Background
Necrotizing enterocolitis (NEC) is an acute devastating intestinal inflammatory disease that mainly occurs in preterm infants shortly after birth [
1]. Epidemiological studies show that 45% of NEC survivors were neurologically impaired at 20 months of age with a higher risk of cerebral palsy, hearing, visual, cognitive, and psychomotor impairments [
2,
3]. Severe NEC requiring surgery is an independent risk factor for severe brain injury detected on MRI, poor mental and psychomotor development at around 2 years of age [
4,
5], and various cognitive deficits at school age [
6]. It is conceivable that NEC may cause acute damage in the developing brain and subsequent lasting neurodevelopmental disorders. Yet, the underlying mechanisms and NEC-associated acute brain effects are largely unknown.
NEC in preterm infants is characterized by excessive gut inflammation, potential pathogen leakage, and systemic inflammation [
1]. Neonatal inflammation may adversely affect the processes of maturation of neuronal and immune systems in a critical period of development [
7,
8]. Studies have shown that early life exposure to lipopolysaccharide (LPS) leads to reduced hippocampal volume, dysregulated neurogenesis, increased number of microglia cells, axonal injury, and memory impairment in rats [
9‐
11]. Similarly, chronic intestinal inflammation reduces hippocampal neurogenesis in mice [
12] and neonatal viral infection induces hippocampal neuroinflammation and learning impairments in piglets [
13]. This may be mediated by inflammatory cytokines which are normally expressed in the developing brain and regulate processes of neurogenesis, neuronal migration, synaptogenesis, and synaptic plasticity [
14‐
16]. A dual role of proinflammatory cytokines in the developing brain may relate to activation of the ubiquitous NFκB signaling pathway, which in neurons is responsible for plasticity and survival, whereas in glial cells, it is involved in mediation of pro-inflammatory responses [
17]. Such neurodevelopmental and neurodegenerative abnormalities may also induce changes in the composition of cerebrospinal fluid (CSF) [
16,
18,
19], but few studies have investigated the links among brain damage, CSF, and NEC in preterm neonates [
20].
Preterm pigs delivered at 90% gestation show impaired gut, immune, and brain development with a high sensitivity to neonatal infections (i.e., NEC and sepsis) and behavioral and learning deficits [
21‐
23]. Thus, a large proportion of preterm pigs develop NEC spontaneously within the first week after birth when fed sub-optimal diets (e.g., infant formula or human donor milk) [
24]. In contrast to rodent models, this model does not require excessive hypothermia and hypoxia treatments [
21]. Using preterm pigs as a model for preterm infants, we hypothesized that NEC lesions would induce immediate changes to the developing hippocampus that might help to explain the later neurodevelopmental deficits in NEC survivors. We show that specifically NEC lesions located to the small intestine are associated with reduced physical activity and upregulation of inflammation-related genes in the hippocampus. Exposure of hippocampal neurons to CSF from pigs with NEC promoted neurite outgrowth in vitro, maybe via NEC-related factors in CSF, such as VEGF, CINC-3, and S100A9 proteins. Thus, our results support the hypothesis that NEC lesions lead to immediate effects on the developing brain in preterm infants.
Methods
Spontaneous NEC model in preterm pigs
One hundred and seventeen preterm piglets were delivered from eight sows by cesarean section at day 106 (90% of gestation, Danish landrace x Large White x Duroc, Askelygaard farm, Denmark). Pigs were housed in individual incubators with regulated temperature (37–38 °C) and oxygen supply (0.5–2.1/min, within the first 24 h). Pigs were inserted with an orogastric feeding tube and an umbilical catheter for parental nutrition and sow plasma infusion, as described previously [
25]. To induce spontaneous NEC [
21], preterm pigs were fed with gradually increasing doses of different types of human donor milk (0–135 mL/kg/day) and gradually decreasing doses of parenteral nutrition (96–48 mL/kg/day) for 8 days, as described previously [
24,
25].
