Gene regulation, genome architecture, and the machinery of transcription

A ground-up overview of how eukaryotic cells decide which genes to express, and how the physical folding of DNA makes that possible.

The problem

A human cell contains ~3 billion base pairs of DNA but expresses only a fraction of its ~20,000 genes at any given moment, and different cell types express radically different subsets. A liver cell and a neuron carry identical genomes. The system that reads the same genome and produces different outputs in different contexts is what this post is about.

Promoters

A promoter is a stretch of non-coding DNA immediately upstream of a gene’s transcription start site (TSS). It is not a protein, it does not “promote” anything actively — it is a docking address. Specific base-pair sequences within it are recognized by the transcription machinery, which assembles there before reading the gene.

Key sequence elements within a core promoter (roughly −35 to +35 bp around the TSS):

TATA box (~−30 bp): recognized by TFIID, the first factor to land
Initiator element (Inr): spans the TSS itself, contributes to precise start-site selection
Downstream promoter element (DPE): found in TATA-less promoters, binds TFIID subunits

The promoter on its own, with only the basal machinery assembled, drives weak background-level transcription. This is called basal transcription — the floor, not the operating level.

Transcription factors

Transcription factors (TFs) are proteins that bind specific short DNA sequences (motifs, typically 6–12 bp) and influence whether a nearby gene is transcribed. There are ~1,600 TFs in the human genome, classified by their DNA-binding domain family: zinc fingers, bHLH, leucine zipper, homeodomain, ETS, nuclear receptors, and others.

Two broad classes:

General TFs (TFIIA, TFIIB, TFIID, TFIIE, TFIIF, TFIIH) are ubiquitous — present in every cell type, assembling at every Pol II promoter. They are not the decision-makers. They are the minimum required machinery.

Sequence-specific TFs bind particular motifs at enhancers and proximal promoter regions. These are the cell-type specificity layer. A liver cell expresses HNF4α; a T cell expresses GATA3; a developing neuron expresses NEUROD1. These TFs recruit coactivators and, via Mediator, stimulate Pol II.

TFs are not permanently glued to DNA. They diffuse in the nuclear volume and transiently bind their cognate motifs via non-covalent interactions — electrostatic, hydrophobic, shape complementarity. Multiple TFs binding nearby motifs on the same enhancer cooperate through direct protein-protein interactions, collectively stabilizing the complex. The word “sits at” in the literature means “is frequently and transiently bound to.”

RNA Polymerase II and the pre-initiation complex

RNA Polymerase II (Pol II) is the enzyme that transcribes all protein-coding genes and most non-coding RNAs. It is a ~500 kDa, 12-subunit complex. It does not find promoters on its own.

Assembly at the TSS proceeds stepwise:

TFIID recognizes the TATA box (and other core promoter elements) and lands first. TFIIA and TFIIB join, stabilizing the nascent complex.
Pol II–TFIIF is recruited as a pre-formed unit, escorted in by TFIIB.
TFIIE and TFIIH complete the pre-initiation complex (PIC). TFIIH has two critical activities: helicase activity to unwind the DNA double helix at the TSS (creating the transcription bubble), and kinase activity to phosphorylate Pol II’s C-terminal domain (CTD).
CTD phosphorylation is the switch. Unphosphorylated CTD = Pol II is assembled but paused. Phosphorylated CTD = Pol II transitions to elongation mode and begins reading the template strand.

This full assembly — six general TFs plus Pol II — is the PIC. It produces basal transcription: a trickle of RNA, with no cell-type specificity.

Enhancers

Enhancers are DNA sequences — typically a few hundred base pairs — that dramatically boost transcription of a target gene. They are defined operationally: if you take the sequence and place it near a reporter gene in a cell, it enhances output. The sequence itself is just DNA; “enhancer” is a functional label. Within an enhancer are clusters of TF binding motifs. Sequence-specific TFs bind these motifs and recruit coactivators. The activated enhancer then contacts the promoter — even if it is tens or hundreds of kilobases away on the linear chromosome — via chromatin looping.

Active enhancers have a characteristic chromatin signature:

H3K4me1: monomethylation of histone H3 lysine 4, enriched at enhancers (as opposed to H3K4me3, which marks active promoters)
H3K27ac: acetylation of H3 lysine 27, associated with active, open chromatin
Open chromatin: nucleosomes are displaced, making the DNA physically accessible — measurable by ATAC-seq

Enhancer RNAs (eRNAs)

Active enhancers are often themselves transcribed by Pol II, producing short, mostly non-polyadenylated RNAs called eRNAs. This does not violate the central dogma — the dogma prohibits RNA-to-protein encoding without a coding sequence, not transcription per se. eRNAs are simply RNA molecules transcribed from enhancer DNA that do not encode protein. Their function is debated: some evidence suggests they stabilize enhancer-promoter loops, other evidence treats them as transcriptional noise that marks enhancer activity without being causally important.

The enhancer-promoter contact

For an enhancer to boost a gene, physical proximity is required. The DNA between enhancer and promoter loops out, bringing the two elements into contact. The bridging molecule is Mediator — a large multi-subunit coactivator complex that physically connects TFs bound at the enhancer to Pol II sitting at the promoter. TF → Mediator → Pol II is the activation chain that converts basal trickle into robust, cell-type-specific transcription.

Genome architecture: TADs, CTCF, and cohesin

If enhancers can loop to promoters over long distances, what prevents a liver enhancer from accidentally activating a brain gene sitting nearby on the chromosome? The answer is three-dimensional genome organization.

