ABSTRACT
Inflammation following neonatal infection is a dominant cause of childhood hydrocephalus in the developing world. Understanding this complex inflammatory response is critical for the development of preventive therapies. In 100 African hydrocephalic infants ≤3 months of age, with and without a history of infection, we elucidated the biological pathways that account for this inflammatory response. We integrated proteomics and RNA sequencing in cerebrospinal fluid, identifying gene pathways involving neutrophil, interleukin (4, 12, and 13) and interferon activity associated with this condition. These findings are required to develop strategies to reduce the risk of hydrocephalus during treatment of infection.
INTRODUCTION
Hydrocephalus is a serious brain disorder in children1,2 and the most common indication for pediatric neurosurgery worldwide3. The most frequent global antecedent of hydrocephalus is infection such as neonatal or infant sepsis4. Postinfectious hydrocephalus (PIH) accounts for up to 60%5 of the nearly 400,000 infants who develop hydrocephalus each year, principally in low- and middle-income countries2. Despite recent clinical efforts to optimize treatment6, PIH remains a leading cause of neurological morbidity and mortality worldwide7,8. Strategies to prevent PIH have been thwarted for two principal reasons. First, the key pathogens responsible for the underlying infectious episodes that precede PIH are often not identified, and thus treatment of the underlying infections is not optimized9. Second, we have not identified the specific underlying inflammatory milieu that leads to hydrocephalus following infection10,11. Unfortunately, even children who undergo surgical treatment for hydrocephalus early in life suffer very poor long-term outcomes7,12. Therefore, major advances in the health of these children will require preventing infection by targeting both the underlying pathogens and their routes of infection13-15, as well as treating infections with adjunctive therapies that can reduce the likelihood of subsequently developing hydrocephalus.
A recent pan-microbial approach uncovered Paenibacillus spp. as a novel PIH pathogen associated with severe ventriculitis in a cohort of Ugandan children16. We hypothesize that the PIH host response is mediated by the activation of specific gene pathways, and that uncovering these pathways may identify therapeutic targets in hydrocephalus prevention.
To test this hypothesis, we performed high throughput proteomics and RNA sequencing (RNA-Seq) of cerebrospinal fluid (CSF) from 64 PIH infants and 36 non-postinfectious hydrocephalus (NPIH) infants to examine the immunopathogenesis of PIH 16-18. Integrating deep-scale proteomics (i.e., high gene coverage) with genomic and/or transcriptomic data, known as proteogenomic analysis19,20, permits the exploration of gene networks and pathways underlying disease. To our knowledge, this is the first proteogenomics study analyzing the interactions between PIH-associated proteome and transcriptome to better understand gene regulatory networks involved in PIH (Figure 1).
RESULTS
Patient characteristics
A total of 100 patients (64 PIH, 36 NPIH), all 3 months of age or less and with weight greater than 2.5 kg, were recruited. PIH infants had either a history of febrile illness and/or seizures preceding the onset of hydrocephalus, or alternative findings such as imaging and endoscopic results indicative of prior ventriculitis (septations, loculations, or deposits of debris within the ventricular system). None of the PIH infants had a history of hydrocephalus at birth. NPIH patients had findings of hydrocephalus that was of non-infectious origin on computed tomography (CT) scan or at endoscopy, including a structural cause (obstruction of the Aqueduct of Sylvius, Dandy-Walker cyst, or other congenital malformation) or evidence of hemorrhage (bloody CSF). Prior to CSF acquisition, none of the 100 patients had undergone surgery on the nervous system (reservoirs, shunt, third ventriculostomy, or myelomeningocele closure), or had evidence of communication between the nervous system and skin such as meningocele, encephalocele, dermal sinus tract, or fistula.
