Contents
05 — Transcription Part 1 — Molecular foundations

05 — Transcription

Before this: Ch 02 · Ch 03 · Ch 04 · Time: ~35 min

What you'll be able to do

The core idea

Replication copies the archive. takes a local, temporary, non-destructive read of one strand, and everything else follows from those three adjectives.

Local, so the machine must be told where to start and stop — and those instructions sit in flanking DNA that never appears in the product. Temporary, so the polymerase can afford to be sloppy: an error in one of thousands of transcripts is a rounding error. Non-destructive, so copies are cheap, which means the cell can regulate almost everything at the point of making them.

The reframe worth carrying:

Expression level is a rate, not a state. A gene is never simply "on". A fires some number of times per hour; each transcript decays with some half-life; abundance is the ratio. Every mechanism in this chapter and all of Part 4 is an intervention on one of those two numbers.


1. The chemistry, and what it inherits from replication

runs the same reaction as : the growing chain's 3'-hydroxyl attacks the α-phosphate of an incoming NTP, releasing pyrophosphate. So the chain grows 5'→3' and the template is read 3'→5'. Two differences from Ch 04 matter.

No primer. RNA polymerase initiates de novo, joining the first two nucleotides with nothing to extend from. This is also why replication uses an RNA primer: the enzyme that can start from nothing makes the starter for the enzyme that cannot.

Far lower fidelity. mis-incorporates at roughly 10⁻⁴–10⁻⁶ per nucleotide, against replication's ~10⁻¹⁰ per base per replication (Ch 04) — four to six orders of magnitude sloppier, and irrelevant, because the product is one disposable instance among many. (The rate of ~1.3 × 10⁻⁸ per bp per generation (verified-facts) is a different unit and not the like-for-like comparator: it accumulates over hundreds of divisions and counts unrepaired chemical damage as well as polymerase error.)

Inside the enzyme sits the transcription bubble: ~12–14 bp held unwound, containing a ~9 bp RNA:DNA hybrid. The RNA is peeled off and threaded out a separate exit channel so the duplex re-anneals behind. The bubble travels with the enzyme, overwinding DNA ahead and underwinding behind — which is why matter as much here as in replication.

2. Template, coding, and the naming disaster

The polymerase physically reads one strand — the template. The RNA is complementary to it, and therefore identical to the other strand with U for T. That untouched strand is the one everyone writes down, because it reads like the product.

Strand actually readStrand the RNA resembles
templatecoding
antisensesense
non-codingnon-template
minus (of the gene)plus (of the gene)

Four synonyms each, and the column called "coding" is the one that mechanistically codes for nothing. Your FASTA file, GenBank, and every paper show it anyway.

                     transcription unit  ──────────────▶
5'- G C T A T G G A A T C C G T A -3'    coding / sense / non-template
    | | | | | | | | | | | | | | |
3'- C G A T A C C T T A G G C A T -5'    template / antisense / non-coding

RNA  5'- G C U A U G G A A U C C G U A -3'

Two consequences that bite in practice:

Template-ness belongs to a unit, not a . Genes point both ways and sometimes overlap, so one gene's template is its neighbour's coding strand. "Which strand is the template?" has no global answer.

The reference + strand is an arbitrary assembly convention carrying no biological meaning. Hence the strand field in every format, the need to reverse-complement a variant on a −-strand gene before reasoning about protein consequence, and the fact that "the C>T variant" is ambiguous until you say relative to what (Ch 41).

Positions are numbered from the transcription start site: +1 is the first nucleotide in the RNA, −1 the base immediately upstream. There is no position 0.

3. Bacteria: one engine, swappable address decoders

Bacteria have one for everything. The core enzyme (α₂ββ'ω) can catalyse and translocate but cannot find a ; add a σ factor and the holoenzyme can. Core is the engine, σ is the address decoder — one worker pool, pluggable dispatch.

The σ⁷⁰ promoter

    −35                 17 bp spacer               −10        6 bp    +1
    TTGACA ─────────────────────────────────────  TATAAT ───────────── A
    σ region 4 reads this                         σ region 2 melts this
    in the duplex, major groove                   and grips unpaired bases

Both are consensus sequences — the modal base at each position across many promoters, not a sequence any real promoter necessarily has. A promoter is properly scored by a position weight matrix, and strength is roughly monotone in that score. But a perfect-consensus promoter is too good: the holoenzyme binds so tightly it cannot escape, and output falls. Affinity and throughput are different objectives.

