What this course is, and how to use it
This is the first course in University Medicine: cell and molecular biology for medicine 医学细胞与分子生物学. It is the preclinical base every later course stands on: biochemistry fills the cell, physiology makes it work, pathology explains when it breaks.
Scope limit, stated once: this is a study course, not a medical degree — no patients, no clinical placements; every case below is fictional and nothing here licenses or guides treatment.
- Units 1.1–1.2 are the cell and its membranes: organisation, transport, and the electrical language of ions.
- Units 1.3–1.5 are the information line: replication, gene expression, and its regulation.
- Units 1.6–1.7 are decision-making: signalling between cells, and the cell's own life-and-death controls.
The medicine habit: mechanism before fact. Every number in this course is something you recompute — a Nernst potential, a compaction ratio, a mutation probability — and every clinical link names the mechanism it teaches.
Cell organisation and organelles
Why it matters. A eukaryotic cell is a factory of compartments: each organelle is a reaction vessel with the right pH, the right enzymes and the right surface. Medicine reads a cell the way an engineer reads a plant diagram — what lives where, and why.
The idea, with a picture. Organelles live on a size ladder spanning three orders of magnitude: a ribosome 核糖体 is 25 nm, a mitochondrion 线粒体 2 µm, the nucleus 细胞核 6 µm, the whole cell 细胞 20 µm. Function follows size: anything the whole cell must do quickly needs a membrane surface to do it on.
Definitions with numbers. The nucleus stores 2.2 m of DNA (6.4 × 10⁹ bp × 0.34 nm/bp) inside a 6 µm envelope — a length about 4 × 10⁵ times the nucleus's width. The rough endoplasmic reticulum 糙面内质网 is ribosome-studded protein export; the Golgi 高尔基体 sorts and glycosylates; lysosomes 溶酶体 keep pH ≈ 4.8 to dismantle cargo; mitochondria make ATP. Ribosomes translate mRNA (unit 1.4).
The derivation. Transport in and out scales with membrane surface area ($4\pi r^{2}$); demand — protein synthesis, ATP turnover — scales with volume ($\tfrac{4}{3}\pi r^{3}$). The ratio $SA/V = 3/r$: double the radius and every unit of cytoplasm gets half the membrane service. That single line is why cells stay small, why the ER folds so much surface into the cytoplasm, and why the intestinal lining is villi rather than flat.
Worked example 1 — the size ladder, verified. A 20 µm hepatocyte: $SA/V = 3/10\ \mu m^{-1} = 0.3$. A 200 µm sphere (an oocyte): $0.03$ — ten times worse ✓, which is why large cells are rare and membrane-folded. Cross-check the ladder: 20 µm / 25 nm = 800 — a ribosome is as small against a cell as a cell is against a 1.6 cm marble ✓.
Worked example 2 — clinical link: the lysosome that fails. A (fictional) patient with progressive muscle weakness is found to lack a single lysosomal hydrolase. Mechanism, not name, is the answer: the substrate cannot be dismantled at pH 4.8, so it accumulates, the lysosome swells and the cell dies by being unable to recycle. One missing enzyme in one compartment = one storage disease: the organelle map is the diagnosis map.
Common mistakes. Treating "the cytoplasm" as one well-mixed bag — compartments differ in pH, calcium and redox state by design. Reading organelle counts as fixed (a liver cell has ~1000 mitochondria; a quiescent lymphocyte, few). And forgetting the membrane is where the cell meets the world.
Summary. Organelles are reaction vessels on a 25 nm–20 µm ladder; $SA/V = 3/r$ caps cell size and explains folded membranes; 2.2 m of DNA packs into a 6 µm nucleus; compartment failure is disease.
Membranes and transport
Why it matters. Every drug target, every nerve impulse and every kidney's work reduces to one question: how does a molecule cross a lipid bilayer? The answer is always one of four doors.
The idea, with a picture. The lipid bilayer 脂双层 is a self-assembled 8 nm sheet impermeable to ions. Crossing is by: simple diffusion 自由扩散 (small, lipophilic), facilitated diffusion 易化扩散 through channels or carriers (down gradient, no ATP), primary active transport 原发性主动转运 (ATP pumps, against gradient), and secondary active transport 继发性主动转运 (an ion gradient drives something else uphill).
