Contents
Question bank — Part 01: Molecular foundations Question bank

Question bank — Part 01: Molecular foundations

Covers Ch 02-08.

Convert to Anki: python reference/to_anki.py question-banks/ --out anki-import.tsv

DNA structure and duplex stability

Q: GC-rich DNA melts at a higher temperature than AT-rich DNA. Why is "because G-C has three hydrogen bonds" the wrong reasoning? A: Base-pair hydrogen bonds are formed in water, and water is itself an excellent hydrogen-bond partner, so melting mostly replaces base-to-base bonds with base-to-water bonds at close to no net cost. What is genuinely lost is stacking -- the face-to-face interaction between adjacent base pairs and the burial of flat hydrophobic surface away from water. G and C stack better, which is why GC-rich DNA melts higher. Pairing supplies the specificity; stacking supplies the stability.

Q: In one careful decomposition of duplex stability, what do pairing and stacking each contribute for A-T and for G-C? A: Yakovchuk and colleagues (2006), at 15 mM Na+: base pairing contributes about +0.6 kcal/mol per A-T pair, which is destabilising, and about -0.1 kcal/mol per G-C pair, which is negligible. contributes about -1.0 kcal/mol per A-T contact and about -1.5 kcal/mol per G-C contact. Almost all the stability is stacking, and A-T pairing on its own is a net liability.

Q: Why must every base pair consist of one and one ? A: To keep the helix a constant width. The C1'-to-C1' distance across a pair must stay near 10.8 Angstroms for the backbone to run smoothly, and purine-purine is too wide while pyrimidine-pyrimidine is too narrow. Among the four purine-pyrimidine combinations, only A-T and G-C put a hydrogen-bond donor opposite an acceptor at every position; A-C and G-T fail because they place donor against donor. They still pair in an offset geometry with ordinary keto bases -- that is what a G-T mismatch in DNA is -- but they reach true Watson-Crick geometry only when a base flips to a rare or ionised form, which is precisely how a mispair slips past a polymerase and becomes a .

Q: Two oligonucleotides of the same length and base composition, and therefore identical hydrogen-bond counts, have that differ by several degrees. What does that establish? A: That duplex stability is a sum over dinucleotide stacks rather than over base pairs. The nearest-neighbour model inside every primer-design tool has no parameter for "a G-C pair" at all; its parameters are indexed by stack. 5'-GCGCATATGC-3' and 5'-GACTGACTGC-3' share composition and 26 hydrogen bonds each, yet give T_m of about 50 C and 47 C at 2 micromolar strand in 1 M Na+.

Q: Why can a sequence-specific DNA-binding protein read sequence without opening the helix, and why does it read the ? A: Base edges are exposed in both grooves, so a protein can hydrogen-bond to them on intact duplex. But only the major groove distinguishes all four orientations, presenting a distinct pattern of donors, acceptors and (for thymine) a methyl group for A-T, T-A, G-C and C-G. The gives only two patterns, AT versus GC, and cannot see orientation at all. Hence a roughly 12 Angstrom recognition helix inserted into a roughly 22 Angstrom major groove, while minor-groove binders read shape rather than base identity.

Q: Every extends the 3' end and none extends the 5' end. What is the chemical reason, and what does it have to do with ? A: The growing chain's 3'-hydroxyl attacks the alpha-phosphate of an incoming deoxynucleoside triphosphate, so the energy for each bond arrives attached to the incoming nucleotide. If synthesis ran 3' to 5', the activating triphosphate would sit on the growing end, and excising a misincorporated terminal base would strip away the reaction's own energy source and dead-end the chain. Extension is 5' to 3' because that is the only direction in which error correction is affordable.

Topology, supercoiling and helix forms

Q: What does Lk = Tw + Wr say, and why can you not unwind DNA locally without consequence? A: In a topologically closed domain the linking number Lk -- the number of times one strand passes through the loop formed by the other -- is an integer that cannot change without breaking a strand. Twist and writhe therefore trade: unwind a region, lowering Tw, and Wr must rise to compensate, so the axis crosses itself and the molecule supercoils. Every process that opens the helix manufactures supercoils that something else must relieve.

Q: Cells hold their DNA at a superhelical density of about -0.05 to -0.07. Does that mean the DNA is wound too tightly? A: The opposite -- it is underwound relative to relaxed B-form. Superhelical density is sigma = (Lk - Lk0)/Lk0, with Lk0 = N/10.5 for N base pairs. The underwinding is a deliberate free-energy store: it pre-pays part of the cost of separating the strands, so open and origins fire more readily.

Q: Quantify the one transcribing polymerase generates, and show why cannot be occasional maintenance. A: II on a 10 kb gene at 2 kb/min moves about 33 bp/s, which at 10.5 bp per turn is 3.2 turns/s -- roughly 3 positive supercoils per second ahead and 3 negative per second behind, and about 950 of each sign over the whole gene. A 100 kb chromatin loop has Lk0 = 9,524, so at sigma = -0.06 its entire torsional budget is only about 570 turns. One polymerase exceeds the whole domain's capacity within minutes, and an active gene carries many.