Home cage activity and neonatal arousal recordings
During the study period, the physical activity for all piglets was recorded using infrared video cameras connected to a motion detection recorder. The proportion of active time was analyzed with PIGLWin application software (Ellegaard System, Faaborg, Denmark), as described previously [
26]. The neonatal arousal of each piglet was registered as the time from birth to the first opening of eyes, first standing, and first walking, as described previously [
26].
Tissue collection, NEC evaluation, and gut cytokine expression
Pigs were anesthetized, and blood was drawn by cardiac puncture, followed by euthanasia by an intracardiac injection of sodium pentobarbital (60 mg/kg). Heparinized plasma fractions were collected and stored at − 80 °C. The CSF samples were collected by sub-occipital puncture immediately after euthanasia, aliquoted, and stored at − 80 °C. Following determination of brain wet weight, the brain was quickly dissected and the left hippocampal formations were snap-frozen in liquid nitrogen and stored at − 80 °C until further processing. The right hemisphere was fixed in 4% paraformaldehyde. Brain dry weight and water content were determined after drying the remaining brain tissues to a constant weight.
The whole gastrointestinal tract (GIT) was excised, and the pathological lesions in the proximal, middle, and distal small intestine, and in the colon, were scored macroscopically by two independent investigators, using an established NEC severity scoring system: score 1: absence of macroscopic lesions; score 2: local hyperemia; score 3: hyperemia, mild hemorrhage, extensive edema; score 4: extensive hemorrhage; score 5: local necrosis and pneumatosis intestinalis; score 6: extensive transmural necrosis and pneumatosis intestinalis (Additional file
1: Figure S1 [
24]). Pigs with score 1 in all regions were diagnosed as without having NEC (No NEC,
n = 51). Pigs with a severity score ≥ 4 in the small intestine (with or without colon lesions) were diagnosed as severe small intestinal NEC (Si-NEC,
n = 13). Pigs with a score ≥ 4 only in the colon were diagnosed as severe colon NEC (Co-NEC,
n = 34). Detailed disease characteristics, group information, and number of animals included in further analyses are shown in Additional file
2: Table S1 and in the corresponding figure legends. Nineteen pigs scored with 3 in minimum one region across the small intestine and the colon showed borderline pathological lesions that may or may not reflect initial NEC lesions. To minimize the effects of such diagnostic uncertainty (which is common also for infants undergoing surgery for NEC), we excluded all these animals from further analyses, resulting in groups of pigs only with clear diagnosis as NEC or no NEC by macroscopic inspection. In preterm pigs, mild clinical symptoms related to NEC are most often located in the colon region while more severe NEC symptoms usually also involve the small intestine (with or without lesions in the colon) [
26‐
28]. Expression of proinflammatory cytokines including IL-1β, IL-6, and IL-8 in the distal intestine and the colon were measured with porcine DuoSet ELISA kits (R&D Systems, Minneapolis, MN, USA), according to the manufacturer’s protocol and presented per milligram of total protein (Pierce BCA Protein Assay Kit, Thermo Fisher Scientific, Rockford, IL, USA).
Plasma and CSF sample analysis
Spectrophotometric measurement of the oxyhemoglobin content of the CSF samples was performed at A450 nm to exclude samples containing blood contamination [
29]. The total protein concentration in the CSF and plasma samples was measured with the Pierce BCA Protein Assay Kit. Plasma and CSF albumin, lactate, and glucose were measured with a GEM premier 3000 whole blood analyzer (Instrumentation Laboratory, Bedford, MA). Plasma C-reactive protein (CRP) concentration was measured by ELISA (DY2648, R&D systems, Minneapolis, MN, USA).
