How to Use Online Tools to Generate New Hypotheses for Mammary Gland Biology Research: A Case Study for Wnt7b

The experimental technology that allows genome wide analyses at the molecular level (genomics, epigenomics, transcriptomics, metabolomics and proteomics – hereafter combinedly referred to as ‘omics’ approaches) continues to evolve at breathtaking speed. Despite the fact that these techniques are becoming more affordable and therefore more widely available for scientists worldwide, they are still quite expensive – a prohibitory factor for those with limited financial resources. This is especially true for sophisticated approaches such as single-cell RNA sequencing (scRNAseq) and other-single cell approaches that are still being developed. Moreover, not everyone will have local access to the required infrastructure. Of course, scientific collaborations can offer a solution. Even then, it can be a challenge to integrate a variety of these technologies into one’s research program [1].

As can be gleaned from the published literature, all too frequently only a few hits or top candidates are followed up in instances where genome-wide datasets are generated. As a consequence, a wealth of data remains unexplored. These datasets constitute a rich and valuable resource for the larger scientific community. As an example, we have previously used published microarray data to identify the most stably rather than the most differentially expressed genes, resulting in a new set of reference genes for qRT-PCR studies in the developing mouse mammary gland [2].

Most ‘omics’ datasets are deposited in public repositories such as the NCBI Gene Expression Omnibus (https://​www.​ncbi.​nlm.​nih.​gov/​geo/​), either in raw format or in a more processed form. While this makes them available to all scientists in theory, in practice not everyone has the bioinformatics skills and expertise to analyze these data from scratch. Fortunately, multiple labs are dedicating their time and effort to the development of online tools that allow easy and intuitive access to these datasets, allowing researchers to explore them via a user friendly graphical interface.

We assume that the reader is familiar with the basic principles behind the different techniques (e.g. scRNAseq, snATACseq, Hi-C), as well as with the way in which these data are commonly presented (e.g. tSNE plots). Please note that for all figures we have kept the exact style and color schemes as generated by the different online tools to aid the reader in recognizing the output when they try out these tools for themselves.

WNT7B is expressed in human breast tissue and its expression has been reported to be altered in breast cancer [3, 4]. Its overexpression has been associated with a poor prognosis and reduced overall survival of breast cancer patients [5]. In breast cancer, WNT7B has not only been shown to be expressed by the tumor cells, but also by myeloid cells present in the local microenvironment. The latter promotes angiogenesis, invasion and metastasis [6].

Its murine counterpart, Wnt7b, is expressed in the ductal epithelium of the mouse mammary gland [7]. The levels of Wnt7b remain unaltered following ovariectomy, suggesting that regulation of Wnt7b expression is estrogen and progesterone independent [7]. During puberty, expression of Wnt7b is enriched in the terminal end bud epithelium, suggesting a role in branching morphogenesis [8]. Wnt7b has been reported to have mild transforming activities in vitro [9, 10] and in vivo [11] although not all studies agree on the extent of this effect [10, 12].

The precise role and regulation of Wnt7b/WNT7B in the mammary gland or breast remain unknown. So far, evidence that WNT7B protein can promote the activation of CTNNB1/TCF transcriptional complexes is lacking, despite the fact that Wnt7b is readily detected and shows prominent expression in luminal cells [13]. This is in contrast to other tissues, such as the skin, where the activities of WNT7B have been linked to CTNNB1/TCF driven processes [14].

Public scRNAseq datasets are an ideal platform to start investigating spatiotemporal gene expression in the mammary gland [15, 16]. We want to highlight three user friendly scRNAseq tools that allow analysis of the in vivo expression patterns of a gene of interest in the postnatal stages of mouse mammary gland development (Box 1). Their combined use reveals extensive details about the expression pattern of any given gene across different stages and cell populations.

Wnt7b expression is absent (or at least below the limit of detection) in the fetal mammary gland (E18, Fig. 1a), but emerges postnatally (Fig. 1b-d, Fig. 2a,d, Fig. 3a-d). Its expression is cell type specific, displaying high gene expression in the luminal compartment, and low or absent expression in basal cells and supportive tissues (fat, endothelial, immune and stromal cells) (Fig. 1b-d, Fig. 2b-e, Fig. 3a-d).

Spatiotemporal expression is dynamically regulated throughout the adult reproductive cycle (Fig. 2a,d, Fig. 3a-d). In nulliparous mice, Wnt7b is expressed in luminal progenitor cells, as well as in more differentiated, hormone-sensing luminal progeny (Fig. 2b-e). During gestation and lactation Wnt7b expression is (mostly) switched off, but it re-emerges post-involution (Fig. 2a,d, Fig. 3a-d). Thus, it is exclusively expressed in the ‘resting’ state, be it nulliparous or post-involution. Of note, although the luminal progenitor population itself re-appears post-involution, Wnt7b expression is lost in this population, becoming restricted to the hormone-sensing luminal lineage post-pregnancy (Fig. 2b-f).