Q: How do type I and type II topoisomerases differ, and why does the mechanism make them drug targets? A: Type I cuts one strand and spends no ATP: type IA passes the intact strand through the break, changing Lk by exactly 1 and relaxing negative supercoils only, while type IB lets the helix swivel and relaxes either sign by one or many turns per event. Type II cuts both strands, passes a whole duplex through the gap, changes Lk by 2 and is ATP-dependent; bacterial gyrase is the special case that actively introduces negative supercoils. All work through a transient covalent enzyme-DNA intermediate, so fluoroquinolones, etoposide and irinotecan kill cells by blocking the resealing step and converting a routine catalytic cycle into a permanent break.

Q: Are A-DNA and laboratory artefacts? A: No. The 2'-hydroxyl of ribose clashes with the C2'-endo sugar pucker of B-form, so every RNA duplex and every RNA:DNA hybrid is obligatorily A-form -- which means every bubble contains an A-form segment. Z-DNA is left-handed and forms in alternating - sequence under negative supercoiling, notably behind a transcribing polymerase, where it acts as a torsional relief valve and is recognised by Zalpha-domain proteins including ADAR1 and ZBP1.

Chromosomes, chromatin and genome content

Q: A nucleus is about 6 micrometres across and holds about 2 metres of DNA. Why is "the nucleus is too small to fit it" the wrong diagnosis? A: Treated as a solid 2 nm cylinder, that DNA occupies only about 6% of nuclear volume, so fitting by volume is easy. The real is polymer conformation: ideal-chain size scales as b times the square root of N, where b is the Kuhn length, twice the , so 100 nm for DNA. 1's 85 mm of contour is N = 8.5 x 10^5 Kuhn segments, giving 100 nm x 920 = about 92 micrometres -- fifteen nuclear diameters. Packaging fights entropy and entanglement, not volume.

Q: What is a , and why can any sequence be wrapped on one? A: An octamer of two copies each of histones H2A, H2B, H3 and H4, with 147 bp of DNA wrapped in about 1.65 left-handed superhelical turns. Histones contact the phosphate backbone and the -- largely by arginine insertion -- and not the bases, so recognition is sequence-independent and the genome is uniformly packageable. Sequence only biases positioning, because bending is cheaper where AA, TT and TA dinucleotides recur with the roughly 10 bp helical period.

Q: How much compaction does the nucleosome achieve, and how much is needed at metaphase? A: With a 200 bp repeat length, 68 nm of DNA contour occupies about 11 nm of axial length -- roughly 6- to 7-fold. Metaphase chromosome 1 packs 85 mm of DNA into a 10 micrometre rod, about 8,500-fold linear compaction. The nucleosome is a down payment; the remaining thousand-fold comes from condensin-driven , and a cell contains about 3.1 x 10^7 nucleosomes, all of which must be duplicated during one S .

Q: What is the evidence for and against the 30 nm chromatin fibre, and what should you believe? A: For: purified nucleosome arrays with magnesium and linker histone H1 fold into 30 nm fibres reproducibly, reconstituted arrays have been solved structurally, and chicken erythrocyte nuclei show them. Against: cryo-electron microscopy of vitrified mitotic chromosomes and small-angle X-ray scattering of nuclei show no 30 nm periodicity, and ChromEMT imaging of human cells in situ (Ou et al., Science 2017) finds disordered chains 5-24 nm across with no repeating unit. Only the 10 nm nucleosomal fibre is established in vivo; above it, order comes from loop extrusion and compartmentalisation.

Q: What distinguishes constitutive from facultative , and what marks each? A: Constitutive heterochromatin is heterochromatic in every cell type -- pericentromeric and telomeric satellite arrays, the Y long arm, the rDNA flanks -- and its job is structural: it keeps repeat arrays from recombining with each other and is required for function. It carries H3K9me3 bound by HP1, which self-associates and therefore spreads along the chromosome and through cell division. Facultative heterochromatin is silenced only in some cells, canonically the inactive X packaged as the Barr body and Polycomb-silenced developmental gene clusters carrying H3K27me3; its job is regulatory, locking in decisions about cell identity. Treat the two-state picture as a coarse-graining -- genome-wide maps resolve a dozen or more chromatin states.

Q: Is a centromere defined by its DNA sequence? A: No -- it is defined epigenetically, by nucleosomes in which histone H3 is replaced by the variant CENP-A. Human centromeres usually sit on alpha satellite DNA (a 171 bp monomer built into chromosome-specific higher-order repeats spanning roughly 0.1-5 Mb), but neocentromeres form and function on ordinary single-copy sequence, and some chromosomes carry a second alpha satellite array that is never used. The DNA is a preferred substrate, not the instruction.

Q: A colleague says the human genome is "98% junk". Give the two-sided correction. A: They are right that only about 1-2% encodes protein and wrong about what follows. Non-coding genes outnumber coding ones roughly 3:1 (58,195 against 19,442 in GENCODE Release 50), ENCODE catalogues on the order of a million candidate cis-regulatory elements, and centromeres, , origins and insulators are non-coding and essential. But do not over-correct: about 46% of the genome is transposable-element derived and largely degraded, low-level is expected of any accessible DNA, and comparative genomics finds only about 5-10% under detectable .

