Immune activation during Paenibacillus brain infection in African infants with frequent cytomegalovirus co-infection.

Inflammation during neonatal brain infections leads to significant secondary sequelae such as hydrocephalus, which often follows neonatal sepsis in the developing world. In 100 African hydrocephalic infants we identified the biological pathways that account for this response. The dominant bacterial pathogen was a Paenibacillus species, with frequent cytomegalovirus co-infection. A proteogenomic strategy was employed to confirm host immune response to Paenibacillus and to define the interplay within the host immune response network. Immune activation emphasized neuroinflammation, oxidative stress reaction, and extracellular matrix organization. The innate immune system response included neutrophil activity, signaling via IL-4, IL-12, IL-13, interferon, and Jak/STAT pathways. Platelet-activating factors and factors involved with microbe recognition such as Class I MHC antigen-presenting complex were also increased. Evidence suggests that dysregulated neuroinflammation propagates inflammatory hydrocephalus, and these pathways are potential targets for adjunctive treatments to reduce the hazards of neuroinflammation and risk of hydrocephalus following neonatal sepsis.

Introduction

Hydrocephalus is a severe brain disorder in children (Isaacs et al., 2018; Dewan et al., 2018) and the most common indication for pediatric neurosurgery worldwide (Simon et al., 2008). Globally, the most frequent antecedent of hydrocephalus is infection such as neonatal or infant sepsis (Warf, 2005). Postinfectious hydrocephalus (PIH) accounts for up to 60% (Warf and East African Neurosurgical Research, 2010) of the nearly 400,000 infants who develop hydrocephalus each year, principally in low- and middle-income countries (Dewan et al., 2018). Strategies to prevent PIH have been thwarted for two principal reasons. First, the key pathogens responsible for the underlying infections that precede PIH are rarely identified and thus treatment of the underlying infections cannot be optimized (Sinnar and Schiff, 2020). Second, we have not identified the specific underlying inflammatory response that causes the hydrocephalus (McAllister et al., 2015; Karimy et al., 2020). Despite recent clinical efforts to optimize surgical treatment (Kulkarni et al., 2017), the outcomes of treatment for PIH early in life remains disappointing (Sinnar and Schiff, 2020; McAllister et al., 2015; Karimy et al., 2020). Characterizing the inflammatory responses that create PIH is necessary to develop adjunctive strategies that can reduce the likelihood of subsequently developing hydrocephalus.

The underlying pathogenic basis for PIH can be complex. A recent pan-microbial approach uncovered Paenibacillus spp. as a PIH pathogen associated with severe ventriculitis in a cohort of Ugandan children (Paulson et al., 2020). However, there were also frequent co-infections with cytomegalovirus (CMV). It is unclear whether the bacteria or virus dominates the inflammatory response or whether they act synergistically. To unravel the immune contributions from multiple pathogens, we performed whole transcriptome and protein mass spectrometry of cerebrospinal fluid (CSF) from 64 infants with PIH and 36 control non-postinfectious hydrocephalus (NPIH) infants without acute inflammation to examine the immunopathogenesis of PIH (Paulson et al., 2020; Rohlwink et al., 2019; Morales et al., 2012). Matching deep-scale proteomics (i.e., high gene coverage) with transcriptomics (Nesvizhskii, 2014; Zhang et al., 2014) enabled optimal characterization of the host response (Figure 1).

Figure 1

Schematic of proteogenomics experimental and data analysis workflow

Output data from concurrent proteomics and RNA-seq on the same samples were preprocessed, normalized, batch-controlled, and explored independently. Differentially expressed transcriptomic and proteomic data were integrated following dimension reduction and feature selection. Gene ontology enrichment was evaluated and common reactome pathways were visualized to identify prominent molecular pathways implicated in the pathophysiology of postinfectious hydrocephalus. Adapted from “PTMScan Workflow,” by BioRender.com (2021). Retrieved from https://app.biorender.com/biorender-templates.

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. Infants with PIH 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 infants with PIH had a history of hydrocephalus at birth. Patients with NPIH had findings of hydrocephalus that was of non-infectious origin on computed tomography (CT) 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). Before 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.