Neurite outgrowth assay
Hippocampal neurons were isolated from Wistar rats (E18; Charles River, Sulzfeld, Germany), plated on eight-well Permanox Lab-Tek chamber slides (Nunc, Roskilde, Denmark) at a density of 10,000 cells/well as previously described [
30] and immediately stimulated with CSF samples. Optimization experiments with serially diluted CSF samples showed that a final concentration of CSF in the culture medium of 1% was the most optimal, which was in accordance with previous work [
31]. For each slide, unstimulated cells and cells treated with 3 μg/ml of the neurotrophic peptide Epobis were used as negative and positive controls, respectively [
30]. In independent experiments, hippocampal neurons were stimulated with recombinant human VEGF, S100A8, S100A9, hetero-complex of S100A8/A9, and rrCINC-3 (CXCL2) (all from R&D Systems, Minneapolis, USA; rhS100A9 and rhS100A8 were kind gifts from Dr. J. Klingelhöfer, University of Copenhagen). Twenty-four hours after the stimulation, neurons were fixed in 4%
v/
v formaldehyde and stained with polyclonal rabbit growth-associated protein (GAP)-43 antibodies (1:1000; Millipore), followed by secondary Alexa Fluor 488- or 546-conjugated goat anti-rabbit antibodies (1:1000; Molecular Probes). Images were acquired using a Zeiss Axiovert 100 microscope connected with AxioCamMRm camera using the ZEN 2012 software. The quantification of neurite outgrowth and number of neurites per cell were performed as described previously [
30].
Hippocampal RNA-seq analyses
Intact frozen hippocampi (n = 5–6 per group) were homogenized by a cryogenic tissue pulverizer in liquid nitrogen, and total RNA was isolated with RNeasy Lipid Tissue Mini Kit (Qiagen, Copenhagen, Denmark). The integrity of RNA samples for RNA-seq and qPCR analyses was evaluated with Agilent Bioanalyzer 2100 and RNA 6000 Nano Chips (Agilent Technologies, Glostrup, Denmark) and resulted in an average RNA integrity number (RIN) of 8.5 (SD ± 0.6). Sequencing libraries were constructed using NEBNext UltraTM RNA library Prep Kit for Illumina (New England BioLabs, Ipswich, MA, USA), following the manufacturer’s recommendations. After amplification, products were purified with the AMPure XP system and library quality was assessed on the Agilent Bioanalyzer 2100 system. The clustering of the index-coded samples was performed on a cBot Cluster Generation System using a HiSeq 4000 PE Cluster Kit (Illumina, San Diego, CA, USA). After cluster generation, constructed cDNA libraries were sequenced on an Illumina Hiseq 4000 platform (Illumina) and 150-bp paired-end raw reads were generated.
Raw reads were trimmed to produce clean reads, including removal of the adapter sequence, and low-quality reads containing either more than 50% bases with
q value < 10 or ≥ 10% N bases, as detected by the FASTX tool kit (v 0.0.13,
http://hannonlab.cshl.edu/fastx_toolkit). Following this cleaning, 273,958,970 clean and paired reads were generated in total. All clean reads were aligned to the Sscrofa 10.2 genome and gene model annotation file (
www.ensembl.org/Sus_scrofa/Info/Index) using Tophat (v2.1.1)-Cufflinks (v2.2.1) pipeline [
32]. The expression level of each gene was estimated using Fragments Per Kilobases per Million reads (FPKM). According to Cuffdiff, the genes with a statistical
q value < 0.2 were considered as differentially expressed genes (DEGs) [
32]. Biological process enrichment was analyzed using the Cytoscape plug-in ClueGO [
33], and enrichment tests with adjusted
p values < 0.05 were considered significant.
