Shining light on dark matter in the genome

The complexity of multicellular organisms requires the genome to be transcribed in a cell-type–dependent manner that is responsive to signals, such as hormones, from the internal environment. This is mediated by the epigenome, which decorates and organizes the genome in a web of modified histone proteins functioning in nucleosomes and chemical modifications to genomic DNA arranged 3-dimensionally in the cell nucleus. Functional features of the epigenome such as acetylation of histone lysine residues are “read” by specialized proteins such as those containing bromodomains (1). Likewise, the genome itself is read by proteins known as sequence-specific transcription factors (TFs), which recognize and bind to specific motifs in genomic DNA. The totality of these sites for a given transcription factor in a given cell is known as its “cistrome” (2). Most of these binding sites occur in the ∼99% of the genome that does not encode for proteins. At some sites, the TFs control the expression of protein-encoding genes, in large part by further modifying the epigenome. These sites, known as enhancers, represent only a fraction of the cistrome, and the mechanisms determining which binding sites are functional are not well understood. In PNAS Fei et al. (3) outline the forward path by reporting a general method to screen for sequences that function as enhancers responsive to a given TF, with a focus on 2 specific TFs that provide great insight into the question of what determines a functional binding site.

Fei et al. (3) first determined the functional binding sites of forkhead box protein A1 (FOXA1), which is a pioneering TF that initiates the transition of enhancer elements from closed chromatin to a primed or poised state (4–6). This allows signal-dependent TFs to gain access to DNA and established nucleosome-free regions and recruit coregulators such as histone acetyltransferases (7) as well as chromatin organization regulators (such as Mediator and Cohesin proteins) that facilitate the formation of the enhancer–promoter loops for engagement of RNA polymerase II, leading to target gene transcription (7, 8). FOXA1 is known to be particularly important for genomic recruitment of steroid hormone receptors in hormone-dependent cancers, such as estrogen receptor (ER) in breast cancer (4, 9, 10) and androgen receptor (AR) in prostate cancer (11, 12). Interestingly, unlike differentiated somatic cells which have identical genomes, DNA mutations are frequently one of the drivers for carcinogenesis and progression, and some of these may cause loss or gain of enhancer function.

Methodologically, Fei et al. (3) used a CRISPR/Cas9-based strategy to delete over 10,000 sites previously determined to be bound by FOXA1 in T47D breast cancer cells or LNCaP prostate cancer cells (Fig. 1A). The functional screen was based on changes in cancer cell survival upon binding-site deletion, and they identified 72 FOXA1 binding sites that are essential for either breast cancer T47D cells or LNCaP prostate cancer cells to survive. Interestingly, some of these binding sites were cell-type specific, indicating the functional plasticity of enhancers and their cell specificity, in part related to differential effects on ER in breast cancer and AR in prostate cancer. Indeed, an advantage of this approach is that it allows functional dissection of different binding sites for the same TF, which cannot be distinguished by the deletion of the TF itself (Fig. 1B).

Fig. 1.

Illuminating our understanding of functional transcription factor binding sites and enhancers. (A) A CRISPR screen designed by Fei et al. (3) to identify TF binding sites that are essential for cancer cell survival. (B–D) This screen provided several mechanistic insights, including the following: (B) identifying which specific TF binding sites function as enhancers, (C) determining which genes are targeted by functional binding sites/enhancers, and (D) predicting which disease-associated SNPs regulate the activity of functional enhancers.

The screen was also applied to CCCTC-binding factor (CTCF), a pleiotropic DNA-binding protein which can function both as a classical sequence-specific TF and as a genomic boundary factor that is critical for the formation of topologically associated domains (TADs), which are critical for genome organization. Fei et al. (3) found 2 classes of CTCF essential binding sites. One class of binding sites was associated with local hyperacetylation of histone H3 on lysine 27 (H3K27Ac), a known mark of enhancer activity and thus consistent with a canonical TF function. The other class of binding sites was associated with lower H3K27ac signal and was enriched at TAD boundary regions, suggesting that, indeed, at these sites the essential role of CTCF is independent of gene transcription regulation per se. This class of cis-regulatory binding sites does not fit the enhancer definition and the basis of its essentiality clearly requires further characterization.

Fei et al.’s (3) study also has implications for the position of enhancers relative to the nearest transcriptional start site of a protein-coding gene. A recent study in which hundreds of enhancers were deleted suggested that the majority of enhancers (about two-thirds) controlled the nearest gene promoter (proximal regulation), with the other one-third governing genes that skipped at least one closer promoter (distal regulation) (13). In this genome-wide screening study of more than 10,000 TF binding sites, Fei et al. (3) also found that functional binding sites are generally closer to essential genes for cancer cell survival compared with all binding sites in the gRNA library and essential genes have a higher likelihood to be located near functional binding sites compared with all genes (Fig. 1C). Together, these studies from 2 independent groups suggest the functional enhancer–promoter interaction is more proximally than distally regulated at the genome-wide level. It would be interesting to combine such analysis with an unbiased study of genome architecture to determine the proportion of functional enhancers that are required for long-range enhancer–promoter interactions and how this differs from that of the entire cistrome.

The power of this large-scale screen for functional cis-acting sequences allowed Fei et al. (3) to build a machine-learning model that trained on FOXA1 primary screening data to predict functional enhancers and TF binding sites (Fig. 1D). The predictive power of functional enhancers defined by the screen is superior to traditional methods using features from either H3K27ac ChIP-seq or other methods that catalog open chromatin such as DNase sequencing in the discovery of functional disease-associated nucleotide polymorphisms (SNPs) in enhancers essential for survival of breast or prostate cancer cells. The power of this technique will increase as more whole-genome sequence information becomes available. Moreover, although cancer cells are characterized by somatically altered genomes, most studies to date have focused on either mutations in protein-coding regions or gene amplifications that predispose to transformation. Fei et al.’s (3) study demonstrates that noncoding mutations in enhancers and other, less well-characterized, TF binding regions also have large effects on cell survival and proliferation. Indeed, it would be interesting if any of the essential binding sites bound in these human breast and prostate cancer cells represent changes from the germline and were actually drivers of the original tumors from which these cells derived. More generally, the insights into enhancers and the predictive value of such data are likely to shine a mechanistic light on unexplored regions of the noncoding genome that are causative for disease and may be mutated in cancer.