Profiling of immune‐related gene expression in children with familial hypercholesterolaemia

Innate and adaptive immune responses are pivotal in atherosclerosis, but their association with early‐stage atherosclerosis in humans is incompletely understood. In this regard, untreated children with familial hypercholesterolaemia may serve as a human model to investigate the effect of elevated low‐density lipoprotein (LDL)‐cholesterol.

Background. Innate and adaptive immune responses are pivotal in atherosclerosis, but their association with early-stage atherosclerosis in humans is incompletely understood. In this regard, untreated children with familial hypercholesterolaemia may serve as a human model to investigate the effect of elevated low-density lipoprotein (LDL)-cholesterol.
Objectives. We aimed to study the immunological and inflammatory pathways involved in early atherosclerosis by examining mRNA molecules in peripheral blood mononuclear cells (PBMCs) from children with FH.

Methods.
We analysed the level of 587 immunerelated mRNA molecules using state-of-the-art Nanostring technology in PBMCs from children with (n = 30) and without (n = 21) FH, and from FH children before and after statin therapy (n = 10).

Introduction
Coronary heart disease (CHD) is a main contributor to the global burden of disease [1]. It is primarily caused by atherosclerosis where a complex interaction between lipids and inflammation is the pathological hallmark [2]. Innate immune responses are pivotal through for example the formation of foam cells from lipid-exposed macrophages, which in turn activate cells in the adaptive immune system, such as T and B cells. These cells may again activate innate immunity † Contributed equally. through for example release of various inflammatory cytokines. All these subsets are important cell types in atherosclerosis [2]. Peripheral blood mononuclear cells (PBMCs), which comprise lymphocytes and monocytes, may thus be considered a unique niche since they are primed by atherogenic factors and directly contribute to the atherosclerosis process. Furthermore, previous studies have shown that alterations in PBMC gene expression level may reflect systemic health [3] and have suggested PBMCs to be a good model system to elucidate the cardiovascular system in relation to inflammation [4].
Due to the nature of CHD, most studies on inflammation and immunity in atherosclerosis are performed in adult patients with established atherosclerosis [5][6][7]. However, these studies may be confounded by co-occurring disease or risk factors, or environmental exposures. Thus, findings in these studies may reflect secondary phenomena and not the molecules that promote the development of atherosclerosis and its complications. It is therefore of major interest for future prevention of CHD to identify and map molecules that could play a pathogenic role in the early stages of this disorder, using study populations free from severe confounding. To this, subjects with familial hypercholesterolaemia (FH) may represent an optimal population.
FH is an inherited disorder caused primarily by a mutation in the gene encoding the low-density lipoprotein (LDL) receptor [8]. Phenotypically, FH individuals have higher total-and LDL-cholesterol (LDL-C) compared to the general population, leading to accelerated atherosclerosis and premature CHD [9]. Accordingly, nonstatin-treated children with FH may serve as a human model to investigate the isolated effect of LDL-C on early development of atherosclerosis caused by a mutation rather than secondary hypercholesterolaemia induced by, for example lifestyle. Indeed, it has been shown that FH children have increased carotid intima-media thickness (IMT) [10] and display signs of inflammation [11] at both protein and mRNA level, including activation of certain pathways such as members of the tumour necrosis factor super family (TNFSF) [10,12]. However, these issues are far from clear and to this end; to the best of our knowledge, no studies have shown comprehensive mapping of the multitude of inflammatory and immune-related pathways that could be involved in the early stages of atherosclerosis in children with FH.
Herein, we aimed to profile the immunological and inflammatory pathways involved in early atherosclerosis in PBMCs from children with FH compared to healthy children.

Subjects
The subjects have been described previously [12]. Briefly, children with FH were recruited from the Lipid Clinic, Oslo University Hospital, Norway. Ageand sex-matched (by median and ranges) healthy control children were recruited through colleagues, employees and friends at the Lipid Clinic and the Department of Nutrition, University of Oslo, Norway. Common inclusion criterion for the two groups was age 18 years or younger. Common exclusion criteria for the two groups were any chronic disease (including inflammatory or metabolic disorders), or use of anti-inflammatory drugs or lipid-lowering therapy (LLT) (at first visit). A nonfasting blood sample was obtained from each individual. Additionally, nonfasting blood samples were available before and after statin treatment in ten FH children, also previously described [12]. The study was approved by the Regional Committee of Medical Ethics and was conducted according to the Declaration of Helsinki. Written informed consent was obtained from all participants, or from one of their parents if the child was below the age of 16 years.

