Translation Initiation in Bacteria
A single E. coli cell contains thousands of mRNA molecules being read at once. Each is cut into hundreds of possible starting points. Yet ribosomes almost never begin translating at the wrong codon. How does a 30S ribosomal subunit, drifting through a cytoplasm packed with RNA, find the one AUG that actually starts a protein? This one AUG must be picked out of dozens scattered through every transcript. The answer is not luck. Bacterial mRNAs carry a short upstream "zip code." A dedicated set of protein factors reads it, checks it, and only then commits the ribosome to a reading frame. Translation initiation matters because it is the single most regulated step in bacterial gene expression at the protein-synthesis level. Transcription decides whether an mRNA exists. Initiation decides how efficiently — and in which frame — that mRNA becomes protein. Small differences in how well a ribosome-binding site is recognized matter greatly. They translate into 100-fold or greater differences in protein output. This is why bacteria (and the biotechnologists who exploit them) tune this step so precisely when engineering expression vectors. The big picture is this: initiation solves two problems simultaneously. First, the ribosome must locate the correct start codon among many possible triplets. It must do this in the correct reading frame. It gets no help from a 5′ cap, unlike eukaryotic ribosomes (a contrast picked up later in Translation Initiation in Eukaryotes). Second, it must load a chemically special initiator tRNA. This is not just any tRNA that happens to recognize AUG. As a result, every protein begins, at least transiently, with a formylated methionine. Three initiation factors, IF1, IF2, and IF3, orchestrate this process. They do so without wasting GTP or misfiring on the thousands of internal AUG codons within coding sequences. This article works through the molecular logic of Shine-Dalgarno pairing. It also covers the special chemistry of fMet-tRNAfMet, and the choreography of the three initiation factors. It examines the classic experiments that established this mechanism — from Shine and Dalgarno's own sequence comparisons to toeprinting and ribosome-binding assays. It closes with worked calculations, common misconceptions, and the biotechnological stakes of getting a ribosome-binding site right.
- Bacterial mRNAs lack a 5′ cap; ribosomes instead locate start codons using the Shine-Dalgarno (SD) sequence, a purine-rich element ~5–10 nt upstream of AUG that base-pairs with the 3′ end of 16S rRNA.
- Initiation always begins with a special initiator tRNA, fMet-tRNAfMet, whose methionine is N-formylated by transformylase before the tRNA ever meets the ribosome.
- Three initiation factors — IF1, IF2, IF3 — bind the 30S subunit, block premature 50S joining, position fMet-tRNAfMet in the P site, and proofread codon–anticodon pairing.
- Assembly proceeds: 30S + IF1/IF2/IF3 + mRNA + fMet-tRNAfMet → 30S initiation complex → 50S joining (GTP hydrolysis by IF2) → 70S initiation complex.
- Classic evidence includes Shine & Dalgarno's rRNA-sequence comparison, Steitz's mRNA-fragment binding studies, toeprinting assays that map the ribosome footprint, and mutational studies of SD complementarity that show proportional changes in expression.
- The strength and spacing of the SD sequence is now a major lever in synthetic biology, used to tune protein expression across orders of magnitude in engineered bacteria.
Bacterial translation initiation is fundamentally a base-pairing recognition problem — solved by direct RNA–RNA hydrogen bonding between the Shine-Dalgarno sequence and 16S rRNA — that three protein factors then convert into a correctly positioned, GTP-licensed 70S ribosome loaded with formyl-methionine tRNA.
