Gene transcription regulation by ER at the single cell and allele level

Abstract

In this short review we discuss the current view of how the estrogen receptor (ER), a pivotal member of the nuclear receptor superfamily of transcription factors, regulates gene transcription at the single cell and allele level, focusing on in vitro cell line models. We discuss central topics and new trends in molecular biology including phenotypic heterogeneity, single cell sequencing, nuclear phase separated condensates, single cell imaging, and image analysis methods, with particular focus on the methodologies and results that have been reported in the last few years using microscopy-based techniques. These observations augment the results from biochemical assays that lead to a much more complex and dynamic view of how ER, and arguably most transcription factors, act to regulate gene transcription.

Introduction

How a cell responds to stimuli and adapts to a continuously changing environment is a central question to understand basic biological processes. The observation that cells in culture, with identical genotypes, react differently to environmental cues like growth factors, nutrients, small molecules, and more (i.e., phenotypic heterogeneity, (1–4)) led to studies designed to identify mechanisms through which this occurs as they become relevant, for example, to understand the insurgence of drug resistance in the absence of additional genetic mutations (5,6). The cell-to-cell variation in responses is usually masked in classical biochemical and molecular biology technologies which average results from thousands to millions of cells. For this reason, single cell technologies, chiefly imaging-based ones, are now gaining more and more traction as they allow for more precise definition and measurement of biological effects.

In terms of gene transcription, differential responses are thought to mainly arise from stochastic on/off switching between gene promoter states and biochemical reactions linked to the transcriptional process (4,7–9). This is complicated by higher level (“cell state”) extrinsic noise that is determined by the instant-by-instant changes in the cellular milieu (i.e., signaling pathways, metabolism etc. (10)), which all contribute to the observed phenotypic differences between cells. Moreover, physical barriers (i.e., nuclear envelope) and neighboring cells have also been shown to buffer and modify the seemingly stochastic events that occur in the nucleus (11,12).

In this review, as a paradigm for studying single cell actions of a stimulus-activated transcription factor, we will focus primarily on the Estrogen Receptor (ER, Figure 1A), a member of the nuclear receptor superfamily of transcription factors (13–15), which is a master regulator of several important pathophysiological processes. Moreover, ER is targeted by several dozen small molecules of natural and synthetic origin which mimic or antagonize the main natural ligand, the steroid hormone 17β-estradiol (E2) (16–18). In this review, we will only focus on canonical nuclear ER actions. it is important to mention that the ER has been shown to be heterogeneously expressed not only in tissues, by immunohistochemistry (e.g., cell-to-cell variation in ER levels is routinely observed and quantified in breast tumors), but also in in vitro cell lines (i.e., the classic MCF-7 breast cancer model) (19,20), a phenomenon usually less studied and that we recently exploited to generate a novel quality control analytical pipeline for single cell image analysis, demonstrating an overall stability of ER levels within large cellular populations over time (21). In Figure 1B we highlight how each cell has its own multidimensional signature that results in cell-to-cell variation in status and response to perturbagens. In the cartoon, each cell and nucleus have different shapes and sizes, each cell contains a different amount of ER protein (shown as shades of grey), ER mRNA (blue dots), and target gene mRNA (shown as red dots for nascent RNA and green dots for steady state RNA). For details on how to perform single molecule RNA FISH for nascent and steady state RNA please see (22).

Figure 1.

A) Estrogen Receptor (ER) domain structure. NTD: N-terminal domain; AF: Activation Function; DBD: DNA Binding Domain; LBD: Ligand Binding Domain; B) Visual cartoon representing the phenotypic heterogeneity (cell-to-cell variation) in ER mRNA, protein, and target gene in response to E2 based on immunofluorescence and single molecule RNA FISH studies (19,21,26,27). The “synthetic cells” are ordered left to right based on ER nuclear levels (shades of grey indicate protein amount/nucleus). Each cell and nucleus are represented with a different size/shape and contain varying numbers of dots, with blue being ER mRNAs, green being target gene steady state mRNAs (defined by exonic probes in RNA FISH experiments), and red being target gene nascent mRNAs marking transcriptionally active alleles in a nucleus (as defined by intronic probes in RNA FISH experiments).

