bioRxiv preprint doi: https://doi.org/10.1101/2021.01.06.425505; this version posted January 6, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. Hierarchical Regulation of Autophagy During Adipocyte Differentiation Mahmoud Ahmed, Trang Huyen Lai, Trang Minh Pham, Sahib Zada, Omar Elashkar, Jin Seok Hwang, and Deok Ryong Kim∗ ∗Correspondence [email protected] Department of Biochemistry and Convergence Medical Sciences and Institute of Health Sciences, Gyeongsang National University School of Medicine Jinju, South Korea January 6, 2021 Abstract We previously showed that some adipogenic transcription factors such as CEBPB and PPARG directly and indirectly regulate autophagy gene expression in adipogenesis. The order and the effect of these events are undetermined. In this study, we modeled the gene expression, DNA- binding of transcriptional regulators, and histone modifications during adipocyte differentiation and evaluated the effect of the regulators on gene expression in terms of direction and magnitude. Then, we identified the overlap of the transcription factors and co-factors binding sites and targets. Finally, we built a chromatin states model based on the histone marks and studied their relation with the factors’ binding. Adipogenic factors differentially regulated autophagy genes as part of the differentiation program. Co-regulators associated with specific transcription factors and preceded them to the regulatory regions. Transcription factors differed in the binding time and location, and their effect on expression was either localized or long-lasting. Adipogenic factors disproportionately targeted genes coding for autophagy-specific transcription factors. To sum, a hierarchical arrangement between adipogenic transcription factors and co-factors drives the regulation of autophagy during adipocyte differentiation. Keywords– transcription-factors/autophagy/differentiation/adipocyte/hierarchical-regulation 1 bioRxiv preprint doi: https://doi.org/10.1101/2021.01.06.425505; this version posted January 6, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. Adipogenic autophagy regulation. Ahmed et al. (2021) 1 Introduction 2 Methods Previous studies suggested one-to-one inter- 2.1 Expression & binding data actions between adipogenic transcription fac- tors and autophagy. CEBPB transactivates We collected two datasets of RNA-seq Atg4b, a key protein in the autophagy ma- and ChIP-seq on 3T3-L1 pre-adipocytes, chinery [1]. The activation of autophagy which were induced to differentiate using 3- through this pathway relieves the repression isobutyl-1-methylxanthine, dexamethasone, of adipogenic activators such as PPARG. and insulin (MDI) and sampled at different FOXO1, a transcription factor with several time points (Table 1, 2,& 3)[4]. We cu- autophagy targets, was suggested to the re- rated the samples’ metadata using a unified press Pparg gene in the presence of insulin language across the studies and processed sensitizers [2]. This repression is likely to the raw data using standard pipelines. The be lifted in early adipogenesis. A previous processed gene expression data were made study from our laboratory showed that au- available as a Bioconductor data package (cu- tophagy gene products are regulated as part ratedAdipoRNA). The data are presented as of the transcription program of adipogene- gene counts at different time points (0 to 240 sis [3]. This regulation is achieved through hr). The processed DNA-binding data of adipogenic transcription factors PPARG and transcription factors, co-factors, and histone CEBPB either directly or indirectly through modifications were made available as a sim- autophagy specific factors. The magnitude ilar package (curatedAdipoChIP). Data are and the ordering of this regulation remain to presented in this package as the reads count be investigated. in a consensus peak set. Moreover, we pro- Here, we used gene expression and DNA- vided links to the identified peaks as well as binding data to model the transcription factor the signal tracks. The packages document the and co-factors binding events during differ- pre-processing and processing pipelines. entiation and their effect on autophagy genes. We obtained two gene expression datasets We used histone modification data to corre- of Cebpb (RNA-seq) or Pparg-knockdown late these events with chromatin states. A hi- (microarrays) from a matching MDI-induced erarchical arrangement of known adipogenic 3T3-L1 pre-adipocytes time-course experi- transcription factors and co-factors emerged ments (Table 4). Gene counts and probe in- in the regulation of autophagy during adipo- tensities were downloaded using GEOquery genesis. We evaluated the spatial and tem- and used to quantify the gene expression from poral aspects of this arrangement. These in- RNA-seq and microarray data, respectively cluded the factors’ contributions to gene ex- [5]. pression, the dependency between the regula- tors, the reliance on chromatin states, and the type of binding targets. 