Sterile CSF samples acquired directly from the cerebral ventricles of all patients during surgical treatment of hydrocephalus (shunt insertion or endoscopic treatment) in Uganda were placed into cryotubes with DNA/RNA preservative (DNA RNA Shield, Zymo Corp). Samples were then frozen in either liquid nitrogen or at -80°C and shipped cryogenically to the USA for further processing. Previous 16S rRNA sequencing had recovered Paenibacillus spp. from 59% of the PIH infants in this study, of which 8 and 27 patients also had Human Herpesvirus 5 (cytomegalovirus, CMV) in the CSF or blood16, respectively. These previous findings directed stratification of our patients into “Paeni-positive” or “Paeni-negative,” “CMV-CSF-positive” or “CMV-CSF-negative,” and “CMV-serum-positive” or “CMV-serum-negative.” Comparison of demographic and clinical attributes between Paeni-positive and Paeni-negative patients showed that Paeni-positive patients had higher white blood cell (WBC) counts in peripheral blood (11.9 vs 9.3 × 103/µL, p=0.002) and CSF (71.6 vs 5.0 × 103/µL, p<.001), and lower blood hemoglobin levels (10.8g/dL vs 12.0 g/dL, p=0.039) than those of Paeni-negative patients. There were no significant differences in age or gender between the groups (Table 1).
Protein expression
Tandem mass spectrometry (MS/MS) spectra obtained from a standard ten-channel peptide mass tagging (TMT-10) system and compared with the current UniRef® database yielded quantitative proteomic data17,21-23. Median values of peptide intensities were assigned to unique proteins and were used to infer relative protein abundances. Peptide identifications that could be assigned to more than one protein were excluded from protein quantification. Batch effects were corrected, and intensity values were quantile normalized. Multi-dimensional scaling (MDS) analysis demonstrated the clustering of PIH and NPIH infants and distinguished Paeni-positive from Paeni-negative patients (Figure 2A). Of 616 proteins identified, 292 were differentially expressed based on Paenibacillus spp. status: 144 and 148 proteins were upregulated or downregulated in the Paeni-positive group, respectively (Figure 2B). Gene set enrichment analysis of the differentially expressed proteins identified the predominant enriched functions were involved in neuroinflammation, particularly neutrophil-mediated inflammation, and negative regulation of proteolysis and peptidase activity, and modulation of extracellular matrix and structure (Figure 2C).
RNA Expression
Gene read counts (25 million reads per sample) were aggregated from expression levels estimated from paired-end RNA-Seq data mapped to the human reference genome hg38 with STAR, and quantified with RESM24,25. Expression of at least 1 count per million in a minimum of 18 samples was required for inclusion in the analysis. MDS analysis demonstrated the clustering of PIH and NPIH infants (Figure 3A). Of 11,114 genes, 2,161 were differentially expressed, and the hierarchical clustering of differentially expressed genes distinguished Paeni-positive from Paeni-negative patients (Figure 3B). Gene ontology analysis of the differentially expressed genes based on Paenibacillus spp. status demonstrated enrichment for genes predominantly related to response to bacteria, host immune regulation, and cell motility, migration, and adhesion (Figure 3C).
Role of CMV co-infection
Adjusting for CMV status at the transcriptome level did not significantly alter the results of the gene ontology enrichment analysis based on Paenibacillus spp. positivity. Sixty-four genes were differentially expressed based on CMV status, and those genes were enriched for functions related to host response to virus (Figure 4A). Principal component analysis (PCA) using RNA abundance of all genes was not able to cluster samples by CMV status (Figure 4B). Adjusting for CMV status on proteomic data did not change the differentially abundant proteins (Figure 2C).
Correlation of clinical traits with gene expression
Network correlation analysis was employed to identify sets of genes that share a pattern of expression among patient RNA data. First, a subset of patients was identified using hierarchical clustering based on expression of genes that were differentially expressed between Paeni-positive and Paeni-negative patients. Then, using Weighted Correlation Network Analysis (WGCNA), 2 gene sets (Modules 1-2) were identified as having similar patterns of expression within this subset of 33 patients. These modules were assessed for network correlation with 4 clinical variables: CSF and peripheral blood WBC counts, Paenibacillus spp. status, and CSF CMV status (Figure 5A). Module 1 contained all but 27 of the 2,205 genes differentially expressed with respect to Paenibacillus spp. status. This module correlated positively with CSF cell count and negatively with Paenibacillus spp. status. Genes in Module 1 were enriched for functions related to host immune response (Figure 5B). Module 2 was positively correlated with Paenibacillus spp. status, but the 27 genes included in this module were not enriched for any gene ontologies, indicating no specific functional enrichment for the small number of genes within this module.
Marker genes of bulk RNA sequencing
Patient bulk RNA expression was analyzed for immune cell-type signatures using reference single-cell RNA sequencing data (Figure 5C). The R package xCell identifies cell type proportions within bulk RNA sequencing data based on enrichment of marker gene expression26. Upon hierarchical clustering, two groups were evident: a mixture of PIH and NPIH samples with a predominance of hematologic cells and PIH-only samples with T-helper and natural killer (NK) cell populations.
