Patient characteristics and sample collection
The study was done retrospectively on neonates with NEC hospitalised from January 2001 to December 2005. All neonates were hospitalised at a single level III Neonatal Intensive Care Unit (NICU) at Rigshospitalet, Copenhagen, Denmark. All neonates had surgical intervention and samples of removed tissue were formalin-fixed and paraffin-embedded at the Department of Pathology, Rigshospitalet. The study was subjected to ethical review and approved by the Ethical Committee for Copenhagen and Frederiksberg, Denmark (KF 01 268923). Patient's records were reviewed in order to characterise the clinical findings, disease progression and clinical outcome. The data gathered from each infant were: gestational age, weight at birth, onset of symptoms, feeding prior to onset of NEC, antibiotic usage prior to operation of NEC, and outcome.
NEC disease evaluation (NEC-score)
Unfortunately, there is no standard pathological characterization of NEC. We decided to characterize the tissue macroscopic from the characterization made by the pathology that originally looked at the tissue and histologically after haematoxylin and eosin (HE) staining. All histologically samples were independently evaluated by two trained pathologists; at the Department of Pathology, Rigshospitalet and at The National Veterinary Institute, Technical University of Denmark.
Perforation was noted not scored, hemorrhagic mucosa +/- necrotic areas, score 5, pneumatosis intestinal score 5. Amount of tissue <10 cm score 1, 10-30 cm score 2, >30 cm score 3.
The formalin-fixed and paraffin-embedded samples were sectioned 3 μm, mounted on slides and stained with HE. The HE slides were graded as follows: (A) Necroses volving; a) luminal epithelia, b) whole mucosa, c) submucosa, d) tunica muscularis; (B) Vascularity; a) oedema, b) bleeding, c) micro-thrombing, d) haemosiderine, (C) Inflammation; a) unspecific (granulocytes), b) eosinophils, c) vasculitis, d) pseudomembranes, e) granulation tissue, f) granulomas, g) granulomas, h) fibrosis, i) atrophy 1)mucosa 2) all other layers; e) and f) was not included in the score but used to graduate the tissue in acute or chronic NEC. (D) Various, 1) ganglion cells 2) non-ganglion cells. All histopathological characteristics were scored one except (D) that was used to distinguish NEC from Hirschsprung's disease. The NEC-score score is the addition of the macroscopic evaluation and the histology evaluation
Bacterial detection by 16S rRNA in situ Hybridization on Formalin-Fixed Tissue Sections
Paraffin was removed of the tissue sections with xylene and dehydrated in 96% ethanol for 30 min. All specimens were hybridized with both a general bacterial probe EUB338 and with selective probes. Probes were synthesized at Eurofins MWG Operon (Ebersberg, Germany) and described in Table 1. Two probes (S-S-C.paraputri-181 and S-S-C. butyricum-663) were designed in ARB http://www.arb-silva.de in this study. The probes were approved for their specificity to closest bacterial type strains by an in silico probe search in RDP release 10 http://rdp.cme.msu.edu/, and experimental verified for signal intensities and specificity by FISH targeting pure culture of C. butyricum CCUG4217T; C. paraputrificum CCUG32755T; C. difficile ATTC17857 and C. perfringens NCTC8449 injected into a piece of pig lung treated as the rest of the tissue samples.
Hybridization was done in 20 μl of hybridization buffer (100 nM Tris, pH 7.2. 0.9 M NaCl, 0.1% sodium dodecyl sulphate) added 100 ng of probe at 45°C for 16 h in a humidified chamber. Slides were washed in 100 ml of preheated (37°C) hybridization buffer for 15 min and subsequently in 10 ml of preheated (37°C) washing solution (100 mM Tris, pH 7.2, 0.9 M NaCl) for 15 min. Slides were rinsed in water and air-dried. All slides were scored as follows: 0) no or low density of bacteria, 1) moderate density of bacteria, 2) high density of bacteria.
NEC tissues used for Laser Capture Micro dissection
Eight intestinal tissue samples were included. The microdissection was performed on tissues excised from 4 neonates that were treated with antibiotics less than 2 days and from 4 neonates treated with antibiotics 10 days or more before surgery. Three μm sections of the tissues were cut (knife was changed between cuts) and mounted on the 0.17-mm PALM® POL-membrane slides (P.A.L.M. Microlaser Technologies AG, Bernried, Germany) and kept at 4°C until use. The slides were hybridized with bacterial probes as previously described.