Two features to derive rather than memorise:

Why is −10 AT-rich? Because it has to melt. σ region 2 doesn't read it as a duplex — it melts ~13 bp, then grips specific unpaired bases, flipping two into pockets on the protein. A:T has two hydrogen bonds, G:C three (Ch 02), so AT-richness lowers the barrier to the one step that must happen here. The eukaryotic is AT-rich for a related but distinct reason — not melting but deformability: it sits well upstream of the Pol II bubble and is never unwound there, but TBP has to kink and unstack it ~90° (§6). Both exploit the same A:T/G:C hydrogen-bonding and base-stacking asymmetry, for two different jobs.

Why does spacer length matter but not spacer sequence? σ regions 2 and 4 are rigidly held apart on one protein, so the spacer must place −35 at the right distance and rotational angle. At ~10.5 bp per turn, one base changes the angle by ~34°. 17 ± 1 bp works; 15 or 20 does not, whatever lies between.

Tuning elements: an extended −10 (upstream TG) rescues a poor −35, and an AT-rich UP element contacted by the α-subunit C-terminal domains raises output roughly 30-fold at very strong promoters such as the rRNA (measured at rrnB P1, where reported values run ~30–70-fold in vivo).

σ factors as global program switches

Swapping σ reprograms genome-wide in one move.

σGeneRegulon
σ⁷⁰rpoDHousekeeping — most genes
σ³²rpoHHeat shock
σ³⁸rpoSStationary , general stress
σ²⁸fliAFlagellar and chemotaxis
σ²⁴rpoEEnvelope stress
σ⁵⁴rpoNNitrogen and alternative metabolism

σ⁵⁴ is the outlier: its holoenzyme forms a stable closed complex that cannot melt DNA alone, requiring an activator bound at an upstream site, typically 80–150 bp away, to loop over and hydrolyse ATP. That is a eukaryotic arrangement — action at a distance, ATP-driven melting — appearing once in bacteria.

The cycle

stateDiagram-v2 [*] --> Closed: holoenzyme binds promoter Closed --> Open: isomerisation, ~13 bp melted, no ATP Open --> Abortive: 2-15 nt made and released Abortive --> Open: retry, DNA scrunched, strain stored Abortive --> Elongation: promoter escape, sigma released Elongation --> Elongation: 40-90 nt/s, with pauses Elongation --> Terminated: hairpin plus U-tract, or Rho Terminated --> [*]: core recycled, new sigma loaded

Abortive initiation is not a defect. The polymerase grips the promoter and pulls downstream DNA into itself — scrunching — storing strain, making and dropping short transcripts until the strain breaks the σ–promoter grip. Escape, not binding, is often the slow step.

4. Bacterial termination: two mechanisms, both derivable

Intrinsic (Rho-independent). The nascent RNA carries a GC-rich inverted repeat followed immediately by 7–9 uridines. The hairpin is GC-rich so it folds fast and holds; the U-tract means the polymerase's grip on the template at that moment is 7–9 rU:dA pairs, the weakest base-paired hybrid that exists. The hairpin forms at the exit channel and tugs the transcript backwards while the grip is at its minimum. It lets go. Neither feature alone suffices: hairpin without U-tract gives a pause, U-tract without hairpin gives a slow patch.

Rho-dependent. Rho is a hexameric ATP-driven 5'→3' RNA translocase. It loads at a rut site — C-rich, G-poor, unstructured, and crucially not covered by — then chases the polymerase and prises the hybrid apart on catching a paused enzyme.

That last clause is the hinge for Ch 21. Rho loads only on naked RNA, and a translating ribosome covers RNA. So a translated message is protected and an untranslated one is not — which is the mechanism of polarity: a nonsense in the first gene of an reduces expression of the downstream genes too, because ribosomes fall off early, Rho loads on the exposed RNA, and the rest of the operon is never transcribed.

5. Eukaryotes: three polymerases

Pol IPol IIPol III
Subunits141217
Locationnucleolusnucleoplasmnucleoplasm
Products47S pre-rRNA → 28S, 18S, 5.8SmRNA, most , precursors, most snRNAtRNA, 5S rRNA, U6 snRNA, 7SL, some SINEs
upstream, dedicated factors (UBF, SL1)core promoter + often internal, downstream of the TSS
α-amanitinresistantexquisitely sensitiveintermediate
Scale~60% of in a growing cell; rRNA is ~80% of cellular RNA by massthe entire proteome and nearly all regulationsmall abundant structural RNAs

All three descend from the bacterial enzyme and share five subunits outright; the catalytic core is the same machine. What multiplied is the initiation apparatus. Two details worth keeping: α-amanitin, from the death cap mushroom, is still what operationally defines "Pol II transcript"; and Pol III type 1 and 2 promoters lie inside the transcribed region — a tRNA gene's promoter is the tRNA, which is why SINEs like Alu carry their own promoter wherever they land (Ch 19).

