Transcriptional Control of Gene Expression

Introduction

Every cell in a multicellular organism carries the same genome, yet a neuron, a liver cell, and a white blood cell are profoundly different in structure, function, and behavior. The difference lies in gene expression — which genes are active and which are silent at any given moment. The most important level of control is transcriptional regulation: the decision of whether or not to transcribe a gene into RNA.

This lesson examines how transcription factors read DNA sequences, how activators and repressors work together in combinatorial logic to create precise expression patterns, how genetic switches generate stable cell states and oscillatory behaviors, and the computational methods for analyzing transcription factor binding, regulatory elements, and gene regulatory networks.

The Different Cell Types Contain the Same DNA

A key insight of modern biology is that all cells in a multicellular organism are genetically identical (with rare exceptions like rearranged immunoglobulin genes in B cells). The genome of a skin cell and the genome of a neuron are the same. What differs is which genes are expressed.

A typical human cell expresses roughly 10,000–15,000 of its ~20,000 protein-coding genes at any given time. The specific set — called the cell’s transcriptome — varies by cell type, developmental stage, and environmental conditions. Even within a single cell type, gene expression changes dynamically in response to signals.

Different Cell Types Synthesize Different Sets of Proteins

The protein complement of a cell (its proteome) is what gives it its identity. A red blood cell is packed with hemoglobin; a pancreatic beta cell makes insulin; a muscle cell is rich in actin and myosin. These proteins are the same in every individual of a species — what differs between cell types is which genes are transcribed and at what levels.

Some proteins are expressed in virtually all cells (housekeeping genes — those encoding metabolic enzymes, ribosomal proteins, cytoskeletal components), while others are expressed only in specific cell types (tissue-specific genes).

Gene Expression Can Be Regulated at Many Steps

Gene expression can be controlled at every step from DNA to protein:

However, for most genes, transcriptional control is the primary regulatory mechanism — it is energetically efficient (no wasted mRNA) and provides stable, long-lasting changes in cell state.

Gene Expression Atlases

Large-scale projects have cataloged gene expression across human tissues and cell types:

• GTEx (Genotype-Tissue Expression) — measures gene expression in 54 human tissues from hundreds of donors, linking expression levels to genetic variation (eQTLs)
• Human Protein Atlas — maps protein expression across tissues and cell types using immunohistochemistry and RNA-seq
• Expression Atlas (EMBL-EBI) — curates gene expression data across species and conditions

These resources enable cell type–specific expression profiling — determining which genes distinguish one cell type from another — and gene co-expression network analysis using methods like WGCNA (Weighted Gene Co-expression Network Analysis), which identifies modules of genes that vary together across conditions, suggesting shared regulation or function.

Transcription Regulators Contain Structural Motifs That Read DNA Sequences

Transcription factors (TFs) are proteins that bind specific DNA sequences to activate or repress transcription. The human genome encodes approximately 1,600 transcription factors — about 8% of all protein-coding genes.

The sequence of nucleotides in the DNA double helix can be read by proteins without unwinding the helix. The edges of the base pairs are exposed in the major groove and minor groove of the double helix, and different base pairs present different patterns of hydrogen bond donors and acceptors. The major groove is wider and more information-rich, making it the primary reading surface for most transcription factors.

Transcription factors typically contain a DNA-binding domain built from one of several well-characterized structural motifs:

Dimerization and Cooperative Binding

Dimerization is a common feature of transcription factors. Forming dimers (homodimers or heterodimers) doubles the number of DNA contacts and dramatically increases binding affinity and specificity. Leucine zipper proteins, for example, dimerize through their coiled-coil region and then grip the DNA with their basic regions (hence the name “bZIP”).

Transcription factors also bind cooperatively to DNA: the binding of one factor facilitates the binding of others nearby. Cooperativity arises from protein-protein contacts between adjacent transcription factors and from chromatin effects (the first factor may recruit a nucleosome remodeler that exposes binding sites for subsequent factors).

Nucleosome positioning near promoter regions varies according to the gene and organism. Many active promoters have a nucleosome-free region (NFR) flanked by well-positioned nucleosomes. Pioneer transcription factors (like FoxA) can bind nucleosomal DNA directly and initiate chromatin opening.

