Blood proteome of acute intracranial hemorrhage in infant victims of abusive head trauma

Abusive head trauma (AHT) is a leading cause of mortality and morbidity in infants. While the reported incidence is close to 40 cases per 100′000 births/year, misdiagnoses are commonly observed in cases with atypical, subacute, or chronic presentation. Currently, standard clinical evaluation of inflicted intracranial hemorrhagic injury (ICH) in infants urgently requires a screening test able to identify infants who need additional investigations. Blood biomarkers characteristic of AHT may assist in detecting these infants, improving prognosis through early medical care. To date, the application of innovative omics technologies in retrospective studies of AHT in infants is rare, due also to the blood serum and cerebrospinal fluid of AHT cases being scarce and not systematically accessible. Here, we explored the circulating blood proteomes of infants with severe AHT and their atraumatic controls. We discovered 165 circulating serum proteins that display differential changes in AHT cases compared with atraumatic controls. The peripheral blood proteomes of pediatric AHT commonly reflect: (i) potentially secreted proteome from injured brain, and (ii) proteome dysregulated in the system's circulation by successive biological events following acute ICH. This study opens up a novel opportunity for research efforts in clinical screening of AHT cases.

increased morbidity and mortality. Pediatric AHT cases rely on medical treatments identical to those of accidental head trauma known to display a better prognosis in young infants [4]. In this context, pediatricians urgently require a screening test for infant victims of AHT.
A novel concept of data-driven discoveries integrating omics approaches is predicted to revolutionize clinical diagnostics in pediatrics [5]. A deep omics profiling of blood proteomes in infant victims of AHT is therefore a prerequisite for the discovery of: (i) diagnostic and prognostic protein markers of AHT cases, and (ii) medical treatments through elucidation of the molecular pathways characteristic of trauma to the developing brain. Over the past decade, developments in the performance of mass spectrometry (MS) based proteomics platforms have resulted in remarkable throughput, reproducibility, and quantification accuracy compared with past techniques [6][7][8]. In particular, MS platforms that rely on a data-independent acquisition (DIA-MS) strategy ensure consistency of proteome analysis across human specimens by creating a permanent record of all ionized peptides from a tested sample [7][8][9][10]. Yet the application of omics approaches for discovery purposes is very rare in the biological specimens of child ICH and inexistent in the AHT cases. To date, postmortem blood (PMB) specimens remain largely unexplored in the biomarker discovery studies of AHT cases in infants. The serum samples of children younger than two with concerns regarding AHT have not been easily accessible, and the animal models that replicate the various aspects of human TBI showed considerable clinical limitations [11,12]. The recent registration of forensic biobanks via medico-legal structures covering the specimens of healthy pediatric populations underrepresented in standard clinical biobanks will open great opportunities for novel discoveries in pediatric TBI [13]. Here, we hypothesized that proteins sourced from brain tissue leakage would likely be elevated in the peripheral PMB and serum of children admitted to hospital with AHT, compared with atraumatic infants. We retrospectively explored blood proteome in AHT cases compared to child who died of sudden infant death syndrome (SIDS), via untargeted proteomics of serum samples collected before the infants' death at intensive care units (ICUs) and PMB specimens collected during medico-legal autopsies of the same infants.
In total, we analyzed 52 distinct specimens collected at two hospital units as ante-or PMB and serum samples corresponding to 26 infants, 7 of whom were victims of AHT. This study is an initial step toward development of clinical screening of ICH in AHT cases in young children.

Clinical samples
AHT cases were evaluated according to the guidelines of the French High Authority of Health [14]. The case-control retrospective study (Project ID 2021-01304) has been approved by the Research Ethics Committee of Geneva, Switzerland. Antemortem serum and blood collected in the ICU and PMB samples from AHT cases and atraumatic controls (infants died of SIDS) were collected from 2013 through 2018 in medico-legal context (Table 1). We analyzed a total of 52 samples,

Significance Statement
Nearly one-third of child victims of AHT with brain lesions was misdiagnosed during previous medical consultations.  Table 1). Proteins were isolated from the different specimens of peripheral blood and subjected to proteomic analysis.
To report the results of technical variations and for objective quality control analysis of clinical samples prior to any proteomics experiment, we used repeated sampling from ante-and PMB pairs of study participants and fresh blood from a single healthy volunteer (i.e., four experimental replicates of whole processes). We also performed five repeated injections from selected serum and blood samples across a measurement queue in the mass spectrometer (i.e., 10 technical replicates). In total, we performed 14 analytical replicates that include four "whole process" and 10 "MS-injection" replicates.