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 United States for further processing. Previous 16S rRNA sequencing had recovered Paenibacillus spp. from 59% of the infants with PIH in this study, of which 8 and 27 patients also had human herpesvirus 5 (CMV) in the CSF and blood (Paulson et al., 2020), respectively. These previous findings directed stratification of our patients into “Paeni-positive” or “Paeni-negative,” “CMV-CSF-positive” or “CMV-CSF-negative,” and “CMV-blood-positive” or “CMV-blood-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 versus 9.3 × 103/μL, p = 0.002) and CSF (71.6 versus 5.0 × 103/μL, p < .001), and lower blood hemoglobin levels (10.8 g/dL versus 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).

Table 1

Demographics and clinical characteristics of patients based on Paenibacillus spp. status

	All patients n = 100	Paeni-negative n = 62		Paeni-positive n = 38
		NPIH n = 36	PIH n = 26	PIH n = 38
Age in days, mean (SD)	57 (24)	43 (27)	75 (11)	59 (17)
Sex
Male (%)	51 (51)	16 (44)	14 (54)	21 (55)
Female (%)	49 (49)	20 (56)	12 (46)	17 (45)
Positive CMV status
Blood	27	9	8	10
CSF	8	0	2	6
CSF WBC [1.0 × 103]/μL, mean (SD)	30 (62)	5 (0.5)	5 (0.0)	72 (86.0)
Peripheral blood WBC [1.0 × 103]/μL, mean (SD)	10.3 (3.6)	8.8 (2.6)	9.8 (3.0)	11.9 (4.1)
Hemoglobin g/dL, mean (SD)	11.5 (2.2)	13.0 (2.8)	10.5 (1.3)	10.8 (1.3)
Hematocrit %, mean (SD)	36.8 (7.4)	41.6 (9.3)	33.6 (4.2)	34.4 (4.1)

CSF, cerebrospinal fluid; CMV, cytomegalovirus; PIH, postinfectious hydrocephalus; NPIH, non-postinfectious hydrocephalus; Paeni positive, CSF Paenibacillus spp.-positive; Paeni-negative, CSF Paenibacillus spp. negative.

Protein expression

Tandem mass spectrometry (MS/MS) spectra obtained from a standard 10-channel peptide mass tagging (TMT-10) system and compared with the current UniRef database yielded quantitative proteomic data (Mertins et al., 2018; Chen et al., 2012; McAlister et al., 2012; Morales et al., 2012). Median values of peptide intensities were assigned to unique proteins and were used to infer relative protein abundances. Peptide identifications that are non-unique by the principle of parsimony were excluded from protein quantification (Koskinen et al., 2011). Batch effects were corrected, and intensity values were quantile normalized. Multidimensional scaling (MDS) analysis demonstrated the clustering of infants with PIH and NPIH 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 up- or downregulated in the Paeni-positive group, respectively (Figure 2B). Gene set enrichment analysis of the differentially expressed proteins identified that the predominant enriched functions were involved in neuroinflammation, particularly neutrophil-mediated inflammation, negative regulation of proteolysis and peptidase activity, and modulation of extracellular matrix and structure (Figure 2C).

Figure 2

Proteomic profile of infants with postinfectious hydrocephalus (PIH) and those without (NPIH), based on 16s rRNA-determined Paenibacillus spp. status

(A) Multidimensional scaling of normalized protein abundances demonstrating clustering of Paenibacillus spp.-positive (Paeni-positive) infants from the remaining groups. The abscissa and ordinate of the scatterplot of individual participants represent the first and second components, respectively. Each oval (not drawn to scale) encircles majority of patients belonging to the color-matched group, with blue as infants in the NPIH group, red as Paeni-negative PIH infants, and green as Paeni-positive infants.

(B) Volcano plot demonstrating differential expression of genes between Paeni-positive and Paeni-negative infants. Adjustment for cytomegalovirus status did not change the differential expressions between groups. The vertical lines crossing the positive and negative abscissae demarcate fold changes of 1 and -1, respectively, and the horizontal dashed line crosses the ordinate at the alpha significance level of 0.05. Each point represents a differentially expressed protein, and those with an absolute fold change greater than 1 that met the significance level (red points) were selected for gene set enrichment analyses. Genes with enrichment below the statistically significant threshold are displayed in gray.

(C) Gene ontology analyses of differentially expressed proteins of infants based on 16s rRNA-determined Paenibacillus spp. status. There was enrichment for functions associated with neuroinflammation, extracellular matrix structure, and cell-cell adhesion among Paeni-positive infants compared with Paeni-negative infants. The abscissae (ratio) of the dot plots correspond to the number of proteins per total number of proteins, and the size of each circle reflects the relative number of proteins expressed that are enriched for the corresponding function (ordinates).

