Problem set 05A — Bacterial and phage genetics
Covers Ch 20A.
Attempt before revealing. This is the most calculable material in classical genetics, and the reason is that bacteria let you select instead of count. Almost every problem below therefore turns on a denominator — how many cells you plated, how many particles were in the lysate, how big a fragment the phage could carry. The set is built to expose four specific habits: reading an absolute clock reading as a map position (problem 1), adding cotransduction frequencies (problem 2), collapsing the two questions "same function?" and "same position?" into one (problems 3 and 4), and trusting a binary search whose central assumption has quietly failed (problem 8).
Every constant used here is pinned in Ch 20A: the E. coli MG1655 is 4,641,652 bp mapped onto 100 minutes, so 1 minute = 46.4 kb; the P1 headful is ~100 kb ≈ 2 minutes; the T4 genome is 168,903 bp over ~1,600 map units, so 1 map unit ≈ 106 bp.
1. Reading a chromosome off a clock
You have an Hfr strain, Hfr B7, of unknown origin and orientation. You mate it with an Str^R F⁻ recipient carrying auxotrophic lesions in all five markers below, blend samples at intervals, and plate each on minimal medium plus streptomycin, selecting one donor marker at a time. Extrapolating each rise back to the axis gives these times of entry:
| Marker selected | Time of entry (min) |
|---|---|
| lac⁺ | 12.0 |
| gal⁺ | 21.2 |
| arg⁺ | 40.0 |
| his⁺ | 49.1 |
| xyl⁺ | 57.6 |
Three of these have known map positions from the MG1655 sequence (Ch 20A §3): lacZ at 7.9 min, galE at 17.1 min, hisG at 45.0 min. The positions of arg and xyl are what you want. Map coordinates run 0 to 100 and increase in the standard direction; all five markers lie in the first 60 minutes of the map, so you need not worry about wrapping.
(a) Give the order of the five markers and the direction in which Hfr B7 transfers. (b) Place arg and xyl on the map, in minutes. (c) You used lac as your anchor. Show that gal and his give the same answer, and say what the agreement demonstrates. (d) A colleague extrapolates the entry times to t = 0 and announces that F integrated at 95.9 min, with an initiation lag of 4.1 minutes. Those two claims cannot both be read off these data. Show that the experiment determines only one quantity, say which, and say what you would have to measure to split it into the two the colleague reported.
Solution
The goal. Turn a set of clock readings into map coordinates, and find out precisely how much of the clock reading is map and how much is overhead.
(a) Order markers by time of entry — that is their order on the , because the strand is pumped from a fixed point at a constant rate:
lac → gal → arg → his → xyl
The three anchored markers enter in the order 7.9 → 17.1 → 45.0, i.e. in the direction of increasing map coordinate. So Hfr B7 transfers "forwards" along the standard map.
(b) Position is an affine function of entry time, so differences transfer directly. Anchoring on lac:
pos(arg) = 7.9 + (40.0 − 12.0) = 7.9 + 28.0 = 35.9 min
pos(xyl) = 7.9 + (57.6 − 12.0) = 7.9 + 45.6 = 53.5 min
arg at 35.9 minutes, xyl at 53.5 minutes.
(c) Repeat from the other two anchors:
from galE: 17.1 + (40.0 − 21.2) = 17.1 + 18.8 = 35.9 ✓
from hisG: 45.0 + (40.0 − 49.1) = 45.0 − 9.1 = 35.9 ✓
from hisG: 45.0 + (57.6 − 49.1) = 45.0 + 8.5 = 53.5 ✓
All three agree exactly. What this demonstrates is the affine model itself, and it does so as a prediction test: two of the three anchors were not needed, so their agreement is a genuine check rather than an artefact of the fit.
You can also read the check directly off the anchors, without ever touching arg or xyl:
t(gal) − t(lac) = 21.2 − 12.0 = 9.2 pos(galE) − pos(lacZ) = 17.1 − 7.9 = 9.2 ✓
t(his) − t(lac) = 49.1 − 12.0 = 37.1 pos(hisG) − pos(lacZ) = 45.0 − 7.9 = 37.1 ✓
Time intervals equal map intervals, minute for minute. That equality is what makes a minute a legitimate map unit, and it is why the E. coli map is additive and never saturates — unlike a , which is a probability of exchange and tops out at 50%.
(d) The lag is not supportable; the impossibility of separating it from the origin is.
Write the model down. If F integrated so that transfer starts at map position O, transfer runs at 1 minute of map per minute of mating, and there is an initiation lag Λ before pumping begins, then for every marker:
entry_time = (position − O) + Λ
Rearranged: entry_time − position = Λ − O. Compute that quantity for all five markers:
| Marker | entry − position |
|---|---|
| lac | 12.0 − 7.9 = 4.1 |
| gal | 21.2 − 17.1 = 4.1 |
| his | 49.1 − 45.0 = 4.1 |
| arg | 40.0 − 35.9 = 4.1 |
| xyl | 57.6 − 53.5 = 4.1 |
The data contain one number, 4.1, and it is Λ − O. There are two unknowns. No amount of extra markers will separate them, because every marker contributes the same equation. The parameters are not identifiable.
The wrong path, and why it is seductive. Extrapolating to t = 0 gives position 7.9 − 12.0 = −4.1, i.e. 95.9 min on the 100-minute circle. That is the answer you get by assuming Λ = 0, and it is wrong by exactly Λ. The chapter's worked example makes the same computation legitimately — but only because that Hfr's origin was already known to be at 0, which turns Λ − O into Λ and lets it be read as a lag. Hfr B7's origin is not known, so the same arithmetic answers a different question and the reader has to notice which.
