YY1 is a cis-regulator in the organoid models of high mammographic density

Abstract

Motivation

Our previous study has shown that ERBB2 is overexpressed in the organoid model of MCF10A when the stiffness of the microenvironment is increased to that of high mammographic density (MD). We now aim to identify key transcription factors (TFs) and functional enhancers that regulate processes associated with increased stiffness of the microenvironment in the organoid models of premalignant human mammary cell lines.

Results

3D colony organizations and the cis-regulatory networks of two human mammary epithelial cell lines (184A1 and MCF10A) are investigated as a function of the increased stiffness of the microenvironment within the range of MD. The 3D colonies are imaged using confocal microscopy, and the morphometries of colony organizations and heterogeneity are quantified as a function of the stiffness of the microenvironment using BioSig3D. In a surrogate assay, colony organizations are profiled by transcriptomics. Transcriptome data are enriched by correlative analysis with the computed morphometric indices. Next, a subset of enriched data are processed against publicly available ChIP-Seq data using Model-based Analysis of Regulation of Gene Expression to predict regulatory transcription factors. This integrative analysis of morphometric and transcriptomic data predicted YY1 as one of the cis-regulators in both cell lines as a result of the increased stiffness of the microenvironment. Subsequent experiments validated that YY1 is expressed at protein and mRNA levels for MCF10A and 184A1, respectively. Also, there is a causal relationship between activation of YY1 and ERBB2 when YY1 is overexpressed at the protein level in MCF10A.

Supplementary information

Supplementary data are available at Bioinformatics online.

1 Introduction

A recent study (Cheng et al., 2016) profiled the organoid model of premalignant cell lines as a function of the stiffness of the extracellular matrix within the range of mammographic density (MD) and concluded that the ERBB2 protein is overexpressed in MCF10A when the stiffness of the microenvironment is increased to that of high MD in the clinical samples. An organoid model is a miniaturized organ that is produced in vitro using one or few cells, and, in 3D, cells can self-organize as a result of self-renewal capacity. Depending upon the culture conditions of ‘ontop’ versus ‘embedded,’ organoids are formed through aggregation of nearby cells or replication of a single cell (Lee et al., 2007), respectively. ERBB2 is an oncogene that is mutated in 14.05% of breast carcinoma patients (Andre et al., 2017). We are interested in identifying key transcription factors (TFs) that are activated as a result of increased stiffness of the microenvironment and whether these TFs have a causal relationship with ERBB2. Toward this objective, we integrate parallel readouts of 3D colony organizations, imaged using confocal microscopy, with transcriptomic data leading to a subset of the enriched gene sets through correlative analysis.

For two premalignant cell lines of MCF10A and 184A1, organoid models are formed, and 3D colony organoids are imaged using confocal microscopy, where each colony is segmented and represented multi-parametrically (Bilgin et al., 2013, 2016) as a function of the stiffness of the microenvironment. MCF10A and 184A1 are non-malignant cell lines; however, since they are immortalized and overcome one of the barriers toward malignant transformation, we refer to them as premalignant. In a parallel assay, transcriptome data are collected, under identical conditions and genes that correlate with computed colony indices (e.g. flatness, colony size) are identified per earlier method (Han et al., 2010). The net result is an enriched gene set based on computed morphometric indices. Subsequently, the enriched gene set is cross-referenced with publicly available ChIP-Seq data for inferring regulatory TFs. Although gene expression profiling, using microarray or RNA-seq techniques, can identify differentially expressed genes between two or more conditions, the gene list is not sufficient to learn about regulation. ChIP-Seq data measure genome-wide distributions of regulatory or epigenetic factors (TFs or histone modifications) and provides direct interactions between regulators and their target genes. By taking advantage of publicly available genomic data, including histone modifications, ChIP-Seq and chromatin accessibility DNase-seq data from various cell types, we can predict TFs and functional enhancer elements in the human genome.