Activators and Repressors Control Gene Expression

Transcription regulators direct RNA polymerase to specific promoters through two basic mechanisms:

Activators promote transcription by:

• Recruiting the general transcription factors and RNA Pol II to the promoter (directly or via Mediator)
• Recruiting chromatin-remodeling complexes (SWI/SNF) that move or eject nucleosomes
• Recruiting histone acetyltransferases (p300/CBP) that open chromatin
• Releasing paused Pol II from the promoter-proximal pause site (through P-TEFb kinase)

Repressors inhibit transcription by:

• Competing with activators for DNA binding sites
• Interacting directly with activators to block their function
• Recruiting histone deacetylases (HDACs) that close chromatin
• Recruiting Polycomb repressive complexes that deposit H3K27me3
• Directing the assembly of condensed chromatin structures

The Lac Operon: A Classic Regulatory Switch

The E. coli lac operon illustrates fundamental principles of gene regulation. Three genes for lactose metabolism (lacZ, lacY, lacA) are controlled by:

• A repressor (LacI) that binds the operator and physically blocks RNA polymerase when lactose is absent
• An activator (CAP/CRP) that enhances transcription when glucose is low (cAMP is high)

The operon acts as a logical AND gate: both conditions (lactose present AND glucose absent) must be met for maximal transcription. This ensures the cell only makes lactose-digesting enzymes when they are needed and no better carbon source is available.

let promoter = "GCGCCCGGTATAATGCGCCGCGCCCCGCGCGCCG"
let intergenic = "ATAAATGATATTAACGATAAAGATATGATCTTACA"
print("TATA boxes in promoter: ")
print(Seq.kmer_count(promoter, 8))
print("CG dinucleotides in promoter: ")
print(Seq.kmer_count(promoter, 2))
print("CG dinucleotides in intergenic: ")
print(Seq.kmer_count(intergenic, 2))

Promoter regions are enriched for specific short sequence motifs. The TATA box (TATAAAT) is found ~25 bp upstream of the transcription start site in many eukaryotic promoters, while CpG dinucleotides are concentrated in CpG island promoters of housekeeping genes. Counting k-mers across a regulatory region reveals these characteristic signatures.

Eukaryotic Transcription Regulators Work in Groups

Unlike bacterial operons, where one or two regulators control a gene, eukaryotic genes are typically controlled by combinations of 5–10 or more transcription factors binding to multiple regulatory elements. This combinatorial control allows a limited number of transcription factors (~1,600 in humans) to generate the enormous diversity of expression patterns seen across hundreds of cell types.

A gene may require:

• Factor A AND Factor B for activation (AND logic)
• Factor A OR Factor C (OR logic)
• Factor A but NOT Repressor R (NOT logic)

This combinatorial approach explains how the same Pax6 transcription factor can activate different target genes in the eye, brain, and pancreas — depending on which other factors are present.

Transcription regulators also act synergistically — the combined effect of multiple activators is often far greater than the sum of their individual effects. This synergy arises from cooperative DNA binding, cooperative recruitment of Mediator and chromatin modifiers, and cooperative relief of nucleosome-mediated repression.

Enhancers, Silencers, and Insulators

Gene expression is controlled by several classes of regulatory DNA sequences:

Enhancers are the primary long-range regulatory elements. They can activate transcription from distances of up to 1 megabase and work in either orientation relative to the promoter. Enhancers function by DNA looping, mediated by the cohesin complex and other architectural proteins, bringing enhancer-bound transcription factors into direct physical contact with the promoter complex.

Silencers are the repressive counterpart of enhancers, binding repressor proteins to inhibit transcription over long distances.

Insulator DNA sequences prevent inappropriate enhancer-promoter interactions. The protein CTCF binds to many insulator elements and, together with cohesin, organizes the genome into topologically associating domains (TADs) that constrain enhancer-promoter communication within defined chromosomal neighborhoods.

Genetic Switches: The Bacteriophage Lambda Decision

The bacteriophage lambda switch is the best-understood genetic decision. When lambda infects E. coli, it “decides” between two fates: lysis (replication and cell killing) or lysogeny (integration into the host genome and dormancy). This decision is controlled by two competing regulatory proteins — CI (repressor, favoring lysogeny) and Cro (favoring lysis) — that bind to the same regulatory region but with different affinities and outcomes.