Gene expression analyses
PBMCs were isolated, and RNA quantity and quality were analysed as previously described [12]. We ran RNA expression analysis on the nCounter â analysis system (Nanostring Technologies), running 12 samples at a time (one 'strip'). The procedure was performed according to the manufacturer´s instructions, applying about 100 ng mRNA. The kit used was a fixed code set for mRNA analysis that comprised genes involved in human immunology: nCounter GX Human Immunology Kit v2 (Nanostring Technologies). The number of mRNA molecules per gene was accounted for detection level average ðnegativecontrolÞ þ 2SDðnegativecontro lsÞ= 10 mRNA molecules, normalized against instrument variations (positive controls) and endogenous control genes found to be stable (ABCF1, EEF1G, HPRT1, RPL19, TUBB, GUSB and POLR2A) using nSolver analysis software 2.5.34 (NanoString Technologies).

Miscellaneous
Oxidized LDL (oxLDL) was measured in serum using an enzyme-linked immunosorbent assay as previously described [12]. Standard blood biochemistry analyses were measured in plasma or serum by standard methods at the accredited medical laboratory, Oslo University Hospital, Rikshospitalet, Oslo, Norway (NS-EN ISO 15189:2007). Not all analyses were performed in all subjects because of the limited volume of blood sample available from each individual.

Statistical analyses
We imported the raw gene expression data (594 genes) into Partek Genomics Suite 6.6, and batchcorrected for strip, age and sex. In principal component analysis (PCA), one individual was a clear outlier and was thus excluded from the statistical analyses. Statistical power analysis based on the entire dataset showed that the minimum sample size for this study was 17 at significance level 0.01 and power 0.90. After exclusion of the used endogenous controls (n = 7, see gene expression analysis section), we performed a 4-way analysis of variance (ANOVA) model using log2-transformed data to test for statistical significance between FH children and healthy children and adjusted for multiple testing using false discovery rate (FDR). FDR-adjusted Pvalue < 0.05 was considered significantly different between the groups. Additionally, based on data from the 4-way ANOVA as input, we ran standard linear regression models for each gene using batchcorrected raw data in R version 3.5.0 [13]. We ran three models per gene, with 1) no covariates, 2) HDL-C and CRP as covariates, or 3) LDL-C as covariate. Also, to further examine associations between gene expression and clinical and biochemical variables, we used Spearman's rank correlation.
For the main cohort, we allocated differentially expressed genes (DEGs) into different immunological pathways using two strategies. First, we manually allocated FDR-adjusted genes to pathways based on their function; main pathways were defined as having ≥ 5 DEGs. Information about gene function was obtained from the following databases NCBI (https://www.ncbi.nlm.nih.gov/ pubmed/, main resource), STRING (https:// string-db.org), GeneCards (https://www.genecard s.org), UniProt (https://www.uniprot.org), OMIM (http://omim.org/) and Ensembl (http://www.e nsembl.org/). Secondly, we created clusters of genes (not FDR-adjusted) in an unsupervised manner; we calculated the Euclidean distance of the correlation matrix for all genes, performed hierarchical clustering (complete linkage) and cut the dendrogram into eight clusters.
We estimated the relative proportion of blood cell types in each sample with the CIBERSORT algorithm [14]. The algorithm takes a whole-genome gene expression matrix as input and returns the relative proportion of 22 leucocyte subtypes by use of feature selection and linear support vector regression (SVR). Because we supplied the algorithm with a limited number of genes (n = 587), the sensitivity of this analysis was expected to be lower than by use of whole-genome datasets. Despite this, all 51 samples had adequate goodness of fit (P < 0.05). Since the PBMC pool should be completely absent neutrophils, we used the estimated proportion of neutrophils as an indicator to remove low-proportion leucocyte variables; hence, we included mainly estimated monocyte and lymphocyte subtypes.
In a subsequent analysis in ten FH children that were not part of the main FH cohort, we analysed gene expression before and after starting statin treatment. The analysis pipeline (gene expression and statistical analyses) was similar as for the main cohort, described above. As for the main cohort, we estimated the change in gene expression (for all genes) corrected for age, sex and strip upon statin therapy. Then, we investigated the association between the two sets of coefficient estimates (for each of the 587 genes): (i) the association between FH and healthy children (exposure variable) and (ii) the change upon statin therapy (outcome variable). We interpreted this association as whether or not statins completely or partially reversed the gene expression differences observed in FH children compared to control children.
Baseline characteristics are given as median (25thÀ75th percentile), mean (standard deviation [SD]) or frequency (%). Mann-Whitney U test or Independent sample t-test was used to test for statistically significant differences in baseline characteristics between the two groups. We used Wilcoxon test to test characteristics before and after statin initiation.