Proteogenomic integration
Integration of proteomics and transcriptomics data was performed following dimension reduction and feature selection of differentially expressed proteins and genes. Unsupervised gene ontology analyses of the concatenated data recapitulated enrichment for functions related to the immune system, metabolism, response to oxidative stress, cell-cell junction interactions, and extracellular matrix structure in association with peptidase activity (Figure 6A).
Panel B is a Box and Whisker plot of the 33 genes that were differentially expressed based on Paenibacillus spp. status. Counts (ordinate) of each gene (abscissa) are shown for each group, with blue representing NPIH, red for Paeni-negative PIH and green for Paeni-positive PIH infants.
There were 33 genes detected by both proteomics and RNA-Seq as significant; Paeni-positive PIH infants had consistently differing counts of those markers than did the Paeni-negative and NPIH infants (Figure 6B). Pathway enrichment analysis of those genes demonstrated a predominance of functions associated with the immune system, particularly those involved with interleukin (IL)-4, IL-12, IL-13, interferon, and neutrophil activity, as well as those relating to Janus kinase/signal transducers and activators of transcription (Jak/STAT) pathway. In addition, there was enrichment for processes involving response to platelet-activating factors and host recognition of microbes including antigen presentation and Class I MHC antigen processing (Figure 7).
DISCUSSION
PIH is a debilitating neurological sequela of neonatal sepsis4,7,8. We identified critical gene products and pathways involved in the inflammatory mechanisms early in PIH. To our knowledge this study is the first to utilize proteogenomics to characterize gene regulatory networks involved in any form of hydrocephalus. These findings inform the development of adjunctive strategies for neonatal sepsis treatments to prevent hydrocephalus.
Large scale proteomics and genomics studies have been instrumental in the discovery of important biomarkers and have provided insights into the pathophysiology of central nervous system disease. CSF proteomics has previously identified clinically useful protein biomarkers such as amyloid precursor protein and cell adhesion molecules in PHH17,27. DNA or RNA sequencing from CSF has been widely used to study infectious diseases28,29. Because disease pathogenesis involves the combinatorial interaction of elements within the proteome, transcriptome, and environmental factors30,31, integrative methods that combine proteomic and genomic data into a network model are essential for thoroughly understanding human disease32. Here we utilized clustering, pathway analysis, and gene ontology enrichment methodologies at the RNA and protein levels to unravel the underlying mechanisms involved in the host response to PIH.
Summary of findings
In this study, genes with differential RNA or protein abundance among PIH patients were enriched for functions related to neuroinflammation, reaction to oxidative stress, cell-cell junction structure, and extracellular matrix organization. Combining proteomics and RNA-Seq approaches narrowed the spectrum of host responses to predominantly those involved with the innate immune system, including neutrophil activity, signaling via IL-4, IL-12, IL-13, and interferon, and Jak/STAT pathways, in addition to platelet response/activating factors. Furthermore, there was enrichment for factors involved with microbe recognition such as Class I MHC antigen-presenting complex. In light of the recent confluence of evidence suggesting that dysregulated neuroinflammation and related ependymal scarring and aqueductal obstruction propagate inflammatory hydrocephalus (including PIH and PHH)4,11, the pathways we identified are potential targets for adjunctive treatments to reduce the risk of neuroinflammation and hydrocephalus following neonatal sepsis.
Host immune responses in PIH
In addition to ventriculomegaly (enlargement of the cerebral ventricles), the PIH phenotype is characterized by CSF fluid loculations, debris within fluid spaces, ectopic calcification, and brain abscesses8, suggesting that severe inflammation occurs locally during the antecedent neonatal sepsis. Typically, once a pathogen breaches a host’s epithelial barriers, immune mechanisms are activated. Although there is a rapid progression of immunologic competence after birth, the innate immune system is the primary active defense for infants less than 3 months of age, since they lack the antigenic experience that informs acquired immunity. Consequently, the neonatal host immune response involves antigen-independent immune components such as neutrophils, phagocytes, NK cells and antigen-presenting cells33,34 35-37. Our RNA-Seq immune cell signature analyses support the role of NK cells in the host response of PIH patients. The Toll-like Receptor (TLR) immune system pathway is commonly implicated in post-inflammatory (including infectious or hemorrhagic) hydrocephalus11,38-40; however, the cytokine and interferon induced Jak/STAT pathway is considered to be the most efficient form of innate immunity, especially with intracellular pathogens37,41. Protein mutations in the Jak/STAT pathway have been implicated in poor inflammation mediation37,41. For example, impaired Jak function in severe combined immunodeficiency is associated with susceptibility to infections due to the absence of NK, B, or T cells42,43. At the protein and RNA levels, we observed differential expression of factors associated with interleukins (IL-4, IL-12 and IL-13) and interferons activity, both of which are obligate mediators of the Jak/STAT pathway37,41. Upregulation of MHC Class I in our cohort could be an indication of presentation to cytotoxic CD8+ cells and IL-12 activity44.