The nuclear receptor superfamily consists of ligand-activated transcription factors that bind and act in response to specific hormones and signaling molecules regulating gene expression. These receptors are characterized by a modular protein structure (Figure 1A), comprising an unstructured N-terminal domain (A/B) usually containing one or more coregulator-interacting activation functions, an α-helical globular domain at the C-terminus (LBD, E domain), which is responsible for the ligand and coregulator binding (AF-2), and a double zinc finger DNA-binding domain (DBD, C domain), flanked by a short hinge domain (D) (15,23–25). The Estrogen Receptor (ER) is part of a subfamily of nuclear receptors, named the steroid receptors subfamily, together with other central transcription factors that regulate many physiological processes, including the androgen receptor (AR), progesterone receptor (PR), glucocorticoid receptor (GR), and mineralocorticoid receptor (MR).

PART1. ER nuclear mechanisms of action based on bulk population and biochemical analysis.

In the canonical mechanism of action, steroid hormones bind to their cognate receptor, which in some cases (i.e., AR, GR, and PR) resides in an inactive state bound to heat shock proteins (HSPs) in the cytoplasm, while in others (ER in breast cancer cells, for example) is primarily localized in the nucleus regardless of hormone binding. Once the hormone binds, a conformational change is induced, resulting in the receptor dimerizing, dissociating from the HSPs, and translocating to the nucleus (when needed). The receptor then binds to specific DNA sequences (i.e., estrogen response elements – EREs) serving as a platform for the recruitment of several classes of coregulators (i.e., p160s, mediator, CBP/p300, etc.) that ultimately regulate gene transcription and modulate the histone code (Figure 2) (28–30).

Figure 2.

Vignette of the canonical mechanism of action of the ER. The indicated inputs on the left impinge upon the recruitment of ER (Estrogen Receptor) to EREs (Estrogen Response Elements) and its coregulators (HATs: histone acetyltransferases; p160: family of coregulators – SRC1, SRC2, and SRC3) to post translationally modify histone tails and cause RNA Polymerase II (PolII) to modulate gene transcription (Txn), ultimately leading to the Output responses on the right-hand side of the cartoon.

Additionally, nuclear receptors activity can influence, and be influenced, by cellular signaling pathways to promote or prevent downstream cellular event, including gene transcription. These activities are often referred to as non-genomic signaling even though more often than not the final output still is modulation of gene transcription, and will not be discussed in this review (31–35).

The ER (36) has been one of the first and most studied receptors in terms of regulation of gene expression, as it elicits a fast (<15 minutes) response to a stimulus (the natural hormone 17β-estradiol - E2 (37) for example) and has been shown to modulate a large part of the cell transcriptome (16,28,34,38–43). The classical view of ER mechanism of action includes ligand binding, followed by receptor dimerization and interaction with thousands of distal enhancers containing EREs, as elegantly shown over the years by ChIP-seq and analog techniques (28,43–47). Next, the ER recruits a large number of coregulator complexes whose identity has been determined by mass-spectrometry, measured by ChIP-seq, and also, more recently, structurally by CryoEM (48–52), all designed to ultimately modify histone tails by methylation, acetylation and other post-translational modifications, allowing gene transcription to proceed (or cease) across hundreds of genes (39).

Many studies enumerated the various contributions of coregulators, signaling pathways, and epigenetic modulators to E2-mediated gene transcription creating a very comprehensive map of the hundreds of potential interrelationships between ER, the proteome, and the transcriptome. An increased level of complexity adding to the canonical mechanistic view arose from concepts that include: 1) the role of pioneering factors (47,53), 2) the potential cycling of ER and coregulators on chromatin (54,55), 3) assisted loading (i.e., collaboration between nuclear receptors to modify specific target genes (56–59)), and, 4) mechanisms regulating ER-mediated transcriptional repression (60–62).