Characteristics
The characteristics of the participants are shown in Table 1. Thirty children with and twenty-one without FH were included in this substudy. All the children with FH had a genetically verified FH diagnosis. Total cholesterol, LDL-C, apolipoprotein (apo) B, oxLDL and C-reactive protein (CRP) were higher (P < 0.001), whereas high-density lipoprotein cholesterol (HDL-C) was lower (nonsignificant, P = 0.059), in the FH children compared to control children.
In order to explore the leucocyte distribution, we predicted the different leucocyte subtypes using a CIBERSORT analysis based on all the mRNA expressions from the nCounter analysis. Children with FH had higher levels of predicted na€ ıve B cells (P < 0.01), but lower levels of predicted monocytes (P < 0.05) and resting mast cells (P < 0.01) compared to healthy children (Fig. S1).

Regulation of immune-related genes after statin therapy
In order to investigate the effect of statin therapy on immune-related genes, we analysed PBMC gene expression before and after statin therapy in ten children with FH. The decrease in total-and LDL-C were 38.5 (34.5-41.4) % and 46.3 (40.6-48.9) % (P < 0.001 for both), median (25th-75th percentile), respectively [12]. In total, 62 genes (FDR < 0.05) were modified by statin therapy (Fig. S2). Figure 3 shows the relationship between differences in gene expression in FH compared to control children and change after statin therapy in FH children. Of the sixty-two genes, ten genes that were higher in FH vs controls at baseline, were reduced after statin initiation IRAK4, ATG16L1, CUL9, IKBKB, CRADD, TRAF5, ENTPD1, C1R, IL1RL1 and CTLA4 (Fig. 3). In contrast, CTSS, PYCARD and C14orf166 were lower in FH vs controls at baseline and increased after statin therapy (Fig. 3). Overall, statins thus seem to partially reverse the DEGs observed in FH children in line with the expression observed in healthy children.

Associations between gene expression and clinical data
To further investigate associations between DEGs in the main pathways and clinical data, we calculated the inter-variable correlation coefficients (Fig. 4). The apoB-containing lipoprotein parameters and CRP were mainly correlated to the immune-related gene expressions. Whereas we found mostly positive correlations between total cholesterol, LDL-C, oxLDL, apoB and CRP, and DEGs related to the pathways: T-cell signalling, Bcell signalling, TNFSF signalling, and cell growth, proliferation and differentiation (Fig. 4), the DEGs related to antigen presentation, apoptosis and TLR pathways were mostly negatively correlated to apoB-containing lipoprotein parameters and CRP (Fig. 4). There were only few correlations between the DEGs and age, HDL-C, apoA1, triglycerides and glucose (Fig. 4).
In addition to LDL-C, CRP and HDL-C were different between the two groups of children (Table 1), thus to investigate the impact of these biochemical variables, we performed a linear regression adjusting for either HDL-C and CRP combined, or LDL-C alone, for all DEGs (n = 176) (Fig. S3). The significance of the DEGs was attenuated or neutralized only when adjusting for LDL-C, but not when adjusting for HDL-C and CRP, suggesting that LDL-C is the main driver for the observed alterations (Fig. S3).
In order to show a direct correlation between LDL-C and mRNA levels, we show a strong correlation in mRNA levels when correlating the difference per mmol/L difference in LDL-C and difference in FH children compared to healthy children (Fig. 5).