RNA expression

Gene read counts (25 million reads per sample) were aggregated from expression levels estimated from paired-end RNA sequencing (RNA-seq) data mapped to the human reference genome hg38 with STAR and quantified with RESM (Table S1) (Dobin et al., 2013; Li and Dewey, 2011; Paulson et al., 2017). 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 infants with PIH and NPIH (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, cell motility, migration, and adhesion (Figure 3C).

Figure 3

Transcriptomic profile of infants with postinfectious hydrocephalus (PIH) and those without (NPIH), based on 16s rRNA-determined Paenibacillus spp. status

(A) Multidimensional scaling of normalized gene expression demonstrating clustering of Paenibacillus spp.-positive (Paeni-positive) infants from the remaining groups. The abscissa and ordinate of the scatterplot of individual participants represent the first and second components, respectively. Each oval (not drawn to scale) encircles the majority of patients belonging to the color-matched group, with blue as infants in the NPIH group, red as Paeni-negative PIH infants, and green as Paeni-positive infants.

(B) Heatmap of the 500 most differentially expressed genes between the Paeni-positive and Paeni-negative infants, with the dendrogram demonstrating hierarchical clustering on Euclidean distance of gene identified in infants based on Paenibacillus spp. status.

(C) Gene ontology analyses of differentially expressed genes of infants based on 16s rRNA-determined Paenibacillus spp. status. There was enrichment for functions associated with neuroinflammation, extracellular matrix structure, and cell-cell adhesion among Paeni-positive infants compared with Paeni-negative infants. The abscissae (ratio) of the dot plots correspond to the number of proteins per total number of proteins, and the size of each circle reflects the relative number of proteins expressed that are enriched for the corresponding function (ordinates).

Differential expression analysis of long non-coding genes

When aligning RNA, we detected 116 long non-coding RNAs (lncRNAs), out of 1,220 lncRNAs annotated in Ensembl. Of the 116, 70 had at least one count (read) in at least 20% of the samples. Ninety samples had at least one lncRNA present. The five most highly expressed lncRNAs were OIP5-AS1, FAM211A-AS1, C1orf132, LINC00963, and ZNF518A. The most variable lncRNAs included OIP5-AS1, FAM211A-AS1, C1orf132, LINC00963, as well as FAM157C. In performing differential expression, we identified 35 lncRNAs significantly associated with Paeni-positive samples with a threshold of false discovery rate <0.05. In Paeni-positive samples, the lncRNAs with the largest fold-changes (upregulation) were FAM157C, SPACA6P, LINC00243, LINC01128, and SEMA3B. No lncRNAs were significantly downregulated in Paeni-positive samples. The nature of how these lncRNAs are associated with disease and Paenibacillus spp. positivity will be of future interest.

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 4C).

Figure 4

Role of CMV co-infection

(A) An interactome of the 64 genes that were differentially expressed based on CMV status, enriched for functions related to host response to virus.

(B) Principal-component analysis plot demonstrating that using RNA abundance of all genes was not able to cluster samples by CMV status. The abscissa and ordinate of the scatterplot of individual participants represent the first and second components, respectively.

(C) Volcano plot demonstrating there are no differentially expressed proteins based on CMV status alone (blue points) that had a fold change greater than 1 or met statistical significance criteria. Genes with enrichment below the statistically significant threshold are displayed in gray. Differential expression of proteins based on Paenibacillus spp. status was similar with or without adjusting for CMV status (Figure 2).

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), two gene sets (Modules 1 and 2) were identified as having similar patterns of expression within this subset of 33 patients (Table S2). These modules were assessed for network correlation with four 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.

Figure 5

Weighted correlation network analysis (WGCNA) and single-cell RNA deconvolution

(A) Gene modules within RNA-seq data for PIH and NPIH cohorts identified by WGCNA. Module 1 positively correlated with CSF cell count and negatively with Paenibacillus spp. status and was enriched for host immune responses, including leukocyte and neutrophil functions. (B) Module 2 was positively correlated with Paenibacillus spp. status, but the associated genes had no specific functional enrichment.

(C) Deconvolution of bulk RNA data into immune cell populations expressed on a scale of 0 to 1 that is calculated for each patient. There was hierarchical clustering of a mixture of NPIH and PIH samples with a hematologic predominance, and PIH-only samples with T helper and NK cell populations.