The core idea
In bacteria, translation initiation is the process by which a small (30S) ribosomal subunit locates the correct start codon on an mRNA, loads a specialized initiator transfer RNA into its peptidyl (P) site, and joins with a large (50S) subunit to form a translation-competent 70S ribosome. Because bacterial mRNAs are typically polycistronic — one transcript can encode several proteins, each with its own start and stop — the ribosome cannot rely on scanning from a capped 5′ end the way eukaryotic ribosomes largely do. Instead, each cistron carries its own upstream recognition element. That element is the Shine-Dalgarno (SD) sequence, a purine-rich stretch (consensus roughly 5′-AGGAGG-3′) located about 5–10 nucleotides upstream of the start codon. It base-pairs directly, in antiparallel fashion, with a complementary pyrimidine-rich sequence at the 3′ end of 16S ribosomal RNA (the anti-Shine-Dalgarno sequence, 3′-UCCUCC-5′ in E. coli, found within helix h45/the 3′ tail of 16S rRNA). This RNA–RNA duplex is the primary docking event of initiation: it is what tells the ribosome "start reading here," independent of reading frame per se, but strongly biasing which downstream AUG (or, less commonly, GUG or UUG) is selected as the start codon. The start codon itself is decoded by a dedicated initiator tRNA, tRNAfMet, charged with methionine by methionyl-tRNA synthetase and then modified by methionyl-tRNA transformylase, which adds a formyl group to the α-amino group of the charged methionine, producing fMet-tRNAfMet. This chemical blocking of the amino group is essential: it means the initiator residue cannot be elongated from — it has no free amino group to accept a peptide bond as an acceptor — so it can only ever occupy the ribosomal P site as the first residue of a nascent chain, never the A site mid-elongation. Three initiation factors — IF1, IF2, and IF3 — bind the 30S subunit to prevent premature association with the 50S subunit, to select and deliver fMet-tRNAfMet specifically to the start codon, and to proofread the codon–anticodon match before committing GTP hydrolysis and subunit joining.
Mechanism
Step 1 — 30S subunit dissociation and factor loading. Following the previous round of translation termination and ribosome recycling, free 30S subunits are prevented from re-associating with 50S subunits by IF3, which binds the 30S platform near the P site and acts as an "anti-association" factor — this is itself part of the evidence that IF3 controls subunit joining (see below). IF1 binds in the A site of the 30S subunit, sterically blocking any tRNA from entering there during initiation and also enhancing IF2's function allosterically. IF2, a GTPase, binds near the P site and, in its GTP-bound form, recruits fMet-tRNAfMet by recognizing the formyl group and the tRNA's unique acceptor-stem sequence — features absent from elongator methionyl-tRNA, which is why fMet-tRNAfMet is never delivered by the elongation factor EF-Tu even though it is aminoacylated methionine.
Step 2 — mRNA docking via Shine-Dalgarno pairing. The 30S subunit, decorated with IF1, IF2•GTP, and IF3, binds the mRNA. The SD sequence anneals to the anti-SD sequence at the 3′ end of 16S rRNA. This interaction is thermodynamically variable across genes — some SD sequences are near-perfect matches to the consensus (strong, efficient sites) and others are weak or nearly absent (weak sites, often on genes needing tight regulation). The spacing between the SD sequence and the AUG (optimally ~5–9 nucleotides in E. coli) is critical: too short or too long a spacer misaligns the start codon relative to the P site, reducing initiation efficiency sharply.
Step 3 — start codon selection. The docked 30S subunit's P site is positioned by the SD interaction close to the true start codon. fMet-tRNAfMet then base-pairs its anticodon (5′-CAU-3′) with the start codon (5′-AUG-3′, occasionally GUG or UUG, which are still read as Met because the initiator tRNA can wobble at the third position in a way elongator tRNAs cannot fully mimic for initiation). IF3 acts as a fidelity factor here: it destabilizes complexes in which fMet-tRNAfMet is mispaired or the wrong codon is engaged, promoting dissociation of incorrect 30S preinitiation complexes and allowing only correctly paired complexes to proceed — a kinetic proofreading step analogous in spirit (though mechanistically distinct) to proofreading described for aminoacyl-tRNA synthetases.
Step 4 — 50S joining and GTP hydrolysis. Correct codon–anticodon pairing triggers conformational changes that promote release of IF3 from the 30S subunit, opening the platform for 50S joining. The 50S subunit associates to form the 70S initiation complex. Association triggers IF2 to hydrolyze its bound GTP to GDP + Pi. GTP hydrolysis, not GTP binding itself, drives release of IF2 (and any remaining IF1) from the ribosome, leaving fMet-tRNAfMet correctly seated in the P site of an assembled 70S ribosome, with an empty A site aligned over the second codon.
Step 5 — handoff to elongation. The 70S initiation complex, with an open A site over codon 2, is now the substrate for elongation factor EF-Tu bringing in the second aminoacyl-tRNA — the topic taken up in Elongation and Codon Recognition. The entire initiation pathway has thus converted a sequence-recognition problem (finding the right AUG among many) into a mechanical, energy-gated commitment (GTP hydrolysis) that only occurs after correct pairing is verified.