Taking Ch 20A's typical few-minute lag, Λ ≈ 6, the origin would be at 95.9 + 6 ≈ 1.9 min — but that borrows a constant from someone else's experiment, which is the habit the chapter warns against.
What to measure. The lag is a property of the mating conditions — time to form a pair and initiate at oriT — not of the map. So calibrate it: run the identical protocol, same medium, same temperature, same cell densities, with a reference Hfr whose origin is independently known. Its entry − position is Λ directly. Subtract, and Hfr B7's origin drops out.
Generalise. Interrupted mating is a time-indexed oracle with an unknown, additive offset. Any quantity you compute from differences of query times is measurable; any quantity that needs an absolute query time needs a separate calibration. That distinction survives well past bacterial genetics — it is the same reason a phylogeny gives you branch ratios for free but needs a fossil to give you dates.
A closing sanity check. arg at 35.9 and his at 45.0 are 9.1 minutes apart, which is 9.1 × 46.4 ≈ 422 kb. That is more than four P1 headfuls, so these two markers will show zero cotransduction — the coarse instrument placed them, and the fine one is useless at this range. Problem 2 is about that boundary.
2. Cotransduction: the ruler and its two dead zones
You grow P1 on a wild-type donor and transduce a recipient mutant at argF, lysA and serB. Selecting one marker at a time and scoring the unselected ones gives:
| Pair | Cotransduction frequency C |
|---|---|
| argF – lysA | 0.614 |
| lysA – serB | 0.422 |
| argF – serB | 0.216 |
Take the P1 headful as L = 2.00 minutes and 1 minute = 46.4 kb.
(a) A student adds up the two inner separations as (1 − C) values and compares with the outer pair. Do that calculation, show that it fails, and say what goes wrong. (b) Convert all three frequencies to distances properly, in minutes and in kb, and test additivity. (c) Does the additivity in (b) establish the gene order? If not, what would? (d) A fourth marker thrB gives C = 0 against all three. A fifth, hisD, gives C = 0.97 against argF, measured with an uncertainty of about ±0.01. What distance does each result give, and how good is it?
Solution
The goal. Get from a co-transfer statistic to a length, and learn where along the range the statistic stops carrying information.
(a) The wrong path. The instinct is that closer markers cotransduce more, so 1 − C ought to behave like a distance:
(1 − 0.614) + (1 − 0.422) = 0.386 + 0.578 = 0.964
observed for the outer pair: 1 − 0.216 = 0.784
The outer "distance" comes out 0.784 against a predicted 0.964 — short by 0.180, or 18.7%.
In a eukaryotic cross you would blame undetected double crossovers, as in Ch 14. That explanation is not available here: there is no second exchange hiding a first. The shortfall is purely functional. C is not linear in d, it is cubic:
C(d) = (1 − d/L)³
so 1 − C is a cubic in d too, and cubics do not add.
(b) Invert the relation instead. This is the step that makes the numbers additive:
d = L (1 − C^(1/3))
| Pair | C | C^(1/3) | d (min) | d (kb) |
|---|---|---|---|---|
| argF – lysA | 0.614 | 0.84994 | 0.300 | 13.9 |
| lysA – serB | 0.422 | 0.75007 | 0.500 | 23.2 |
| argF – serB | 0.216 | 0.60000 | 0.800 | 37.1 |
Additivity test:
0.300 + 0.500 = 0.800 min observed outer: 0.800 min ✓
Exact to three decimals — the residual is 0.000033 min, about 1.5 bp, which is rounding in the input frequencies and nothing else. In kb: 13.9 + 23.2 = 37.1 ✓.
Contrast that with (a)'s 18.7% shortfall on the identical data. The cube root does for cotransduction exactly what a mapping function does for : it converts an observable that saturates into a length that adds. The mechanism is completely different — one is packaging geometry, the other is multiple crossovers — and the remedy has the same shape, which is worth noticing rather than memorising.
(c) No. Additivity is consistent with the order argF – lysA – serB, but a lucky set of estimates can be additive by accident, and additivity alone can never distinguish an order from its mirror image.
The order comes from a three-factor transduction: transduce a triple mutant, select an outside marker, score the other two, and look at the rarest class. Under the correct order, three of the four classes need two crossovers and one needs four; the four-crossover class is rare, and in it, the marker that disagrees with the other two is the middle one. Crucially, you then check the alternative orders explicitly — a class requiring two extra exchanges cannot be the second most common, so any order that predicts a common class to be rare is excluded by the data rather than assumed away. Ch 20A's worked example runs this on tonB–trp–cysB.
(d) The two dead zones.
thrB, C = 0 — the saturated end. C = 0 means d ≥ L. All you can say is
d ≥ 2.00 min = 92.8 kb
and that is a bound, not a measurement. A marker 95 kb away and a marker on the opposite side of the give the identical reading, because neither ever shares a headful with your selected marker. This is the RF = 50% ceiling of Ch 14 in different clothing. The remedy is also the same: change instrument. Conjugation covers the whole 100 minutes and will place thrB to within about a minute, after which you hunt for a nearer marker to link it to by P1.
hisD, C = 0.97 — the compressed end.
d = 2(1 − 0.97^(1/3)) = 2(1 − 0.98990) = 0.02020 min = 938 bp
Now propagate the ±0.01 uncertainty:
C = 0.98 → d = 0.01342 min = 623 bp
C = 0.96 → d = 0.02703 min = 1,255 bp
A ±0.01 in the frequency swings the answer over a 632 bp window — two-thirds of the estimate itself. The reason is visible in the derivative of C near d = 0:
dC/dd = −3/L = −1.5 per minute
so 1 kb (= 0.0215 min) moves C by only 1.5 × 0.0215 ≈ 0.032, about three percentage points. Once your kilobases are worth three points of frequency each, ordinary plating noise is kilobase-scale noise. P1 bottoms out around 1–2 kb, and 938 bp is below that floor: quote it as "under 1 kb, and not resolvable further by P1".