Q: What is the , and what does genome size actually track? A: Genome size correlates with neither organismal complexity nor gene number. A pufferfish carries roughly the same gene count as a human in about an eighth of the DNA, an onion has around five times the human genome, and the record holder is the fork fern Tmesipteris oblanceolata at 160.45 Gb. Size tracks transposable-element accumulation and whole-genome duplication history, not complexity.

Q: A diploid human cell sits in G2, after replication and before division. Is it 2n or 4n, and is it 2C or 4C? A: Still 2n, but 4C. n counts complete chromosome sets while C counts DNA content, and they are different axes: every chromosome now consists of two sister chromatids, so the DNA content has doubled while the number of chromosome sets has not. Keeping those apart is the difference between following mitosis and and being permanently lost in them. Note also that "human = diploid" is an approximation -- hepatocytes are frequently tetraploid or octoploid, megakaryocytes reach 64n, and roughly a third of solid tumours have undergone whole-genome doubling.

Q: Why does a of mitochondrial disease look nothing like a Mendelian one? A: Two features combine. mtDNA is maternally inherited -- paternal mitochondria are actively destroyed after fertilisation -- so an affected male transmits nothing whatever his own load, giving transmission through females only. And a cell holds thousands of mtDNA copies rather than two , so a has a load (): disease appears only above a threshold, typically 60-90% depending on mutation and tissue, and the bottleneck means each oocyte is seeded by a small sample, so sibling severity is a sampling distribution rather than a Mendelian fraction.

Q: Give the basic specification of the human mitochondrial genome, including its . A: 16,569 bp, circular, no histones, with 37 genes -- 13 protein-coding (all oxidative-phosphorylation subunits), 22 tRNA and 2 rRNA -- plus a roughly 1.1 kb control region and essentially no . Copy number is about 10^3-10^4 per somatic cell. Its code deviates from the standard table: UGA is Trp rather than stop, AUA is Met rather than Ile, and AGA and AGG are stops rather than Arg, which is why every tool takes a -table parameter.

Q: Repeats are the majority of the genome. Which class breaks alignment and which breaks assembly? A: Interspersed repeats break alignment: a 150 bp short read from an has on the order of a million plausible origins among about 1.1 million copies, so the aligner emits it with mapping quality 0 and downstream callers discard it. Tandem repeats break assembly: a cannot resolve a cycle longer than the read, so a megabase satellite array stays a tangle no amount of short-read fixes. This is why roughly 8% of the genome had no reference sequence at all until reads got longer than the repeats.

DNA replication

Q: Why was the second generation of the Meselson-Stahl experiment necessary? A: Generation 1 gave a single band at intermediate density, which excludes the conservative model (it requires a surviving fully heavy parental duplex) but fits semiconservative and dispersive replication equally well, since a hybrid duplex and a uniform patchwork are both half-heavy. They diverge at generation 2: semiconservative predicts two discrete bands, half hybrid and half light, with the hybrid band not moving, while dispersive predicts one band drifting to quarter-heavy. Two bands appeared, so dispersive was excluded.

Q: Why must a human genome be replicated from tens of thousands of origins? A: A fork moves about 1.5 kb/min, so copying 1 (249 Mb) from a single central origin would take roughly 58 days against an S of about 8 hours. Each fork covers at most 720 kb in one S phase, so a bidirectional origin covers at most 1.44 Mb and a 6.2 Gb genome needs a floor of about 4,300 origins; the measured figure is 30,000-50,000 fired per cell cycle. Licensing is deliberately generous -- MCM2-7 loads at 10-20 times more sites than ever fire, and only about 5-10% of licensed origins fire in any given cycle, the rest being dormant reserve for stalled forks.

Q: Is the synthesised 3' to 5'? A: No strand, anywhere, in any organism, is synthesised 3' to 5'. Each is made 5' to 3' like everything else; what runs backwards is the series of fragments, each primed near the fork and extended away from it into the previous fragment. The direction in which the fragments are laid down and the direction in which any single chain grows are different things.

Q: Does unwind the double helix? A: No. unwinds -- DnaB in bacteria, the CMG complex of Cdc45, MCM2-7 and GINS in eukaryotes -- and polymerase can only copy a template that is already single-stranded, so removing helicase stops replication before polymerase is even relevant. Single-strand binding protein (SSB in bacteria, RPA in eukaryotes) then coats the exposed strands to stop them re-annealing or forming hairpins, and RPA-coated single-stranded DNA doubles as the signal that a fork is in trouble.

Q: A bare replicative polymerase falls off its template after about 10 nucleotides. What raises processivity to over 50 kb, and why is the clamp loader one of the busiest machines in the cell? A: The sliding clamp -- the beta clamp dimer in bacteria, PCNA trimer in eukaryotes -- a closed ring threaded onto the DNA. The polymerase grips the ring rather than the DNA, so it slides freely along the template while being topologically unable to leave: the difference between holding a rope and being handcuffed to it. The clamp loader (the gamma/tau complex, or RFC) is ATP-driven and cracks the ring open around each new primer, and since the lagging strand needs a fresh clamp for every Okazaki fragment it runs on the order of 40 million times per S phase.