Proteomic analysis
Proteomic analysis of either 6 µL crude blood (post-and AM) or 4 µL serum samples was performed as described elsewhere [9].    [15,16]. For differential analysis, Spectronaut DIA proteomics experiments were created either for serum or PMB cohort. Quantitative data matrices for each sample type were generated with input of corresponding spectral library and respective DIA-MS raw data files. The proteins and peptide matrices were generated by default settings (i.e., BGS factory settings). Crossrun data normalization was performed on the whole data sets of PMB and serum based on global median normalization of precursor intensities ( Figure S1A). Normalized protein and peptide data were exported from Spectronaut software as csv. files for further analysis. Technical replicates from selected serum and blood samples display high positive correlation (Spearman's rho ≥ 0.9) across proteins quantified between the same patient samples injected randomly across a measurement queue in the mass spectrometer ( Figure S1B).

Analysis of generated MS data
Generation of Spectral library. MS files acquired in parallel in DDA mode (i.e., 33 raw MS files) were used to create a crude blood-or serum-specific spectral library. For creation of spectral libraries we used default parameters against the ex_sp_9606_decoy.fasta database (the reviewed canonical Swiss-Prot complete proteome database for human, released 2014-01-24, entries 40′544) appended with common contaminants, reversed sequence decoys and iRT peptides.
Included were trypsin digestion allowing two missed cleavages and 'Carbamidomethyl (C)' as static and 'Oxidation (M)' as variable modifications while minimum and maximum peptide length was set to seven and 52 amino acids, respectively. The mass tolerances were set to mode "dynamic" that is software determined tolerance based on extensive mass calibration and one-time correction factor was applied for precursor-and for fragment-ions. Unique proteins identified at 1% of protein false discovery rate (FDR) were included in libraries.

The generation of protein data matrices for group comparisons
The initial protein matrices corresponding to serum (i.e., 1051 Protein Groups and 7664 tryptic peptides) or PMB (i.e., 1133 Protein Groups and 10′504 tryptic peptides) were generated at 1% protein and peptide FDR based solely on tryptic peptide quantities. For data comparison we avoided imputation of missing values and the unique proteins identified at 1% of protein FDR were filtered from each data according to criteria that each protein (i.e., UniProt ID) be quantified in more than half of the samples per each respective condition (i.e., 4/7 and 10/19 samples, in AHT and in the control group, respectively). Based on these criteria, we found a relatively low percentage of missing values in total protein matrices, 13% and 7% of quantified data points in PMB and serum cohort, respectively ( Figure S1C). We quantified 862 and 518 confidently detected proteins across PMB and serum cohorts, respectively.

2.6
Quantification, statistical analysis, and data visualization R software for statistical computing and graphics (version: 3.6.1) was used for the data analysis and visualization. The computation of Spearman (Spearman's rho) or Pearson correlation coefficients (Pearson's r) between respective samples was performed with package ggplot2 (ver- For protein differential analysis, we first examined data distribution of the given data set by a generation of histogram plots in R and then performed statistical testing. We performed a 2-tailed Student t-test with significance set at nominal p < 0.05 for PMB and serum cohort. In order to adjust for sample size at minimum 80% of statistical power of the analysis, we estimated the minimal expected effect size measure (i.e., Cohen's d) for significant and upregulated proteins. The power of the analysis for each protein was estimated based on the sample size, degree of freedom, effect of size and statistical changes by R packages library(pwr) and library(rstatix). Due to protein variability potentially related to age differences, linear regression (LR) adjusted for infant age was performed only for selected brain-specific proteins upregulated in the serum or PMB.
Protein fold change (FC) was computed from a ratio of log2transformed value of protein intensities (i.e., Log 2 FC (AHT/Ctrl)).
Volcano plots were generated from log2-transformed FC and -log10 transformed p-values of proteins detected in the AHT and control comparison of specimens of interest. Boxplots of individual proteins were generated from log2-transformed protein intensities across tested samples.
The initial list of brain specific proteins was downloaded from the human protein atlas database [17]. Gene ontology (GO) candidate classification and data visualization was performed with clusterProfiler R package via DOSE (version 3.14.3). Manhattan distances were used for hierarchical clustering of log2-transformed relative protein intensities, while non-scaled original data were used in pheatmap R package (version 1.0.12). Proteins upregulated in the serum of AHT cases with statistical significance were analyzed by The KEGG (Kyoto Encyclopedia of Genes and Genomes) database [18] and KEGG map search results for "pathway" classifications with the highest number of mapped candidates (i.e., hsa04610) were exported in KGML format.
The network was visualized in Cytoscape [19] by using KEGGscape applications and pathway KGML format as input file.

