**Abstract**

Astrocytes are critical for the metabolic, structural and functional modulatory support of the brain. Lipotoxicity or high levels of saturated fatty acid as Palmitate (PA) has been associated with neurotoxicity, the loss or change of astrocytic functionality, and the etiology and progression of neurodegenerative diseases such as Parkinson or Alzheimer. Several molecular mechanisms of PA's effect in astrocytes have been described, yet the role of epigenetic regulation and chromatin architecture have not been fully explored. In this study, we developed a multi-omic epigeneticbased model to identify the molecular mechanisms of lipotoxic PA activity in astrocytes. We used data from nine histone modifications, location of Topological Associated Domains (TADs) and transcriptional CTCF regions, where we identified the basal astrocyte epigenetic landscape. Moreover, we integrated transcriptomic data of astrocytic cellular response to PA with the epigenetic multi-omic model to identify lipotoxic-induced molecular mechanisms. The multi-omic model showed that chromatin conformation in astrocytes treated with PA have response genes located within shared topological domains, in which most of them also showed either repressive or enhancing marks in the Chip-Seq enrichment, reinforcing the idea that epigenetic regulation has a huge impact on the lipotoxic mechanisms of PA in the brain.

**Keywords:** epigenetic landscape, lipotoxicity, inflammation, astrocyte-neuron interaction, neurodegeneration

#### **1. Introduction**

Obesity is referred to as the excessive accumulation of body fat. It has become a worldwide public health issue which several studies have linked hormonal impairment to other diseases like coronary pathologies, diabetes, hypertension, atherosclerosis, and certain types of cancer among others [1, 2]. Studies using insulin growth factor-1 (IGF-1) receptor, insulin receptor substrate-1 (IRS-1), insulin receptor substrate-4 (IRS-4), glial fibrillary acidic protein, as well as an increase in β-actin protein have been associated with fatty acid excess in the brain [3]. Additionally, recent evidence has linked adiposity and high fatty acid concentrations to significant brain region-specific dysfunction, atrophy, inflammation, and cognitive decline [2, 4, 5], as well as an increased risk in developing the accumulation of amyloid β and Tau associated with Alzheimer's disease [3].

Astrocytes are the most versatile glial cells in the central nervous system (CNS) constituting from 20 to 40% of neuroglia, protecting the brain through so many signaling [6], demonstrating that these cells effectively engulf dead cells, synapses and protein aggregates of amyloid β (Aβ) and ɑ-synuclein, typical of Alzheimer's disease (AD) and Parkinson's disease (PD), respectively. Additionally, astrocytes have been shown to regulate K+ levels [7] and prevent excitotoxicity in Huntington's disease (HD) [8–10]. Nonetheless, evidence suggests that elevated concentrations of fatty acids can trigger a pro-inflammatory response altering the correct functioning of astrocytes [11–13]. Recently, authors have proved that metabolic insults produced by fatty acids can trigger a pro-inflammatory response in astrocytes, due to their high recruitment and metabolic capacity. Among them is PA, a long-chain saturated fatty acid, that can trigger an increase in inflammatory cytokines [5] such as Interleukin (IL)-1B, IL-6 and tumor necrosis factor alpha (TNFα), leading to accelerated cognitive decline, decreased cell viability, increased endoplasmic reticulum stress, inhibition of autophagy, finally compromising the Blood–Brain Barrier (BBB) integrity and promoting dementia-like progression in humans and animal models [5, 7, 14, 15].

Recent evidence supports epigenetic responses in astrocytes followed by PA-lipotoxic exposure [10, 16]. Epigenetic transcriptional regulation such as chromatin accessibility by histone modifications and chromatin architecture modulate euchromatin/heterochromatin equilibrium has shown the great potential of providing groundbreaking insight into the effects of neurotoxic compounds such as PA [4, 17, 18]. Additionally, the epigenetic modulation in astrocytes produced by lipotoxic compounds like PA can trigger inflammation, neurotoxicity, astrocyte reactivity, and cell fate determination in the CNS [16, 19]. In this case, the epigenetic landscape regulatory role and its response in the PA-induced astrocyte lipotoxicity are both the key to comprehend the loss of cellular function. Furthermore, several authors have also demonstrated that multi-omic models have proved to be more efficient than conventional astrocytic models in the evaluation of non-linearity in chromatin regulation considering regulatory mechanisms such as enhancers, isolators, epigenetic marks, and non-coding RNA [20–22].

It has been demonstrated that epigenetic data such as Chip-Seq and Hi-C with transcriptomics allows the detailed identification of specific molecular mechanisms associated with impairment conditions. In the present study, we report a multiomic model to describe the epigenetic baseline of astrocytes as well as the astrocytic response to PA-lipotoxicity over specific astrocytic processes such as inflammation, autophagy, endoplasmic reticulum stress, energetic metabolism, mitochondrial dysfunction, and astrocyte-neuron interaction pathways, herein described here as astrocytic PA response (APAR) mechanisms.