Q: How is a per-base error rate of about 10^-10 achieved, and why does replication fidelity not degrade gracefully? A: As the product of three individually mediocre filters in series: polymerase base selection by active-site geometry gets to about 10^-5, a 3' to 5' exonuclease step to about 10^-7, and post-replicative to about 10^-9 to 10^-10. Because the rates multiply, losing any one filter costs its entire factor rather than some fraction of it. Humans confirm this: mismatch-repair loss gives Lynch syndrome with 100- to 1000-fold elevated , and a POLE exonuclease-domain gives ultramutated tumours carrying over 100 mutations per megabase.

Q: Mismatch repair must know which strand is the new one. What goes wrong if it does not, and how is the new strand identified? A: If it cannot tell, it "corrects" the template half the time and makes the error permanent, so the filter contributes nothing. Bacteria label the strand by Dam methylation: GATC sites are methylated on both strands, but for a few minutes after the fork passes only the parental strand is, and repair reads that hemimethylated state. Eukaryotes have no such mark and use the nicks transiently present in the nascent strand -- the un-ligated Okazaki junctions themselves -- plus the orientation of PCNA.

Q: Replication fidelity is about 10^-10 and the mutation rate about 1.1-1.3 x 10^-8. Are these comparable numbers? A: No -- they are different quantities in different units and differ by roughly 100-fold. Fidelity of ~10^-10 is per base per replication; the germline rate of ~1.1-1.3 x 10^-8 is per base per generation, summing hundreds of cell divisions and also counting unrepaired chemical damage that was never a polymerase error. Multiplying 10^-10 by a few hundred germline divisions gives about 3 x 10^-8, which overshoots the measurement by a factor of two or three -- agreement within a small factor is all that calculation is entitled to claim.

Q: Why is the replication primer made of RNA rather than DNA? A: A DNA polymerase cannot start a chain, only extend an existing 3'-hydroxyl, so primase -- an , which can start from nothing -- lays the first few nucleotides. Those are also the least accurate nucleotides in every fragment, laid down with no template-matched predecessor to check against, and RNA is chemically marked as foreign inside a DNA molecule. Making them of RNA guarantees they will be found and excised rather than left in the archive.

Q: Why do shorten, and is "polymerase cannot copy the last bit" the whole story? A: Only half of it. The last Okazaki fragment on the lagging strand is primed some distance from the terminus, and when its RNA primer is removed the gap cannot be filled because no 3'-hydroxyl exists to its right. But the leading-strand end, which could in principle be copied to the last base, is also deliberately resected by nucleases to generate the single-stranded 3' overhang that forms the t-loop. Both ends shorten, and part of the shortening is an active choice rather than a limitation.

Q: Longer telomeres are healthier and is an anti-ageing enzyme. What is wrong with that? A: Telomere length is a division counter with a genuine two-sided trade-off. Germline variants that shorten telomeres cause bone-marrow failure and pulmonary fibrosis, while germline variants that lengthen them raise the risk of melanoma, glioma and other cancers. Reactivated telomerase is a near-universal feature of tumours: 85-90% of cancers restore it and 10-15% use the recombination-based ALT mechanism, and TERT mutations are among the most frequent non-coding known.

Q: A method a 2.3 Mb gene as significantly mutated across a tumour cohort. What should you check before believing it? A: Whether the background mutation rate was modelled locally. Late-replicating, gene-poor, heterochromatic regions accumulate substantially more somatic mutations, and chromatin state plus replication timing explain up to about 86% of the variance in mutation density at megabase scale (Polak et al., Nature 2015). A test using a flat genome-wide background will nominate long, late-replicating, lowly expressed genes as drivers purely for sitting in high-background neighbourhoods -- a statistical modelling failure that produced a well-known crop of spurious cancer genes.

Transcription

Q: reads the , but databases print the coding strand. Why is "a gene's template strand" still an ill-formed idea? A: Template-ness belongs to a unit, not to a . Genes point both ways and sometimes overlap, so one gene's template is its neighbour's coding strand, and the reference plus strand is an arbitrary assembly convention with no biological meaning. This is why every format carries an explicit strand field, why a variant on a minus-strand gene must be reverse-complemented before reasoning about protein consequence, and why "the C>T variant" is ambiguous until you say relative to what.

Q: Why is the bacterial -10 element AT-rich, and why does the spacer's length matter but not its sequence? A: The -10 element is where the duplex must melt, and sigma region 2 grips the unpaired bases rather than reading a duplex; AT-rich sequence has the lowest duplex-opening -- the three weakest nearest-neighbour steps (5'-TA, 5'-AT, 5'-AA) are the AT-only ones, and A:T pairing is itself net destabilising -- so it lowers the barrier to the one step that has to happen there. The spacer matters only geometrically: sigma regions 2 and 4 are held rigidly apart on one protein, so the spacer must place the -35 element at the right distance and rotational angle. At 10.5 bp per turn one extra base rotates the -35 by about 34 degrees, so 17 plus or minus 1 bp works and 15 or 20 does not, whatever lies between.