2.7
Validation of Brain acid soluble protein 1 (BASP1) and Ectonucleotide pyrophosphatase/phosphodiesterase family member 2 (ENPP2) serum levels by ELISA assay For validation experiments, we prioritize the available sera of respective infants as the gold clinical specimen for quantification of laboratory markers [20]. The serum levels of AHT and atraumatic control infants were determined by ELISA commercial assays for two selected proteins with brain-enhanced expression (Human Brain

PMB reflects AMB proteome
We anticipated that PM specimens routinely collected upon medicolegal examination represent a valuable source of biological information for future research on infant victims of AHT [13] (Figure 1). We adapted the serum proteomic workflow as described elsewhere [9] to full blood specimens. Then, based on quantified proteome, we defined the window of postmortem delay (PMD) for blood samples eligible for proteomics.
We primarily proceeded with fresh blood from a single healthy volunteer, and which corresponds to the reference sample collected

PMB cohort analysis identifies proteins that distinguish AHT cases
The differential analysis of PMB proteomes highlighted 115 proteins (2-sided p ≤ 0.05, Student t-test) as differentially expressed in AHT cases compared with atraumatic controls ( Figure 3A, left), of which as many as 94 were upregulated in AHT cases (Table S1).
Differential changes were consistent with the clinicopathological picture of AHT or TBI [1,2,24,25], revealing enrichment of proteins of acute-phase response, complement activation, hypoxia, reduced oxygen transport, catecholamine uptake, and midbrain (GO analysis for Biological process, Fisher exact test, p < 0.01; Figure   3A, right).
To inspect for potentially abundant proteins related to brain trauma, we overlapped 115 differentially expressed proteins with the proteins showing brain-elevated expression. We obtained the list of organenriched proteins (e.g., brain-enhanced proteins, 2685, Figure 3A, pie chart) from The Human Protein Atlas (HPA) [17], to date, the most comprehensive resource of protein expression data across human tissues. We used The Brain Atlas (i.e., part of the HPA) containing protein profiles across 13 anatomically defined brain regions [26] and selected the subset of brain-enhanced proteins defining the genes that are either exclusively enriched in the brain or are expressed in the brain and very few other tissues (i.e., 2 to 5) with at least four-fold enhanced mRNA expression compared with other body tissues (Table S2). Remarkably, we found four proteins (Volcano plot, Figure 3A) with brain-enhanced expression, as upregulated in the PMB of AHT cases compared with atraumatic controls: BASP1

AM analysis of infant serum
We retrospectively analyzed the AM sera of the same study partici-  Figure 4A).
Among them, we detected 26 upregulated proteins (i.e., out of 40 PMB proteins that we again detected with upregulated trend, Figure   S2A) that maintain statistically significant upregulation in both PMB and sera cohort ( Figure S2A and Table S3). Importantly, 139 novel  (Table S3). For clarity of the results, we distinguish below two types of blood serum proteome changes related to infant AHT: (i) generic blood changes, and (ii) brain trauma-specific proteome changes.

Generic proteome changes in the serum of infant AHT
Previous studies demonstrated that the endogenous brain coagulation cascade and complement system (CS) as major part of innate immunity are instantly activated upon mechanical brain injury [25,27]. The activation is additionally enhanced by an influx of complement components from the peripheral circulation due to increased blood brain barrier (BBB) permeability [28].
There exists, however, a lack of systematic serum proteome profiling of AHT cases in infants. As shown in Figure 4A, we identified 165 upregulated and statistically significant proteins at a power level of 0.80 (i.e.,  Figure 4A-B-C and Figure S2B). Indeed, the study on animal models demonstrated that mannose binding initiators of the LP disrupt the BBB through formation of lytic MAC [29]. Moreover, LP components play a central role in adult cerebral contusions and associate with brain injury severity [30].
The highest protein difference in AHT infant serum compared with atraumatic controls was detected for Coagulation factor 5 (F5) (p < 0.0001, Cohen's d = 4.39, Figure 4A-B), known as activated protein C (APC) cofactor, proaccelerin or labile factor. F5 is an essential part of the prothrombinase complex (i.e., F10a-F5a) [31,32] with coagulation factor 10 (i.e., F10, Figure 4C) that results in increased formation of thrombin from prothrombin or coagulation factor 2 (i.e., F2, Figure 4B-C). Thrombin already in physiological concentration exhibits C5 convertase activity and cleaves C5 complement component [33] activating the MAC by an auxiliary route ( Figure 4C). Interestingly, a previous study revealed that a large fraction of F5 coagulation factor is secreted by the choroid plexus in the brain [34]. A number of previous studies also emphasized the importance of molecular crosstalk between coagulation and complement cascades in order to understand posttraumatic complications following TBI [35] and for the potential discovery of novel therapeutics [27].