Generalise. The instrument is trustworthy over roughly 1 kb to 100 kb — a range of two orders of magnitude, bounded above by geometry (the headful) and below by counting statistics (the flat slope). Both boundaries are properties of the measurement, not of the chromosome, and the fix at either end is a different assay, not a better fit.
3. Two tests that disagree ★
Four independently isolated rII mutants of phage T4 — m1, m2, m3, m4 — none of which reverts detectably. You run both of Benzer's tests on all six pairs.
Complementation. Coinfect E. coli K-12(λ) with each pair at high multiplicity and ask whether the cell lyses. + = lysis (complementation), − = no burst.
| m1 | m2 | m3 | m4 | |
|---|---|---|---|---|
| m1 | — | − | + | + |
| m2 | — | + | + | |
| m3 | — | − | ||
| m4 | — |
Recombination. Cross each pair on the permissive host E. coli B, plate the burst on K-12(λ), and count r⁺ plaques. is the percentage of progeny that are r⁺, with no doubling, as in Ch 20A §8; 1 map unit = 106 bp.
| Pair | RF (%) |
|---|---|
| m1 × m2 | 1.10 |
| m1 × m3 | 1.17 |
| m1 × m4 | 7.05 |
| m2 × m3 | 0.07 |
| m2 × m4 | 5.95 |
| m3 × m4 | 5.88 |
You are told independently that m1 lies in rIIA, and that rIIA precedes rIIB on the map.
(a) How many cistrons are represented, and which mutants fall in each? (b) Build the fine-structure map: linear order and spacing in bp. (c) m1 and m2 fail to complement; m2 and m3 complement. Which pair is physically closer? State plainly what question each test answered. (d) A reader concludes from the matrix that m1 and m2 are "the same " and from the 0.07% that m2 and m3 are "in the same gene". Diagnose both errors.
Solution
The goal. Run the two tests on one dataset and watch them return contradictory-looking verdicts, then see that they were never answering the same question.
(a) Complementation is an equivalence-like partition: mutants that fail to complement each other share a function. Reading the matrix, the "−" cells are m1×m2 and m3×m4, and every cross-group pair is "+".
Two cistrons: {m1, m2} and {m3, m4}. Since m1 is in rIIA, that group is rIIA and {m3, m4} is rIIB. This is Benzer's headline partition — all rII mutants fall into exactly two complementation groups, and he named the unit the cistron after the cis–trans test that defines it.
(b) Convert every frequency with 1 map unit = 106 bp:
| Pair | RF (%) | Separation (bp) |
|---|---|---|
| m1 × m2 | 1.10 | 116.6 |
| m1 × m3 | 1.17 | 124.0 |
| m1 × m4 | 7.05 | 747.3 |
| m2 × m3 | 0.07 | 7.4 |
| m2 × m4 | 5.95 | 630.7 |
| m3 × m4 | 5.88 | 623.3 |
The largest separation, m1–m4 at 747.3 bp, identifies the outer pair. Test the order m1 – m2 – m3 – m4:
m1–m2 + m2–m3 = 116.6 + 7.4 = 124.0 vs observed m1–m3 = 124.0 ✓
m2–m3 + m3–m4 = 7.4 + 623.3 = 630.7 vs observed m2–m4 = 630.7 ✓
m1–m2 + m2–m3 + m3–m4 = 116.6 + 7.4 + 623.3 = 747.3 vs observed m1–m4 = 747.3 ✓
Perfectly additive. The map:
|-------------- rIIA ----------------|------------- rIIB --------------|
m1 -------116.6 bp------- m2 --7.4--- m3 ------623.3 bp------ m4
^
the A/B cistron boundary, in the 7.4 bp gap
(Recombination fixes an order, not a direction; the mirror image is equally consistent, and it is the external statement "m1 is in rIIA" that orients the map. Compare problem 1(d) — again the data determine differences, and an absolute has to come from outside.)
(c) The pair that fails to complement is 16 times further apart.
m1 – m2 fail to complement 116.6 bp apart
m2 – m3 complement 7.4 bp apart
Say plainly what happened:
- The asked do these lesions damage the same function? It answered: m1/m2 yes, m2/m3 no. Its unit is the cistron.
- The recombination test asked are these lesions at the same position? It answered: all six pairs are at different positions, and gave the distances. Its unit is the recon, down to a base pair.
There is no contradiction because there was never one question. A cistron boundary is a discontinuity in function — the point where one reading frame stops and the next begins — and it has no obligation whatsoever to be a large physical distance. m2 sits near the end of rIIA and m3 near the start of rIIB; they are seven base pairs apart and in different genes, while m1 and m2 are a hundred and sixteen base pairs apart and in the same one.
That is the whole of Benzer's result, stated as sharply as the data allow: a gene is not the unit of recombination. It is an interval of many mutable sites, and happens between them.
(d) Both errors are the same error, applied in opposite directions.
"m1 and m2 are the same mutation." This reads a complementation failure as a statement about position. It is not — and the data refute it in the same table: m1 × m2 yields r⁺ recombinants at 1.10%, which is 116.6 bp of separation, comfortably above Benzer's ~0.02% (≈2 bp) detection floor. Two mutations at the identical site can never give a wild-type recombinant, because no parent carries wild-type sequence there. They did, so they are not.