## **2. Materials and methods**

#### **2.1 Hi-C, ChiP-Seq and transcriptomics datasets acquisition**

Hi-C has been adopted as a method to measure pairwise contacts between pairs of genomic loci and allows a mapping of the three-dimensional conformation of

*Multi-Omic Epigenetic-Based Model Reveals Key Molecular Mechanisms Associated… DOI: http://dx.doi.org/10.5772/intechopen.100133*

chromatin within a population of cells, as well as to detect the structural variation and corrects assembly of chromosomal missed junctions [23]. Chip-Seq data also allows the analysis of histone marks interaction with DNA in an activation/repression mechanism. In this study, we analyzed nine treatments which were controls, H3K36me3, H3K27me3, H3K9ac, H3K9me3, H3K79me2, H4K20me1, H3K4me1 and CTCF (entry: GSM733678, GSM733751, GSM733729, GSM1003534, GSM1003491, GSM1003490, GSM1003525, GSM733710 and GSM733765 respectively). Tissuespecific datasets for astrocyte Hi-C from cerebellum and spinal cord were downloaded from the Encyclopedia of DNA Elements (ENCODE), as part of the ENCODE project consortium with ID numbers 200105194 and 200105957, respectively [24–26]. From ChiP-Seq data we obtained nine astrocyte datasets from NHA cells culture from the ENCODE database. Moreover, the whole human genome GRC version hg19 was obtained from ENSEMBL (https://www.ensembl.org/index. html) to map and enrich all the datasets. Transcriptomic data was experimentally obtained in the laboratory of Experimental and Computational Biochemistry of the Pontificia Universidad Javeriana, Bogotá D.C, Colombia.

#### **2.2 Transcriptomic data**

We used Normal Human Astrocytes (NHA; Lonza, CC-2565) divided in three different batches (#0000612736, #00005656712, #0000514417), which were cultured according to the manufacturer's specifications. All batches were cultured in a supplemented medium with SingleQuots supplements. In order to induce PA toxicity, NHA cells were seeded in 48, 24, 12, and 6-well plates and incubated in a humidified incubator for 12 days at 37°C and 5% CO2. Then the NHA cells were washed with PBS 1x and starved in medium with serum-free DMEM without phenol red, L- and supplements (Lonza, Basel, Switzerland) for 6 h.

RNA extraction was performed using mini kit RNeasy (Qiagen, USA). The RNA quantification of the preparations was performed with NanoDrop (Thermo Fisher Scientific, Waltham, Ma, 174 USA). To remove possible DNA contamination, RNA was treated with DNase I. The RNA integrity (RIN) and 28S/18S ratio were determined with the RNA Nano 6000 Assay Kit of the Agilent Bioanalyzer 2100 system (Agilent Technologies, CA, USA). Transcriptomic datasets were obtained for NHA astrocytes treated in DMEM medium. RNA-seq libraries were prepared using the TruSeq Stranded mRNA library prep kit following the manufacturer's protocol (Illumina, Cat# RS-122- 2101) [27, 28]. The RNA-seq libraries were sequenced in HiSeq platform (Illumina) using protocol 2x150bp paired-end configuration, single index per lane. Scores and nucleotide composition were assessed with FastQ to evaluate accuracy using the Nextflow (V18.10.1) pipeline QUARS (https://github.com/TainVelasco-Luquez/QUARS).

Salmon package was used for mapping and quantifying the expression level (https://combine-lab.github.io/salmon/, V0.13.1) on the genome assembly GRCh38 (patch 12) from the NCBI without ALT regions using Gencode [26, 29]. NOISeq was used to import data into R (V3.6.1) and assessed sequence plot quality diagnosis [5]. Gene level Ensembl IDs were used with tximport function to create the count matrix. We used DESeq2 for modeling the average expression in function of the treatments correcting for sex (design formula: ~ sex + condition) [30]. Moreover, DESeq2 was used for normalization by size, variance shrinkage, outliers filtering, and hypothesis testing. The Wald test was used for assessing genes differentially expressed above |LFC| > = 0.5 with an alpha cutoff at 0.05.