The Nernst equation. For ion X at 37 °C the Nernst equation 能斯特方程 gives the voltage at which the ion stops moving:
Typical mammalian values — K⁺ (140 in / 5 out): $E_{K} = 61.5\cdot \log_{10}(5/140) = -89.0$ mV ✓; Na⁺ (12/145): $+66.6$ mV; Cl⁻ (10/110, $z=-1$): $-64.0$ mV; Ca²⁺ (100 nM/1.2 mM, $z=2$): $+125.4$ mV.
The derivation. At equilibrium the electrical force balances the chemical: $zF E = RT\ln([out]/[in])$ — that IS the Nernst line. With several ions permeable at once, replace the single ratio with a permeability-weighted sum (Goldman–Hodgkin–Katz): with $P_{K}:P_{Na}:P_{Cl} = 1:0.04:0.45$,
✓ the resting potential 静息电位 — about 20 mV above $E_{K}$, because Na⁺ leaks in and the Na⁺/K⁺-ATPase (3 Na⁺ out, 2 K⁺ in, 1 ATP) holds the gradients it needs.
Worked example 1 — hyperkalaemia arithmetic. A (fictional) patient's extracellular K⁺ doubles from 5 to 10 mM. New Nernst: $E_{K}$ shifts by $61.5\log_{10}(10/5) = +18.5$ mV, from −89 to −71 mV ✓ — the depolarised $E_K$ flattens the resting voltage, which is why raised potassium is a cardiac emergency: the mechanism is one logarithm.
Worked example 2 — clinical link: a broken chloride door. The most common cystic-fibrosis variant deletes one phenylalanine (ΔF508) from a Cl⁻ channel, so the channel misfolds and never reaches the membrane. Follow the mechanism: Cl⁻ cannot leave → less salt reaches the airway surface, so less water follows it by osmosis → the surface liquid dehydrates → mucus cannot clear. One folding defect, one door, one disease — and small molecules that hold the fold open are the therapy rationale.
Common mistakes. Using the Nernst equation on non-equilibrium gradients (it is an equilibrium statement; GHK is the honest steady approximation). Forgetting $z$ in the denominator (Ca²⁺ halves the slope). Calling channels "active" — only pumps burn ATP.
Summary. Four doors cross the bilayer; $E_X = (61.5/z)\log_{10} ([out]/[in])$ mV at 37 °C; GHK with $P_K{:}P_{Na}{:}P_{Cl} = 1{:}0.04{:}0.45$ gives −67 mV; 3 Na⁺ out / 2 K⁺ in per ATP keeps the battery charged.
DNA replication and repair
Why it matters. A cell must copy 6.4 billion letters with almost no mistakes, once per division, forever. The machinery that manages this — and the repair systems behind it — is where hereditary disease, mutagenesis and most oncology begin.
The idea, with a picture. The replication fork 复制叉: helicase unzips, polymerases extend 5′→3′ only, so one strand is copied continuously (leading 前导链) and the other backwards in Okazaki fragments 冈崎片段 (lagging 后随链), each restarted by a primer.
Definitions with numbers. Rate: bacterial polymerase ~1000 nt/s; human ~50 nt/s per fork. Fidelity: raw polymerase error ~$10^{-5}$ per bp; proofreading 校对功能 buys ×100 → $10^{-7}$; mismatch repair 错配修复 another ×100 → $10^{-9}$ per bp ✓ — three systems deep, one billionth.
The derivation (the timing argument). Human S phase copies $6.4\times 10^{9}$ bp in 8 h. A single origin with two forks at 50 bp/s would need $6.4\times10^{9}/100 = 6.4\times10^{7}$ s ≈ 2 years. Replication works because ~30 000 origins fire in timed clusters: $6.4\times10^{9} / (100 \times 3\times10^{4}) \approx 2100$ s ≈ 35 min of pure synthesis ✓, wrapped in chromatin time to fill S phase.