From these analyses we would conclude that Wnt7b is expressed exclusively in the luminal compartment of the nulliparous mammary gland, is lost during pregnancy, and is re-established post-involution (Fig. 4). Indeed, this is supported by other studies showing that Wnt7b is expressed in the virgin mammary gland, but drops at Preg12.5 of pregnancy to undetectable levels [7]. This underscores the validity of this approach and illustrates the usefulness of interactive in silico tools to determine spatiotemporal patterns of in vivo gene expression.

Little is known about the molecular signals and cis-regulatory elements that control mouse Wnt7b or human WNT7B gene expression. In ER-/HER2 + breast tumors, WNT7B was shown to be a direct transcriptional target of the androgen receptor (AR) [42] and predicted to be regulated by Nuclear respiratory factor 1 (NRF1) [43]. Although Wnt7b is also expressed in hormone-responsive cells (Fig. 2 and [13]), at present there is no experimental evidence to support that its expression is regulated by steroid hormones, in particular progesterone [44]. Wnt7b expression is not limited to the mammary gland, however. It is required for lung [45, 46], and kidney development [47] to name but a few and can therefore be regulated by a myriad of signals.

One way to gain understanding into tissue-specific gene expression, is to identify cis-acting enhancer elements. Using ChIPseq analysis, a recent study predicted 440 mammary-specific super-enhancers [48]. Super-enhancers can be classified as dense clusters of transcriptional enhancers that are likely to control genes important for cell type specification [47‐49]. Only one of these was followed up in more detail in that particular study. However, a supplementary file listing all 440 of these putative regulatory elements is available. We were particularly intrigued by a sequence that spans more than 24 kb on chromosome 15 (published mm9 coordinates chr15: 85,475,778–85,500,063, mm10 coordinates chr15: 85,645,348–85,669,633), which was assigned as a putative regulator of the nearest gene: Wnt7b (Fig. 5a). While it is common to do so, linear proximity alone is not an accurate measure for functional interaction between an enhancer and its putative target gene [49, 50]. Other genes in this region – including two miRNAs (Mirlet7c-2/Mirlet7b) and a protein coding gene (Ppara) – might also be regulated by this particular super-enhancer. A region on the edge of this super-enhancer (mm9 coordinates chr15:85,473,689–85,478,592, published mm10 coordinates chr15: 85,643,259–85,648,162) was recently indeed associated with Wnt7b, albeit not in the mammary gland but in a mouse model for hair-follicle derived skin tumors, and based on strain-specific polymorphisms rather than on having been shown to directly regulate Wnt7b expression [14]. These results show that association of this super-enhancer with Wnt7b in the mammary gland is worthy of follow-up analysis.

The term “super-enhancer” is used to define a larger chromatin area that contains clusters of smaller, individual enhancers and that is enriched for active chromatin marks (e.g. H3K27ac) or occupied by transcriptional activators (e.g. MED1) and master regulatory transcription factors (e.g. STAT5A) [48, 51, 52]. More than 80,000 super-enhancers (combined numbers for the mouse and human genome) can be accessed through the online dbSuper database [19]. An updated version of the Super Enhancer Archive (SEA 3.0) provides another entry point [20] (Table 1).

A first screen of the dbSuper database shows the tissue-specificity of super-enhancers: a putative Wnt7b super-enhancer has also been identified in the murine heart, lung and testis. However, this sequence does not overlap with the mammary-specific super-enhancer described by Shin et al. [48]. Instead, the dbSuper database predicts this particular location to contain two super-enhancers, identified in hair follicle stem cells, linked to Mirlet7c-2/Mirlet7b [53]. Additional super-enhancers in this region, identified in the kidney and the liver, are tentatively associated with Ppara (Fig. 5b). It should be noted that also in dbSuper, super-enhancers and their associated genes are linked based on a simple proximity rule to the nearest transcriptional start site (TSS) [19]. Out of the genes located in this ~ 500 kb area on chromosome 15, only Atxn10 and Wnt7b show prominent expression in one or more mammary gland cell subpopulations, although Ppara, Mirlet7c-2/b and a non-coding RNA, Lncppara, may be differentially expressed at low levels (Fig. 5c,d).

In recent years, it has become generally accepted that regulatory elements control target gene expression within the confines of larger, structurally ordered regions of the chromatin known as topologically associating domains (TADs) [54]. Specific DNA sequences (i.e. regulatory elements and their target genes) are much more likely to interact within a TAD, than across a TAD boundary. A logical next step in exploring the potential regulation of Wnt7b by the aforementioned mammary-specific super-enhancer would therefore be to determine the boundaries of the Wnt7b TAD.