Validation of gene transcription by qPCR analyses
Expression of DEGs of interest and other related genes were further measured by microfluidic qPCR analyses. To ensure valid data, for each RNA sample, two separate technical cDNA replicates were synthesized and a non-reverse transcriptase control was included. Pre-amplification of each cDNA was carried out using TaqMan PreAmpMasterMix (Applied Biosystems, Foster City, CA, USA) followed by exonuclease treatment (Exonuclease 1, New England biolabs, PN MO293L), as described previously [
34]. Porcine-specific primers were designed whenever possible over introns (Primer3:
http://frodo.wi.mit.edu; Sigma-Aldrich, Broendby, Denmark). Gene symbol, primer sequences, and amplicon lengths are shown in Additional file
3: Table S6. The amplification efficiencies of all primers were between 85 and 115%. Quantitative PCR of pre-amplified cDNA samples, including non-reverse and non-template controls, was performed using 96.96 Dynamic Array Integrated Fluidic Circuits on a BioMark thermocycler (Fluidigm, CA, USA). The cycling conditions were 2 min at 50 °C, 30 min at 80 °C for thermal mix, then 2 min at 50 °C and 10 min at 95 °C, followed by 35 cycles of 15 s at 95 °C and 1 min at 60 °C for the signal detection. Melting curves were generated after each run (from 60 to 95 °C, increasing 1 °C/3 s). Acquired Cq values were uploaded to the online PCR analysis tool (
http://dataanalysis.sabiosciences.com/pcr/arrayanalysis.php) and analyzed as previously described [
35]. Using GenEx, the expression levels of target genes were normalized to the three reference genes including
GAPDH,
RPL13A, and
ACTB.
Stereology
Entire hippocampi were dissected from fixed hemispheres, embedded in paraffin using a Leica ASP300 S tissue processor, and sectioned with a Jung HN40 sliding microtome for 5-μm exhaustive sagittal sections. Sections were sampled using uniform random sampling [
36], by which every 70th section pair was collected herein yielding 8–12 section pairs per hippocampus. For immunohistochemistry, the sampled sections were dewaxed and hydrated followed by incubation in 3% H
2O
2 to block endogenous peroxidase activity. Antigens were retrieved by boiling the sections for 15 min in 10 mmol/l citrate buffer (pH 6). Sections were stained with anti-Iba1 antibodies (1:1000; ab5076, Abcam, Denmark) followed by HRP-conjugated secondary antibody (1:500; Polyclonal Rabbit Anti-Goat/HRP, P0449, Dako, Denmark). The reaction was developed using 3,3′-Diaminobenzidine (Sigma-Aldrich, Denmark), then sections were counter-stained with Mayer’s hematoxylin, mounted with Pertex, and cover slipped (Leica Microsystems, Ballerup, Denmark).
The total number (
N) of microglia was estimated using the physical disector-design [
37] for three hippocampi and counting on series of single-sections. The numerical density (
NV) was estimated by dividing the total number of particles (∑
Q) by the volume in which they were counted, e.g., area of the frame (
a(frame)), height of the section (
h), and number of disectors (∑
P) (Eq. 1:
N
V
= ∑Q/(
a(frame)
*h*∑P)
. To estimate the total number (
N) of microglia in the region, the numerical density (
N
V
) was multiplied by the volume of the region of interest (ROI), i.e., the reference volume (
Vref) (Eq. 2:
N=N
V
*Vref). The
Vref was estimated by Cavalieri’s method, where the number of points hitting the ROI (∑
p) was multiplied by the area per point (
a(point)) and the block advance (BA) (Eq. 3:
Vref = a(point)
*BA
*∑p). As the estimated bias introduced by counting cells in only one of the sections in a section pair was 3.1%, the number of microglia (
N) on a series of single sections was calculated as described above in Eqs. 1–3. From the systematic uniform random sampled images, the ratio of amoeboid to total microglia was estimated using following morphological criteria: cells with round or amoeboid shapes, with no processes, were classified as amoeboid microglia, which were distinguished from cells with a small or large cell body with visible thin or thick ramifications (ramified or primed microglia, respectively [
38]). Cell counting was performed using the NewCast software (Visiopharm, Hoersholm, Denmark) and a Nikon Eclipse 60i microscope (Olympus, Ballerup, Denmark) equipped with a Heidenhain electronic microcator measuring the
z-axis and a ProScan II motorized stage system (Prior Scientific Instruments., Cambridge, UK). Digital live microscope images were visualized by a high-resolution camera (Olympus DP72, Nikon Nordic AB, Copenhagen, Denmark).