Worked example 1 — the fidelity budget. After proofreading and mismatch repair the residual rate is ~$10^{-9}$/bp/division. For a 6.4 × 10⁹ bp diploid genome: $6.4\times10^{9} \times 10^{-9} \approx 6$ new mutations per daughter cell ✓ — the number oncology (unit 1.7) and genetics both lean on. Skip mismatch repair and the rate rises ×100: hundreds of mutations per division, which is hereditary non-polyposis colon cancer's mechanism.
Worked example 2 — clinical link: sunlit skin without repair. (Fictional patient) severe sunburn blisters after minutes outdoors, skin cancers by twenty. Mechanism: ultraviolet light fuses adjacent thymines into dimers; nucleotide excision repair 核苷酸切除修复 is the system that cuts them out; without it every sunbeam is a mutagen with no auditor. The lesson generalises: each repair pathway owns one lesion class, and its absence names one syndrome.
Common mistakes. Reading polymerases as reading 3′→5′ — they MOVE 3′→5′ along the template while MAKING 5′→3′. Calling S phase "copying once" without the origins argument. Quoting $10^{-9}$ as the raw polymerase rate (it is the post-repair rate).
Summary. Semiconservative, 5′→3′, one leading + one lagging strand; $10^{-5} \to 10^{-7} \to 10^{-9}$ fidelity ladder; ~30 000 origins turn 2 years into 8 hours; ~6 mutations survive per division; each repair pathway owns one lesion class.
Gene expression: transcription and translation
Why it matters. The gene is a recipe, not a machine: between DNA and protein stand RNA polymerase, processing, splicing and the ribosome — and most human disease mutations, it turns out, break the in-between steps, not the recipe itself.
The idea, with a picture. The central dogma 中心法则 with its eukaryotic detours: DNA → pre-mRNA 前体mRNA (transcription 转录 by Pol II at ~20 nt/s) → 5′ cap + poly(A) + splicing 剪接 (introns out, exons in) → mature mRNA → ribosome (translation 翻译) → protein (~3 amino acids/s). One gene, alternative exons, several proteins.
Definitions with numbers. The genetic code: 64 codons 密码子 = 61 sense + 3 stop; 20 amino acids from 4 bases in threes ($4^{3} = 64$ ✓ — the arithmetic OF the code). Degeneracy averages ~3 codons per amino acid; AUG starts (methionine), UAA/UAG/UGA stop.
The derivation (the counting argument). Why triplets? Two bases give $4^{2} = 16 < 20$: too few. Three give $4^{3} = 64 > 20$: enough, with surplus spent on degeneracy — mostly at the third position, which is why third-base substitutions are so often silent. The code's redundancy is not waste; it is damage tolerance.
Worked example 1 — timing a protein. Preproinsulin is 110 amino acids: translation at ~3 aa/s takes ~37 s; its 0.5 kb mRNA took ~25 s to transcribe at 20 nt/s. The numbers set expectations: making one insulin molecule costs about a minute of machinery, which is why secretion systems pre-build and store (unit 1.6 signalling triggers release, not synthesis).
Worked example 2 — clinical link: the splice that breaks. (Fictional pedigree) anaemia, and sequencing shows no coding mutation — the defect is a single base that creates a new splice site inside a β-globin intron. Mechanism: the spliceosome obeys consensus sequences; a new site lures it, a stretch of the intron is kept in the mRNA, the reading frame shifts and a premature stop follows. β-thalassaemia without a single amino-acid change: the dogma's middle step is the disease.
Common mistakes. Calling introns "junk" (regulatory and splicing-adjacent function, unit 1.5). Confusing the template strand with the coding sequence (the mRNA matches the coding strand, T→U). Expecting one gene = one protein (alternative splicing).
Summary. Pol II ~20 nt/s → processing (cap, poly(A), splice) → ribosome ~3 aa/s; $4^{3} = 64$ codons for 20 amino acids with third-base degeneracy; one insulin molecule ≈ one minute; splice-site mutations break genes without touching their protein code.
Regulation of gene expression
Why it matters. Every one of your ~20 000 genes exists in nearly every cell; what makes a neuron a neuron is which genes are ON, when, and how much. Regulation is the actual genome — and the most common site of both evolution and disease.