2.2 Mouse genome annotations Gene ontology (GO) terms in mouse bi- ological processes were used to identify the gene products relevant to autophagy and lipogenesis [6]. The Bioconduc- tor package org.Mm.eg.db was used to access the GO annotations [7]. The gene accessor IDs were mapped be- 2 bioRxiv preprint doi: https://doi.org/10.1101/2021.01.06.425505; this version posted January 6, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. Adipogenic autophagy regulation. Ahmed et al. (2021) Table 1: MDI-induced 3T3-L1 gene expres- Table 2: Transcription factors binding data. sion data by RNA-seq. SRA ID N Antibody Ref. GEO ID N Time (hr) Ref. SRP000630 12 PPARG/ [26] GSE100056 2 24 [10] RXRG GSE104508 3 192 [11] SRP002337 2 PPARG [27] GSE35724 3 192 [12] SRP002507 2 CEBPB [28] GSE50612 4 0/144 [13] SRP006001 9 CEBPB/ [29] GSE50934 6 0/168 [14] CEBPD/ GSE53244 3 0/48/240 [15] RXRG/ GSE57415 4 0/4 [16] PPARG GSE60745 12 0/24/48 [17] SRP028367 3 PPARG/ [30] GSE64757 6 168 [18] MED1 GSE75639 3 0/48/168 [19] SRP041249 3 RXRG/ [31] GSE84410 5 0/4/48 [20] MED1/ GSE87113 5 0/2/4/48/168 [21] EP300 GSE89621 3 240 [22] SRP100871 28 CTCF/ [24] GSE95029 8 0/48/144/192 [23] MED1/ GSE95533 10 4/0/24/48/168 [24] NCOR1/ GSE96764 6 0/2/4 [25] EP300 Table 3: Histone modification data. tween gene symbols and Entrez IDs using SRA ID N Antibody Ref. TxDb.Mmusculus.UCSC.mm10.knownGene SRP002337 11 H3K4me3/ [27] [8]. The same package was used to extract H3K27me3/ gene coordinates in the mouse genome. Fi- H3K36me3/ nally, the GO terms for molecular functions H3K4me2/ terms were used to identify the transcription H3K4me1/ binding targets’ functional categories. GO.db H3K27ac was used to access these terms [9]. SRP041249 6 H3K27ac/ [31] H3K4me1/ 2.3 Differential gene expression H3K4me2 SRP064188 3 H3K27me3/ [32] RNA-seq reads were aligned to the mm10 H3K9me3 mouse genome and counted in known SRP078506 6 H3K4me3 [20] genes using HISAT2, and featureCount [34, SRP100871 6 H3K27ac/ [24] 35]. Gene counts were filtered, normal- H3K4me1/ ized, transformed, and subjected to batch ef- H3K4me2 fects removal. Microarrays probe intensities were filtered and collapsed to correspond- ing known genes, normalized, and trans- or the probe intensities were compared be- formed. To identify gene expression changes tween conditions (#hr vs. 0 hr or knockdown over time or in response to transcription fac- vs. control). Fold-change and p-value for ev- tors genes knockdown, we applied differen- ery gene in each comparison were calculated. tial gene expression analysis using DESeq2, False-discovery rate (FDR) was used to ad- or LIMMA [36, 37]. Briefly, the gene counts just for multiple testing. 3 bioRxiv preprint doi: https://doi.org/10.1101/2021.01.06.425505; this version posted January 6, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. Adipogenic autophagy regulation. Ahmed et al. (2021) Table 4: Perturbed MDI-induced 3T3-L1 gene expression data by RNA-seq. 2.6 Gene set enrichment and GEO ID N KD Ref. over-representation GSE57415 8 Cebpb [16] To calculate GO terms’ enrichment scores at GSE12929 18 Pparg [33] different times of differentiation, we ranked all genes by fold-change, performed a walk of the gene set members over the ranked 2.4 Binding peaks analysis list, and compared it to random walks. The enrichment score is the maximum distance ChIP-seq reads were aligned to the mm10 between the gene set and the random walk mouse genome using BOWTIE2 [38]. Bind- [42]. ChromHMM calculates the enrichment ing peaks were identified using MACS2 with of states as (C/A)/(B/D) where A is the num- the annotation file of the same genome [39]. ber of bases in the state, B is the number of Peaks were annotated and assigned to the bases in external annotation, C is the num- nearest gene using ChIPSeeker [40]. The ber of basses in the state, and the annota- numbers of binding sites and targets were cal- tion and D is the number of bases in the culated in each sample. When more than genome. clusterProfiler calculates the over- one sample was available for a given factor, representation as the number of items in the only replicated binding sites or targets were query and subject groups compared to the included. The intersections of binding sites groups’ total number [43]. and targets among the samples were calcu- lated and visualized using ggupset. 2.7 Software & reproducibility 2.5 Hidden markov chain models The analysis was conducted in R lan- guage and environment for statistical Multi-states hidden Markov chain models of computing and graphics [44].