteristics [27]. Furthermore, genetically similar clones of Enterobacter sakazakii were distinguished from powdered milk formula and neonates with NEC [28]. A few other bacterial species—such as UPEC and C. paraputrificum—have been suspected of being associated with NEC [4,29].

The use of whole-genome sequencing allowed us to further compare and characterize isolates even if the number of strains and available genomes was limited. This study disclosed genetic similarity between strains isolated from the same NICU. Through phylogenetic analysis, we identified three distinct clusters, highlighted by their geographic areas of isolation, suggesting the same clone of C. neonatale spreading in the same NICU-3 as we previously observed with C. butyricum [7]. This is in agreement with the discovery of this species, as it was first isolated from a Canadian NICU (strains LCDC), and clonality between isolates was proved by pulsed-field gel electrophoresis from stool and blood cultures [8]. Geographic and temporal clustering of NECs were described in several studies with various aetiologies. Temporal clustering was reported by Faustini et al. in diverse NICUs [30]. The similarity between control (C25) and NEC-associated C. neonatale in cluster [3] presupposes the existence of asymptomatic carriage. This same kind of mechanism was described in the case of C. difficile-associated pseudomembranous colitis [31], and we suggested a similar mechanism for C. butyricum in neonates [7]. It should be noted that there appears to be a strong association between C. neonatale and neonates/children as it is a species that has never been isolated from adults in our microbiota studies [32]. On the contrary, C. neonatale has been identified by sequencing of 16S rRNA in the microbiota of young children and has a strong association with developing asthma [33].

If the association between clostridia and NEC is increasingly being reported, the pathogeny of this disease remains elusive in spite of the usual suggestion of toxin production, as in the case of C. difficile infections (CDIs) [33]. Herein, genes encoding the secretion of bacterial toxins were predicted, especially haemolysin and C. difficile toxins A/B (TcdA/B). These latter are a leading cause of CDI, where the pathogenic mechanism is the consequence of TcdA (enterotoxin) and TcdB (cytotoxin) production inducing colonic tissue damage [25,26]. The idea of toxin-mediated disease rather than invasion-mediated disease is supported by the work of Heida et al. [34]. Furthermore, haemolysin sequences with a highly conserved domain found in a toxin of Brachyspira hyodysenteriae showed a cytolytic effect in several cell lines [13]. Cassir et al. identified cytotoxic activity of C. butyricum supernatant on Jurkat cells [4].

In conclusion, this study highlights the association between C. neonatale and NEC and the possible existence of NEC-associated geographic clones. Further genomic analysis is required on a larger number of sequenced genomes.

Conflict of interest

None of the authors have any conflicts of interest to report. This work was supported by the French Government under the Investissements d’Avenir programme managed by the Agence Nationale de la Recherche (ANR) (reference: Méditerranée-Infection 10-IAHU-03), by Région Provence-Alpes-Côte d’Azur and European funding FEDER PRIMI. M. Hosny was supported by Fondation de Coopération Scientifique Méditerranée-Infection (Infectiopôle-Sud 2015).