The idea, with a picture. Control stacks: transcription factors 转录因子 bind promoters and enhancers 增强子 (which can act from 100 kb away by DNA looping); chromatin 染色质 state — histone marks and CpG methylation 甲基化 — decides access; epigenetics 表观遗传学 is the heritable part of that access without changing sequence.
Definitions with numbers. Only ~1.5 % of the genome codes protein; regulatory sequence dwarfs it. 70–80 % of CpG dinucleotides in a differentiated cell are methylated — the switched-off baseline. The 2.2 m of DNA (unit 1.1) wraps nucleosomes (~200 bp each, about 7-fold shorter) folded into a ~30 nm fibre (about 40-fold); loops and further folding bring a mitotic chromosome near $10^{4}$-fold.
The derivation (the combinatorics of specificity). A transcription factor recognises ~6–8 bases. Random 6-mers occur every ~4096 bp — about 800 000 times along each strand of the 3.2 Gb haploid genome: far too often for clean regulation. Specificity comes from COMBINATIONS: several factors binding adjacent sites cut false hits multiplicatively ($4^{-6}$ to the power of cooperating sites) — the same logic that makes passwords long. Enhancers are cooperative logic gates, which is why regulation is network-shaped.
Worked example 1 — lactase persistence. The enzyme LCT falls after weaning in most of the world's population; northern Europeans carry a single base change ~14 kb UPSTREAM — inside an intron of a neighbouring gene, where it creates new enhancer binding. Mechanism chain: one regulatory base → one enhancer strengthened → LCT stays on into adulthood → milk digestible. A zero-protein-change trait that swept a continent: regulation evolves fastest.
Worked example 2 — clinical link: silencing gone wrong. (Fictional twin pair) identical genotypes, different severity: one carries methylation of the disease gene's promoter, the other does not. Same letters, different access — an epigenetic 表观遗传的 modifier of severity. The mechanism explains why "identical genes" never promises identical disease, and why methylation assays sit next to sequencers in the diagnostic laboratory.
Common mistakes. Equating epigenetic with permanent (marks divide with the cell but can be reprogrammed). Treating enhancers as promoters-with-a-different-name (orientation- and distance-independent vs the fixed start site). Reading "gene ON/OFF" as binary — expression is graded and tissue-timed.
Summary. ~20 000 genes, ~1.5 % coding; enhancers are cooperative combinatorial switches acting by looping; methylation (70–80 % of CpG) and histone marks gate access heritably; one upstream base can rewire a phenotype.
Cell signalling
Why it matters. Cells are not self-sufficient: hormones, neurotransmitters and growth factors are instructions arriving at the membrane. Signal transduction is how a message at the surface becomes biochemistry inside — and roughly half of all drugs act at this layer.
The idea, with a picture. Two canonical cascades: a GPCR G蛋白偶联受体 (the hormone never enters; a G-protein switch relays to second messengers 第二信使 like cAMP) and an RTK 受体酪氨酸激酶 (receptor phosphorylates itself; adaptor proteins pass the signal to RAS→MAPK growth commands). Both AMPLIFY: one surface event becomes thousands of intracellular events.
Definitions with numbers. Resting cAMP ~$10^{-8}$ M; a maximal hormone pulse drives it to ~$10^{-6}$ M — a hundredfold switch in seconds. One receptor can activate tens of G-proteins; each adenylyl cyclase makes ~1000 cAMP/s: amplification ~$10^{4}$ from hormone to second messenger ✓. GTPase activity of the Gα subunit is the built-in OFF timer (seconds).
The derivation (the amplification product). Signal output = $\prod$ (stage gains): hormone → receptor (1 : ~20 G-proteins) → cyclase (each ~1000 cAMP/s) → PKA (each phosphorylates many targets). The gains MULTIPLY, so the cascade is exponential in gain and, crucially, every stage is a control point: phosphodiesterase destroys cAMP, GTPase switches off Gα, phosphatases strip phosphates. Amplification and reset are the same machinery seen from two sides.