Discussion
In the present study, we found that FH children display several changes in mRNA molecules related to pathways including, for example T-and B-cell signalling, TNFSF, interleukin-signalling and cell growth, proliferation and differentiation, compared to healthy control children. Interestingly, the DEGs reflected both anti-inflammatory (e.g. transforming growth factor-b-related molecules) and pro-inflammatory (e.g. TNFSF and various chemokines) molecules. Furthermore, LDL-C, as expected, seems to be the main driver of these alterations further underscoring the role of LDL-C as a trigger  of inflammation also in the early stage of atherosclerosis, before any overt disease.
Human atherosclerotic plaques contain many immune cells, such as monocytes/macrophages, antigen-presenting cells, and T and B lymphocytes, of which monocytes and macrophages are the most common cell type [15]. Recently we showed that children with FH were characterized by monocytosis with a shift towards more circulating proinflammatory/nonclassical monocytes when compared to healthy controls, supporting a central role of monocytes/macrophages in early atherosclerosis [11]. Furthermore, in the same study, we found a tendency to difference in distribution of certain Tcell subsets [11]. Herein we further extend this finding by showing an upregulation of genes related to T-and B-cell signalling in PBMCs from children with FH compared to control children, suggesting an involvement of these cells in an early stage of atherogenesis. Moreover, whereas mRNA molecules related to antigen presentation, apoptosis and TLR pathway were mostly negatively correlated with lipid parameters and CRP, an opposite pattern was seen for genes related to T-and B-cell signalling suggesting that these pathways are differently related to lipid parameters and systemic inflammation.
The intracellular signal transducers of the TNF receptor-associated factor (TRAF) family regulate cell survival, proliferation, differentiation and activation in addition to cytokine production and autophagy [25]. Pathways activated through CD40 or BAFFR via TRAF1/2/3/5/6 may subsequently activate mitogen-activated protein kinases or trigger BCL2 transcription, respectively, in B cells [25]. Regarding prevention of atherosclerosis, targeting TRAF6 through CD40 has drawn attention [26], but enhanced expression of TRAF1/2/3/ 5 in human atherosclerotic plaques also indicate a role of these TRAF members in atherosclerosis [27]. We found higher expression of BAFFR (TNFRSF13C), TRAF1/3/5, MAP4K2 and BCL2 in children with FH, possibly suggesting LDL-C-induced activation of several TRAF members and pathways in PBMCs, and again, some of these could be targeted therapeutically [26].
Previously, we and others have shown increased systemic inflammation in children with FH, in particular, an imbalance between TNFSF members and IL-10 [28], CRP [29,30], P-selectin [29] and neopterin [30,31]. Indeed, using a single-gene approach, we have previously observed altered gene expression of certain TNFSF-related genes in the same children with FH compared to control children [12]. However, a holistic approach to map not only single genes but a wide range of candidate genes is important to expand our understanding of atherosclerosis and is of clinical relevance in the prevention of CHD. The present study is to the best of our knowledge the first study to perform a comprehensive analysis of immune-related gene expression in young FH children, representing an early stage of atherosclerosis without any overt symptoms, compared to healthy children, contributing to characterize the early inflammatory profile in this disorder.
Our multivariate analyses suggest that LDL-C is the main driver of the different expression of immune-related genes in FH children compared to control children. Indeed, ten genes were reduced and three were increased after statin initiation, indicating normalization towards the expression level in healthy controls. This further supports a crucial impact of LDL-C on these immune genes. In line with others, we have, in the same FH children, shown changes in inflammatory molecules at gene expression and circulating level after statins [12,30] or apheresis [32] in children with FH. In contrast, we have previously shown a lasting presence of low-grade inflammation in adults with FH despite long-term LLT [33], which questions whether statins or other LLT completely can reverse the detrimental effects of elevated LDL-C on the immune system for many years in childhood. Nonetheless, our findings further support a role for LDL-C as a trigger of inflammation and show that statins could modulate several pathways that are dysregulated in PBMC from FH children. Most data suggest that the anti-inflammatory effect of statins is mediated through the lipid-lowering effect [34][35][36]; however, anti-inflammatory effects of statins cannot be ruled out. Hence, if these effects in FH children are secondary to downregulation of LDL-C, direct immunomodulatory effects or a combination thereof is at present not clear.
The major strength of this study is the comprehensive analysis of about 600 immune-related mRNA molecules in PBMCs by using state-of-the-art Nanostring technology, and subsequent statistical and bioinformatic analyses. Secondly, genetically verified FH children unbiased by LLT are suitable for studying the pathophysiological role of isolated, elevated levels of LDL-C. The main limitations are the relatively low number of subjects, which increases the probability of type 1 and 2 errors, and lack of measurement of subclinical atherosclerosis, for example IMT. Furthermore, a previous study has shown that neither leucocyte, lymphocyte nor monocyte count did differ between FH subjects and healthy controls [37]; however, we have recently shown altered distribution of leucocytes in FH children [11]. Moreover, our CIBER-SORT analysis shows a slightly different distribution in some predicted subtypes; thus, another limitation is the lack of leucocyte count as we cannot account for confounding association with leucocyte subset. Since PBMCs are exposed to atherogenic factors and are pivotal in the atherosclerotic process, we have used these cells as a model system to investigate inflammation in atherosclerosis. However, this model does not necessarily reflect the in vivo situation and other methods could have been used, such as macrophage stimulation in vitro.
In conclusion, FH children display higher PBMC expression of immune-related genes mapped to several pathways, including T and B cells, TNFSF and cell growth, proliferation and differentiation than healthy children. These novel data support the notion that several immune cells and related pathways play a role in the early stage of atherosclerosis before any disease manifestation. Our results suggest that LDL-C plays a key role in modulating expression of different immune-related genes. Some of the genes that were characterized in PBMC from FH children could potentially also represent future therapeutic targets for prevention of atherosclerosis progression.

Supporting Information
Additional Supporting Information may be found in the online version of this article: Figure S1. Predicted leucocyte subtypes using CIBERSORT analysis based on all the mRNA expressions from the nCounter analysis in children with (n = 30) and without (n = 21) FH. Figure S2. Differential gene expression before and after statin initiation in FH children (n = 10) at P < 0.05. Figure S3. Association between fold difference and nominal P-value for all genes in FH (n = 30) and healthy children (n = 21).