We use a recently developed method called Model-based Analysis of Regulation of Gene Expression (MARGE; Wang et al., 2016, 2018) to infer the cis-regulatory network. MARGE uses a compendium of published ChIP-Seq data for H3K27ac and applies a regression-based semi-supervised learning approach to predict the functional enhancers of a given gene set. The key TFs can then be predicted by associating the MARGE-predicted enhancer profile with thousands of existing TF ChIP-Seq datasets. MARGE (Wang et al., 2016) defines a regulatory potential for each gene by summarizing nearby H3K27ac ChIP-Seq signals, where we have shown that regulatory potentials scaled across the public H3K27ac profiles are more predictive of genes that are repressed or activated by BET inhibitors or the super-enhancer-based approach. MARGE offers three distinct advantages. First, compared with the existing master-regulator prediction methods such as super-enhancer analysis (Loven et al., 2013; Whyte et al., 2013), MARGE uses a more quantitative model to associate H3K27ac signals to each gene, and, as a result, MARGE has been shown to outperform ROSE (Wang et al., 2016). Second, MARGE uses a compendium of published ChIP-Seq data for H3K27ac, a histone mark associated with the active enhancer regions in the genome (Creyghton et al., 2010; Rada-Iglesias et al., 2011), which have been reported to provide more informative for inference of gene regulation than looking at the promoters alone. Third, MARGE is designed to capture important information that can explain variations in gene expression, regardless of up/down-regulation, or the origin of cell type or tissue.

2 Results

2.1 The frequency of aberrant phenotypes is increased as a function of increased stiffness

Two immortalized HMEC lines, MCF10A and 184A1, were cultured in 3D, and the stiffness of the microenvironment was modulated between 250 and 1800 Pa., which is the stiffness values at low and high MD (Cheng et al., 2016). At each stiffness value, colony formation was imaged using confocal microscopy, and from the same passage, RNA was collected for transcriptomic profiling. Colony formation indices, such as flatness, elongation, colony size and surrogate markers of lumen formation were quantified using BioSig3D. Colony flatness and elongation are surrogate markers of dysplasia or loss of lumen formation. However, evaluating lumen formation requires the addition of multiple polarity markers, which is not tractable for high content screening. BioSig3D can also profile the heterogeneity through consensus clustering, where the frequency of occurrence of each cluster can then be quantified. One example of heterogeneity analysis is presented in Figure 1, where the elongation index is shown to have two subpopulations, and the frequency of occurrence of the aberrant subpopulation increases as a function of the stiffness of the microenvironment. Yin-Yang 1 (YY1) is predicted as a regulatory TF for premalignant human mammary cell lines.

Fig. 1.

Colony shapes are heterogeneous: (a) Consensus clustering identifies two morphometric subtypes quantified with the colony elongation index as a function of increased stiffness of the microenvironment from 250 to 1800 Pa. (b, c) Representative of colony organization for each of the subtypes imaged through confocal microscopy with each nucleus segmented with BioSig3D. (d) Frequency of occurrence of colony elongation increases as a function of increased stiffness of the microenvironment. Scale bar is 20 μm. * denotes P < 0.001 between the two clusters

Genes that correlate with four indices of elongation, flatness, colony size and the number of cells per colony were identified and a subset of genes, per morphometric index, was selected based on their p-values. The correlative analysis essentially is a gene enrichment protocol. Each set of the enriched genes was applied to MARGE, and the relevant publicly available H3K27ac ChIP-Seq datasets and the functional enhancer-profiles that regulate these genes were predicted, as shown in Figure 2a. The relevant H3K27ac ChIP-Seq datasets, selected by MARGE, include samples from the breast cancer cell line MCF-7, which indicates biological similarity in gene regulation profiles between the cell lines, as shown in Supplementary Table S1. An example of the MARGE-predicted enhancer profile is shown in Figure 2b.

Fig. 2.

MARGE predicts YY1 as a functional TF regulating phenotype-associated genes. (a) MARGE schematic. MARGE predict functional cis-regulatory (enhancer) profiles of a given gene list, using a compendium of public H3K27ac ChIP-Seq profiles. Key TFs are further predicted by associating the cis-regulatory (enhancer) profiles generated by MARGE with each of 2264 collected public TF ChIP-Seq datasets. For each TF dataset, a receiver operating characteristic (ROC) curve is generated, and the area under the curve (AUC) is used to assess the association of the regulatory profile with this TF in the genome. (b) Sample genome browser snapshot of a public YY1 ChIP-Seq profile (upper track) and an MARGE-predicted cis-regulatory (enhancer) profile (lower track). YY1 ChIP-Seq dataset is from K562 cell line (Pope et al., 2014). The enhancer profile is predicted from genes correlated with stiffness. (c–f) YY1 is highly enrich regardless of the phenotypic indices for MCF10A. Cumulative distributions of AUC scores for predicting all 2264 ChIP-Seq datasets for different TFs’ binding (blue) and 18 datasets for YY1 binding (red). Having significantly higher AUC scores than the all-TF background, YY1 is predicted as a functional regulator of the input gene sets for cell line MCF10A. (P-value < 0.001, by K-S test). (Color version of this figure is available at Bioinformatics online.)