Worked example 1 — the cAMP ledger. Basal $10$ nM in a $2\times10^{-12}$ L hepatocyte is $10^{-8}\ \mathrm{mol/L} \times 2\times10^{-12}\ \mathrm{L} \times N_A \approx 12\,000$ molecules ✓. One second of 10 activated cyclases at 1000/s would produce 10 000 cAMP — the entire resting pool is replaced in about a second, so the OFF enzymes matter exactly as much as the ON ones.
Worked example 2 — clinical link: the switch that cannot turn off. (Fictional outbreak vignette) profuse secretory diarrhoea within hours. Mechanism: cholera toxin ADP-ribosylates the Gαs subunit, killing its GTPase — the OFF timer is disabled, adenylyl cyclase runs constitutively, cAMP stays at maximum, and the intestinal cell pumps Cl⁻ (unit 1.2's door) and water into the lumen. Therapy is fluid replacement while the toxin-modified proteins decay: you cannot inhibit a switch that is already welded on.
Common mistakes. Saying the hormone "enters the cell" (GPCR ligands do not; steroid hormones do — different mechanism entirely). Forgetting second messengers are destroyed continuously (PDE vs caffeine and other inhibitors). Reading kinase cascades as linear (feedback everywhere).
Summary. GPCR → G-protein → cAMP → PKA; RTK → RAS → MAPK; stage gains multiply to ~$10^{4}$; every ON stage has an OFF enzyme; cholera is the GTPase welded open.
Cell cycle, cell death and cancer biology
Why it matters. Division and death are the cell's most expensive decisions, gated by checkpoints. Cancer is not one disease but the accumulated failure of those gates — and every gate is a named molecular system you have already met (repair, signalling, regulation).
The idea, with a picture. The cell cycle 细胞周期: G1 (growth, ~11 h) → S (DNA replication, ~8 h, unit 1.3) → G2 (~4 h) → M (mitosis, ~1 h), with checkpoints 检查点 at G1/S (the restriction point), G2/M and the spindle assembly check. Cells can leave to G0 quiescence — or to apoptosis 细胞凋亡, programmed death, executed by caspases.
Definitions with numbers. A typical cycling human cell: 24 h total (11 + 8 + 4 + 1) ✓. p53 (guardian of the genome) pauses the cycle after DNA damage for repair — or commits the cell to apoptosis. Cancer's requirement: ~4–7 rate-limiting driver events, plus replicative immortality (telomerase) and escape from apoptosis.
The derivation (the doubling count). One cell doubling unchecked: after $n$ divisions there are $2^{n}$ cells. A 1 cm³ tumour masses about 1 g ≈ $10^{9}$ cells: $2^{30} = 1.07\times10^{9}$ ✓ — a detectable tumour is already THIRTY generations old. The clinical implication is arithmetic: screening intervals and "why did it appear so fast" (it did not — it was silent for years) live on this exponential.
Worked example 1 — the driver-event lottery. Unit 1.3: ~6 mutations survive per division. Most hit passengers (vast non-coding space, unit 1.5); the chance any given mutation disables a specific gate is small — but $10^{9}$ cells × decades buys many trials, each driver raising the rate further. Cancer incidence rising with age is this arithmetic: ~$10^{16}$ divisions in a lifetime, each holding a lottery ticket.
Worked example 2 — clinical link: a family born one gate down. (Fictional pedigree) early sarcoma, breast cancer and a brain tumour across three generations — one inherited mutated p53 allele (Li–Fraumeni pattern). Mechanism: every cell starts one gate down; a single somatic hit on the second allele removes the checkpoint entirely. That is the two-hit logic of a tumour suppressor gene 抑癌基因, and it ties the course together: repair (1.3), signalling (1.6) and regulation (1.5) are the gates cancer must pass.
Common mistakes. Treating apoptosis as passive death (it is active, programmed, ATP-dependent — necrosis is the messy one). Reading oncogenes and tumour suppressors as the same (gain-of-function vs loss-of-function; one hit vs two). Equating fast growth with fast cycles (usually fewer cells dying, not faster divisions).
Summary. 24 h cycle, three checkpoints, G0 and apoptosis as exits; p53 is the damage gate; $2^{30} \approx 10^{9}$ cells = 1 cm³ = thirty silent doublings; 4–7 drivers plus immortality.