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 expression (Aran et al., 2017). 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).

Figure 6

Proteogenomic integration of proteins and genes expressed in RNA-seq and/or proteomics among infants with postinfectious and non-postinfectious hydrocephalus stratified by cerebrospinal fluid 16s rRNA Paenibacillus spp. status

(A) Alluvial plot demonstrating the most prominent gene ontological (GO) functions and interactions for PIH pathophysiology between Paenibacillus spp.-positive and Paenibacillus spp.-negative infants. Each ontological clustering occupies a column in the diagram and is horizontally connected to preceding and succeeding significance clustering by stream fields, representing similar gene involvement. Each stacked bar is color-coded based on the assay being assessed, with red representing RNA-seq data, purple for proteomics, and green for genes common to both RNA-seq and proteomics. The ordinate shows the number of genes represented in each cluster.

(B) 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; infants with Paeni-positive PIH had consistently differing counts of those markers than did the Paeni-negative infants and infants with NPIH (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 major histocompatibility complex (MHC) antigen processing (Figure 7, Table S3).

Figure 7

Pathway analysis of 33 genes that were differentially expressed in both RNA-seq and proteomics among infants with postinfectious and non-postinfectious hydrocephalus stratified by cerebrospinal fluid 16s rRNA Paenibacillus spp. status

Corresponding pairs of upregulated (red boxes) and downregulated (green boxes) proteins (blue columns) and genes (purple columns) in the Paenibacillus spp.-positive group that were identified with proteomics and RNA-seq, respectively, are listed on the abscissa. The 33 common genes demonstrated predominant involvement of the immune system, particularly the innate system and those associated with neutrophil-mediated activity, interleukins, interferon, and the Janus kinase/signal transducers and activators of transcription (JAK-STAT) pathway (ordinate). Differential expression was defined as log2 fold change of >1 or < -1 at an alpha significance level of 0.05.

Discussion

Disease pathogenesis involves the combinatorial interaction of the proteome, transcriptome, and environment (Rédei, 2008; Heintzman and Ren, 2006). By matching deep-scale proteomics (i.e., high gene coverage) with transcriptomics (Nesvizhskii, 2014; Zhang et al., 2014), the interplay within the host immune response can be optimally reduced into network models representing disease (Ruggles et al., 2017).

In this study, genes with differential mRNA or protein abundance among patients with PIH were enriched for functions related to neuroinflammation, reaction to oxidative stress, cell-cell junction structure, and extracellular matrix organization. Combining proteomics and RNA-seq results narrowed the spectrum of responses to the innate immune system, including neutrophil activity and signaling via IL-4, IL-12, IL-13, interferon, and Jak/STAT pathways, in addition to platelet-activating factors. Furthermore, there was enrichment for factors involved with microbe recognition such as Class I MHC antigen-presenting complex. Our findings are consistent with a major role for up-regulation of genes and proteins associated with neutrophil activation. Nevertheless, in addition to the direct role in fighting the infection, we note that there appears to be substantial concomitant parenchymal loss in these patients, and we cannot rule out innate-immunity-mediated tissue damage caused by this immune response. One of the difficulties of ascribing infection causality to the presence of pathogen genomes is the ubiquitous problem of environmental and reagent contamination in low-biomass samples, the sensitivity of sequencing techniques, the potential artifacts and biases contributed by sequence amplification when investigating bacterial 16S, as well as the presence of microbes within nominally sterile body spaces that may not be causing active infection. Our findings demonstrated a patterned innate immune response associated with the presence of Paenibacillus 16S DNA that strengthens the causal association of Paenibacillus with infection. In light of the recent confluence of evidence suggesting that dysregulated neuroinflammation propagates inflammatory hydrocephalus (including PIH and post-hemorrhagic, PHH post-hemorrhagic) (Karimy et al., 2020; Warf, 2005), the pathways we identified are potential targets for adjunctive treatments to reduce the hazards of neuroinflammation and risk of 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 abscesses (Ciurea et al., 2005), suggesting that severe inflammation occurs locally during the antecedent neonatal sepsis. Typically, once a pathogen breaches a host's endothelial and epithelial barriers, immune responses are activated. Although there is a rapid accumulation 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 cells (Bell, 2003; Kawai and Akira, 2006; Villarino et al., 2017; PrabhuDas et al., 2011; Ygberg and Nilsson, 2012). Our RNA-seq immune cell signature analyses support the role of NK cells in the host response of patients with PIH. The toll-like receptor (TLR) immune system pathway is commonly implicated in post-inflammatory (including infectious or hemorrhagic) hydrocephalus (Karimy et al., 2017, 2020; Sokol et al., 2016; Lattke et al., 2012); however, the cytokine- and interferon-induced Jak/STAT pathway is considered the most efficient form of innate immunity, especially with intracellular pathogens (Villarino et al., 2017; O'Shea et al., 2015). Protein mutations in the Jak/STAT pathway have been implicated in inadequate inflammation mediation (Villarino et al., 2017; O'Shea et al., 2015). For example, impaired Jak function in severe combined immunodeficiency is associated with susceptibility to infections due to the absence of NK, B, or T cells (Noguchi et al., 1993, 2008). At the protein and transcript levels, we observed differential expression of factors associated with interleukins (IL-4, IL-12 and IL-13) and interferon activity, both of which are obligate mediators of the Jak/STAT pathway (Villarino et al., 2017; O'Shea et al., 2015). Upregulation of MHC Class I in our cohort could be an indication of presentation to cytotoxic CD8+ cells and IL-12 activity (Adiko et al., 2015).