However, most, if not all, of these studies were performed on large populations of cells (thousands to millions) and using biochemical methods that can destroy the elegant cellular/spatial context without focusing on hormonal effects at the individual cell level or taking into consideration the dynamics of the system. The model for transcriptional regulation by ER that has evolved from all these efforts (Figure 2) points to the formation of large and stable macromolecular complexes that interact for minutes to hours on DNA to perform their biological functions. In the next part of the review, we will explore the findings that proposed alternative mechanistic avenues based on imaging, single cell analysis and live cell dynamics.

PART 2. ER mechanisms of action using single cell techniques.

In this part of the review, we will highlight the efforts that have been done regarding the analysis of ER mechanisms of action at the single cell level focusing on single cell RNA sequencing, ER dynamics, formation of liquid-liquid phase separated nuclear condensates, and imaging transcriptional responses in live and fixed samples. An overview of the technologies used, including basic advantages and disadvantages, can be found in Table 1.

Table 1.

Single cell methods that have been utilized to analyze ER mechanisms of action.

Technique	Goal	Advantages	Disadvantages
Single Cell RNA sequencing	Target gene expression	Provides information on the target genes expression landscape	Loss in intra- and inter-cellular spatial information
Fluorescence Recovery After Photobleaching (FRAP)	Protein mobility	Relatively simple and quantitative, live dynamics	Bulk protein, requires specific controls and knowledge of laser- scanning confocal microscopy. Requires engineering a model with XFP-tagged protein
Fluorescence Correlation Spectroscopy (FCS) and Single Molecule Tracking (HiLO)	Protein mobility	Better measure single particle mobility in a smaller volume of investigation and at higher frequency (nanosecond scale), live dynamics	Requires extensive technical knowledge and quantification can be challenging. Requires engineering a model with XFP-tagged protein. Requires specific instrumentation.
Immunofluorescence	Protein localization and level	Easy protocol, multiplexable, higher throughput, easy automated quantification for single cell analysis	Not single molecule, no dynamics, good antibodies might be missing for target protein of interest
Live Cell Imaging	Protein dynamics and localization	Easy setup, easy analysis	Requires engineering a model with XFP-tagged protein, no single molecule information
smFISH/MERFISH	Target gene activation/expression	Measure response at the single cell and allele level; from one gene to the whole transcriptome	No dynamic information, expensive, requires complex informatics and design (MERFISH), requires specialized instrumentation
MS2/PP7 Repeats Live Imaging	Live dynamics of target gene activation	Measure dynamic response of a target gene in live cells	Requires engineering a model system, difficult to multiplex and to tag multiple alleles in a cell, could be prone to variation in interpretation depending on where tagging is, complex mathematical interpretation/modeling

Single-cell RNA-seq (scRNA-seq).

scRNA-seq is the closest method that conceptually moves away from bulk population analysis, by employing next-gen sequencing of barcoded RNAs from single cells after library prep and amplification (63,64). scRNA-seq was developed to capture the transcriptomic variation between single cells either in vitro or, perhaps more interestingly, in tissue samples. It has been used to measure changes in the transcriptome over time after estrogen treatment across multiple cell models (65,66). Zhu et al., performed a time-course E2 analysis in MCF-7 and T47D breast cancer cell lines and, not surprisingly, identified a large cell-to-cell variation in the transcriptional response, including almost 25–40% of the target genes having a bimodal distribution, a feature that would be completely missed by population aggregation. While interesting, these early efforts did not dive into the reasons underlying the cell-to-cell variability.

Single cell imaging studies – ER dynamics by live imaging, FRAP, and FCS.

One of the aspects that biochemistry-based assays are unable to cover is ER dynamics in a live cell. Several studies over the years employed fluorescent protein-tagged ER in live cells to study the movement of the receptor in the nucleus upon different ligand treatments and the results overall have demonstrated that ER moves around the nucleus at different rates of speed, its movement being directly linked to transcriptional regulation (67,68).