6. Pol II initiation: the pre-initiation complex

Pol II cannot recognise a at all. General deliver it.

Core promoter elements. The (TATAWAWR, roughly −31 to −24) is the famous one and the misleading one: depending how strictly the motif is defined, only about 10–25% of human core promoters have a recognisable one. Most are TATA-less, sit in , and initiate in a dispersed way — a cloud of start sites spread over tens of base pairs rather than one. Inr, BRE, DPE and MTE appear in various combinations; many promoters match none of them strongly.

TBP is recruited even where there is no TATA box — brought in as part of TFIID, positioned by TAF subunits reading other elements or by protein contacts alone. "TATA-less" describes the DNA, not the machinery.

graph LR D["TFIID<br/>TBP + TAFs"] --> A["TFIIA"] A --> B["TFIIB<br/>sets the start site"] B --> P["Pol II + TFIIF"] P --> E["TFIIE"] E --> H["TFIIH<br/>XPB translocase, CDK7 kinase"] H --> O["open complex<br/><i>ATP consumed</i>"]

TBP does something structurally violent: it wedges phenylalanines into the between base pairs and bends the DNA ~90°, tolerable only in AT-rich sequence — the same A:T/G:C energetic asymmetry that makes the bacterial −10 meltable, spent here on bending rather than on melting. TFIIB bridges TBP and Pol II and fixes where +1 falls. TFIIH does two jobs: its XPB subunit is a double-stranded DNA translocase that pumps DNA in and forces the bubble open — unlike bacteria, this costs ATP — while its CDK7 subunit begins phosphorylating the polymerase's tail.

Mediator is the interface to everything regulatory: a 26-subunit core plus a dissociable 4-subunit kinase module (CDK8, cyclin C, MED12, MED13). -bound activators contact its tail; its head contacts Pol II and stimulates CDK7. It converts "a transcription factor is bound nearby" into "this polymerase initiates more often" (Ch 22).

One caveat textbooks drop: that tidy assembly order is an in vitro reconstitution result. In cells much of it arrives pre-associated. Read it as a dependency graph, not a schedule.

7. The CTD: a state machine bolted to the polymerase

Pol II's largest subunit ends in a C-terminal domain of tandem Y-S-P-T-S-P-S heptads — 52 in mammals, 26 in budding yeast. That is ~360 residues of disordered tail whipping about outside the enzyme, and it is the best-engineered thing in this chapter.

Kinases write marks on it, erase them, and the pattern changes systematically with position along the gene. Processing factors bind the marks, not the polymerase.

StateWritten byErased byWhat it licenses
Unphosphorylated—Fcp1PIC entry, Mediator binding
Ser5-PCDK7 (TFIIH)RPAP2 / Rtr1Capping enzymes; escape
Ser2-PCDK9 (P-TEFb), CDK12/13Fcp1, cleavage/polyadenylation
Ser7-PCDK7RPAP2Integrator, snRNA 3'-end processing
stateDiagram-v2 [*] --> Unphos: PIC assembly Unphos --> Ser5P: CDK7 phosphorylates Ser5P --> Paused: escape, then pause near +30 Paused --> Ser2P: P-TEFb fires Ser2P --> Ser2P: elongation, Ser5-P decays Ser2P --> Unphos: termination, Fcp1 resets Unphos --> [*]: recycled for reinitiation

The design insight: the CTD carries positional information along the gene, and each processing machine subscribes to the tag rather than to a clock. Capping enzyme binds Ser5-P, which peaks near the promoter — so the cap is added when the nascent RNA is only ~20–25 nucleotides long, before the fragile 5' end is exposed for any length of time. Splicing and 3'-end factors bind Ser2-P, which accumulates downstream, exactly where their substrates appear. RNA processing is therefore not a stage after but co-transcriptional, implemented as mutable tags on a moving worker. All of Ch 06 hangs off this.

8. Promoter-proximal pausing

At most metazoan genes, initiation is not the committed step.

Pol II initiates, escapes, transcribes 20–60 nucleotides — peaking around +30 — and stops, held stably for minutes by DSIF (SPT4/SPT5) and NELF. The transcript is capped and goes no further. Genome-wide this is the normal state of a metazoan , not an exception.

Release requires P-TEFb (CDK9 + cyclin T1), which phosphorylates CTD Ser2, SPT5 and NELF-E together. NELF dissociates; DSIF changes sign and becomes a positive elongation factor; with PAF1C and SPT6 the polymerase turns processive. P-TEFb is itself regulated — most is sequestered in the 7SK snRNP (Ch 24), released on demand by BRD4 and activator complexes. So for many genes the regulated quantity is not how often a polymerase is recruited but how often a recruited one is allowed to leave.