We used the 3D Genome Browser (Box 2) to visualize TAD predictions of the Wnt7b locus using publicly available Hi-C datasets [55]. In this browser, TAD boundary predictions are calculated according to the so-called directionality index, which is a method that looks at the degree of up- and downstream interaction bias for DNA regions [56]. It was noted that DNA regions at the periphery of TADs are highly biased in their direction of interaction. Upstream regions in a TAD are highly biased towards interacting with downstream regions and vice versa. Using this directional bias, the boundaries of adjacent TADs can be predicted. Their coordinates are provided by the 3D Genome Browser, which also includes an intuitive visual reference (Fig. 6).

Only one mammary-specific Hi-C dataset is currently available, derived from human mammary epithelial cells (HMEC) [57]. However, TADs have been reported to be stable across cell types and even species [56, 60]. Although not all TAD boundaries are equally stable [61], TAD organization can therefore also be investigated using Hi-C datasets generated from a different tissue as input.

According to this analysis, the Wnt7b TAD boundary lies immediately upstream of the Wnt7b TSS in both HMECs and mouse lymphoma cells (Fig. 6a,b). This would imply that the mammary-specific super-enhancer identified by Shin et al. lies outside of the predicted Wnt7b TAD, which makes it less likely that this particular super-enhancer directly regulates the expression of Wnt7b. However, in other Hi-C datasets this TAD boundary is less well defined (Fig. 6c,d).

To gain a better understanding of how the spatiotemporal expression of Wnt7b is regulated in the adult mammary gland, we can start by probing the epigenetic state of the Wnt7b locus in an R shiny app published by the Wahl lab (Box 3). This tool not only allows chromatin accessibility and relevant histone modifications to be examined, but also can be used to make predictions about specific promoters and their regulatory sequences of interest. An attractive graphical interface allows intuitive interpretation of the data (Fig. 7).

If we focus our attention on the Wnt7b promoter and gene region (i.e. the center portion of Fig. 7), snATACseq reveals that the chromatin is relatively accessible in all mammary cell type subpopulations irrespective of Wnt7b gene expression levels (Fig. 7, top 5 rows). In contrast, H3K27ac of the Wnt7b promoter and gene region is exclusively enriched in the luminal compartment (Fig. 7, bottom 4 rows in red). This suggests that Wnt7b is ‘primed’ and open in all epithelial cells in the mammary gland, but its potential for increased gene expression is only realized in the luminal compartment where the chromatin displays the proper histone acetylation marks.

Combining the Cicero algorithm (see Box 3) with snATACseq data, this online tool can also be used to infer co-accessibility of distal sites and the promoter of their putative genes in individual cells. In this manner, Cicero can predict cis-regulatory elements that would be able to interact with the Wnt7b promoter in vivo. At a co-accessibility threshold of 0.15, Cicero identifies 10 regions within 150 kb up- or downstream of the viewpoint that interact with the promoter of Wnt7b. Of these, 4 are located upstream of Wnt7b in an area dense with H3K27ac that encompasses, but extends beyond, the super-enhancer region, and 6 are located downstream of Wnt7b (Fig. 7).

The interacting regions depicted to the left of the Wnt7b promoter (regions 1–6, located 3′ distal to the TSS) all fall within in the predicted Wnt7b TAD (compare Fig. 6,7). These distal sites are either somewhat enriched for chromatin accessibility or H3K27ac, or a combination of both epigenetic features, in adult luminal progenitor and mature luminal cells compared to the adult basal subpopulation (Fig. 7,8). The 4 regions downstream of Wnt7b (7–10) do not display evident changes in chromatin accessibility or H3K27ac when luminal cells are compared to the basal compartment, except for region 8 (Fig. 7,8). Note that the distance between region 9 and 10 spans more than 60 kb, which is considerably larger than the reported size of the mammary-specific super-enhancer. This entire stretch of 60 kb shows characteristic marks of active and open chromatin, suggesting that a much larger collection of regulatory elements may exist in this area (Fig. 7).

In previous studies, highly conserved sequences were associated with developmental and transcriptional regulators [67‐73]. Given the fundamental role of Wnt signaling not only in vertebrate development [74], but also specifically in mammary gland development and maintenance [75‐77], focusing on conserved sequences could be another criteria for the selection of candidate Wnt7b enhancers. To identify conserved regions in the vicinity of Wnt7b, we used the evolutionary conserved region (ECR) browser (Box 4).