The earlier reports used mostly fluorescence recovery after photobleaching (FRAP) (68–70), a well-established live cell technique that relies upon confocal microscopy to photobleach a region of interest and measure the half time of recovery of the tagged protein (71). These studies showed that the ER half time of recovery was short (few seconds) even after hormone treatment (in control cells ER mobility is even faster), a quantity that can also be used to classify different ligand classes (72). Interestingly, some antagonists (i.e., ICI182,780/fulvestrant) work by immobilizing the receptor to the nuclear matrix (68) ultimately targeting it for degradation, which began a new mechanistic class of ER antagonists (73). The same time scale of cell dynamics (seconds) has also been reported for other transcription factors and coregulators; most notably the glucocorticoid receptor, which, through a biosensor cell line, gave rise to the theory of hit-and-run as a mechanism of action for nuclear receptor mediated gene transcription (70,74,75) on promoters, which depends upon the transcription factor moving rapidly in the nuclear microenvironment and transiently interacting with DNA and cofactors, questioning the overall stability of the biochemically-defined multiprotein complexes. While FRAP studies provided seminal ideas in the field, including similar dynamic results on integrated arrays containing thousands of EREs (76), the technique has not advanced enough to allow for visualization and analysis of single ER molecules and their direct interaction with chromatin. Newer techniques allowed to perform single molecule tracking (SMT) (77–80) to directly measure transcription factors behavior in the dense nuclear environment by using highly inclined and laminated optical sheet (HILO) microscopy. These very complex studies further highlight that nuclear receptors are very dynamic yet can slow down somewhat when interacting with chromatin; however, the time scales measured are also in the order of milliseconds to seconds, akin to the original FRAP studies. Another important note is that the fraction of DNA bound receptor is probably only on the order of 5–10% of the nuclear ER pool, at any given time (77). Additionally, a recent paper validated these concepts using an orthogonal method, intensity-sorted fluorescence correlation spectroscopy (FCS). FCS is a methodology that calculates fluctuations of fluorescence emitted by single molecules moving in and out of a very small observation volume (67,81). In this unique study (67), the authors measured the mobility of GFP-ER in the nucleoplasm and compared it with GFP-ER interacting with an engineered multicopy array of the prolactin promoter/enhancer region containing thousands of ER binding sites (76).

ER and nuclear condensates.

Nuclear condensates (liquid-liquid phase separated droplets - LLPSs) are membraneless micro-compartments that could play a role in regulating gene expression by increasing the local concentration of specific biomolecules, facilitating protein-protein interactions and enzymatic reactions (82–84). Boija et al. (85) suggested that intrinsically disordered regions in the activation domains of transcription factors can have phase separation capability, and one of the targets studied was ER. Interestingly, ER formed LLPSs were promoted by the presence of estrogen and included the coactivator MED1. In addition, Nair et al. (86) demonstrated that, at ER regulated enhancers, transcription of eRNA mediated the assembly of ribonucleoprotein (RNP) granules that show phase separated condensate properties. These initial studies paint the possibility that also ER is a transcription factor capable of modulating gene transcription via the formation of dynamic LLPSs. However, a direct link between LLPSs and ER-mediated gene transcription in intact cells is still missing.

Single cell imaging studies – ER regulation of gene transcription in live and fixed samples.

The intrinsic cell-to-cell variability in hormone-dependent stimulation of gene transcription is best understood and queried by microscopy-based methods, which facilitate spatial analysis of nascent and mature RNA per cell (via spectrally-separated fluorescently labeled oligonucleotides designed to specifically recognize introns and exons of target genes of interest), with methods ranging from single or few genes analysis (single molecule RNA FISH, (8,22,87)) to the whole transcriptome (MERFISH and seqFISH, (88–91)), and allow for studying transcriptional dynamics at endogenous loci (92–94).