"m2 and m3 are in the same gene." This reads a small recombination frequency as a statement about function. It is not — and again the table refutes it: m2 and m3 complement, so a cell carrying both makes both products and lyses. Proximity is not membership. If it were, the cistron would be defined by a distance threshold, and no threshold exists that puts 7.4 bp inside one gene while keeping 116.6 bp inside another.
Generalise. Whenever two tests on one pair of mutants disagree, the first move is not to distrust either result — it is to write down what each test's output type is. Here one returns a set partition and the other returns a metric. A partition and a metric can disagree about everything and both be right. (Ch 20A's check-yourself 3 the reverse pattern — complementation plus zero recombination — which is a different animal: usually a mutation or intragenic complementation of a multimer, and Ch 11 §7 lists the three ways a complementation test lies.)
Scale check. rIIA is 2,178 bp and rIIB is 939 bp, so the region is 3,117 bp — about 1.8% of the 168,903 bp T4 genome, or 29.4 map units at 106 bp each. The four mutants span 747 bp, roughly a quarter of it.
4. Designing the merodiploid
You have four independently isolated histidine auxotrophs of E. coli: his-1, his-2, his-3, his-4. All four are point somewhere in the histidine biosynthetic . You want to know which pairs damage the same cistron. E. coli is haploid.
(a) Design the test. Say what you have to construct, what you plate on, and — before looking at any data — what result you would call "complementation". (b) You build the and plate 2.0 × 10⁸ cells of each on minimal medium without histidine:
| Construct | Colonies |
|---|---|
| his-1 / F′his-2 (trans) | confluent lawn |
| his-3 chromosome / F′his-4 (trans) | 236 |
| his-3 chromosome / F′his⁺ (control) | confluent lawn |
| his⁺ chromosome / F′ carrying both his-3 and his-4 (cis control) | confluent lawn |
| his-3 haploid, no F′ | 38 |
| his-4 haploid, no F′ | 45 |
Interpret every row. (c) A colleague looks at the 236 colonies and concludes that his-3 and his-4 complement weakly. Compute what is actually going on. (d) Why must the two mutations sit on separate DNA molecules? What is the cis control for, and what would you conclude if it had failed to grow?
Solution
The goal. Build a inside a haploid, and then be disciplined about what counts as a positive result.
(a) The design. A requires both mutations in one cell, on separate DNA molecules, each molecule otherwise complete. In a diploid eukaryote that is a cross; in a haploid it is impossible — until an F′ supplies a second copy of one region and nothing else.
So:
- Obtain an F′ carrying the his region (F′his), derived from imprecise excision of an integrated F in an Hfr whose integration site is near his. F is single-copy, so the merodiploid is a clean 2n for this region and n everywhere else.
- Introduce F′his-2 by conjugation into an F⁻ his-1 recipient, and so on for each pair. Select for the F′ using an unselected marker on it — never using His⁺, which is the thing you are about to measure.
- Plate the merodiploid on minimal medium without histidine.
- Controls, all at the same cell density: each parent alone (reversion), the F′his⁺ merodiploid (the F′ can supply the function at all), and a cis control.
The positive result, defined in advance: every cell grows. Complementation is a property of the , so if the two lesions are in different cistrons, each molecule supplies what the other lacks in every single cell and the plate is a confluent lawn. This definition is the whole problem, and it is why you state it before you look.
(b) Row by row.
| Construct | Reading |
|---|---|
| his-1 / F′his-2 | Confluent — complementation. Different cistrons. |
| his-3 / F′his-4 | 236 colonies from 2.0 × 10⁸ = 1.18 × 10⁻⁶. Not complementation (see (c)). Same cistron. |
| his-3 / F′his⁺ | Confluent — confirms the F′ carries a functional his region and that his-3 is recessive to his⁺. Without this row, a negative result is uninterpretable: it could just mean your F′ was broken. |
| cis control | Confluent — both lesions on one molecule, wild type on the other. Confirms neither is or . |
| his-3 haploid | 38 revertants per 2.0 × 10⁸ = 1.9 × 10⁻⁷. |
| his-4 haploid | 45 revertants per 2.0 × 10⁸ = 2.25 × 10⁻⁷. |
(c) The arithmetic that kills "weak complementation".
First, compare with what complementation looks like. If his-3 and his-4 were in different cistrons, every one of the 2.0 × 10⁸ cells would grow — a frequency of 1. Observed:
236 / 2.0 × 10⁸ = 1.18 × 10⁻⁶
That is smaller than the complementation prediction by a factor of about 8.5 × 10⁵, nearly six orders of magnitude. Nothing about "weak" covers a millionfold gap. A quantitative complementation would show up as slower or smaller colonies everywhere, not as a handful of colonies on an otherwise empty plate.
Second, account for the 236 colonies properly. Two processes make His⁺ cells in a his-3/F′his-4 merodiploid that has failed to complement:
reversion of his-3 38 per 2.0 × 10⁸
reversion of his-4 45 per 2.0 × 10⁸
─────────────────
expected from reversion 83 per 2.0 × 10⁸ = 4.15 × 10⁻⁷
observed 236
excess 236 − 83 = 153 per 2.0 × 10⁸ = 7.65 × 10⁻⁷
The excess is recombination between the two copies. Both mutations are in the same cistron, one on the chromosome and one on the F′; a crossover between the two mutant sites reconstructs an intact his sequence on one molecule. This is precisely Benzer's point again — two mutations in one cistron still recombine — and it is why you would need to plate ~1.3 × 10⁶ cells just to see one such event.
So the verdict is: his-3 and his-4 fail to complement — same cistron — and the 236 colonies are 83 revertants plus ~153 recombinants.