We then investigated the association between the MARGE-predicted functional enhancer profile and 2264 ChIP-Seq datasets collected from the public domain (Mei et al., 2017) representing genomic profiles of over 400 TFs in various cell types. Table 1 lists the top 10 TFs, with a P-value of < 0.01, and with a full list shown in Supplementary Table S2 as a spreadsheet. Most of these TFs are involved in biosynthesis or regulation of RNA polymerase and expressed in breast cancer by cross-referencing against The Human Protein Atlas pathology database. However, we were interested in TFs that are also associated with ERBB2, and YY1 is potentially relevant. YY1 has been associated with the MARGE-predicted enhancer profile based on the criteria that YY1-binding sites, in YY1 ChIP-Seq datasets, align with the top-ranked enhancers in the genome with a P-value < 0.001, as shown in Figure 2c–f. YY1 has been shown to cooperate with the AP-2 TF (cell growth and differentiation) to stimulate ERBB2 (Begon et al., 2005) and negatively regulate p27 (a cell cycle inhibitor) (Wan et al., 2012). However, MARGE analysis did not predict AP-2 with a significant P-value. It is important to note that YY1 was predicted with only 15 human breast samples from H3K27ac ChIP-Seq.

Table 1.

Top 10 TFs predicted by MARGE with a P-value <0.01

TF	Correlation coefficient	Role
POLR2A	0.92	Synthesizing mRNA
TAF1 Myers	0.91	Initiating transcription by polymerase
PHF8	0.91	Tumor growth and epithelium–mesenchyme transition
TBP Snyder	0.91	Initiating RNA polymerase
SNAPC1	0.91	Initiating RNA polymerase
BRD2	0.89	Transcription and cancer
GMEB2	0.89	DNA replication
CREBBP	0.89	Growth control
YY1	0.89	Differentiation, proliferation and cancer
GMEB1	0.90	DNA replication

2.2 YY1 is activated, at the protein level, in MCF10A, at high stiffness values of the microenvironment, and there is a causal relationship between activation of YY1 and ERBB2

Having identified YY1 as a TF regulator using MARGE, we then designed three experiments with the necessary controls for validating the overexpression of YY1 in the organoid model of MCF10A. To begin, control cells were engineered by overexpressing and knocking down YY1 using CRISPR technology. The ERBB2⁺ SKBR3 and the triple-negative MDA-MB-231 breast cancer cell lines are used as a control with results shown in Supplementary Figures S1 and S2. The first experiment was performed with MCF10A, which indicated that (i) if YY1 is turned off by CRISPR, then ERBB2 is not activated at low or high stiffness values of the microenvironment, (ii) if YY1 is activated with CRISPR then ERBB2 is upregulated at low and high stiffness of the microenvironment and (iii) there is a causal relationship between activation of YY1 and ERBB2 only at high stiffness of the microenvironment. These results are shown in Figure 3a–c, where co-expression of YY1 and ERRB2 are shown on different colonies and two corresponding antibodies. Co-expressions of YY1 and ERBB2 are not heterogeneous with six more examples shown in Supplementary Figures S3 and S4. The second experiment was performed with the non-transformed 184A1; however, neither overexpression of ERBB2 nor YY1 was observed at the protein level. The third experiment was performed with the ERBB2 positive cell line of SKBR3, where control cells were engineered to overexpress and knockdown YY1, as shown in Supplementary Figures S1 and S2. Regardless of the stiffness of the microenvironment and YY1 being up or down, ERBB2 was always expressed in this cell line.

Fig. 3.

YY1 has a causal relationship with ERBB2 in MCF10A monitored by fluorescent microscopy. DAPI stain is used in all samples, organoids are fixed on Day 10, and approximately the middle of each organoid is imaged with confocal microscopy. (a, b) Control experiments. (a) If YY1 is knocked down then ERBB2 is turned off. (b) If YY1 is overexpressed then ERBB2 is turned on. (c) YY1 and ERBB2 expression are correlated at high stiffness. The first two and the second two rows group YY1 and ERRB2 expressions at low and high stiffness of the microenvironment. Scale bar is 10 µm