The evidence
1. Shine and Dalgarno's sequence comparison (1974). John Shine and Lynn Dalgarno noted that the 3′ end of E. coli 16S rRNA terminates in the sequence 3′-UCCUCUAG...OH-5′ (pyrimidine-rich). Comparing the 5′ untranslated regions of several bacteriophage and bacterial mRNAs, they found a recurring purine-rich sequence just upstream of the start codon that was complementary to this rRNA tail. They proposed that this complementarity, rather than any special recognition of AUG alone, positions the ribosome at the correct start codon. This was a bioinformatic/comparative discovery — pattern recognition across sequences — later confirmed biochemically.
2. Steitz and Jakes: direct RNA–RNA hybridization (1975). Joan Steitz and Karen Jakes tested the pairing hypothesis directly. They isolated 30S ribosome-protected mRNA fragments (using ribonuclease digestion of ribosome–mRNA complexes, a precursor to modern footprinting) and showed that these fragments could form RNase-resistant hybrids with the 3′ end of 16S rRNA — direct physical evidence of Watson-Crick pairing between mRNA and rRNA, exactly where Shine and Dalgarno had predicted it. This moved the SD hypothesis from sequence comparison to demonstrated molecular interaction.
3. Toeprinting assays mapping the ribosome's footprint. Later work (extending classic primer-extension inhibition, or "toeprinting," pioneered by Hartz, McPheeters, Traut, and Gold in the 1980s) used reverse transcriptase to extend a labeled primer along mRNA until it ran into a bound 30S initiation complex. The reverse transcriptase stalls at a characteristic position (+15 to +16 nucleotides downstream of the A of AUG), producing a "toeprint" band whose position confirms precisely where the ribosome sits relative to the SD sequence and start codon, and how IF3 affects the stability of that complex at different codons. Toeprinting assays showed that IF3 selectively destabilizes complexes formed on non-AUG or mismatched codons, providing direct kinetic evidence for its fidelity-checking role described above.
4. Mutagenesis of the SD sequence and spacer length. Studies (e.g., by Gold and colleagues, and later systematic synthetic-biology screens) mutated the SD sequence itself, varied its distance from the start codon, and measured translation efficiency using reporter genes (β-galactosidase, GFP). These experiments showed a graded, quantitative relationship: closer matches to the consensus AGGAGG and optimal spacing (~5–9 nt) increased protein yield by up to two orders of magnitude relative to weak or absent SD sequences, while restoring complementarity in the 16S rRNA anti-SD sequence via compensatory mutation could restore initiation on an SD sequence, directly proving the base-pairing model (rather than some indirect protein-mediated recognition) by classic complementation logic — analogous in spirit to compensatory base-change experiments used to prove RNA secondary structures generally.
5. Reconstitution of initiation from purified components. Biochemists (following early work by Ochoa, and later refined in vitro translation systems) showed that purified 30S subunits, purified IF1, IF2, IF3, synthetic mRNA, GTP, and chemically defined fMet-tRNAfMet are sufficient to reconstitute a functional 70S initiation complex, without any additional unknown factors. Omitting any one of the three IFs, or replacing GTP with a non-hydrolyzable analog (GDPNP/GDPCP), stalls the reaction at defined intermediate steps — IF2-GDPNP still recruits fMet-tRNAfMet and allows 50S joining to occur (poorly), but blocks IF2 release, directly demonstrating that GTP hydrolysis drives factor release, not subunit joining itself.
Worked examples
Example 1 — Predicting relative translation efficiency from Shine-Dalgarno strength. Suppose you are comparing two synthetic ribosome-binding sites (RBS) cloned upstream of the same reporter gene in E. coli:
- RBS-A: sequence `AGGAGG`, spacer of 7 nt to the AUG (matches consensus, optimal spacing).
- RBS-B: sequence `AGGtGc` (two mismatches relative to consensus), spacer of 13 nt to the AUG (too long).
The anti-SD sequence in 16S rRNA is 3′-UCCUCC-5′ (paired against 5′-AGGAGG-3′ for a perfect match). RBS-A pairs across all six bases with correct spacing, giving a strong, thermodynamically favorable duplex positioned to place the AUG precisely in the P site. RBS-B loses two base pairs (weakening the ΔG of hybridization) and has a spacer roughly double the optimal length, which geometrically misaligns the docked 30S subunit relative to the start codon — even a chemically capable duplex may fail to deliver the P site to the right nucleotide. Empirically, studies varying SD strength and spacer length (e.g., systematic RBS libraries used in synthetic biology, following the logic of Gold's original spacer experiments) show that:
- Perfect SD match + optimal spacing (5–9 nt): initiation efficiency near maximal (say, defined as 100%).