The named trap: rare colonies are recombinants, complementation is a lawn. The two are separated by six orders of magnitude and by kind, not degree. Anyone who scores a complementation test by counting colonies rather than by looking at whether the population grows will eventually call a a weak complementation.
(d) Why separate molecules, and what cis buys you.
The trans configuration is the informative one because it forces each molecule to supply a whole product on its own. If the two lesions sat on the same molecule with a wild-type partner — the cis configuration — the wild-type molecule alone would supply everything, and the cell would grow whether the lesions were in one cistron or fifty. The cis arrangement cannot distinguish the hypotheses; it is a control, not a test. That asymmetry is what "cis–trans test" names.
What the cis control rules out is dominance. If his-3 or his-4 made a poisonous product — a subunit that spoils the complex it joins — then the trans merodiploid would fail to grow for a reason having nothing to do with which cistron the lesion is in, and you would wrongly score "same cistron". A cis control that failed to grow would say exactly that: at least one mutation is dominant, the complementation test is invalid for this pair, and the negative trans result must be discarded.
Generalise. Two instances of one object in a single address space, independently mutable, is the enabling structure — and it is the reason Ch 21 can turn cis versus trans into an algorithm for telling a broken diffusible product from a broken sequence element. A diffusible product can be supplied by either copy; a sequence element can only ever affect the molecule it sits on.
5. Which markers move, and how often
Two lysates are prepared from the same wild-type E. coli donor.
- Lysate X — phage P1 grown lytically on the donor.
- Lysate Y — phage λ, obtained by UV-inducing a λ lysogen of the donor.
Each is used to transduce a recipient mutant in every marker, and transductants are counted per plaque-forming unit:
| Marker | Map position (min) | Lysate X | Lysate Y |
|---|---|---|---|
| thr | 0.0 | 1.8 × 10⁻⁶ | none detected |
| lac | 7.9 | 2.4 × 10⁻⁶ | none detected |
| gal | 17.1 | 2.1 × 10⁻⁶ | 1.0 × 10⁻⁶ |
| bio | just beyond attB | 2.0 × 10⁻⁶ | 1.0 × 10⁻⁶ |
| trp | 28.5 | 1.6 × 10⁻⁶ | none detected |
| his | 45.0 | 2.2 × 10⁻⁶ | none detected |
A Gal⁺ transductant from lysate Y, which also carries a normal λ prophage, is then induced. The resulting lysate Z transduces gal at 1.0 × 10⁻¹ per pfu and nothing else at all.
λ's attachment site attB is at 17.4 min. The P1 headful is ~100 kb.
(a) Name the mechanism behind each lysate and give the diagnostic feature in the table. (b) Why does lysate Y move gal and bio and nothing else? How far is gal from attB? (c) Explain lysate Z's 10⁵-fold jump, and why bio dropped out of it. (d) Taking 2.0 × 10⁻⁶ as the typical lysate X frequency, estimate what fraction of P1 particles carry host DNA rather than phage DNA. Is your estimate an upper or lower bound? (e) What kind of cell is a transductant from each lysate?
Solution
The goal. Diagnose two mechanisms from a table of which markers move, and turn a transduction frequency into a statement about the particle population.
(a) Lysate X is generalized transduction (P1). Lysate Y is specialized transduction (λ). The diagnostic is the shape of the marker list, not the frequencies — both are ~10⁻⁶:
- X moves every marker tested, at similar frequency, regardless of position. That is what you expect if the mechanism is a packaging error: pseudo-pac sites are scattered around the , so any 100 kb window can end up in a head.
- Y moves exactly two markers, and they are the two flanking attB. That is what you expect if the mechanism is an excision error: the prophage is at one site, and imprecise excision can only pick up DNA immediately adjacent to it.
Note the trap the table is built to catch: comparing the frequencies (2 × 10⁻⁶ vs 1 × 10⁻⁶) tells you nothing at all. Both mechanisms are rare errors. Only the distribution over markers separates them.
(b) λ integrates at a single site by site-specific recombination, and excision reverses that integration. When excision happens between mismatched sites, the circle that comes out carries a stretch of whatever lies immediately outside the prophage — and leaves an equivalent length of phage behind, because a head holds a fixed amount of DNA. In E. coli, attB sits between the gal and bio , so the only host genes λ can ever carry are gal (one side) and bio (the other). The products are named for it: λdgal, λdbio.
Distance from gal to attB:
17.4 − 17.1 = 0.3 min = 0.3 × 46.4 kb ≈ 13.9 kb
Two consequences follow from that number, and they run in opposite directions. It is small enough that λ can reach gal by an excision error at all. And it is well under the 2-minute P1 headful, so P1 can move gal too — the two mechanisms overlap at this locus and are told apart only by what else each lysate moves. trp at 28.5 min is 11.4 minutes from gal, more than five headfuls, so no P1 particle carries both.
(c) The Gal⁺ transductant from lysate Y carries a defective λdgal plus a normal helper λ — it is a double lysogen. Induce it and the helper supplies every function the defective genome lacks, so roughly half the burst is λdgal. The lysate goes from LFT (low-frequency transducing, ~10⁻⁶) to HFT (high-frequency transducing, ~10⁻¹):
1.0 × 10⁻¹ / 1.0 × 10⁻⁶ = 10⁵-fold
bio dropped out because a λdgal particle carries gal. It never carried bio: an individual imprecise excision goes one way or the other, not both. Lysate Z is a clonal amplification of one particular excision accident, which is exactly why it is so much better at the one marker and useless at everything else. Specialized transduction trades generality for efficiency, and lysate Z is that trade taken to its limit.