The lambda switch demonstrates several general principles:

• Positive feedback: CI activates its own transcription, stabilizing the lysogenic state
• Mutual repression: CI represses cro and Cro represses cI, creating a bistable switch
• Noise-driven decision: stochastic fluctuations in early gene expression tip the balance toward one fate or the other

Gene Regulatory Circuits: Feedback Loops and Oscillators

Regulatory circuits built from transcription factors create complex dynamic behaviors:

• Positive feedback loops — a transcription factor activates its own expression, creating bistable switches that lock cells into stable states (e.g., the MyoD loop in muscle differentiation)
• Negative feedback loops — a transcription factor represses its own expression, creating homeostasis (dampening fluctuations) or oscillations when combined with time delays
• Feed-forward loops — a transcription factor activates a target both directly and indirectly (through an intermediate), creating filters that respond only to sustained signals

Circadian clocks are the most familiar example of oscillatory gene expression. The mammalian circadian oscillator involves a negative feedback loop in which the CLOCK/BMAL1 complex activates transcription of Per and Cry genes, whose protein products accumulate, form complexes, and eventually repress CLOCK/BMAL1 activity — creating an oscillation with a ~24-hour period.

Master Transcription Regulators Drive Cell Differentiation

Master regulators are transcription factors that, when expressed, are sufficient to redirect cell fate. The best-known example is MyoD, which can convert fibroblasts into muscle cells when ectopically expressed. Similarly, Yamanaka factors (Oct4, Sox2, Klf4, Myc) can reprogram differentiated cells into induced pluripotent stem cells (iPSCs).

Master regulators typically work by activating a cascade of downstream transcription factors and establishing self-reinforcing positive feedback loops that stabilize the new cell state.

Position Weight Matrices and Sequence Motifs

Transcription factor binding specificity is described computationally using position weight matrices (PWMs) — matrices that specify the probability of each nucleotide at each position in the binding site. PWMs are derived from collections of experimentally determined binding sites.

Tools for discovering and analyzing sequence motifs include:

• MEME (Multiple Em for Motif Elicitation) — discovers novel motifs in a set of DNA sequences using expectation maximization
• HOMER (Hypergeometric Optimization of Motif EnRichment) — identifies enriched motifs in ChIP-seq peaks or other genomic regions, comparing to background sequences

Databases of known transcription factor binding motifs include:

• JASPAR — an open-access, curated, non-redundant database of TF binding profiles
• TRANSFAC — one of the oldest and most comprehensive motif databases

let expression = [
  {"gene": "BRCA1", "liver": 2.1, "brain": 0.4, "blood": 5.8},
  {"gene": "HBB", "liver": 0.1, "brain": 0.0, "blood": 98.3},
  {"gene": "GAPDH", "liver": 45.2, "brain": 41.7, "blood": 43.9},
  {"gene": "NEUROD1", "liver": 0.0, "brain": 32.6, "blood": 0.1},
  {"gene": "ALB", "liver": 87.4, "brain": 0.0, "blood": 0.3}
]
print("Gene expression across tissues (TPM):")
print(Viz.heatmap(expression, "gene"))

A gene expression heatmap reveals tissue-specific expression patterns at a glance. Housekeeping genes like GAPDH show similar levels across tissues, while tissue-specific genes like HBB (blood), NEUROD1 (brain), and ALB (liver) show dramatic enrichment in one tissue. These patterns reflect the combinatorial activity of transcription factors unique to each cell type.

ChIP-seq and Transcription Factor Binding Analysis

ChIP-seq (chromatin immunoprecipitation followed by sequencing) is the standard experimental method for mapping transcription factor binding sites genome-wide:

1. Cross-link proteins to DNA (formaldehyde)
2. Fragment chromatin (sonication)
3. Immunoprecipitate with an antibody against the TF of interest
4. Sequence the enriched DNA fragments
5. Map reads to the genome and call peaks (enriched regions)

MACS2 (Model-based Analysis of ChIP-seq) is the most widely used peak caller. It models the local background and identifies regions significantly enriched for reads.