Topologically associating domains (TADs)

The genome is partitioned into TADs — regions of roughly 100 kb to 1 Mb within which DNA contacts are frequent, and between which contacts are rare. Enhancers and promoters within the same TAD contact each other readily; those in different TADs are largely insulated from each other. This is the regulatory neighborhood structure of the genome.

CTCF

CTCF is a zinc finger protein with 11 zinc fingers, giving it flexible sequence recognition. It binds ~20,000 sites across the human genome and is strongly enriched at TAD boundaries. CTCF binding sites are directional — the motif has an orientation, and this orientation matters for what happens when cohesin meets it.

CTCF is often called an insulator protein: it defines the edges of regulatory neighborhoods. Disrupting a CTCF boundary (by mutation, deletion, or inversion of the binding site) can allow an enhancer to loop into an adjacent TAD and ectopically activate a gene it normally never contacts — a documented mechanism of cancer gene activation.

Cohesin and loop extrusion

Cohesin is a ring-shaped SMC (structural maintenance of chromosomes) complex. Its core is a heterotrimer: SMC1 and SMC3 form long coiled-coil arms joined at a hinge, and RAD21 (the kleisin subunit) closes the ring by linking the ATPase heads of SMC1 and SMC3. A fourth subunit, either STAG1 or STAG2 (mutually exclusive), associates with RAD21 and provides additional DNA contacts and docking surfaces for regulatory proteins including CTCF.

Cohesin’s canonical job is holding sister chromatids together after replication. In interphase, it also organizes chromatin by extruding DNA loops.

Loop extrusion: cohesin loads onto chromatin and reels in flanking DNA on both sides simultaneously, growing a loop in an ATP-dependent manner. This continues until the complex encounters a roadblock. The primary roadblock is CTCF — but only when oriented convergently relative to the incoming cohesin. Pairs of convergently oriented CTCF sites define stable loop anchors, which correspond to TAD boundaries.

Within a TAD, smaller cohesin-mediated loops bring specific enhancers into direct contact with their target promoters.

The cohesin loader: NIPBL/MAU2

Cohesin does not load itself. The heterodimeric complex NIPBL/MAU2 is the cohesin loader — required to initiate topological loading of cohesin onto chromatin, and to stimulate cohesin’s ATPase activity to drive loop extrusion. NIPBL is a large protein with a long disordered N-terminal domain; MAU2 wraps around this N-terminus, stabilizing NIPBL and enabling loader function.

A direct link between the TF/enhancer story and the cohesin/architecture story: sequence-specific TFs physically recruit NIPBL/MAU2 to their target enhancers. TF, NIPBL, and MAU2 can form a ternary complex in vivo. This explains how cohesin — which cannot itself read DNA sequence — gets concentrated at specific active enhancers. The chain is:

TF binds its motif at the enhancer
TF recruits NIPBL/MAU2
NIPBL/MAU2 loads the cohesin ring
Cohesin extrudes a loop
Enhancer contacts promoter; Mediator bridges TF to Pol II; transcription fires

The cohesin unloader: WAPL

Cohesin is not permanently resident on chromatin — it turns over. WAPL is the cohesin unloader: it opens the cohesin ring and releases it from DNA, keeping residence times short and preventing pathological over-accumulation. The balance between NIPBL/MAU2 (loading) and WAPL (unloading) determines how much cohesin is on chromatin at any given time and therefore how much loop extrusion occurs.

Mutations in cohesin subunits (SMC1A, SMC3, RAD21) or its regulators cause Cornelia de Lange syndrome (CdLS), a multisystem developmental disorder — illustrating how essential proper cohesin dosage is for normal development.

The full picture

Sequence-specific TF
  ↓ binds motif at enhancer DNA
  ↓ recruits NIPBL/MAU2 (cohesin loader)
  ↓ cohesin loads and extrudes a loop
  ↓ CTCF sites at TAD boundaries constrain the loop
  ↓ enhancer physically contacts promoter

At promoter:
  TFIID recognizes TATA box
  → TFIIA, TFIIB join
  → Pol II–TFIIF recruited
  → TFIIE, TFIIH complete PIC
  → TFIIH unwinds DNA, phosphorylates CTD
  → Pol II fires

TF at enhancer → Mediator → Pol II at promoter
  = boost from basal trickle to robust cell-type-specific output

CTCF and cohesin set up the permissive spatial context — they determine which enhancers are in the same regulatory neighborhood as which promoters. TFs and Mediator provide the instructive signal — they determine whether Pol II at a given promoter fires at all, and at what level. Both layers are required for normal gene regulation.

Key proteins at a glance

protein	what it is	what it does
Pol II	12-subunit RNA polymerase	transcribes DNA → RNA
TFIID/A/B/E/F/H	general transcription factors	assemble PIC at every promoter
sequence-specific TFs	DNA-binding proteins (zinc finger, bHLH, etc.)	read enhancer motifs; provide cell-type specificity
Mediator	large coactivator complex	bridges enhancer TFs to Pol II
CTCF	11-zinc-finger insulator protein	binds TAD boundaries; stalls cohesin
cohesin (SMC1/SMC3/RAD21 + STAG1 or STAG2)	ring-shaped SMC complex	extrudes chromatin loops; organizes TADs
NIPBL/MAU2	cohesin loader heterodimer	loads cohesin; stimulates ATPase; recruited to enhancers by TFs
WAPL	cohesin unloader	releases cohesin from chromatin; controls residence time