A polymerase sitting on a promoter tells you nothing about whether the gene is expressed. A strong Pol II peak at a TSS is fully compatible with a silent gene, because the peak is the paused polymerase. Occupancy and output are different measurements, and conflating them is among the commonest errors in reading functional genomics data.

Why the cell bothers:

Measuring it needs nascent- assays (PRO-seq, NET-seq), which sequence the RNA still inside the polymerase (Ch 49).

9. Enhancers, distance, and the nearest-gene fallacy

Bacterial regulation is local: an operator overlapping the , a physically in the way. Eukaryotic are not.

An enhancer is a cluster of -factor binding sites that raises initiation at a promoter in either orientation, upstream or downstream, from tens of kilobases to about a megabase away, frequently past several intervening genes. The canonical case: the ZRS enhancer driving SHH in the limb bud sits roughly 850 kb–1 Mb from SHH, inside 5 of an unrelated gene, LMBR1. Point in it cause extra digits. Reading the sequence would never tell you what it controls.

Mechanically, the intervening DNA loops out, bringing activators into contact with Mediator and the PIC. Which loops are possible is constrained by and into topologically associating domains (Ch 50) — which is why deleting a domain boundary causes disease by connecting elements that were never meant to meet. Enhancers are themselves transcribed, yielding short unstable eRNAs that mark activity reliably and whose own function is still argued about.

The consequence is inescapable in Part 11: a non-coding variant need not act on its nearest gene, so assigning hits by proximity is systematically biased (Ch 52).

10. Rates, coupling, and the nuclear envelope

Pol II elongates at roughly 1–2 kb per minute (measurements span ~0.5–4 kb/min). Take that seriously:

DMD, the dystrophin gene, spans ~2.2 Mb. 2,200 kb ÷ 2 kb/min ≈ 1,100 min ≈ 18 hours to transcribe one molecule end to end.

That is a real . It exceeds many cell-cycle lengths, so the gene cannot be fully transcribed in a rapidly dividing cell, and the 3' end of a message exists many hours after its 5' end.

Bacterial runs at ~40–90 nt/s (≈2.5–5 kb/min), and bacterial translate at ~15–20 amino acids/s, consuming ~45–60 nt/s of message. The rates match — and that is not a coincidence.

BacteriaEukaryotes
Polymerases1 (+ swappable σ)3
recognitionσ factor, −10/−35GTFs, TFIID, variable core elements
Open complexspontaneousATP-driven (TFIIH/XPB)
Regulation acts atinitiation, mostlyinitiation, pause release, elongation, processing, export
Regulator distanceadjacent (bp)up to ~1 Mb
Transcriptoften polycistronic ()almost always monocistronic
coupled, same compartmentseparated by the nuclear envelope
Processingminimalcapping, , polyadenylation — co-transcriptional

In bacteria the ribosome loads onto the nascent mRNA while the polymerase is still transcribing, the two travelling as one assembly. Hence: no room for elaborate processing; ribosomes shield the transcript from Rho, producing polarity; and protein appears before the message is finished. (A nuance: coupling is tight in fast-growing E. coli, but in B. subtilis the polymerase can run ahead. It is a strong tendency, not a law.)

Eukaryotes cannot do this — the nuclear envelope enforces separation, and the gap it creates is where capping, splicing, polyadenylation, export licensing and nonsense-mediated surveillance all moved in. The envelope did not merely tidy things up; it created a pipeline stage into which new kinds of control could be inserted, and eukaryotes filled it.

11. Pervasive transcription

Point a sensitive enough assay at the genome and is nearly everywhere: ENCODE reported at least ~75% of the human genome transcribed in some cell type under some condition. GENCODE 50 annotates 35,885 genes against 19,442 protein-coding ones (verified-facts) — nearly 2:1 for lncRNA alone, and 58,195 non-coding genes in total (lncRNA + small ncRNA + ), close to 3:1, all before counting unannotated material.

Where it comes from: divergent initiation (most mammalian fire both ways; the upstream product, a PROMPT, is short-lived and destroyed by the nuclear exosome — directionality is imposed after initiation, the sense direction protected by U1 snRNP while the antisense direction hits polyadenylation signals and terminates early); eRNAs; antisense and intronic transcription; and read-through past termination sites, which rises sharply under stress.

Whether any of it is function is genuinely unresolved — the argument is laid out in the last Check-yourself question, and it is the same argument as in Ch 39. The operational consequence is immediate regardless: in you will see reads outside annotated genes, and the abundance threshold at which you stop calling them noise is a modelling decision you are making, not a fact you are discovering (Ch 47).