The overlap analysis was performed using Fisher's exact test using alpha at 0.05 implemented in the package GeneOverlap. To correct for multiple testing, p-values were adjusted using the Benjamini-Hochberg method [31]. Additionally, we leveraged the GeneOverlap's Odds Ratio and Jaccard Index as measures of the strength of association and the similarity, respectively. Odds ratio equal to or less than 1 means no associations and greater than 1 represents strong associations. Jaccard Index is a measure of similarity that varies between 0, no similarity, and 1, completely identical lists. The package WGCNA from the platform iDEP (V0.90) was used to perform the co-expression analysis, employing the 1000 more variable genes across all samples, with a soft threshold of 16 and minimum module size of 20 [32]. The Pearson's correlation was used on the count matrix, normalized and regularized using the log transformation of the DESeq2 library.

## **2.3 Epigenetics data analysis**

To create the multi-layer model, first we obtained the ChIP-Seq data of an astrocyte in homeostatic conditions from the ENCODE database, then re-analyzed the BED/BAM files using ChIP-Seq model-based analysis implemented in MACS2. We integrated H3K4me3, H3K27ac, H3K27me3, H3K9ac, H3K4me1, H4K20me1, H3K36me3, and H3K79me2 to the human genome, in order to identify the core active regulatory and repressed genes in astrocyte, H3K27me3, H4K20me1 and H3K9me3. The active genes were also identified by the integration of H3K4me3, H3K4me1, H3K79me, 2H3K9ac, H3K27me3, and H3K27ac [33] to the same genome (hg19) and those who shared both repressing and enhancing modifications were identified as bivalent genes. All the individual samples, the core activation, repression and bivalent samples were enriched using the cutoff p-value set at 0.01 for both molecular function and biological process excluding redundant gene ontology (GO) terms.

Hi-C data was obtained from the ENCODE database. In this database, it can be found up to 80% of the annotated genome, in which for the interest of the investigation cerebellum and spinal cord data with ID numbers 200105194 and 200105957 respectively [24]. Hi-C data of spinal cord and cerebellum were compared to identify the potential tissue-specific differences in astrocyte functioning. All the individual samples, the core activation, repression, and bivalent samples were enriched using the cutoff p value set at 0.01. Molecular enrichment was performed using ShinyGO [34].

#### **2.4 Data integration**

In order to identify euchromatin and heterochromatin regions in astrocyte, we overlapped all activation/repression specific histone modifications covering regions. With this approach, it was possible to identify activation, repression and bivalent core genomic regions. Thus, both the omic and epigenetic integrations were performed through the adjudication of the data described above into a multi-omic model. The model consisted in three different layers of the Chip-seq and Hi-C of an astrocyte under homeostatic conditions and the transcriptomic data of an astrocyte under the lipotoxic effects of PA, where these three layers were used to make an inference about the possible epigenetic effects of PA in an astrocyte. Considering that there are no Hi-C or Chip-Seq data for astrocytes under the effects of PA, the integration of the transcriptomic data allows the identification and analysis of genes associated with PA response. Accordingly, the identification of changes in the epigenetic regulation of genes was performed as follows: first the differentially expressed genes in the transcriptome presented within the TADs of euchromatin and with the groups of genes were identified. Then, the proteomic data was sought to identify whether gene expression patterns are correlated with histone modification data sets [35–37]. Also, both the core regions and the specific histone marks with the Hi-C were overlapped with the TADs of an astrocyte in order to identify patterns between chromatin architecture and modulation of histone expression [37, 38]. Later, the Chip-Seq core regions integrated with TADs were compared with gene expression and proteomic data.

*Multi-Omic Epigenetic-Based Model Reveals Key Molecular Mechanisms Associated… DOI: http://dx.doi.org/10.5772/intechopen.100133*

To identify the role of chromatin conformation in the expression and the effect of TADs in PA activity, gene expression and gene localization were associated with each other. This approach allowed us to identify additional epigenetic regulatory events related with TAD genes [39]. Subsequently non-coding regions such as enhancers and promoter regions were identified to be able to explain the patterns of expression of PA lipotoxicity. All data analysis was developed using R-Bioconductor suite (https://www.bioconductor.org/) as well as publicly available databases to ensure reproducibility and robustness. All the resulting sets of genes were enriched for molecular processes and biological functions.

## **3. Results**

#### **3.1 Chromatin/Histone expression regulation**

Considering the functional importance of histone modifications in the cellular behavior [40], we identified a set of 34852 genes with known activation roles across the seven ChiP-Seq samples. Histone marks H3K4me3, H3K27ac, H3K9ac, H3K4me1, H4K20me1, H3K36me3, and H3K79me2 and the number of genes per activation mark were 2432, 3034, 3773, 5133, 8228, 6766 and 5486, respectively, with a non-homogenous pattern. We obtained a set of 11214 genes enhanced between the 7 studied active histone modifications with at least each gene included in two or more of the samples. Moreover, samples H3K27me3 and H3K9me3 showed 6276 and 6747 repressed genes, respectively. We identified a set of 9796 genes repressed in astrocytes based on H3K27me3 and H3K9me3 data with the condition that each gene should be present in at least one of the ChiP samples. Considering that a bivalent region is due to the presence of a repressor and an enhancer in the modifications of histones H3K4me3 and H3K27me3, we identified 7608 genes in bivalent sites. Moreover, shared genes for both activation and repression were identified as coding regions present in at least one of the mark datasets in each group for the specific markers [33].