Statistics
Data analyses were performed using the software package R (version 3.3.2). Continuous outcomes (e.g., brain weight, pro-inflammatory cytokine levels) were analyzed using the
lm function, and repeated measurements (i.e., physical activity) were analyzed using the
lmer function. The normality and variance homogeneity of the residuals and fitted values were tested, and data transformation was performed if necessary. To determine any potential sex bias on the processes of brain development and neuroimmune responses, all above models were adjusted for potential covariates and confounders (i.e., birth weight, litter, and sex). Data was subsequently treated by Dunnett’s post hoc test with the No NEC group as the reference group using
glht function. Neuritogenesis was analyzed by one-way ANOVA. Data were autoscaled and applied to principal components analysis (PCA) using the R package “pcaMethods” [
39]. Data were presented as means ± SEMs, unless otherwise stated, and
p < 0.05 was considered significant, with
p < 0.15 indicated as a tendency to an effect.
Discussion
In contrast to term infants, a preterm infant has an immature gastrointestinal tract, brain, and immune system and such infants have high risks of both NEC and brain damage in early life. Severe NEC is known to be strongly associated with abnormalities in white and gray matters and poor neurodevelopmental outcomes which develop later in life [
7,
59,
60]. For obvious reasons, the brain samples from neonates are inaccessible; thus, it remains unclear whether NEC lesions are associated with immediate adverse responses in the developing brain. Understanding the early NEC-related events in the brain is important to identify new protective strategies for preterm infants during the immediate neonatal period. Taking advantage of a clinically-relevant preterm pig NEC model, we now document that acute NEC located in the small intestine is associated with decreased physical activity and changes to hippocampal gene expression. NEC lesions that include the small intestine (with or without colon lesions) were associated with more profound responses in the developing hippocampus, relative to NEC lesions present exclusively in the colon region. This observation is consistent with the experience from human infants in which NEC diagnosis requiring subsequent tissue resection usually involves the presence of severe NEC lesions in the small intestine [
27].
Exposure of rodent hippocampal neurons to CSF samples from NEC pigs promoted neuronal differentiation in vitro. The differentiation of neuronal precursor cells towards polarized neurons occurs via well-defined morphological steps and includes neurite outgrowth followed by neurite differentiation into dendrites and axons, and synapse formation. This differentiation is orchestrated by a number of factors, including immune-related molecules [
61,
62], whose basal expression is much higher in the developing brain, compared with the mature brain [
14,
63,
64]. In our study, the observed increase in neurite outgrowth and branching of neurites induced by CSF from NEC-affected animals may therefore be a consequence of elevated levels of pro-inflammatory cytokines (e.g.,VEGF [
45], MMP-8 [
43], CINC-3 [
65] (called CXCL2 in humans) and INF-γ [
42]). The promotion of neurite outgrowth from hippocampal neurons was confirmed specifically for VEGF and CINC-3, which further supports the hypothesis that at least some inflammation-induced factors may affect neuronal plasticity. Consistent with this, two hippocampal inflammation-related genes, coding for S100A8 and S100A9, were upregulated in Si-NEC pigs. These proteins form homodimers (Calgranulins) and a heterocomplex (Calprotectin) and are shown to interact and regulate the biological activities of receptors for advanced glycation end products (RAGE) and TLR4 (gram-negative bacterial liposaccharide receptor). Although S100A8 and S100A9 were upregulated within the CNS in response to different pathological conditions [
66,
67], their role in brain development is poorly understood. We now first show that the extra-cellular application of S100A9 or S100A8/A9 (but not S100A8 alone) dose-dependently promotes neurite outgrowth from primary hippocampal neurons, further pointing to the stimulatory effect of inflammatory factors on neuronal plasticity. In addition to S100A8/A9, the neuritogenic and neuroprotective effects may also be induced by