Brain trauma-specific protein changes in the serum of AHT cases
Consistent with our initial hypothesis, we confirmed several brainenhanced proteins classified in the web-based HPA database [17] as upregulated with statistical significance in the peripheral circulation of AHT infants compared with atraumatic controls (i.e., ENPP2, CLU, MAN2A1, LANCL1, NCHL1, CRTAC1 in AHT serum, Figure 5A). Most of the brain proteins detected by our analysis were formerly related to the anatomical regions of brain, or to the mental and neurodegenerative disorders: Glutathione S-transferase LANCL1 (LANCL1) (p < 0.02, Cohen's d = 2.06) an essential protein in neuronal function [40], Clusterin (CLU) also referred to as Apolipoprotein J (p = 0.01, Cohen's d = 1.80) [41], and glycolytic enzyme Alpha-mannosidase 2 (MAN2A1) (p = 0.001, Cohen's d = 5.19) [42,43].
Remarkably, three serum upregulated proteins were enriched in the anatomical region of the fetal brain and, according to literature, evidence showed their involvement in brain development: Ectonucleotide pyrophosphatase/phosphodiesterase family member 2 (ENPP2), Neural cell adhesion molecule L1-like protein (NCHL1) and Brain acid soluble protein 1(BASP1).
ENPP2, known as Autotaxin (ATX), displays elevated concentration in the serum of AHT cases compared with atraumatic SIDS controls (p < 0.001, Cohen's d = 3.57, Figure 5B). It is a major enzyme involved in the synthesis of Lysophosphatidic Acid (LPA), a bioactive lipid essential in neuronal development, as best illustrated by the early embryonic lethality of Enpp2 null mice due to neuronal tube defects [44,45].
Similarly to ENPP2, Neural cell adhesion molecule L1-like protein (CHL1 or NCHL1) also known as a close homolog of L1 was found elevated in AHT serum samples collected upon hospitalization (p = 0.02, Cohen's d = 1.79) ( Figure 5C). CHL1 is a cell adhesion molecule of the immunoglobulin superfamily expressed by neurons and glial cells in all regions of the brain [47] with evidence of protein enrichment in the cerebral cortex [48]. It is implicated in: (i) nervous system development promoting axonal guidance and neurite outgrowth of cerebellar and hippocampal neurons [47,49] and (ii) synaptic plasticity and neuronal regeneration after brain trauma [49].  Figure S2A). Of note, we observed variability in BASP1 protein concentrations across the individual sera of AHT cases ( Figure 5D). To inspect if BASP1 protein levels can be influenced by the differences, in months, in the ages of infants detected between tested groups (p = 0.047, Figure S3A), we correlated the age and BASP1 levels across the serum and PMB of reference cases-atraumatic controls. Interestingly, our analysis demonstrated that circulating levels of BASP1 and even ENPP2 display rather negative correlation with infant age in months (e.g., Spearman's rho serum = -0.36, Figure S3B-C). This is in accordance with a recent study that showed that BASP1 displays a higher brain concentration during the development-, as opposed to adult phase [50] and even labels neuronal stem and progenitor cells (NPC) in the embryonic mouse brain similarly to known NPC markers, glial fibrillary acidic protein (GFAP) and vimentin. In line with this, the serum differences of BASP1 protein levels are more pronounced if the analysis accounts for infant age in months as a confounding factor (coefficient P (age) = 0.094) between atraumatic controls and AHT cases (linear regression p = 0.053, Figure S3D).
To validate our results by orthogonal analytical methods, we used commercially available ELISA tests of selected markers and the serum of respective infants ( Figure 5E). We investigated BASP1 and ENPP2, two proteins with strong literature evidence of protein expression in the developing brain. Remarkably, while both BASP1 and ENPP2 showed upregulation in AHT cases, BASP1 did not reach statistically significant results probably due to the individual variability of the samples related to patient characteristics, as observed with MS data ( Figure 5D). Moreover, the validation results of absolute serum concentrations of both markers were in strong accordance with our large scale MS proteomics data as demonstrated by significant positive correlation across individual cases (R (ENPP2) = 0.85; R (BASP1) = 0.6; Figure   S3E-F), indeed confirming higher levels of both proteins in the serum samples of AHT cases compared with atraumatic cases. F I G U R E 5 Brain specific proteins detected in sera of AHT cases. (A) Volcano plot visualizes serum proteins expressed in brainstem as gene names in red font. Brain-enhanced markers with serum enhanced expression depicted by red circles. (B-C) Boxplots of brain specific and significant serum markers, ENPP2 (B) and NCHL1(C). (D) Boxplots of brain-tissue specific BASP1 protein detected with upregulated trend in the serum (Cohen's d > 0.8) and statistically significant in initial PMB analysis (on the right). (E) Validation of brain-related markers of AHT in the serum by orthogonal method, ELISA test quantification