In general, we performed a functional enrichment of the dataset where it was possible to identify the biological process and molecular functions associated with GO terms. As a result, 30 biological processes relevant to APAR biological mechanisms were presented (**Table 1**). Among these biological processes identified for the activation of ChIP-Seq datasets, glutamate-cysteine ligase activity, CD4 receptor binding, ion channel binding, and extracellular matrix binding were the top-enriched functions. Besides, functional groups associated with the homeostatic astrocytic activity were identified as highlighting transferase activity, carbohydrate derivative binding, hydrolase activity, DNA-binding transcription factor activity, molecular function regulator, transporter activity, oxidoreductase activity, enzyme regulator activity, transmembrane transporter activity, signaling receptor activity, extracellular matrix structural constituent, lipid binding, extracellular matrix binding, electron transfer activity, and lyase/ligase activity. The genes that encode the molecular functions mentioned above, are associated with metabolic support in astrocytic activity and neural functionality present in euchromatin regions.

It was also possible to identify active coding regions tightly regulated for cellular ion maintenance and response to stimuli that are essential for astrocytes wellfunctioning [7]. Additionally, we were able to associate the presence of constant euchromatin regions with genes that encode for metabolic and cellular exchange mechanisms necessary for astrocyte function [41]. Therefore, the model demonstrated that the presence of genes in regular euchromatin regions are often associated with many regulatory elements such as promoters, enhancers, insulators and silencers, all related with cell adhesion, support and exchange processes [42, 43].


#### **Table 1.**

*Top biological processes associated with the core activation dataset from the histone markers H3K4me3, H3K27ac, H3K9ac, H3K4me1, H4K20me1, H3K36me3, and H3K79me2. Accordingly, biological processes were separated into functions and groups, in terms associated with GO for a more detailed analysis.*

In the case of bivalent expression regions, 30 biological processes were identified corresponding with APAR mechanisms, highlighting transcription regulation, sequence-specific DNA binding, RNA polymerase II/III distal enhancer, regulatory region, and proximal promoter sequence-specific DNA binding, gamma-aminobutyric acid transmembrane transporter activity, nerve growth factor binding, ubiquitin-protein transferase activator activity, mannosyl-transferase activity and cofactor, corepressor and coactivator transcription binding (**Table 2**). Additionally, these same genes that were also associated with specific functional groups such as macromolecule binding (*i.e.,* carbohydrates, sulfurates, lipids, amides), DNAbinding transcription factor activity, cofactor binding, extracellular matrix structural constituent, structural constituent of ribosome, structural constituent of cytoskeleton, extracellular matrix binding, neurotransmitter binding and structural constituent of myelin sheath and activity of hydrolase, transferase, peroxidase, oxidoreductase, isomerase, lyase/ligase signaling, transmembrane transporters, enzyme regulation, antioxidant, electron transfer, neurotransmitter transport, cytochrome-c oxidase, MAPKK, and glutathione dehydrogenase functionality.

Further, we identified all the specific genes associated with the APAR mechanisms in astrocytes under homeostatic conditions (**Figure 1**). It was possible to clarify the epigenetic basal response of astrocytes and identify the gene activation/ expression profiling under homeostatic conditions of these cells to elucidate the

#### *Multi-Omic Epigenetic-Based Model Reveals Key Molecular Mechanisms Associated… DOI: http://dx.doi.org/10.5772/intechopen.100133*

potential response to PA detrimental conditions. Both activation and bivalent processes and functions made it possible to understand the expression and regulatory mechanisms associated with an epigenetic chromatin landscape in astrocytes [4]. For both activation and bivalent datasets, the biological processes and functions were consistent with the basal astrocytic activity (**Tables 1** and **2**).


#### **Table 2.**

*Biological processes associated with histone markers H3K4me3 and H3K27me3 were separated between functions and groups. All the terms were associated with GO terms for further analysis [33].*

#### **Figure 1.**

*Graphical representation of activation, repression, and bivalent genes present in every APAR categories. Specifically, AN interaction has 42%, 25% and 31%; autophagy has 42%, 20% and 37%; ER Stress has 33%, 23% and 43%; inflammation has 39%, 28% and 31% all for activation, bivalent and repression genes respectively. Note: the colors correspond to Green-Activation, Orange-Bivalent, and Purple-Repression.*