ATAC-seq footprinting provides an alternative approach: open chromatin regions are identified by ATAC-seq, and the specific “footprints” left by bound transcription factors (protected from transposase insertion) can be detected computationally, revealing binding events without the need for specific antibodies.

Motif enrichment analysis examines ChIP-seq peaks or other regulatory regions for over-representation of known motifs, identifying which transcription factors are likely to be active at those sites.

Enhancer Identification and Enhancer-Promoter Interaction Mapping

Identifying enhancers and linking them to their target genes is a central challenge in regulatory genomics:

• Enhancer identification relies on characteristic histone marks (H3K4me1 and H3K27ac mark active enhancers) and chromatin accessibility (ATAC-seq or DNase-seq peaks away from promoters)
• eQTLs (expression quantitative trait loci) link genetic variants in regulatory regions to changes in gene expression, functionally connecting non-coding variants to target genes
• pcHi-C (promoter capture Hi-C) identifies physical contacts between promoters and distal regulatory elements
• The ABC model (Activity-by-Contact) predicts enhancer-gene connections by combining enhancer activity (accessibility × H3K27ac signal) with 3D contact frequency from Hi-C

CRISPR Screens and Massively Parallel Reporter Assays

Functional testing of regulatory elements at scale uses:

• CRISPR screens — systematically perturb regulatory elements (by deletion, CRISPRi repression, or CRISPRa activation) and measure effects on gene expression, identifying functional enhancers and their target genes
• MPRA (Massively Parallel Reporter Assays) — test thousands of regulatory sequences simultaneously for enhancer activity by cloning them upstream of a reporter gene with unique barcodes

Regulatory Variant Interpretation

Interpreting the functional impact of non-coding genetic variants is critical for human genetics:

• CADD (Combined Annotation Dependent Depletion) — scores the deleteriousness of every possible SNV in the genome by integrating diverse annotations
• LINSIGHT — estimates the probability that a non-coding variant is under negative selection
• RegulomeDB — annotates non-coding variants with regulatory information from ENCODE and other sources

Gene Regulatory Network Inference

Computational methods reconstruct the regulatory relationships between transcription factors and their target genes:

• SCENIC (Single-Cell rEgulatory Network Inference and Clustering) — infers gene regulatory networks from scRNA-seq data by combining co-expression analysis with motif enrichment
• GENIE3 and ARACNe — use machine learning and mutual information to infer regulatory relationships from expression data
• Boolean and ODE-based models — formalize regulatory networks as mathematical models (Boolean networks for discrete on/off logic; ordinary differential equations for continuous dynamics)
• Trajectory inference (Monocle, RNA velocity) — reconstructs developmental trajectories from scRNA-seq data, revealing how gene expression changes as cells differentiate
• Pseudotime analysis — orders cells along an inferred developmental trajectory, enabling analysis of gene expression dynamics without time-series experiments

let tp53_expr = [0.5, 1.2, 3.8, 7.1, 9.3, 8.7]
let mdm2_expr = [0.3, 0.9, 3.1, 6.4, 8.8, 8.1]
let gapdh_expr = [5.1, 5.3, 4.9, 5.0, 5.2, 5.1]
let r_tp53_mdm2 = Stats.pearson(tp53_expr, mdm2_expr)
let r_tp53_gapdh = Stats.pearson(tp53_expr, gapdh_expr)
print("TP53 vs MDM2 correlation: " + r_tp53_mdm2)
print("TP53 vs GAPDH correlation: " + r_tp53_gapdh)

Co-expression analysis identifies genes whose expression levels rise and fall together across conditions or tissues. A high Pearson correlation between TP53 and MDM2 reflects their well-known regulatory relationship (p53 activates MDM2 transcription; MDM2 targets p53 for degradation). In contrast, a housekeeping gene like GAPDH shows no correlation because its expression is stable regardless of p53 activity. Methods like WGCNA exploit these co-expression patterns to discover modules of co-regulated genes.