The non-genetic heterogeneity evident in a cell population has been described across multiple models, with a few underlying principles being highlighted by many studies. In a simplistic way, the single cell milieu of gene expression can be considered the combination of intrinsic and extrinsic noise that are important in creating the “cell state” (4,8,10–12). While extrinsic noise is thought to depend on multiple factors determining the “cell state” and equally influencing the expression of all the genes in a cell, intrinsic noise appears to be the main determinant of expression variation of each gene. Arguably the main parameter contributing to intrinsic noise is transcriptional bursting, describing transcription as a phenomenon that occurs through usually short periods of ON time followed by longer OFF periods, an evolutionary conserved observation (92), and has been effectively mathematically modeled (8,95,96). In mammalian cells it is noteworthy to mention that the intrinsic noise in gene transcription is buffered by post-transcriptional mechanisms, including RNA processing, nuclear export, and RNA degradation (11).

In order to study the live dynamics of estrogen-mediated gene stimulation on endogenous genes, two pioneer studies (93,94) used CRISPR/Cas9 to integrate MS2 or PP7 repeats (97,98) in the endogenous loci of the prototype ER-target genes GREB1 and TFF1, which allow for monitoring active transcription over time by single cell live imaging.

Briefly, in these seminal papers the authors inserted 24x MS2 or PP7 stem loops in the 3’ UTR (93) or exon 2 (94) of TFF1 and GREB1, respectively, that can be visualized live by the concurrent expression of a fluorescently-tagged viral coat protein that specifically binds to RNA loops (99).

These studies highlighted that individual alleles in a nucleus are often turned ON for a short period of time (minutes) followed by extended OFF times (hours), in a cell state-independent manner (i.e., alleles in the same nucleus behave somewhat independently). From these studies it appears clear that the calculations of the ON/OFF times may depend on experimental considerations, including the gene structure, the positioning of the engineered repeats, the relative abundance of the fluorescently tagged coat protein, the analysis methods and more, which make the observations challenging to generalize. However, E2 addition is shown to not activate transcription simultaneously in all cells, and to cause a dose-dependent increase in transcriptional burst frequency, a common phenomenon in enhancer-driven transcriptional responses, coupled with an increase or little effect on burst size (i.e., number of RNAs/burst) depending on the gene studied (9,100).

The E2 effects are overlayed on global cellular kinetic variations in transcription initiation and elongation (i.e., extrinsic noise) that allow for some correlation between alleles in the same cell or, at minimum, a higher probability that additional alleles become active in the same nucleus after an initial response, as shown by live imaging at the GREB1 and TFF1 loci (93,94).

These observations were possible as Rodriguez et al. succeeded in tagging three out of the five TFF1 alleles, and Fritzsch et al., two out of the four GREB1 alleles, in MCF-7 cells, which gave them the opportunity to compare the response to hormone at individual alleles (93,94).

Fritzsch et al., also analyzed correlations between response to E2 and some basic morphological features that could be extracted from the images, including nuclear shape, cell area, local cell density, and cell cycle, without finding any correlation; the only tracked feature showing some linkage was cellular volume, as also identified by Raj and colleagues in different cell models (9).

In our recent manuscripts (27,101), we utilized single molecule RNA FISH (smFISH, (22,87)) in several breast cancer cells, to study the ER-dependent activation of GREB1 and MYC gene transcription at the single cell and allele level. We accomplished this by using spectrally separated probe sets (22,87,102) designed to hybridize to either GREB1 introns or exons, allowing for simultaneously measuring the steady state mRNA (i.e., how much mature RNA is a cell) and the nascent RNA (i.e., how many alleles are active at a particular moment). The obvious disadvantage of this approach, as compared to the ones previously described, is the loss of dynamic analysis; however, there is a specific advantage of this approach: gaining information about individual allele regulation in a much larger population of cells (hundreds). Collectively, addition of spatial information by robust imaging provides a larger view of how hormonal responses develop over time. In addition, the fixed-cell approach is more amenable to perturbations and screening campaigns. In this study (27), we demonstrated that response to hormone happens rapidly (<15 minutes, but likely much faster as the probe design limits temporal resolution), although not in every cell in the population, and also not all alleles in the same nucleus respond synchronously. This held true when we tested multiple cell lines with different allele numbers, multiple hormone doses, and various ligands impinging on the ER (e.g., bisphenol A or genistein) or activating other nuclear receptors (e.g., AR and GR). We further demonstrated, perhaps more surprisingly, that the nuclear level of ER, which is highly heterogenous in a cell population (19–21,27), was not tightly correlated with allele-by-allele responses to hormone; a phenomenon that is likely to be transcription factor and intracellular signaling specific, as a stronger correlation has been reported for different pathways (100,103).