Often, conservation is scored across vertebrate species. However, in an attempt to identify regions that are specifically conserved in mammals, we specifically selected candidate sequences in a region of ~ 100 kb up- and downstream of the Wnt7b TSS that are conserved across mammalian, but not necessarily in non-mammalian vertebrate species available in the ECR browser (Fig. 9).

Of course, none of these approaches (sequence conservation, histone modification, transcription factor ChIPseq), either by themselves or in combination, are sufficient to definitively link any of these putative regulatory elements to Wnt7b. This requires further experimental validation and specific follow up. However, as a prediction tool these combined analyses provide an excellent starting point for dissecting this super-enhancer in more detail. If we put all of the different pieces of information together (Fig. 10), we can draft some hypotheses regarding the regulation of Wnt7b expression in the mammary gland.

First, we propose that in mammary epithelial cells the proposed TAD boundary immediately upstream of Wnt7b (Fig. 6) is not very stable, given that the Cicero algorithm predicts four interactions between the Wnt7b promoter and regions to the right of this presumed TAD boundary (i.e. regions 7–10 in Fig. 7). Of note, two of these interactions (Cicero regions 7 and 8) occur in the direct vicinity of this presumed TAD boundary. The other two interactions (Cicero regions 9 and 10) border a large area of active chromatin, which extends beyond the super-enhancer region previously identified by Shin et al. [48]. This 60 kb area harbors an annotated lncRNA (Lncppara) and two microRNAs, MirLet7b/MirLet7c-2, which are broadly expressed and implicated in cancer formation [78‐80]. Moreover, this region also contains multiple conserved sequences that could represent functional enhancer elements (including ECR_6, ECR_7 and ECR_8 from Fig. 9).

Second, if we do take the TAD boundary prediction into account, it may be wise to prioritize the interactions that occur between Wnt7b and more downstream sequences (i.e. regions 1–6 in Fig. 7). Although the coordinates from the Cicero prediction algorithm deserve further scrutiny of the original datasets, these downstream interacting regions also lie in close vicinity to conserved sequence elements.

Third, in combination with the expression data analysis (Fig. 1-3), the published literature and the active enhancer marks (Fig. 7,8), we can make a further prioritization of putative Wnt7b enhancer sequences that are worthy of experimental validation and follow up. In this case, region 2 is particularly interesting as is has the highest Cicero score and displays both differential chromatin accessibility and H3K27 acetylation in the luminal compartment.

To summarize, by using publicly available online tools we assessed the genomic conformation of the Wnt7b locus, and how this relates to the previously identified putative Wnt7b super enhancer. By examining the epigenetic status of the Wnt7b locus more closely, we noticed that although the Wnt7b promoter is predicted to interact with the super-enhancer region, this is likely not cell type specific as both chromatin accessibility and H3K27ac do not change between the basal and luminal lineages in this region. However, regions downstream of Wnt7b do change their epigenetic status in accordance to Wnt7b gene expression and are also predicted to interact with the Wnt7b promoter. This entire area would be worthy of experimental follow up to definitively associate specific regulatory elements with Wnt7b and/or other nearby genes – in particular the miRNAs and Ppara.

At the same time, we call for joint efforts from our community to ensure that the repertoire of tools as well as of accessible datasets continues to grow and remains of high quality and value to investigators worldwide. As others have undoubtedly noticed, mammary gland and breast tissue datasets are often notoriously absent from public, large-scale -omics efforts. Generating and curating additional genome wide datasets (e.g. Hi-C and others) for both epithelial and stromal cells of multiple species, including mouse and human, would be a tremendous resource for our community as a whole. The careful generation of such datasets in combination with user-friendly online tools provide a valuable resource for researchers, and could in the long run also help to reduce animal experimentation. Certain features will enhance the user experience and promote the wide use of such tools, including the ability to export high resolution graphs (ideally allowing further customization, e.g. PDF format as offered by [15, 55]) and the ability to easily download specific sequences or genome coordinates (as offered by [55, 65]). Given the challenges associated with keeping these databases up to date and operating smoothly, international and consortium efforts that provide sufficient support infrastructure may, in the long term, prove to be essential in this regard.

Here we have shown how the combined use of different online tools can be applied to generate novel hypotheses. Of course, the same tools can also be used to complement existing projects by providing additional data. Ideally, in the not too near future, researchers will have a broad compendium of resources available to them that are of such high quality that they will allow in vivo analyses to be performed in silico, thereby bringing such genome-wide analyses within reach of all scientists. This will only be possible, however, if sufficient tissue-specific datasets can be accessed. Especially in the case of the mammary gland, great care should be taken to include different timepoints to cover both embryonic and postnatal developmental stages, as well as the entire gestational cycle. Here, biological and computational expertise will continually need to go hand in hand to ensure that such online tools can meet the demands of the scientific questions that are being asked.