Exercise: Detect Promoter Elements

Use k-mer counting to identify regulatory motifs in a promoter sequence. Count all 4-mers and determine whether the TATA box motif (TATA) is present. Then count CG dinucleotides to assess CpG density.

let promoter = "GCGCGTATAATCGCGGCCGCGATCGCGGC"
let dimers = Seq.kmer_count(promoter, 2)
let tetramers = Seq.kmer_count(promoter, 4)
print("Dinucleotide counts:")
print(dimers)
print("Tetramer counts:")
print(tetramers)
// Classify this promoter based on the k-mer results
let answer = "CpG island with TATA box"
print(answer)

Exercise: Measure Expression Correlation

Two genes, MYC and its known target CDK4, are measured across six experimental conditions. Calculate their Pearson correlation to determine whether they are co-expressed. Compare with an unrelated gene to confirm specificity.

let myc = [1.2, 3.5, 6.8, 9.1, 7.4, 2.3]
let cdk4 = [0.8, 2.9, 5.7, 8.3, 6.9, 1.8]
let actb = [12.1, 11.8, 12.3, 12.0, 11.9, 12.2]
let r_target = Stats.pearson(myc, cdk4)
let r_control = Stats.pearson(myc, actb)
print("MYC vs CDK4: " + r_target)
print("MYC vs ACTB: " + r_control)
// Based on correlation values, is CDK4 co-expressed with MYC?
let answer = "Co-expressed"
print(answer)

Exercise: Identify Gene Regulation Patterns

A gene expression matrix records expression levels of four genes across three tissues. Use a heatmap to visualize the data. Then compute the correlation between the two most tissue-specific genes to determine whether they share a regulatory program or are independently regulated.

let data = [
  {"gene": "INS", "pancreas": 95.0, "liver": 0.2, "brain": 0.0},
  {"gene": "ALB", "pancreas": 0.1, "liver": 88.5, "brain": 0.0},
  {"gene": "SYP", "pancreas": 0.3, "liver": 0.0, "brain": 47.2},
  {"gene": "GAPDH", "pancreas": 40.1, "liver": 42.3, "brain": 39.8}
]
print(Viz.heatmap(data, "gene"))
let ins_vals = [95.0, 0.2, 0.0]
let alb_vals = [0.1, 88.5, 0.0]
let r = Stats.pearson(ins_vals, alb_vals)
print("INS vs ALB correlation: " + r)
// Are INS and ALB co-regulated or independently regulated?
let answer = "Independent"
print(answer)

Knowledge Check

Summary

In this lesson you covered transcriptional regulation in depth:

• All cells in a multicellular organism contain the same DNA; gene expression determines cell identity
• Gene expression can be regulated at many steps; transcriptional control is the primary mechanism
• Gene expression atlases (GTEx, Human Protein Atlas) and co-expression networks (WGCNA) characterize expression across tissues
• Transcription factors read the major groove of DNA using structural motifs (helix-turn-helix, zinc finger, leucine zipper, helix-loop-helix)
• Dimerization and cooperative binding increase TF affinity and specificity; nucleosome positioning influences access
• Activators recruit Pol II, Mediator, chromatin remodelers, and histone acetyltransferases; repressors recruit HDACs and Polycomb
• The lac operon demonstrates how repressors and activators create an AND-gate regulatory switch
• Combinatorial control by 5–10+ TFs per gene enables ~1,600 TFs to generate diverse expression patterns
• Enhancers act over megabase distances via DNA looping; insulators (CTCF/cohesin) define regulatory boundaries (TADs)
• Genetic switches (lambda, bistable circuits) use mutual repression and positive feedback
• Circadian clocks use negative feedback with time delays to generate ~24-hour oscillations
• Master regulators (MyoD, Yamanaka factors) can redirect cell fate
• PWMs describe TF binding specificity; MEME and HOMER discover motifs; JASPAR catalogs known motifs
• ChIP-seq with MACS2 maps TF binding genome-wide; ATAC-seq footprinting detects binding without specific antibodies
• Enhancer identification uses H3K4me1/H3K27ac marks; eQTLs, pcHi-C, and the ABC model link enhancers to target genes
• CRISPR screens and MPRAs functionally test regulatory elements at scale
• Regulatory variant interpretation (CADD, LINSIGHT, RegulomeDB) assesses non-coding variant impact
• Network inference (SCENIC, GENIE3, ARACNe) reconstructs regulatory networks; trajectory inference (Monocle, RNA velocity) maps developmental dynamics