By using a CRIPR/Cas9 engineered MCF-7 cell line harboring a constitutive active ER mutation (Y537S, (104)), we also showed that, even when all receptor molecules are considered in an ON state, the response to hormone at the single cell level was similar, indicating that neither the receptor levels nor its “activation status” are sufficient to cause synchrony in all alleles of a target gene in all the cells of the population.

These results implied that other mechanisms, if any, could be employed in a cell to control the magnitude of response to hormones. To search for candidate pathways, we performed a high throughput microscopy screen combining ER immunofluorescence and GREB1 intron and exon smFISH on MCF-7 cells pre-treated with a library of epigenetic small molecule inhibitors. This complex approach allowed us to filter the results based on ER levels, which are well known to significantly change upon treatment with histone deacetylase inhibitors (105). Indeed, the largest number of hit compounds prevented E2-mediated induction of GREB1 transcription by either reducing ER levels or via other mechanisms, including BRD4 inhibition, as previously shown (106). Surprisingly, a small handful of inhibitors increased the number of active GREB1 alleles/cell and led us to focus on the CARM1/PRMT6 inhibitor MS049 (107). This was particularly intriguing as both CARM1 and PRMT6 have been considered as canonical nuclear receptor coactivators (108,109) for a long time and clearly play a central role in the formation and modulation of ER-containing complexes (48,49,110). Through a set of experiments, including the use of CARM1 knock-out MCF-7 cells, we established that the mechanism of action was likely not through epigenetic modifications but involved arginine methylation of non-histone targets, notably MED12 (27,110).

This was the first demonstration in the nuclear receptor field that the frequency of active alleles after hormonal stimuli can be modified by cellular pathways, making this phenomenon regulatable and not exclusively stochastic.

Conclusions

The observations from both live (68,93,94) and fixed (27) cell experiments clearly show that there is a disconnect between the imaging-based measurements at the single cell and allele level as compared to the more classical, population-based and biochemistry-driven view that hypothesized a coordinated, rhythmic, and synchronous response to hormone (28,48,54,55). A variation of the model that includes information derived from live imaging experiments is shown in Figure 3.

Figure 3.

Adjusted mechanistic model of ER action based on live imaging and single cell analysis experimental studies. We want to highlight how all the components of the cartoon are highly mobile and their time scale dynamics variable.

The variation of responses between alleles in the same nucleus poses the question of how ER levels and mechanism of action act at different alleles. In principle, one could assume that perhaps the variation is simply due to the absolute number of ER molecules in the cell (though unlikely, (27)), or to random Brownian diffusion of ER and/or its coregulators, or it could depend on ER/coregulator post-translational modifications that define an active/competent pool that is modulated by intracellular signaling pathways. This highly dynamic environment, followed by formation of equally dynamic and transient phase separated condensates at enhancers/promoters of target genes (86), could bridge the gap between the two current models of ER-mediated gene transcription. The next decade of exploration and technological advances will be instrumental in fully understanding the intricacies of gene transcription regulation. The impact of ER dynamics and the allele-by-allele regulation of target genes is an area that will require a large effort in the next years to identify if these factors are important in modulating normal tissue biology (i.e., mammary gland development, menstrual cycle etc.) and whether this information can be used as an enhanced therapeutic avenue, which is plausible as ER “immobilizers” like fulvestrant are already in the clinic.