(d) A P1 particle carrying host DNA holds a random ~100 kb window. The fraction of the chromosome in one window:
100,000 / 4,641,652 = 0.02154 → 2.15% of the chromosome
If a fraction f of particles carry host DNA, the chance that a given particle both carries host DNA and that its window contains your marker is f × 0.0215. Setting that equal to the observed transduction frequency:
f × 0.02154 = 2.0 × 10⁻⁶
f = 9.3 × 10⁻⁵ → about 1 particle in 11,000
This is a lower bound, and the reason matters. The calculation assumed that every delivered fragment containing the marker produces a transductant. It does not: the fragment must also recombine into the resident chromosome, with a crossover on each side of the marker — the same requirement that makes cotransduction a cube in problem 2. Recombination succeeds well under 100% of the time, so the true fraction of host-DNA-carrying particles is higher than 9.3 × 10⁻⁵, possibly by an order of magnitude.
Get the direction of that bound the wrong way round and you conclude that transduction is more efficient than it is. The rule: when your model omits a step that can only lose events, your estimate of the input is too low.
(e) They are different objects, and the difference is testable.
- From lysate X: an ordinary haploid cell. The particle carried host DNA and no phage genes, so it can neither lyse nor lysogenise. The donor fragment recombined in, replacing the recipient . The transductant is stable, is not a lysogen, and releases no phage.
- From lysate Y: typically a . The λdgal genome integrates carrying the donor gal⁺ while the recipient's own gal⁻ stays in place — two copies of the region in one cell. It is a lysogen, it is immune to superinfection by λ, and inducing it releases phage.
So a one-line diagnostic: streak the transductant and induce it. Lysate X's transductants do nothing; lysate Y's release a burst. And because lysate Y's product is a partial , it is usable for exactly the of problem 4 — reached by a completely different route from an F′, by the same underlying accident. Imprecise excision of an integrated element, carrying flanking host DNA with it, is one mechanism with two names, depending on whether the integrated element was a plasmid or a virus.
6. Which plasmids can share a cell
Five plasmids, characterised in E. coli:
| Plasmid | Size | Copy number | Replication control | Other |
|---|---|---|---|---|
| pAmp | 5.2 kb | ~20 | ColE1-type antisense RNA I, family α | Amp^R, non-conjugative |
| pTet | 28 kb | ~19 | ColE1-type antisense RNA I, family α | Tet^R, conjugative |
| pCam | 5.4 kb | ~18 | ColE1-type antisense RNA I, family β | Cm^R; 99% identical in sequence to pAmp outside the control region |
| pKan | 99 kb | ~1 | F-type iterons | Kan^R, conjugative |
| pStr | 4.8 kb | ~18 | ColE1-type antisense RNA I, family β | Str^R, non-conjugative |
(a) There are 10 pairs. Which can be stably co-maintained and which cannot? Give the rule you used in one sentence. (b) What is the largest number of these plasmids one cell can stably carry, and how many distinct such sets are there? (c) pAmp and pCam are 99% identical in sequence. Explain, mechanistically, why that is irrelevant. Then explain why pAmp and pTet — differing more than fivefold in size — have essentially the same copy number. (d) pKan is a conjugative plasmid carrying Kan^R plus, on an integron cassette, three further resistance determinants. A patient is treated with kanamycin only. Explain what happens to the other three determinants, and why a Klebsiella isolate can be resistant to a drug it has never met.
Solution
The goal. Predict coexistence from a controller, not from a sequence — and see why the whole Inc classification is a behavioural one.
(a) The rule: two plasmids are incompatible if and only if they share a replication control system. Nothing else in the table matters.
Grouping by the control column: Inc-α = {pAmp, pTet}, Inc-β = {pCam, pStr}, Inc-F(iteron) = {pKan}. Two plasmids from the same group are incompatible; any two from different groups coexist.
| Pair | Verdict |
|---|---|
| pAmp + pTet | incompatible — both Inc-α |
| pCam + pStr | incompatible — both Inc-β |
| pAmp + pCam · pAmp + pKan · pAmp + pStr · pTet + pCam · pTet + pKan · pTet + pStr · pCam + pKan · pKan + pStr | compatible — 8 pairs |
Two of the ten pairs fail.
The mechanism is worth restating because it explains why the rule has that exact form. A replication controller measures its own inhibitor concentration — for ColE1 types, the antisense RNA I that pairs with the replication primer — and throttles initiation above a set point. The inhibitor is diffusible and made by every copy. Two plasmids sharing a controller are therefore counted as a sum: the cell holds the total at the set point but has no way to allocate it, so which molecule fires at any moment is a coin flip, and random partitioning at division drives one lineage to . Incompatibility is not exclusion at the door and not a fight; it is under a shared sensor.
(b) Pick at most one plasmid from each Inc group. Three groups, so:
Maximum = 3 plasmids. The number of distinct maximum sets is 2 × 2 × 1 = 4:
{pAmp, pCam, pKan} {pAmp, pStr, pKan} {pTet, pCam, pKan} {pTet, pStr, pKan}
pKan is in all four because it is alone in its group. Note that the answer is a purely combinatorial one — the graph is a union of cliques, one per Inc group, and a maximum compatible set is a transversal.
(c) Both halves of this part exist to break a size-and-similarity intuition.
99% identity is irrelevant because the test is functional, not homological. pAmp and pCam differ precisely where it counts: the control region. Family α's RNA I does not pair with family β's primer and vice versa, so each plasmid's controller sees only its own copies and holds only its own set point. Two independent controllers, two independently maintained replicons, stable coexistence — even though 99% of their sequence is shared. Run the argument the other way: pKan shares essentially no sequence with pAmp, and they also coexist, for the same reason. Sequence identity is neither necessary nor sufficient.