Q: Is a strong or weak like a switch, and would a perfect-consensus promoter be the strongest possible? A: Neither. Expression level is a rate, not a state: a promoter fires some number of times per hour and each transcript decays with some half-life, and abundance is the ratio. Promoter strength is well approximated by a position weight matrix score, but a perfect-consensus promoter is counterproductive -- the holoenzyme binds so tightly it cannot escape, and output falls. Affinity and throughput are different objectives.

Q: An intrinsic terminator needs a GC-rich hairpin and a U-tract. Predict the effect of deleting each, and say why neither composition is arbitrary. A: Delete the U-tract and the hairpin still tugs but the polymerase grips through an ordinary hybrid it can hold, so you get a strong pause rather than termination -- exactly how regulatory pause hairpins work. Delete the hairpin and the weak rU:dA tract merely slows the enzyme, giving mostly read-through. The stem is GC-rich so it folds faster than the polymerase can move past it and then holds; the tract is uridine because rU:dA is the weakest RNA:DNA pairing there is. Termination is the coincidence of maximum pull and minimum grip.

Q: How does Rho-dependent termination work, and why does a nonsense in the first gene of an silence the genes downstream? A: Rho is a hexameric ATP-driven 5' to 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 RNA:DNA hybrid apart when it catches a paused enzyme. Because Rho loads only on naked RNA and a translating ribosome covers RNA, a translated message is protected and an untranslated one is not. That is polarity: the nonsense makes ribosomes fall off early, Rho loads on the newly exposed RNA, and the rest of the operon is never transcribed.

Q: Does transcription start at the , and do most human promoters have one? A: No on both counts. The TATA box sits roughly 31 to 24 bp upstream of the start site rather than at it, and depending how strictly the motif is defined only about 10-25% of human core promoters carry 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 position. TBP is recruited regardless, arriving as part of TFIID and positioned by TAF subunits reading other elements or by protein contacts alone, so "TATA-less" describes the DNA and not the machinery.

Q: How does the Pol II C-terminal domain make RNA processing co-transcriptional? A: The CTD is 52 tandem Y-S-P-T-S-P-S heptads whose phosphorylation pattern changes systematically with position along the gene, and processing machines bind the marks rather than the polymerase or a clock. Ser5 phosphorylation, written by CDK7, peaks near the promoter and recruits the capping enzymes, so the cap is added when the nascent RNA is only about 20-25 nucleotides long. Ser2 phosphorylation, written by P-TEFb and CDK12/13, accumulates downstream and recruits and cleavage/polyadenylation factors -- exactly where their substrates appear.

Q: A strong Pol II peak sits at a promoter but shows essentially no transcript. What is the most likely explanation, and how would you test it? A: At most metazoan genes Pol II initiates, transcribes 20-60 nucleotides with a peak near +30, and stops, held for minutes by DSIF and NELF until P-TEFb phosphorylates CTD Ser2, SPT5 and NELF-E and releases it. The ChIP peak usually is that paused polymerase, so occupancy and output are different measurements. The competing explanation is healthy synthesis with rapid transcript decay; a nascent-transcription assay such as PRO-seq or GRO-seq distinguishes them, showing a sharp +30 peak with an empty gene body under pausing and signal throughout the gene under rapid decay.

Q: Why is assigning a non-coding regulatory variant to its nearest gene a biased heuristic? A: Because act in either orientation, upstream or downstream, from tens of kilobases up to about a megabase, and routinely skip intervening genes -- the intervening DNA loops out to bring activators to Mediator and the pre-initiation complex. The ZRS enhancer that drives SHH in the limb bud sits roughly 850 kb to 1 Mb away, inside 5 of an unrelated gene, LMBR1, and point mutations in it cause extra digits. Reading the sequence would never tell you what it controls.

Q: About 75% of the human genome is transcribed in some cell type. Why is "transcribed" therefore a very weak functional claim? A: Because low-level transcription is expected of any accessible DNA, and most of that signal sits under one copy per cell and is poorly conserved. Its sources are mundane: most mammalian promoters fire in both directions, and the upstream product (a PROMPT) is short-lived and destroyed by the nuclear exosome, so directionality is imposed after initiation rather than at it; add enhancer RNAs, antisense and intronic transcription, and read-through past termination sites, which rises sharply under stress. The operational consequence is immediate -- in RNA-seq 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.

RNA processing and splicing

Q: forms as many phosphodiester bonds as it breaks, so the chemistry is near energetically neutral. Why then does the hydrolyse ATP at eight separate steps? A: Not to drive the reaction. Eight DExD/H-box RNA each drive an irreversible conformational rearrangement and act as a kinetic checkpoint: a correctly paired substrate advances to the next state faster than the ATPase can reject it, while a poorly paired one is discarded. The energy buys ordering and fidelity for a reaction that would otherwise be reversible and error-prone -- accuracy from imperfect filters in series, paid for in ATP.