Inflammation and barrier integrity in PIH
The activated immune pathways found in PIH support the perspective that the neonatal immune response to toxins generated by infection or hemorrhage inadvertently leads to ependymal gliosis or denudation, scarring of CSF conduits, and disruption of CSF physiology to cause hydrocephalus. Given that infants presenting with PIH are often remote from the original infection and clinically well for surgery, persistent inflammation in their CSF may be pathological rather than beneficially adaptive11. While the host-immune response is necessary for clearing microbial pathogens and fighting infections, prolonged or overstimulated immune activation may be detrimental to the host45. In PIH, it is possible that the same host immune activity that clears pathogens also causes hydrocephalus by damaging local tissues. Our finding of a differential abundance of RNA and proteins involved in cell-cell integrity pathways is consistent with recent evidence from both experimental46-51 and clinical46,52,53 studies demonstrating that host immune response is associated with ependymal cell-cell junction protein disruption as a critical pathogenetic mechanism of hydrocephalus. We also observed differential abundance of RNA and proteins associated with platelet-activating factors and response to reactive oxidative species. Elevations in platelet-activating factors have been associated with compromise of the blood brain barrier54, and their signaling pathways are an important link between inflammatory and thrombotic processes, including in sepsis55. Neutrophil-generated reactive oxidative species have also been implicated in disruptive blood-brain barrier changes during sepsis56. During an acute infection there is a high rate of oxidative metabolism and response to reactive oxidative species such as hydrogen peroxide, superoxide radicals and nitric oxide57, which typically sets off a cascade of cellular excitotoxicity and secondary brain injury that impairs brain oxygenation and perpetuates cerebral vascular dysfunction and ischemia58. Such vascular pathology might be associated with our finding of a differential expression of factors involved in platelet activity including PFN1, ITIH4 and A2M. The blood-brain barrier disruption may herald the extravasation of microbes and inflammatory mediators from the systemic compartments into the ventricles and other intracranial compartments.
Potential pathologic overlap with NPIH
Inflammation is a component of other forms of hydrocephalus38,59, which informed our decision to use non-infectious hydrocephalic infants as controls; doing so allowed us to be able to detect molecular mechanisms that are more specific to PIH. In a recent review, Karimy et al. discussed evidence that PIH shares common host immune pathways with PHH11. In our cohort, we identified IL-12 signaling, which is typically involved in augmentation of CD8+ T cell cytotoxicity, as an important cytokine pathway underlying PIH. Interestingly, Habiyaremye et. al. recently reported PHH is associated with significantly elevated levels of IL-12, in addition to other cytokines including IL-1, IL-10, CCL-3 and CCL-1359. In an experimental model, TLR4-mediated host immune response was found to lead to choroid plexus hypersecretion of CSF38, and potential facilitation of PHH. Neuroinflammation has been linked to the ependymal barrier damage found in humans and several experimental models of hydrocephalus. In a human postmortem histological study, McAllister et. al. implicated glial fibrillary acidic protein over-expression, a non-specific surrogate marker of neuroinflammation and glial scarring in the diffuse ependymal damage they identified in PHH infants60. The ependymal damage was shown to result from cleavage of cell adhesion proteins60,61. Indeed, many forms of human congenital hydrocephalus result from genetic alterations of proteins involved in cell-cell junctional integrity including N-cadherin, connexin, and L1CAM62, that are critical for the differentiation of the ventricular and subventricular zones neural stem cells into mature ependyma63. In addition, a recent exome sequencing of 381 human congenital hydrocephalus infants identified a predominance of de novo mutations in genes associated with neural stem differentiation64. Experimentally, a series of studies on hyh mice49,51,65-67 that develop perinatal aqueductal stenosis also support the concept of a defect in cell junction complexes as an underlying cause of hydrocephalus. Collectively, these findings emphasize a critical association between host-immune response, cell-cell junction regulatory molecules and hydrocephalus, and raise the question as to whether modulation of the inflammatory signals common to PIH and NPIH might be beneficial.