This is why Inc typing is done by experiment: introduce one plasmid into a strain carrying the other, then grow without selection for ~20 generations and ask whether the resident is lost. The scheme classified plasmids by behaviour long before anyone could sequence them, and it survives sequencing because behaviour is what matters clinically.
Copy number tracks the controller, not the size. pTet is 5.4 times the size of pAmp and sits at ~19 copies against pAmp's ~20 — indistinguishable, because they share family α's set point. Meanwhile pKan is 19 times the size of pAmp and sits at 1 copy, but that is because its iteron controller has a different set point, not because it is large. If you were handed a sixth plasmid of 30 kb and asked to predict its copy number, the honest answer is that the size tells you nothing; you need to know which control system it carries.
(d) Selecting for one determinant selects for all of them, and the accounting is not vertical.
The four determinants sit on one replicon. They are not merely linked — they are co-replicated. Kanamycin kills cells that lack pKan, so every survivor carries the whole plasmid, and the three unused determinants ride to fixation alongside the one under selection. There is no recombinational route by which a cell keeps Kan^R and discards the others, because losing them means losing the replicon.
The cross-species part follows from pKan being conjugative and broad-host-range. A resistance determinant travels as cargo on a self-transmissible plasmid over genus boundaries, so a Klebsiella isolate can acquire resistance to a carbapenem no one ever gave it. "Never exposed to the drug" is not evidence against resistance — it is the expected situation when the determinant travels on a replicon rather than in a lineage. Resistance frequency is the frequency of a replicon in a community, not the frequency of an in a , and treating it as the latter is the modelling error behind a public-health problem the GRAM analysis puts at 1.14 million deaths attributable to bacterial antimicrobial resistance in 2021.
Generalise. Every property in this problem — copy number, incompatibility, co-selection, host range — is a consequence of one definition: a plasmid is a replicon, a molecule with its own origin and its own control over how often that origin fires. Start from the controller and the rest is derivation.
7. Perturbing the switch ★
The λ lysis/lysogeny decision is implemented by cI and Cro competing for three operators, OR1, OR2 and OR3, between the divergent PRM (makes cI) and PR (makes Cro). Three features matter: mutual repression (cI at OR1/OR2 blocks PR; Cro at OR3 blocks PRM), positive autoregulation (the cI dimer on OR2 recruits to PRM), and negative autoregulation at the top (very high cI also fills OR3).
For each mutant, say which way the switch goes and which circuit feature was removed.
(a) A cI protein that binds OR1 and OR2 normally but cannot bind OR3 at any concentration. (b) A cI protein that binds all three operators normally but whose OR2-bound dimer can no longer contact RNA polymerase. (c) A phage with no functional Cro. What plaque does it make, and what happens if you try to induce a lysogen of it with UV? (d) A lysogen of wild-type λ, treated with mitomycin C (a DNA-damaging agent) — and the same experiment in a recA⁻ host. (e) Spontaneous induction runs at about 10⁻⁵ per cell per generation. In a culture of 10⁹ lysogens with a burst size of ~100, how much free phage appears per generation? What fraction of lineages have induced after 1,000 generations, and how many generations until half have? At a 20-minute generation, how long is that in real time?
Solution
The goal. Read a circuit diagram as a dynamical system rather than a wiring list, and see that "bistable" is a claim with quantitative consequences.
(a) Lysogeny, and an unusually deep one. Negative autoregulation is removed.
Binding to OR1 and OR2 is intact, so cI still shuts off PR (no Cro) and still recruits polymerase to PRM (more cI). Both arms of the lysogenic programme work, so the decision goes to lysogeny.
What is lost is the ceiling. In wild type, once cI gets very high it also occupies OR3 and represses its own promoter, capping the level. Without that, cI accumulates far above the normal set point. The lysogen becomes harder to induce, because RecA-stimulated self-cleavage now has more to get through before PR opens. A circuit with a broken cap is still stable — it is stable in a way that no longer responds properly to its input.
(b) Lysogeny cannot be maintained. Positive autoregulation is removed.
This is the sharpest one. The protein still represses PR, so what remains is mutual repression alone. Ch 20A puts the consequence in one line: mutual repression alone gives a switch you can flip; mutual repression plus positive feedback on one arm gives two locally stable steady states with an unstable one between them — bistability.
Strip the feedback and cI's own promoter is no longer self-sustaining. Whatever cI is present is diluted by growth and division with no mechanism amplifying it back, so the high-cI state is not an attractor: fluctuations downward are not pulled back, they run away. The prediction is a lysogen that is unstable — it establishes, then loses repression at a high rate, and the culture throws lytic events far above the 10⁻⁵ spontaneous background.
The general lesson is the one worth keeping: "switchable" and "stable" are different properties, produced by different features of the circuit. Mutual repression buys you the first; only feedback buys you the second.
(c) Lysogeny becomes the default. The plaque is turbid or absent, and UV induction is inefficient.
Cro's job is to take OR3 and shut off PRM, which is how the lytic arm wins. Without Cro there is no way to shut down cI synthesis, so infection resolves to lysogeny at close to 100% and the phage barely lyses: an extremely turbid plaque, or no visible plaque at all. (Note this is the opposite of a clear plaque, which is what a cI⁻ mutant makes — lysogeny broken. Kaiser's 1957 clear-plaque collection is the complementary .)
UV then exposes the second half of Cro's job. RecA is activated, cI cleaves itself, PR opens — but with no Cro to occupy OR3, PRM is never shut off. As soon as the damage signal decays, cI is resynthesised and the cell can slide back into lysogeny. Cro is not just the lytic repressor; it is the latch that makes the transition irreversible. Induction without it is leaky and reversible.