Common misconceptions

What people thinkWhat's actually true
reads the coding strandIt reads the template. The RNA merely resembles the coding strand — which is why that's the one printed in databases
A gene has "a" Template-ness belongs to a unit. Genes run both ways; one strand's template is the neighbouring gene's coding strand
Transcription starts at the The TATA box is ~25–30 bp upstream of the start site, and most human lack one — though TBP is recruited regardless
A promoter is strong or weak, like a switchStrength is a rate, well approximated by a score. And perfect consensus is counterproductive: the polymerase binds too tightly to escape
Once Pol II initiates, the gene is expressedAt most metazoan genes it initiates, moves ~30 nt and pauses. Release by P-TEFb is a separate, separately regulated decision
Pol II ChIP signal at a promoter means the gene is onIt usually means a paused polymerase. ChIP measures presence; PRO-seq and NET-seq measure output
Transcription is high-fidelity like replicationPer event it is roughly 10⁴–10⁶ times sloppier — ~10⁻⁴–10⁻⁶ per base against replication's ~10⁻¹⁰ per base per replication — and that is fine: the product is one disposable copy of thousands, not the archive
The RNA is made, then processedCapping, and 3'-end processing are co-transcriptional, recruited by CTD phosphorylation state as the polymerase moves
regulate the nearest geneThey routinely skip genes and act from up to ~1 Mb. Proximity is a biased heuristic, and knowing that is most of Ch 52
Transcribed means functional~75% of the genome is transcribed, mostly at under one copy per cell and poorly conserved. Transcription is a very weak functional claim
Bacteria and eukaryotes differ mainly in complexityThe load-bearing difference is compartmentalisation. Coupling transcription to , or separating them, determines what regulation is even possible

Worked example: from duplex to transcript

Here is a stretch of E. coli DNA, coding strand 5'→3'. Find the , write the RNA, explain where it stops.

AAAAAATT TTGACT GCTAACGTTCGATCAGT TATAAT GCTAGC ATGGCAGAATTCGTTAAACAG GCCCGCC TTCAAG GGCGGGC TTTTTTT

Step 1 — find the −10. Scan for TATAAT. It occurs once, exactly on consensus (6/6).

Step 2 — find the −35. By convention the −10 hexamer occupies −12 to −7. Look 17 bp upstream of its first base: TTGACT. Against consensus TTGACA that is 5/6, one mismatch at the final position. A good but imperfect promoter, compensated by a perfect −10.

Step 3 — check the spacer. Between the hexamers lies GCTAACGTTCGATCAGT = 17 bp, the optimum. Its sequence is irrelevant; its length is not.

Step 4 — number the positions.

 −43      −35    −30                    −13 −12   −7 −6   −1 +1
 AAAAAATT TTGACT GCTAACGTTCGATCAGT       TATAAT GCTAGC  A...
          └─−35─┘└──── 17 bp ────────────┘└−10─┘└ disc.┘  ↑
                                                         TSS

Check: 6 (−35) + 17 (spacer) + 6 (−10) + 6 (discriminator) = 35 positions, spanning −35 to −1. Consistent.

Step 5 — write both strands over the transcribed region.

5'- A T G G C A G A A T T C G T T A A A C A G ...   coding (given)
3'- T A C C G T C T T A A G C A A T T T G T C ...   template (read 3'→5')

Step 6 — write the RNA. It matches the coding strand with U for T, from +1:

5'- AUGGCAGAAUUCGUUAAACAG GCCCGCC UUCAAG GGCGGGC UUUUUUU -3'

Step 7 — fold the 3' end. GCCCGCC and GGCGGGC are reverse complements, so they pair with UUCAAG as the loop:

                                              U  U
  5'- ...A A A C A G   G C C C G C C                C
                       | | | | | | |   ← 7 bp stem,  A
  3'- U U U U U U U    C G G G C G G     all G:C  G  A
      └─────┬─────┘
   these 7 rU:dA pairs are the polymerase's
   entire grip on the template right now

Step 8 — explain termination. A 7 bp all-GC stem folds essentially the moment it clears the exit channel. Behind it, the enzyme holds the transcript by 7 rU:dA pairs — the weakest RNA:DNA hybrid available. The hairpin pulls, the hybrid cannot resist, the complex dissociates. Delete the U-tract and you get a strong pause but no release; delete the hairpin and the U-tract merely slows the enzyme.

Step 9 — now put the same sequence in a human cell. Almost nothing carries over:

Same chemistry, same catalytic core, an entirely different control system.

Connections