Q: The spliceosome contains roughly 150 proteins. What actually performs the catalysis, and what does that imply about where the machine came from? A: RNA does. U6 snRNA folds a triplex that coordinates the two magnesium ions doing the chemistry, and the proteins are scaffolding, motors and proofreaders. That U6 catalytic core is superimposable on domain V of a group II self-splicing -- an RNA that performs the identical two transesterifications, through the identical lariat, with no spliceosome at all. The economical reading is that the spliceosome is a group II intron broken into trans-acting pieces, and once the recognition elements became separable from the sequence they act on, they became regulatable and became possible.

Q: A variant destroys the 5' splice site of intron 4 in a human gene. Predict the mRNA, and say why the prediction differs in budding yeast. A: In humans exon 4 is skipped and exon 3 joins exon 5. Human exons are short (median about 120 bp) inside long introns (mean about 5.4 kb), so initial recognition is by -- U1 on an exon's downstream 5' site pairs across the exon with U2AF on its upstream 3' site -- and killing the donor leaves the exon without half its defining pair. In yeast, introns are short and recognition spans the intron, so losing the donor strands the downstream acceptor and the usual outcome is intron retention.

Q: Why can you not find splice sites by pattern-matching the GU-AG consensus? A: A 5' splice site carries only about 8 bits across 9 degenerate positions, so one position in 2^8 = 256 matches by chance. A 3.1 Gb genome therefore contains roughly 1.2 x 10^7 consensus-matching decoys per strand against roughly 1.5 x 10^5 real 5' splice sites -- about eighty decoys for every true site. The consensus is necessary and nowhere near sufficient; discrimination comes from auxiliary sequence elements, the proteins bound to them, exon-definition geometry and the order in which the polymerase emits the sequence.

Q: Which splice-disrupting variants do standard pipelines label benign and discard? A: Everything except the invariant GU and AG dinucleotides. A synonymous or change inside an exon can destroy an exonic splicing and delete the whole exon; a deep intronic variant can create a new GU and activate a pseudoexon; and donor +3 to +6 or acceptor -3 to -20 changes weaken rather than abolish a site, giving dose-dependent partial skipping. CEP290 c.2991+1655A>G lies 1,655 nt inside an intron and activates a 128 bp pseudoexon causing Leber congenital amaurosis type 10, and every exome-only pipeline is blind to it. A variant's consequence annotation is a statement about a transcript model, not about a molecule.

Q: How does a know that a stop is premature, and where does the 50-55 nucleotide rule come from? A: Each splicing event deposits an exon-junction complex about 20-24 nucleotides upstream of the junction, sequence-independently, and the first round of sweeps them off as the ribosome passes. A normal stop sits in the last exon, so every complex has been cleared by termination; terminate with one still downstream and UPF1/UPF2/UPF3 contact it and the message is destroyed. The threshold follows from geometry: a terminating ribosome protects about 25-30 nt downstream of the stop, which physically masks a complex sitting 20-24 nt upstream of a junction, so only stops more than about 50-55 nt upstream of the final junction trigger decay.

Q: A gene has 12 exons. Compare a nonsense variant in exon 4 with one 200 nt into the 400 nt final exon. Which makes protein, and which is more likely to be ? A: Exon 4 lies far upstream of the last junction, so the exon-junction complex is still present at termination, fires, the message is degraded and no protein is made -- a clean , typically . The exon 12 variant sits downstream of the last junction, so no complex remains, decay does not fire and a truncated protein is actually made; if it still assembles into complexes but does not work, it poisons the product of the normal and behaves as a dominant negative. Same class, opposite genetic behaviour, decided by position.

Q: on one sample gives about 40,000 A-to-G calls with no DNA support, clustered in 3' UTRs and introns inside inverted repeats. What are they? A: A-to-I editing by ADAR. Inosine is read as G by reverse transcriptase, so an edited adenosine appears as an A-to-G mismatch while the genome still reads A. The clustering is diagnostic: ADAR needs double-stranded RNA, and the commonest endogenous source is two nearby in opposite orientation folding back on each other. These are not DNA variants and must be filtered against a known-editing database rather than reported as .

Q: Where does an mRNA's 3' end come from, and what sets the message's clock? A: Not from where stopped -- Pol II has no crisp terminator, so the 3' end is where the transcript was cut. CPSF reads the AAUAAA hexamer (canonical, in roughly half of human 3' ends, with AUUAAA the common second) while CstF binds a downstream G/U-rich region; after cleavage, poly(A) polymerase adds about 200-250 adenosines with no template at all, so the tail is not encoded in the genome. Poly(A)-binding protein coats the tail and blocks 3' to 5' decay, and because deadenylation is the first and usually rate-limiting step of turnover, tail length is the message's clock.

The genetic code and translation

Q: What does "the is degenerate" mean, and why is degenerate not the same as ambiguous? A: Three bases per group is the smallest that works, since two give only 16 combinations for 20 amino acids, and three give 64 -- 61 sense codons plus 3 stops -- for 20 outputs, so the mapping is many-to-one. Degenerate means several codons share one meaning; ambiguous would mean one codon has several meanings, which never happens. The map is a function, and the redundancy is unevenly distributed: Leu, Ser and Arg have six codons each while Met and Trp have one.