Inflammation and barrier integrity in PIH

The activated immune pathways found in PIH support the perspective that the neonatal immune response to infection or hemorrhage leads to ependymal gliosis or denudation, scarring of CSF conduits, and disruption of CSF physiology to cause hydrocephalus. Although the host-immune response is necessary for fighting infection and clearing microbial pathogens, prolonged or overstimulated immune activation may be detrimental (Tchessalova et al., 2018); indeed, recent evidence from both experimental (Guerra et al., 2015a; Ortloff et al., 2013; Rodriguez et al., 2012; Jimenez et al., 2009; Paez et al., 2007; Batiz et al., 2005) and clinical (Ortega et al., 2016; Guerra et al., 2015a; Sival et al., 2011) studies demonstrates that host immune response associated with ependymal cell-cell junction protein disruption is a critical pathogenetic mechanism of hydrocephalus.

We also observed platelet-activating factors and response to reactive oxidative species. Elevations in platelet-activating factors have been associated with compromise of the blood-brain barrier (Brailoiu et al., 2018), and their signaling pathways are an important link between inflammatory and thrombotic processes, including in sepsis (de Oliveira et al., 2017; Yost et al., 2010; Pun et al., 2009). Reactive oxygen species set off a cascade of cellular excitotoxicity and secondary brain injury that impairs brain oxygenation and perpetuates cerebral vascular dysfunction (Lehner et al., 2011). Such vascular pathology is consistent with our finding of a differential expression of factors involved in platelet activity including PFN1, ITIH4, and A2M.

Potential pathologic overlap with NPIH

Inflammation is a component of other forms of hydrocephalus (Habiyaremye et al., 2017; Karimy et al., 2017), which informed our decision to use noninfectious hydrocephalic infants as controls. In a recent review, Karimy et al. discussed evidence that PIH shares common host immune pathways with PHH (Karimy et al., 2020). In our PIH cohort, we identified upregulation of IL-12 signaling, which is typically involved in augmentation of CD8+ T cell cytotoxicity. PHH is also associated with significantly elevated levels in other cytokines including IL-1, IL-10, CCL-3, and CCL-13 (Habiyaremye et al., 2017). Neuroinflammation has been linked to the ependymal barrier damage found in humans and several experimental models of hydrocephalus (McAllister et al., 2017). Such ependymal damage was shown to result from cleavage of cell adhesion proteins (McAllister et al., 2017; Castaneyra-Ruiz et al., 2018). Indeed, many forms of human congenital hydrocephalus result from genetic alterations of proteins involved in cell-cell junctional integrity including N-cadherin, connexin, and L1CAM (Guerra et al., 2015b), which are critical for the differentiation of the ventricular and subventricular zone neural stem cells into mature ependyma (Jin et al., 2020). In addition, a recent exome sequencing of 381 infants with congenital hydrocephalus identified a predominance of de novo mutations in genes associated with neural stem cell differentiation (Jin et al., 2020). Experimentally, a series of studies on hyh mice (Jimenez et al., 2009; Batiz et al., 2005, 2006; Perez-Figares et al., 1998; Wagner et al., 2003) that develop perinatal aqueductal stenosis also support the concept of a defect in cell junction complexes as an underlying cause of hydrocephalus. Although our use of NPIH as controls focused on the acute inflammatory responses that lead to PIH, there are linkages to common mechanisms in other forms of infant hydrocephalus.