2.3 YY1 is activated, at the mRNA level, in 184A1 cells, at high stiffness values of the microenvironment

qPCR was performed to determine mRNA expression of YY1 in MCF10A and 184A1 cells at low and high stiffness (250 and 1800 Pa) of the microenvironment. All validation data were converted to fold changes using 2^−ΔC_T and 2^−ΔΔC_T methods for raw and normalized data. Figure 4 shows that the relative expressions of YY1 in MCF10A remain unchanged at low and high stiffness of the microenvironments (0.97- and 1.03-fold); however, the relative expression of YY1 in 184A1 is increased from 0.15- to 0.72-fold. The mRNA expression, collected from 184A1 cells, indicates no change in the ERBB2 expression between low and high stiffness of the microenvironment.

Fig. 4.

Relative mRNA expression of YY1 is increased in 184A1 because of increased stiffness of the microenvironment. In MCF10A, the relative expression remains unchanged

3 Discussion

The following observations are made using the two human mammary epithelial cell lines of MCF10A and 184A1. These two cell lines were selected because MCF10A originates from a patient with fibrocystic disease, while 184A1 originates from a normal patient following reduction mammoplasty. Furthermore, at high stiffness of the microenvironment, MCF10A expresses ERBB2 protein, while 184A1 does not (Cheng et al., 2016). Our data illustrate that (i) the increased stiffness of the microenvironment, within the normal MD range also increased the frequency of aberrant colony formation for both cell lines. (ii) Having enriched the gene expression data with computed colony indices, MARGE predicted several TFs, with YY1 being a relevant cis-regulatory hub for both cell lines. (iii) The expression of YY1 correlates with ERBB2 in MCF10A, at the protein level, when the stiffness of the microenvironment is increased to that of high MD. Yet, overexpression of YY1, in 184A1, was observed at the mRNA level at high stiffness of the microenvironment. It is highly plausible that once sufficient protein is expressed, the mRNA production is turned off in the case of MCF10A. These observations also correlate with previous results that high levels of YY1 transcriptional activation overexpresses ERBB2 (Allouche et al., 2008). However, in the ERBB2 positive cell line of SKBR3, overexpression or knock out of YY1 had no impact on the expression of ERBB2, indicating cell line independency and presence of alternative pathways to overexpress ERBB2.

A potential mechanism can be summarized as follow. YAP and TAZ have been identified as transcriptional coactivators that function as mechanosensory (Harvey et al., 2013) that interact with the Rho/Rock pathway(Halder et al., 2012), and transfer information about the stiffness of the microenvironment from the membrane to the nucleus for a transcriptional response. These coactivators are overexpressed as a result of increased stiffness of the microenvironment and lead to increased cellular proliferation (Lin et al., 2015); thus, requiring higher metabolism. However, there is growing evidence that NF-κB, an important regulator in cancer progression, is also responsive to the stiffness of the microenvironment (Ishihara et al., 2013; Mammoto et al., 2012; Sero et al., 2015). For example, ChiP analysis has shown that NF-κB subunits bind directly to YAP and TAZ promoters to activate their transcription (Ferraiuolo et al., 2018), and inactivation of NF-κB leads to the downregulation of YAP and TAZ. Finally, the link between NF-κB and YY1 has been long established through knockout experiments, where inhibition of NF-kB also leads to inhibition of YY1 and dysregulation of the epithelial-mesenchymal transition circuitry (Atchison et al., 2011; Kashyap and Bonavida, 2014). YY1 has been shown to regulate metabolism (Park et al., 2017) by activating genes involved in mitochondrial bioenergetics and a number of oncogenes that are responsible for increased proliferation such as c-Myc (Riggs et al., 1993). Increased proliferation is also associated with PCNA, a necessary component of the DNA replication machinery (Wood and Shivji, 1997), which is regulated via higher expression of YY1 (Weng and Yung, 2005). These observations lead to a possible set of interactions, as shown in Supplementary Figure S5.

In conclusion, integrative morphometric and transcriptomic data have enriched the molecular signature for the inference of the cis-regulatory networks. Our integrative analysis indicates that YY1 is overexpressed in the organoid models of two human mammary epithelial cells when the stiffness of the microenvironment is increased to that of high MD. In addition, YY1 is expressed at protein and mRNA levels for MCF10A and 184A1, respectively, as a function of increased stiffness of the microenvironment. This is potentially because MCF10A is more differentiated and is derived from a benign proliferative breast tissue, which is spontaneously immortalized without defined factors (Soule et al., 1990).