- Two mismatches: efficiency typically drops to roughly 10–30% of maximal, depending on which bases are altered (loss of G-C pairs is more damaging than loss of A-U/G-U pairs).
- Spacer doubled beyond optimal range: efficiency can fall further, often below 10%, because both duplex stability and geometric register are compromised simultaneously.
So we predict RBS-A will produce substantially more reporter protein per mRNA molecule than RBS-B — potentially an order of magnitude or more — even though both constructs carry identical coding sequences and identical promoters. This is exactly the logic exploited when engineers use RBS calculators to dial in a target expression level: they are, in effect, quantitatively modeling the thermodynamics of SD:anti-SD pairing and spacer geometry established by the classic experiments above.
Example 2 — Distinguishing initiator tRNA from elongator tRNA experimentally. You purify total tRNA from E. coli and separate methionine-charged species by chromatography. You find two distinct methionyl-tRNA species: tRNAfMet and tRNAmMet (the elongator methionine tRNA). You want to determine, using a simple biochemical criterion, which one initiates translation and why the cell cannot simply use one tRNA for both jobs. Reasoning through it: 1. Both tRNAs can be charged with methionine by the same methionyl-tRNA synthetase (both have the CAU anticodon and the correct identity elements — recall Aminoacyl-tRNA Synthetases and the Second Genetic Code). 2. Only tRNAfMet is a substrate for methionyl-tRNA transformylase, because it has a distinguishing structural feature (a mismatched first base pair at the top of the acceptor stem, C1:A72 in E. coli, rather than a Watson-Crick pair) that the transformylase specifically recognizes. Formylation converts Met-tRNAfMet into fMet-tRNAfMet. 3. Formylated fMet-tRNAfMet is recognized by IF2, which binds only formylated initiator tRNA with high affinity and delivers it to the ribosomal P site during initiation — it is excluded from binding by EF-Tu, which normally delivers all elongator aminoacyl-tRNAs to the A site. 4. Conversely, tRNAmMet (uncharged of any formyl group) is recognized efficiently by EF-Tu and delivered to the A site during elongation, but it is a poor substrate for IF2 and cannot support initiation. 5. Experimentally, if you strip the formyl group off fMet-tRNAfMet (using formyltransferase inhibitors or a deformylase treatment) and try to use it in an in vitro initiation reaction, initiation efficiency collapses — direct proof that the formyl group, not merely the methionine or the anticodon, is what IF2 (and downstream, the P site machinery) is reading. The logical takeaway: the cell partitions its two initiation/elongation identities not by anticodon (both are CAU) but by post-transcriptional/post-translational-on-the-tRNA chemistry (acceptor-stem mismatch → transformylase substrate → formylation → IF2 recognition), a beautiful example of how molecular identity is enforced by structure, not sequence alone.