DISCUSSION
In the last two decades, very few untargeted proteomic studies that employed methods based on in-gel protein quantification analyzed CSF and serum samples of pediatric AHT cases [51,52], revealing a potential for novel discoveries. These non-hypothesis-driven studies are rare, and, to our knowledge, our current study is the first recent analysis to apply "state of the art" untargeted LC-MS based proteomics to characterize the blood proteomes of infants with AHT. Previous efforts related to serum biomarkers of pediatric TBI mostly involved successive targeted studies that monitored sets of 3-5 serum biomarkers across larger sample cohorts, leading to important clinical conclusions [53][54][55]. However, preselected brain-specific markers based on adult data might lack specificity in very young infants, probably due to the process of rapid CNS maturation and proteome variability in the postnatal period affecting the protein composition in the serum [54][55][56][57].
Here, we generated a unique data set of blood proteomes of infants under age two recorded concurrently across the cases of pediatric AHT and atraumatic controls by large scale proteome profiling. Our differential analysis of infant PM and corresponding AM samples delivered a set of 165 accessible serum proteins upregulated in AHT, of which 26 were also elevated in PM specimens, paving the way for the clinical screening of infant victims of AHT.
We observed that the several proteins upregulated in the serum of AHT cases (i.e., ENPP2, BASP1, LCAT) including the highly elevated F5 protein display an overlap of their CNS expression with the anatomical lesions found in the brain of infant victims of AHT, but not accidental head trauma [58,59].
For example, recent data on BASP1 protein levels in PM human brain demonstrated the highest expression in the hippocampus, brainstem, and spinal cord [50], the CNS regions that display known lesions in AHT cases [58]. Along the same lines, ENPP2 displays the highest CNS expression in the corpus callosum, spinal cord and pigmented layer of the retina [60]. As mentioned above and in line with data in the HPA, F5 is dominantly expressed in the corpus callosum, the pons, and the medulla oblongata, brain regions proximate to the spinal cord, and which frequently report injuries in AHT but not in accidental head trauma in infants [2,58,59].
Moreover, we speculate that permeability of the BBB is potentially induced by the activated MAC [29] through high F5 serum levels sourced from injured neuronal regions, which results in increased formation of thrombin. This, in turn, leads to cerebral accumulation of components from the peripheral circulation, increasingly worsening the clinical picture of those infants.
Noteworthy, however, those proteome changes related to coagulopathy are rather common to all traumatic injuries and represent generic changes unspecific to AHT cases in young infants. By contrast, the serum of infants with AHT reflects proteome changes related to midbrain and brainstem (e.g., F5, BASP1, ENPP2, LCAT) and supports the statement that infant victims of abusive injuries by shaking are subjected to a specific pattern of forces [58,59]. The limitations of our study are the modest size of the cohorts and the absence of accidental ICH cases such as are required to better estimate the proteome characteristics specific to AHT in infants with acute ICH.
Curiously, ENPP2, also enriched in brainstem, display increased protein concentration in serum of AHT cases upon hospital admission, but was almost undetectable in PMB samples, suggesting a period of protein short detectability following trauma and opening new possibilities for research on living patients. Former studies associated ENPP2 to brain trauma [61,62] and some indeed reported a transient ENPP2 increase in the brain lesions affected by trauma [63] followed by reduced mRNA levels in PM brain samples hours after injury [60,64].
This short window of serum detectability of only several hours after TBI was seen previously for ENPP2 [64] and also for other TBI markers [57].
Here, we also performed an extensive quality control analysis on PMB, comparing proteomes from PM and AM specimens taken from the same subjects. Our study strongly encourages the future investigation of PM specimens from ICH cases in infants; first, because of the lack of proper animal models to study TBI in infants, and second, because we found sufficient proteome stability for all PMB specimens stored at 4 • C within 72 h of PMD. These unique proteome datasets of AHT cases and atraumatic controls together represent a remarkable opportunity to improve prognosis for these infants by early detection of AHT and medical care, and could be explored further in larger prospective studies.

ASSOCIATED DATA
Raw data files and individual spectral libraries (.kit file extension) are available via ProteomeXchange with identifier PXD028801.