Role of immunomodulation for PIH prevention
This study supports the role of the immune response in PIH pathophysiology. Host immune response was the most predominant biological process among proteins and genes with differential abundance in PIH. However, it would be premature to suggest that our findings support the general use of anti-inflammatory or immunosuppressive agents during neonatal sepsis to mitigate the risk of developing hydrocephalus. There have been many attempts to use corticosteroids in treating infants with acute bacterial meningitis; but except for some notable exceptions such as Hemophilus influenza68-72 and pneumococcal meningitis68- 73, there have been many failures with some devastating outcomes69,70,72-76. The use of corticosteroids remains a controversial topic in perinatal medicine, and in particular, neonatal sepsis77-80. Our work however, points to novel intervention pathways and suggests that selective and targeted modulation of aspects of the immune response pathways may be a more successful approach to treating brain infections to prevent hydrocephalus. For example, in addition to pro-inflammatory immune factors, we found differential expression of markers associated with IL-4 and IL-13, which generally have anti-inflammatory effects. IL-4 and IL-13 can share a common receptor81-83 and typically work synergistically to decrease inflammation by counteracting the activity of pro-inflammatory cytokines such as IL-1284,85. IL-4 and IL-13 are also considered neuroprotective as they can induce death of microglial cells that mediate neuronal damage86-88. However, Park and colleagues89,90 report IL-4 and IL-13 potentiate oxidative stress-related injury to hippocampal neurons in experimental models. Thus, one hypothesis is that the balance of IL-12 and IL-4/IL-13 within the CSF contributes to risk of PIH, suggesting that their targeted modulation can be a potential therapeutic target. Further studies are needed to examine the role of cytokine/chemokine modulators in neonatal sepsis to prevent hydrocephalus.
Viral co-infection in PIH
Identifying CMV in the CSF of a subset of infants in our cohort16 raises the important question of whether the host-immune response we observed was, at least in part, attributable to CMV. While there were differentially expressed transcripts detected in CSF and/or serum consistent with the host response to central nervous system viral infection in CMV-positive infants, we did not detect any functional proteins associated with a response to CMV, nor was there any effect of CMV status on the differential expression of host-immune markers at the protein level. The inability to detect a CMV-related host immune response at the protein level may be attributed to our small sample size (8/100 and 27/100) of CMV-CSF-positive and CMV-serum-positive infants, or the ability of the virus to modulate protein synthesis in the host91. Alternatively, CSF CMV status may underestimate CMV burden within the cohort, as intracellular CMV may be difficult to detect in CSF samples with low cell counts. Another possibility is that because CMV tends to cycle through latent or non-lytic and active states in affected individuals92, sampling CSF at any single point in time may not capture an active state. Thus, it remains unknown whether CMV was a risk factor for developing PIH, whether CMV can be implicated in the persistent host-immune response we observed, or whether the presence of this virus will have long-term effects in these infants.
Limitations
Limitations to this study include small cohort size for subset analyses such as CMV positive participants, low RNA abundance in NPIH samples, and reduced availability of clinical covariates to assess potential correlations with gene abundance patterns. Furthermore, the majority of activated gene networks we identified in this cohort were pathogen-stratified based on Paenibacillus spp. status, possibly limiting generalizability for other pathogens. However, the concentrated, high incidence of PIH in Sub-Saharan Africa makes our findings directly relevant and applicable to PIH not just in globally under-resourced areas but also as it occurs in other developed regions. Nevertheless, important future work will test the generalizability of host response observed here to PIH related to other pathogens.
CONCLUSIONS
Inflammation following neonatal infection is a dominant cause of childhood hydrocephalus in the developing world. Proteome and transcript expression analysis identified gene pathways involving neutrophil, interleukin (4, 12, and 13) and interferon activity critically associated with PIH in African infants. These findings enable the development of preventive hydrocephalus risk reduction strategies.
Data Availability
All relevant data for this manuscript have been included in the methodology and results