Malladi_2020: Total functional score of enhancer elements identifies lineage-specific enhancers that drive differentiation of pancreatic cells

keywords

Enhancer, epigenome, gene regulation, pancreas, tissue-specific Transcription, transcription factor

Total functional score of enhancer elements identifies lineage-specific enhancers that drive differentiation of pancreatic cells

Abstract

Introduction

Lineage specification is dependent on the interactions of Transcription Factor (TF) and chromatin states at enhancers

Enhancers have been shown to share several common features

• Increased chromatin accessibility (as measured by DNase-seq or ATAC-seq)

• enrichment of posttranslational modification of the amino-terminal tails of core histone proteins (as assessed by ChIP-Seq)

• histone H3 lysine 4 monomethyl (H3K4me1)

• histone H3 lysine 27 acetyl (H3K27ac)

Recent genomic assays have shown that active enhancers are bound by RNA polymerase II (Pol II) and are transcribed, producing noncoding RNAs known as enhancer RNAs (eRNAs)

Enhancer Transcription (as measured by total RNA-seq, GRO-seq, or PRO-seq) can be used in the absence of any other genomic information to predict enhancer activity

Advances in technology have facilitated the large-scale functional characterization of enhancer activity and the annotation of TF-binding sites (TFBSs)

genome-wide in various cell types and tissues

Analyses that predict TFBSs

• Fail to consider that such sequences occur frequently

by chance throughout the genome and that TF binding is cell type specific

TFBSs, which are usually 4 to 12 nucleotides in length

Total Functional Score of Enhancer Elements (TFSEE)

Aimed to (1) evaluate TFSEE as an enhancer-calling algorithm and (2) understand the TF-driven transcriptional programs differentiating human embryonic stem cells (hESCs) into pancreatic cells

Figure 1.

Materials and Methods

Genomic data curation

• GRO-seq, ChIP-seq, and RNA-seq data from time course differentiation of human embryonic stem cells (hESCs) to pancreatic endoderm (PE)

Analysis of ChIP-seq data

• GRCh37/hg19

• Aligned with Bowtie v 1.0.0

Analysis of RNA-seq data

• GRCh37/hg19

• Aligned with STAR v 2.4.2a

• Quantification of genes with RSEM

Analysis of GRO-seq data

• Trimmed to the first 36 bases to the trim adapter and low-quality sequence using fastx_trimmer

• GRCh37/hg19

• Aligned with BWA v0.7.12

Kernel density

• calculated in Python (ver. 2.7.11) using the kdeplot function from seaborn version 0.7.1

Defining Transcription Start Site (TSS) and promoters

• Made TSSs for protein-coding genes using MakeGencodeTSS

• Indetified active promoters using H3K4me3 enrichment

• RPKM cutoff of ≥1 for H3K4me3

Enhancer calling by GRO-seq

Calling a universe of transcripts from GRO-seq data.

• groHMM

• built a universe of transcripts by merging the groHMM-called transcripts from individual cell lines and stratifying the boundaries to remove overlaps/redundancies occurring from the union of all transcripts.

Calling active enhancers using GRO-seq-defined enhancer transcripts

• <9 kb in length and >3 kb away from known TSSs

• protein-coding genes AND H3K4me3 peaks

• Classified into

  • short paired eRNAs

    • cutoff of RPKM ≥ 0.5

  • short unpaired eRNAs

    • cutoff of RPKM ≥ 1

• The comprehensive universe of expressed eRNAs (short paired and short unpaired) assembled using the cutoffs noted above for each cell line was used for further analyses.

Motif analyses for GRO-seq-defined enhancers

• De novo motif analyses was preformed on a 1kb region surround the overlap center, or the TSS using MEME

Enhancer calling by ChIP-seq

Calling active enhancers using histone modification ChIP-seq data

• Built a universe of peak calls by merging the peaks from individual cell lines for histone modifications (H3K4me1 and H3K27ac)

• Potential enhancers were defined as peaks that were >3 kb from known TSSs, protein-coding genes from Gencode version 19 annotations, 41 and H3K4me3 peaks

• RPKM cutoff of ≥1 for H3K4me1 and H3K27ac in at least one cell line

Motif analyses for ChIP-seq-defined enhancers

• De novo motif analyses were performed on a 1 kb region (±500 bp) surrounding the peak summit for the top 10000 enhancers using MEME

TODO Generating heatmaps and clusters

TODO Nearest neighboring gene analyses and box plots

TODO Overlapping enhancer analysis

Results

The TFSEE model

• Used both GRO-Seq and ChIP-Seq enhancer identification methods for input

Step 1 - Method 1: enhancer calling based on enhancer transcripts defined by GRO-seq

Enhancer calling by GRO-seq

Step 1 - Method 2: enhancer calling based on histone modification defined by ChIP-seq

Enhancer calling by ChIP-seq

Step 2: Calculating enrichment and activity profiles

Figure 2. Data processing for Total Functional Score of Enhancer Elements (TFSEE) method

• 5 data processing steps

Steps 3 to 5: De novo motif searching and TF expression

Calculating the TFSEE score by data integration

Figure 3. Overview of Total Functional Score of Enhancer Elements (TFSEE) method

GRO-seq expression + H3K27ac enrichment + H3K4me1 enrichment = Enhancer Activity

Enhancer Activity x Motif prediction

Comparison of enhancer calls by methods 1 and 2

• greater than 84% of the enhancers identified based solely on enhancer transcription were not called based on enrichment of H3K27ac or H3K4me1

Figure 4. Comparison of approaches for genome-wide prediction of enhancers during pancreatic differentiation

Decided to focus on the enhancers identified based on enhancer transcription using GRO-seq data (method 1)

TFSEE identifies lineage-specific enhancers and their cognate TFs during pancreatic differentiation

TFSEE scores determined by using inputs from method 1

Figure 5. TFSEE identifies cell type-specific enhancers and their cognate TFs that drive gene expression during pancreatic differentiation

1. Heatmap of the 5 stages of pandcreatic differenitation

2. Box plots of normalized TFSEE score for clusters identified in pancreatic differentiation

3. Cluster TFSEE scores

Figure 6. TFSEE-predicted TFs are enriched in pre- and late pancreatic differentiation

1. TF expression

2. Enhancer Trascription

3. Nearest Neighboring Gene Expression

4. Cluster 3 Rank Order of Enriched TFs

5. Cluster 4 Rank Order of Enriched TFs

TODO TFSEE scores determined using inputs from method 2.

TODO Comparison of TF identification using inputs from method 1 or method 2

Discussion

• TFSEE enables analysis of driver TFs using a limited amount of data

• model was able to identify lineage-specific TFs with as little as 5 cell types and with only 2 data types, RNA-seq and ChIP-seq (for H3K4me3, H3K4me1, and H3K27ac)

• A limitation of the TFSEE method is that while the model can be used with a reduced number of data types for enhancer identification, it fails to identify additional subtype- or stage-specific drivers with reducde data input

Integrating additional genomic data into TFSEE

• Integrate genomic data indicating open regions of chromatin (ATAC-seq, DNase-seq, or MNase-seq)

• ChromHMM could be used to annotate alternate chromatin states with additional histone modifications

• Chromatin Looping data for enhancer-promoter interactions (as measured by 4C, ChIA-PET, or Hi-C)

Conclusions