Systems Biology: Networks, Circuits, and Emergent Life

1. Types of Biological Networks

Different molecular interaction types form different network classes, each with characteristic topology and function:

Gene regulatory networks (GRN): Directed graph where transcription factors (TFs) bind to promoters and activate or repress target gene expression. Nodes = genes, edges = regulatory interactions (+/−). Human GRN: ~1,500 TFs regulating ~25,000 targets. Key property: scale-free topology (most genes few connections, few "hub" TFs thousands).
Protein-protein interaction networks (PPI): Undirected graph of physical protein binding. Human PPI network: ~300,000 interactions (estimate, highly incomplete). Hubs often essential genes — knock them out → cell death.
Metabolic networks: Directed bipartite graph (metabolites ↔ reactions). Stoichiometric matrix S: S·v = 0 at steady state (S = stoichiometry, v = flux vector). Flux balance analysis (FBA) optimises objective function (growth rate) subject to flux constraints.
Signalling networks: Proteins transmit signals from cell surface to nucleus via post-translational modifications (phosphorylation, ubiquitination). Dynamic, fast (seconds to minutes).

2. Network Motifs: Building Blocks of Circuits

Alon (2007) — network motifs: patterns that appear far more frequently in biological networks than in random networks with same degree distribution. Key motifs: 1. Negative self-regulation (autorepression): TF represses its own transcription. ~50% of E. coli TFs autorepress. Function: speeds up response time, reduces expression noise (variance). Alon lab: responding protein concentration rises and overshoots less with autorepression \to faster, more precise adaptation. 2. Feedforward loop (FFL) — 8 subtypes, 2 coherent / 2 incoherent main types: X activates Y, X activates Z, and Y activates Z: Type 1 coherent FFL (majority in E. coli gene networks): Z turns on only when BOTH X and Y are present. Function: sign-sensitive delay — Z responds to persistent X signal but ignores brief pulses of X (because Y has slow dynamics). ~ low-pass filter / noise filter. Type 1 incoherent FFL: X activates Z but Y inhibits Z. X also activates Y. Function: pulse generator — Z turns on when X activates, then turns off when Y accumulates. Transient response. 3. Positive feedback / bistable switch: X activates Y, Y activates X. Two stable states: BOTH off or BOTH on. Bistability \to decision-making, cell fate commitment. Example: Mos/MAP kinase in Xenopus oocyte maturation.

3. ODE Models of Gene Expression

Simple gene expression model: dm/dt = α_m - β_m \cdot m (mRNA production - degradation) dp/dt = α_p \cdot m - β_p \cdot p (protein synthesis - degradation) where m = mRNA concentration, p = protein concentration α_m = transcription rate (with/without activator) β_m = mRNA degradation rate (t½ mRNA ~3-10 min in E. coli; ~30-60 min mammals) α_p = translation rate β_p = protein degradation rate (t½ protein ~hours-days; ~20 hours in E. coli on avg) Steady state: m_ss = α_m/β_m, p_ss = α_p \cdot α_m / (β_m \cdot β_p) Response time (to step change in transcription): t_response \approx ln(2) \cdot max(1/β_m, 1/β_p) Determined by slowest degradation rate (usually protein). Hill function (transcriptional activation): f(X) = X^n / (K_d^n + X^n) (activator) f(X) = 1 / (1 + (X/K_d)^n) (repressor) K_d = dissociation constant (concentration at half-maximal activation) n = Hill coefficient (cooperativity; sigmoidal for n > 1) n = 1: Michaelis-Menten (no cooperativity), graded response n = 2-4: switch-like response; n = \infty: perfect binary switch Bistability requires: n \geq 2 in a positive feedback loop (approximately)

4. Signalling Cascades: The MAPK Pathway

The MAPK (Mitogen-Activated Protein Kinase) cascade is a conserved signalling module controlling cell proliferation, differentiation, and stress responses. Its architecture in mammalian cells:

Level 1 — Ras (GTPase): Activated receptor tyrosine kinase (e.g. EGFR) recruits GEF (Sos) → Ras-GTP. Inactivated by GAP proteins.
Level 2 — Raf (MAPKKK): Ras-GTP recruits and activates Raf kinase. Raf phosphorylates MEK.
Level 3 — MEK (MAPKK): Dual-specificity kinase — phosphorylates ERK on both Thr and Tyr residues (unusual dual requirement).
Level 4 — ERK (MAPK): Doubly phosphorylated ERK (ppERK) is active. Targets: transcription factors (Elk-1, c-Fos), ribosomal enzymes (RSK), cell cycle regulators (cyclin D1).