Q: Why is the third codon position tolerant, and why is that really a statement about tRNA rather than about the code? A: The small ribosomal subunit inspects the of positions 35 and 36 and rejects anything non-canonical, but position 34 sits at the end of the anticodon loop and is not inspected that way. Non-standard pairings tolerated only there -- G with U, U with G, and inosine with A, C or U -- let one tRNA read two or three codons. Third-position in the table is therefore a downstream consequence of physical looseness in the decoding site, and it is why 31 tRNAs suffice for 61 codons.

Q: Transitions are about twice as common as transversions. Show that the code exploits this. A: Every two-fold degenerate family splits its third-position options into {U, C} against {A, G} -- against , exactly the transition partition. A third-position transition is therefore synonymous in every two-fold family and trivially so in every four-fold family. The only exceptions are AUA/AUG and UGA/UGG, involving Met and Trp, the two amino acids with a single codon each, where no arrangement could have helped. Scored against random codes with realistic error weighting, fewer than one in a million does better (Freeland and Hurst, 1998).

Q: Why is the second codon position the one to examine hardest? A: Because it is the hydrophobicity axis and has no degeneracy at all. Every codon with U at position 2 encodes a hydrophobic residue, every codon with A at position 2 encodes a polar or charged one or a stop, and C at position 2 gives the small residues. A second-position change always alters the amino acid, and usually carries it out of its class -- but usually, not necessarily. The U (hydrophobic), C (small) and A (polar or charged) columns are chemically coherent; the G column (Cys, Trp, Arg, Ser, Gly) is not, so first-position changes inside it are uncushioned too (UGU Cys to CGU Arg), and Arg to Lys (AGA to AAA) and Ser to Thr (AGU to ACU) are second-position changes that stay in class.

Q: Is AUG "the start codon"? A: AUG is methionine everywhere it occurs; what makes one AUG a start is context, and the two domains solve it differently. Bacteria use a Shine-Dalgarno sequence (consensus AGGAGG, about 8 nt upstream) that base-pairs with the 3' end of 16S rRNA and drops the AUG straight into the P site, so a can initiate internally and each cistron of an gets its own start. Eukaryotes use cap-dependent scanning -- eIF4E binds the 5' m7G cap, the 43S complex loads at the 5' end and scans until it meets an AUG in good Kozak context, gccRccAUGG, where a purine at -3 and a G at +4 do most of the work -- so eukaryotic mRNAs are functionally monocistronic, which is why rather than operon structure is the eukaryotic route to several products from one locus.

Q: You chemically convert a cysteine-anticodon tRNA so that it carries alanine instead, then add it to a reaction. What comes out, and what does it show? A: Alanine at every cysteine codon -- Chapeville's 1962 experiment. The ribosome inspects codon-anticodon geometry and nothing else; a tRNA's two ends are 70-80 Angstroms apart and functionally uncoupled, so it cannot check which amino acid is attached. The genetic code is therefore enforced by the twenty aminoacyl-tRNA synthetases at the charging step, not by the ribosome, which enforces only the frame and the pairing.

Q: How does a synthetase discriminate between amino acids that differ by a single methyl group? A: By a double sieve, because binding-pocket geometry alone is not enough -- isoleucyl-tRNA synthetase discriminates against valine only about 1 in 150 on shape. A synthetic site excludes anything larger than the correct amino acid, and a separate editing site hydrolyses anything smaller that slipped through. Two imperfect filters in series give a combined error rate near 10^-4, the same architecture as replication fidelity.

Q: A 4 bp and a 3 bp deletion occur at the same position in a coding exon. Predict the difference, and give the expected length of the abnormal peptide. A: The 3 bp deletion removes one codon and preserves the frame -- one residue lost, everything downstream unchanged, severity comparable to a change (CFTR p.Phe508del is this case). The 4 bp deletion shifts the frame, so every downstream codon is unrelated sequence. In the new frame 3 of 64 codons are stops, so codons-before-stop is geometric with p = 3/64 and the expected run is 64/3, about 21 codons; the transcript is then usually destroyed by , so the practical result is no protein.

Q: Give four distinct mechanisms by which a synonymous change can be pathogenic. A: It can destroy an exonic and cause exon skipping, which is the entire difference between SMN1 and SMN2 (c.840C>T at position +6 of exon 7, no amino acid changed, exon 7 skipped in 80-90% of transcripts). It can swap a common codon for a rare one, slowing translation and altering co-translational folding. It can change local mRNA structure and stability, particularly near the 5' end where structure occludes the start. And the same bases can carry overlapping information -- target sites, RNA-binding-protein sites, or in compact genomes a second reading frame.