Laser Capture Microdissection
A PALM Robot-Microbeam system (P.A.L.M. Microlaser Technologies AG) consisting of an Axivert 200 M microscope (Carl Zeiss, Oberkochen, Germany) equipped for fluorescence with a 100-W Hg lamp, a 40x/1.30 oil Fluar objective (Carl Zeiss), filter set XF53 (Omega Optical, Brattleboro, VT, USA) and the PALM RoboSoftware version 1.2 (P.A.L.M Microlaser Technologies AG) was used. Bacteria were visualized by FISH using the general bacterial probe EUB338 and dissected from both the intestinal lumen and mucus of the surgical tissue by the cutting and catapulting function, RoboLPC as previous described . The micro-dissected area from the lumen and mucus associated tissues were never in contact with any external contaminators because the micro-dissected area is cut by a laser and "transported" to the tube by a photonic force and against gravity as described by Carl Zeiss AG, Deutschland http://www.zeiss.de/. The risk for external contaminators is therefore minimal.
The catapulting material was collected in the cap of a 200 μl Thermo-Tube (ABgene, Epsom, UK) containing 20 μl proteinase K buffer. The microdissected material was digested in proteinase K buffer (10 mM Tris-HCl, pH 8.0, 150 mM NaCl, 10 mM EDTA, 0.1% sodium dodecyl sulphate, 1 U proteinase K) at 55°C for 72 h. Subsequently, the proteinase K was inactivated at 95°C for 15 min. Two μl of solution were subsequently used as template for the polymerase chain reaction (PCR).
Clone library and sequencing of intestinal bacteria
The primers Bact64f and Bact109r1 (Eurofins MWG Operon ) were used for 16S rRNA gene amplification of the hyper variable region V1 from the small subunit ribosomal RNA gene (Table 1). PCRs (always including a non template control) were done in 20 μl volumes containing 1 × PCR buffer [20 mM Tris-HCl (pH 8.4) and 50 mM KCl], 200 μM dNTP, 500 nM each primer, 3.3 mM MgCl, and 1 U of Pfu DNA polymerase (Invitrogen Corporation, Carlsbad, CA), which creates blunt end fragments. The thermal profiles were as follows: an initial denaturation step at 94°C for 3 min; 30 cycles of 94°C for 30 s, 50°C for 30 s, and 72°C for 30 s; and a final elongation step at 72°C for 5 min. The amplicons were purified by using phenol/chloroform (P/C, pH 8.0), and the DNA was precipitated with 2.5 M ammonium acetate in ethanol. After two washes with 80% (v/v) ethanol, the DNA pellet was dried and resuspended in 10 μl, 0.2 μl filtrated, double-distilled water. Following the manufacturer's descriptions the cloning was done by using a Zero blunt TOPO cloning kit (Invitrogen Corporation). Fifty to hundred colonies from each cloning were picked and sequenced by pyrosequencing. A PYROMark Q96 ID was used to short DNA sequencing of the approximately 40-60 bp clone insert using the recommended protocol (Biotage AB, Uppsala, Sweden) as described previously using the primer PyroBact64f . The sequences (tags) were imported into the software BioNumerics 4.61 and manually checked, aligned and filtered for high quality sequences. Sanger sequencing with an Applied Biosystem 3130 Genetic Analyzer (Foster City, CA, USA) was used to check consensus tags for the pyrosequencing accuracy. The Sequence match analysis tool in the Ribosomal database project 10 http://rdp.cme.msu.edu/ was used to assign the Phylogenetic position of each consensus tag. The search criteria were for both type and non-type strains, both environmental (uncultured) sequences and isolates, near-full-length sequences (>1200 bases) of good quality. If there was a consensus at the genus level the tag was assigned this taxonomic classification. If no such consensus was found, the classification proceeded up one level to family and again if no taxonomic affiliation could be assigned the tag continued to be proceeded up the tree as described by Huse et al., . In some cases it was not possible to assign a domain and these sequences might represent new novel organisms or the sequences might be biased, in these cases the tags were excluded from the dataset. In total 364 sequences were finally included in the alignment.
The phylogenetic analysis was done by downloading 16S rRNA gene sequences longer than 1,200 base pair from the RDP database of the Ralstonia type strains http://rdp.cme.msu.edu. The RDP alignment was used and a phylogenetic tree was constructed by using the Ward algorithm in the software Bionumerics. Burkholderia cepacia (GenBank accession no. AF097530) was used as an out-group.
The statistical analysis was done in two steps: First, the association between one predictor at a time and the NEC score was analysed by robust least squares methodology adjusting for gestational age. This is equivalent to a normal linear GEE modal with working independence correlation structure on child level. For each predictor the estimated change in expected NEC score is reported with Wald 95% confidence limits in parentheses. The overall association between the predictor and the NEC score is evaluated by a robust score-test.
Second, we formulate a normal linear GEE model including gestational age and all predictors with a robust score-test p-value below 0.1 in the above analyse. This multivariable model is then reduced by backwards elimination using 0.05 as a cut-off level.
All analyses were performed using PROC GENMOD in SAS version 9.1 (SAS Institute, Cary, NC).