Ultrasensitivity and bistability: The cascade's three-tier architecture, combined with the dual phosphorylation requirement for ERK activation, creates a highly ultrasensitive (steep input-output) response. Goldbeter and Koshland (1981) showed that opposing kinase/phosphatase cycles can generate apparently cooperative behaviour (apparent Hill coefficient >1) without actual cooperativity in individual reactions. This "zero-order ultrasensitivity" may be common in signalling and explains how cells make switch-like decisions.

5. The p53-Mdm2 Oscillator

p53 ("guardian of the genome") oscillator after DNA damage: Feedback structure: p53 \to activates \to Mdm2 (transcriptionally) Mdm2 \to inhibits \to p53 (ubiquitin-mediated degradation) DNA damage \to disrupts Mdm2-mediated degradation of p53 ODE model (simplified Geva-Zatorsky et al. 2006): dp53/dt = α_p53 - β_p53 \cdot p53 \cdot Mdm2 ... (production - Mdm2-mediated degradation) dMdm2/dt = α_Mdm2 \cdot p53 - β_Mdm2 \cdot Mdm2 (transcription by p53 - degradation) [Plus delay term for Mdm2 nuclear transport ~30-40 min] Result: pulses of p53 and Mdm2 every ~5-6 hours after DNA damage. Observed in: single-cell live imaging of fluorescent p53 in MCF7 cells. Each pulse corresponds to one "attempt" to assess damage and trigger apoptosis. p53 digital response: Number of pulses encodes damage severity. Small damage: 1-2 pulses \to DNA repair, survival. Severe damage: many pulses \to apoptosis decision. Cancer relevance: p53 mutated in ~50% of all human cancers (most common mutation in cancer). Loss of p53 \to cells bypass apoptosis after DNA damage \to uncontrolled growth. MDM2 overexpressed in ~10% of cancers \to too much p53 degradation.

6. Boolean Networks and Attractor States

When kinetic parameters are unknown, Boolean networks offer a coarser but computationally tractable model. Each gene is either ON (1) or OFF (0), with logical update rules:

Boolean network: State: vector (x₁, x₂, ..., x_N) ∈ {0,1}^N Update: x_i(t+1) = f_i(x₁(t), ..., x_N(t)) — logical function 2^N possible states. With sequential or synchronous updates, state space trajectories are deterministic → converge to attractors: Fixed-point attractor: single state (stable cell type / developmental fate) Limit cycle: oscillating trajectory (cell cycle, circadian oscillator) Kauffman's NK model (1969): N genes, each with K inputs (randomly assigned). K=2: ordered regime — few attractors of small length (robust to perturbations) K>2: chaotic regime — exponential number of long attractors (fragile) Real GRNs: K ~ 2 (sparse connectivity) → robust gene regulation Example — Drosophila segment polarity network (Albert & Othmer 2003): 15-gene Boolean network recapitulates correct gene expression patterns for all segments of fly embryo. 7 attractors matching 7 observed cell types. Robust to >90% single-gene perturbations.

7. Synthetic Biology Applications

Synthetic biology applies engineering design principles — standardised parts, modular design, abstraction — to build novel biological circuits:

Toggle switch (Gardner et al., 2000): Two repressors, each repressing the other's promoter. Two stable states: R₁-on/R₂-off AND R₁-off/R₂-on. Switching by transient induction. First engineered bistable circuit. Implemented in E. coli.
Repressilator (Elowitz & Leibler, 2000): Three repressor genes (lacI, tetR, cI) in a cycle — each represses the next. Result: oscillating protein levels with ~150-minute period in E. coli. First engineered genetic oscillator. Demonstrated that periodic dynamics can be designed synthetically.
Biosensors: Cells engineered to detect pollutants, metabolites, or pathogens. Transcription factor activated by target chemical → drives reporter gene. Applications: water contamination detection, diagnostics.
Metabolic engineering: Artemisinin (antimalarial drug, originally from A. annua plant) produced in engineered S. cerevisiae at industrial scale (Keasling lab / Amyris). Farnesyl pyrophosphate → artemisinic acid via 10 heterologous genes. Reduces production cost 10-fold.
CAR-T cell engineering: T cells engineered with chimeric antigen receptors — synthetic receptor proteins binding to tumour antigens → activate T cell killing. FDA-approved for certain blood cancers. Logic-gated CARs (AND, NOT gates using two receptor types) under development for solid tumours.