Proteins and gene function

Q: Why is usually , and what characterises the genes where it is not? A: Not because the intact works harder -- it is generally not upregulated. Because 50% of normal product is usually enough: flux through a multi-step pathway is controlled jointly by all its enzymes, so each has a small flux control coefficient and the flux-activity curve saturates strongly (Kacser and Burns). Recessivity is a by-product of pathway kinetics, not an adaptation. The exceptions are haploinsufficient genes -- rate-limiting products, dosage-sensitive , and subunits needed in fixed stoichiometry -- and roughly 3,000 human genes score as strongly loss-of-function-intolerant in metrics.

Q: In some disease gene, variants cause severe disease while whole-gene deletions cause almost nothing. What is the mechanism, and what does that imply for a frameshift found in a patient? A: Not loss of function, since losing the gene entirely is tolerated. The missense variants must be doing something positive: a (constitutive or novel activity, or a toxic species) or a effect, and whether the protein works in a complex distinguishes the two. The implication is that the frameshift is probably not the cause -- it produces the same state as the benign deletion, so applying "predicted null means strong evidence of pathogenicity" here would be a mechanistic error.

Q: For a complex containing n copies of an affected subunit, what fraction is fully wild-type in a heterozygote, and why does that make a missense worse than a null? A: The fraction of fully normal complexes is (1 - f)^n with mutant fraction f, so for a heterozygote at f = 0.5 it is 50% for a monomer, 25% for two copies, 12.5% for a homotrimer and 6.25% for a homotetramer. A null allele floors effective activity at 50% because the molecules it does make are perfect; a dominant-negative allele drives it far below that, because defective subunits are incorporated and poison the assembly. Type I collagen draws two alpha1(I) chains from the pool, so a glycine substitution leaves three quarters of all molecules defective, which is why severe osteogenesis imperfecta comes from missense and mild disease from a null.

Q: A protein has a 20-hour half-life. Can you halve its level within 30 minutes by shutting off ? A: No. With dP/dt = s - kP, the steady-state level is P = s/k and the approach to any new steady state has half-time ln2/k -- the protein's own half-life -- regardless of what you do to the synthesis rate s. Even setting s to zero leaves 20 hours to lose half the molecules, because the existing pool must clear. Responding in 30 minutes requires raising k by targeted ubiquitination and proteasomal degradation, which is why fast-acting regulators such as cyclins and transcription factors are short-lived and pay for it in continuous resynthesis.

Q: Ubiquitination tags a protein for destruction. What does that summary miss? A: That ubiquitin is a typed tag, not a boolean . Chains are built by linking through one of ubiquitin's own lysines, and the carries the meaning: K48-linked chains route the substrate to the 26S proteasome, while K63-linked chains degrade nothing and act instead in signalling, and trafficking. Substrate specificity comes from the E3 ligase at the end of the E1-E2-E3 cascade, of which humans encode more than 600. The proteasome itself is a barrel with its catalytic sites inside, so a substrate must be unfolded and threaded through a narrow pore -- destruction is sequestered, and nothing is degraded by accident.

Q: Do chaperones tell a protein how to fold? A: No -- they add no structural information and do not know the target fold. Anfinsen showed that all the information specifying the structure is in the sequence, and Levinthal showed why exhaustive search is impossible: 3^100 is about 5 x 10^47 states, which at 10^-13 s each would take 1.6 x 10^27 years. Folding works because the energy landscape is funnelled; chaperones such as Hsp70 and GroEL/GroES only suppress aggregation in a crowded cytoplasm and give the chain repeated attempts at that funnel.

Q: In what sense is a gene an rather than a thing? A: Because every physical definition fails. gives one locus many products (GENCODE 50 annotates 78,733 genes and 644,292 transcripts, about eight per gene); non-coding RNA genes never make protein; genes nest inside other genes' on the opposite strand (OMG, EVI2A and EVI2B inside NF1); put several products under one transcription unit; regulatory elements sit tens of kilobases away inside other genes; and RNA editing means the transcript sequence is not the genomic sequence. A gene is a curation project's claim that a set of overlapping transcripts is one coherent unit, revised at each release.

Q: A structure predictor returns a model of your missense variant essentially identical to wild type at high per-residue confidence. What have you learned? A: Almost nothing about the variant. These models predict the structure a sequence adopts if it folds; they were not trained on stability changes and characteristically return the wild-type fold, confidently, for destabilising substitutions, and pLDDT expresses confidence in the fold rather than in thermodynamic stability. What the model is genuinely good for is context -- whether the residue is buried or surface-exposed, in an active site, at a domain interface, or in a disordered region -- and a buried charge or volume change is far more likely to be damaging than the same substitution on the surface.

Q: Why are glycine, proline and cysteine substitutions unusually informative when reading a variant? A: Glycine has no side chain, so it occupies almost no space and adopts backbone geometries nothing else can; a conserved glycine usually marks a tight turn or a packing constraint, and collagen's obligatory glycine at every third position is the extreme case. Proline's side chain loops back onto its own backbone nitrogen, so it cannot donate a backbone hydrogen bond and its rotation is locked -- it breaks helices and terminates sheets. Cysteine forms the disulfide bond, the only covalent cross-link in ordinary protein structure, and because the cytoplasm is reducing, disulfides are effectively confined to secreted and extracellular proteins.