Role of immunomodulation for PIH prevention

It would be premature to suggest that our findings support the 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 Haemophilus influenza (Schaad et al., 1993; Syrogiannopoulos et al., 1994; King et al., 1994; Wald et al., 1995; Kilpi et al., 1995) and pneumococcal meningitis (Kanra et al., 1995; Kilpi et al., 1995; King et al., 1994; Schaad et al., 1993; Syrogiannopoulos et al., 1994; Wald et al., 1995), there have been many failures with some devastating outcomes (Kanra et al., 1995; Kilpi et al., 1995; King et al., 1994; Lebel et al., 1988, 1989; Odio et al., 1991; Syrogiannopoulos et al., 1994) and this remains an area of controversy (Schaad et al., 1995; McIntyre et al., 1997; Coyle, 1999; Goodman et al., 2002). Our work points to critical intervention pathways and suggests that selective and targeted modulation of aspects of the immune response might be studied when treating brain infections to prevent hydrocephalus. For example, in addition to pro-inflammatory immune factors, we found differential expression of IL-4 and IL-13, which generally have anti-inflammatory effects. IL-4 and IL-13 can share a common receptor (Hilton et al., 1996; Jiang et al., 2000; Callard et al., 1996) and typically work synergistically to decrease inflammation by counteracting the activity of pro-inflammatory cytokines such as IL-12 (Minty et al., 1993; Mori et al., 2016). IL-4 and IL-13 are also considered neuroprotective as they can induce death of microglial cells that mediate neuronal damage (Won et al., 2013; Deboy et al., 2006; Yang et al., 2002). However, IL-4 and IL-13 can potentiate oxidative stress-related injury to neurons (Nam et al., 2012; Park et al., 2008). Thus, one possibility is that the balance of IL-12 and IL-4/IL-13 within the CSF contributes to risk of PIH, implying that their targeted modulation is a potential therapeutic target.

Viral co-infection in PIH

Identifying CMV in the CSF of a subset of infants in our cohort (Paulson et al., 2020) raises the important question of whether the host immune response was in part attributable to CMV. Although there were differentially expressed transcripts detected in CSF 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. This may be due to our small sample size (8/100 CSF and 27/100 blood) of CMV-positive infants, or the ability of the virus to modulate protein synthesis in the host (Marshall and Geballe, 2009). Alternatively, CSF CMV status using PCR underestimates CMV burden as intracellular CMV may be difficult to detect in CSF samples with low cell counts, and because CMV cycles through latent and active states in affected individuals (Cheeran et al., 2009). It remains unclear whether CMV was a risk factor for developing PIH or whether the presence of this virus will have long-term effects in these infants.

Conclusions

Inflammation following neonatal infection is a dominant cause of childhood hydrocephalus. An integrated proteogenomic strategy identified gene pathways involving the innate immune system, cell-cell junction structure, platelet activation, and microbial recognition in African infants with PIH. These findings enable the development of preventive hydrocephalus risk reduction strategies during the treatment of neonatal sepsis.

Limitations of the study

Limitations to this study include small cohort size for subset analyses such as CMV-positive participants and low RNA abundance in NPIH samples. Because NK cell activity, IL-12 signaling, and JAK/STAT pathway may indicate viral immune response, it is possible that co-infection with CMV was more common in our patients than our conservative approach to CMV detection permitted. The observation of Th2 cell activation by IL-4/IL-13 needs confirmation in an independent cohort. In addition, attribution of RNA expression to specific cells cannot be completed for the current samples, limiting the ability to confirm the role of specific immune cells in PIH. Furthermore, the majority of activated gene networks identified were pathogen-stratified based on Paenibacillus spp. presence.

Resource availability

Lead contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Steven J. Schiff (steven.j.schiff@gmail.com).

Materials availability

This study did not generate new unique reagents.

Data and code availability

The code generated during this study is available at our public Github repository https://github.com/Schiff-Lab/iScience_Paper_Isaacs_Morton_2021.

RNA sequencing and sample metadata are available in the NCBI archive under project ID #PRJNA605220. Proteomic data are available in the Proteome Exchange under project ID #PDX024842.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.