Common misconceptions & traps
| Misconception | Why it's wrong |
|---|---|
| "The ribosome scans the mRNA from the 5′ end looking for the first AUG, like in eukaryotes." | Bacterial ribosomes do not scan from the cap (there is no cap). They are guided directly to an internal SD-proximal AUG, and polycistronic mRNAs have multiple independent initiation sites, not one 5′-most start. |
| "fMet-tRNAfMet is just a regular methionine tRNA used at the start." | It is a chemically and structurally distinct species: formylated, transformylase-selected, IF2-specific, and structurally distinguished by its acceptor-stem mismatch — it cannot function in elongation. |
| "IF1, IF2, IF3 all do the same generic job of 'helping the ribosome.'" | Each has a distinct, separable function: IF1 blocks the A site and stimulates IF2; IF2 is the GTPase that delivers fMet-tRNAfMet; IF3 is the anti-association/fidelity factor that dissociates incorrect complexes. |
| "A strong Shine-Dalgarno match guarantees high protein expression." | mRNA secondary structure that sequesters the SD sequence or start codon, spacer length, and downstream codon context can override SD strength; SD complementarity is necessary but not sufficient. |
| "GTP hydrolysis by IF2 is what joins the 50S subunit." | Subunit joining occurs first (driven by correct codon-anticodon pairing and factor conformational changes); GTP hydrolysis instead triggers release of IF2 (and IF1) from the assembled 70S complex. |
| "The start codon must be AUG." | E. coli also initiates, less frequently, at GUG (~14% of genes) and rarely UUG or even AUU, all still decoded as formyl-methionine by the initiator tRNA machinery, not as their "normal" elongator meaning. |
The scientist behind it
Work establishing the Shine-Dalgarno mechanism emerged from a productive Australian-American collaboration in the mid-1970s. John Shine, then a graduate student at the John Curtin School of Medical Research at the Australian National University, working with Lynn Dalgarno, compared the 3′-terminal sequence of 16S ribosomal RNA against the 5′ untranslated regions of several known bacteriophage mRNAs (including R17 and QB phage transcripts) that had already been sequenced. Their insight — that a short, purine-rich stretch just upstream of the start codon was complementary to the pyrimidine-rich rRNA tail across multiple, unrelated mRNAs — was a triumph of comparative sequence analysis at a time when RNA sequencing was laborious and only a handful of complete mRNA sequences existed. The biochemical confirmation came soon after from Joan Steitz, working at Yale, whose ribonuclease-protection and RNA–RNA hybridization experiments with Karen Jakes provided direct physical evidence that ribosome-protected mRNA fragments could hybridize to the 16S rRNA 3′ end exactly as predicted.
Origins · 1 minJohn Shine (b. 1946) and Lynn Dalgarno (1935–2023) and how this was discovered
Shine and Dalgarno worked at a moment when only a small number of complete mRNA sequences (mostly from RNA bacteriophages) were available, forcing them to rely on careful manual comparison rather than the automated bioinformatics of later decades — a reminder that foundational discoveries often precede the tools that later make them routine.
Where this matters today
Recombinant protein production. Every plasmid used to overexpress a protein in E. coli — insulin, growth hormone, vaccine antigens, industrial enzymes — depends on engineering a strong, well-spaced Shine-Dalgarno sequence into the expression construct. Computational "RBS calculators," built directly from the thermodynamics of SD:anti-SD pairing established by the classic experiments above, let bioengineers predict and tune expression levels before ever running an experiment.
Antibiotic development. Several clinically important antibiotics act at or near the initiation step. Aminoglycosides and other ribosome-targeting drugs can interfere with accurate initiation-complex formation, and understanding the precise conformational choreography of IF1/IF2/IF3 has informed structure-based design of novel antibacterials that spare the mechanistically distinct eukaryotic initiation pathway (see Translation Initiation in Eukaryotes) — a crucial selectivity requirement for any drug meant to kill bacteria without harming the human host.
Synthetic biology and metabolic engineering. Because SD strength and spacer length give quantitative, tunable control over protein output, synthetic biologists routinely build libraries of ribosome-binding sites with graded strengths to balance expression of multiple enzymes in an engineered metabolic pathway, avoiding toxic overexpression of one enzyme and rate-limiting underexpression of another — for example, in engineered pathways producing biofuels, pharmaceuticals, or specialty chemicals in bacterial chassis.
Basic understanding of translational control. Riboswitches, small regulatory RNAs, and RNA thermometers frequently work by occluding or exposing the Shine-Dalgarno sequence in response to metabolite levels, temperature, or antisense RNA binding — meaning the initiation mechanism described here is the direct molecular target of an entire layer of bacterial post-transcriptional gene regulation, connecting this mechanism forward to broader themes in gene expression control.
Ask the tutor: How does the strength and spacing of a Shine-Dalgarno sequence quantitatively predict protein expression level, and how do synthetic biologists exploit this to engineer bacterial gene circuits?
Why don't bacterial ribosomes accidentally initiate at internal AUG codons within a coding sequence?
Internal AUG codons lack an upstream Shine-Dalgarno sequence positioned at the correct spacing, so the 30S subunit has no strong, correctly registered docking site there. Even if an internal AUG happens to sit near a purine-rich stretch by chance, the geometry (spacer length, accessibility, and any competing mRNA secondary structure) usually fails to reconstruct a productive preinitiation complex. In addition, once a ribosome is elongating through a coding sequence, that region is occupied and translocating, physically preventing new initiation events from ribosomes trying to load internally. This is why polycistronic mRNAs need a genuinely independent, well-formed SD/AUG pair for each cistron rather than relying on any internal AUG being read as a new start.