ADAM8 [
68] which was upregulated specifically in the Si-NEC group. Likewise, hippocampal upregulation of transferrin (
TF), particularly in the Co-NEC group, confirms previous results on neurite outgrowth in vitro [
69]. Apart from affecting neural maturation, pro-inflammatory cytokines are also well known to interfere with oligodendrocyte maturation and myelination processes [
47]. In line with this notion, Si-NEC preterm pigs showed downregulation of
OPALIN, a marker of mature oligodendrocytes [
70]. Based on our results, we suggest that inflammation-induced factors may have effects on immature neurons to enhance their maturation and confer neuronal plasticity which might lead to redirection of the neural network formation and consequently to neurodevelopmental disorders. This is consistent with the notion that neonatal inflammation is closely associated with neurological disorders such as schizophrenia, autism spectrum disorder, and Rett syndrome [
64,
71]. We speculate that peripheral inflammation in the gut may lead to dysregulation of neuronal differentiation and formation of synaptic networks and have long-term neurodevelopmental consequences.
It is widely accepted that inflammation and oxidative stress responses often occur simultaneously, but it is unclear if hypoxia-activated genes have consecutive, correlative, or causal roles. The Si-NEC DEGs were highly enriched with upregulated genes related to hypoxia (see the “
Results” section) and downregulated transcripts encoding
HBB and its synthetase
ALAS2. Hemoglobin beta (HBB) is known to be expressed in neurons [
72], hippocampal astrocytes, and mature oligodendrocytes [
73] and is neuroprotective against oxidative and nitrosative stresses [
74]. HBB may also support neuronal metabolism via epigenetic control of histones [
75] and neuronal mitochondria functions [
72]. Thus, Si-NEC-related downregulation of
HBB may therefore uncover neurons for potential oxidative and NO stress and inhibit hippocampal mitochondrial functions. Together, inflammation- and hypoxia-related events appear responsible for effects of Si-NEC on the developing hippocampus.
We observed an increase in the amoeboid microglia population in the hippocampus of NEC pigs, which constituted about 10–20% of the total microglial cells. However, we did not observe difference between groups in the total number of microglia. The relative low abundance of activated microglia may be explained by the upregulation of
ADM, as adrenomedullin downregulates LPS-induced microglia activation and decreases the production of pro-inflammatory cytokines in vitro [
76]. Further, upregulated expression of
S100A8 and
S100A9, and of their RAGE receptor, may play a role, as soluble RAGE can function as a decoy receptor attenuating the pro-inflammatory effects of S100 proteins. Finally, NEC-related upregulation of
USP18 (Fig.
3a, Additional file
3: Table S3), encoding ubiquitin-specific protease 18, a negative regulator of microglia activation [
77], therefore may counteract effect of NEC on microglia activation.
Overall, we observed more profound changes in hippocampal gene expression in Si-NEC pigs than in Co-NEC pigs. These changes suggest potential neuroinflammation, hypoxia, and oxidative distress, which are more evident in the pigs with NEC lesions in the small intestine than in the colon. This may be because the small intestine is more vulnerable to insults in early life than the colon [
78,
79], and small intestinal NEC is more often linked to increased disease severity and mortality, less nutrient absorption, and high risk of bacterial translocation [
28]. Some of the pigs in Si-NEC group also had NEC lesions in the colon, yet the hippocampal gene expression profile of these pigs were not different from the pigs only having NEC in the small intestine (Additional file
1: Figure S4). Apart from neonatal inflammation, NEC-associated malnutrition and poor growth are also risk factors for brain injuries and neurodevelopmental impairments in preterm infants [
80‐
82]. However, nutritional factors may not play a critical role in the present study considering the continuous parenteral nutrition supply for all animals and body weight was not different among the groups.