Is the Shine-Dalgarno mechanism universal across all bacteria?
No. It is widespread and well characterized in E. coli and many other bacteria, but a substantial fraction of bacterial and archaeal genes — including many "leaderless" mRNAs that begin directly with AUG at the extreme 5′ end with no upstream untranslated region at all — initiate by SD-independent mechanisms. Leaderless mRNAs are especially common in some archaea and in a subset of bacterial genes, and are thought to be recognized by a distinct pathway involving direct 70S (rather than 30S) initiation in some cases. So while SD pairing is the dominant and best-studied mechanism, it is one strategy among several, and its relative usage varies substantially across bacterial lineages.
Why does the initiator tRNA carry a formyl-methionine rather than plain methionine?
The formyl group blocks the free α-amino group of the initiator methionine, which has two consequences. First, it prevents fMet-tRNAfMet from being mistaken for an elongator aminoacyl-tRNA by EF-Tu, since EF-Tu's binding pocket is shaped to recognize free aminoacyl groups on elongator tRNAs. Second, because a peptide bond requires a free amino group on the incoming (A-site) amino acid, blocking the initiator's amino group ensures it can only ever function as the amino-donor at the very start of the chain, never as an acceptor mid-elongation — reinforcing that it must occupy the P site as the first residue and cannot appear anywhere else in the growing polypeptide, notwithstanding subsequent removal of the formyl group (and often the methionine itself) by deformylase and methionine aminopeptidase after initiation.
How does the strength of a Shine-Dalgarno sequence affect protein expression quantitatively?
Empirically, ribosome-binding sites with sequences close to the consensus AGGAGG and spacers around 5–9 nucleotides from the start codon support the highest initiation rates, while progressively weaker matches or non-optimal spacing reduce expression, sometimes by more than 90%. This relationship is exploited directly in synthetic biology: RBS libraries with computationally predicted strengths (based on estimated SD:anti-SD binding free energy and spacer geometry) let engineers select a ribosome-binding site that yields a target expression level for a given gene, without altering the coding sequence or promoter at all. It is one of the cleanest examples of a molecular recognition rule translated directly into an engineering design parameter.
What role does mRNA secondary structure play in modulating initiation, beyond SD sequence and spacer length?
If the Shine-Dalgarno sequence or the start codon is sequestered within a stable stem-loop, the 30S subunit (and IF3, IF1, IF2) cannot access it efficiently, dramatically reducing initiation regardless of how "strong" the SD sequence looks in isolation. Cells exploit this directly: RNA thermometers unfold at higher temperatures to expose a previously hidden SD sequence, allowing temperature-dependent translation (relevant, for instance, to some heat-shock and virulence genes), and riboswitches can fold differently depending on metabolite binding to expose or occlude the ribosome-binding site. This means predicting real-world initiation efficiency requires considering RNA folding energetics together with SD complementarity, not SD complementarity alone.
How does bacterial initiation differ fundamentally from eukaryotic cap-dependent initiation?
Bacterial ribosomes locate start codons through direct, sequence-specific RNA–RNA pairing (SD:anti-SD) at each individual cistron, permitting efficient translation of polycistronic mRNAs with several independent start sites. Eukaryotic ribosomes instead typically load at the 5′ cap and scan linearly until encountering a start codon in good context (the Kozak sequence), a fundamentally different search strategy suited to predominantly monocistronic eukaryotic mRNAs. Eukaryotic initiation also uses a substantially larger set of initiation factors (eIFs) and an uncharged (non-formylated) initiator Met-tRNA, reflecting the different problem being solved — this contrast is developed further in Translation Initiation in Eukaryotes.
Why is GTP hydrolysis by IF2 placed after subunit joining rather than driving it?
Reconstitution experiments using non-hydrolyzable GTP analogs (GDPNP or GDPCP) showed that 50S subunits can still join a 30S preinitiation complex even when IF2 cannot hydrolyze GTP, meaning joining itself does not require hydrolysis. What fails under these conditions is the subsequent release of IF2 (and IF1) from the assembled 70S complex, which normally follows GTP hydrolysis. This places GTP hydrolysis functionally after joining, as a factor-release and complex-maturation signal rather than as the driving force of association — a distinction that matters because it shows the ribosome's own conformational rearrangements upon correct codon recognition, not GTP hydrolysis, are what license subunit joining.