(d) Wild-type host: synchronous lysis. recA⁻ host: nothing.
Mitomycin C damages DNA, which activates RecA (the SOS response, Ch 17). Activated RecA stimulates cI to cleave itself. cI collapses, Cro takes OR3, PRM stays off, the lytic programme runs, and the phage abandons a host whose genome is falling apart.
In a recA⁻ host the signal transducer is missing. The DNA is damaged just as badly, but nothing stimulates cI cleavage, so the prophage sits there and the culture does not lyse. This is the cleanest possible demonstration that the prophage does not detect damage — it subscribes to the host's damage detector. No phage-encoded sensor exists; the phage wired its escape hatch to a signal the host was already generating.
(e) A culture of 10⁹ lysogens, 10⁻⁵ induction per cell per generation, burst 100:
cells inducing per generation = 10⁹ × 10⁻⁵ = 10⁴
phage released per generation = 10⁴ × 100 = 10⁶ pfu
So a "quiescent" lysogenic culture continuously carries about 10⁶ free phage, one particle per thousand cells. Any experiment that assumes a lysogen releases nothing is wrong by six orders of magnitude in absolute terms — which is why lysogen supernatants always plate.
Fraction of lineages that have not induced after n generations is (1 − 10⁻⁵)ⁿ:
n = 1,000: (1 − 10⁻⁵)^1000 = 0.99005
so 1 − 0.99005 = 0.00995 → about 1.0% have induced
Generations until half have induced:
(1 − 10⁻⁵)ⁿ = 0.5
n = ln(0.5) / ln(1 − 10⁻⁵) = 69,314 generations
At a 20-minute generation:
69,314 × 20 min = 1,386,287 min = 963 days ≈ 2.6 years of continuous growth
Why bistability is the point. "Stable for thousands of generations" is not rhetoric — it is the 69,314 above, and it is only achievable because cI sits in an attractor. Thermal and partitioning noise perturbs cI constantly; positive autoregulation pulls those perturbations back. A merely switchable circuit, like the mutant in (b), has no restoring force, so the same noise is a random walk that escapes quickly. Bistability is what converts a decision into a memory.
And note the asymmetry, which is what separates this circuit from a thermostat: a thermostat tracks its input, while this one commits. Once Cro occupies OR3, restoring cI function does not restore cI synthesis — the trajectory has crossed the separatrix and there is no route back. That hysteresis is the same phenomenon as the escape hatch in (d), read in the other direction.
8. Deletion mapping an unknown ★★
You are mapping point mutants inside rIIA of phage T4. You hold a nested set of ten deletions that divide rIIA into eight consecutive intervals A1 … A8, left to right:
rIIA: |--A1--|--A2--|--A3--|--A4--|--A5--|--A6--|--A7--|--A8--|
D1 ############################
D2 ##############
D3 ##############
D4 ############################
D5 ##############
D6 ##############
D7 #######
D8 #######
D9 #######
D10 #######
The cross is Benzer's: coinfect the permissive host E. coli B with the point mutant and a deletion, plate the burst on E. coli K-12(λ), and ask whether any r⁺ plaques appear. This is a binary predicate, not a frequency: a lying inside a deletion can never yield a wild-type recombinant, because no wild-type sequence exists at that position on either parent.
Four mutants, none of which reverts detectably, are crossed against all ten deletions. − = no r⁺ recombinants; + = r⁺ recovered.
| D1 | D2 | D3 | D4 | D5 | D6 | D7 | D8 | D9 | D10 | |
|---|---|---|---|---|---|---|---|---|---|---|
| x1 | − | + | − | + | + | + | + | − | + | + |
| x2 | + | + | + | − | − | + | + | + | + | + |
| x3 | − | − | + | + | + | + | + | + | + | + |
| x4 | − | − | − | + | + | + | − | − | + | + |
(a) Localise x1, x2 and x3. (b) x4 is not a point mutant. Prove it from the table alone, say what it is, and give its extent. What one further experiment would confirm your answer? (c) Only three crosses per mutant are needed if you choose the deletions adaptively. Show the decision tree. Then run x4 through it and say what it returns — and what that tells you about redundancy in a search. (d) You have 2,000 rII point mutants. Compare the total number of crosses needed to map them all pairwise against the number needed if you first assign each to one of the eight intervals. Assume mutants are evenly distributed. (e) Assume the ten deletions cut rIIA into eight intervals of equal length. What resolution does interval assignment give, and what do you have to do to go finer?
Solution
The goal. Turn a mapping problem from a measurement into a sequence of yes/no containment queries — and then discover the assumption that the trick smuggles in.
(a) For a point mutant, "fails to recombine with D" means "lies inside D". So the mutation lies in the intersection of every deletion it failed against, minus every deletion it succeeded against.
x1 fails against D1, D3, D8.
D1 = {A1,A2,A3,A4} D3 = {A3,A4} D8 = {A4}
intersection = {A4}
Cross-check with the "+" rows: it recombined with D2 = {A1,A2} ✓ (A4 not in D2), with D4 = {A5..A8} ✓, and with all the rest ✓. x1 is in A4.
x2 fails against D4, D5.
D4 = {A5,A6,A7,A8} D5 = {A5,A6}
intersection = {A5,A6}
It recombined with D9 = {A6}, which removes A6. x2 is in A5.
x3 fails against D1, D2.
D1 = {A1,A2,A3,A4} D2 = {A1,A2}
intersection = {A1,A2}
It recombined with D7 = {A2}, removing A2. x3 is in A1.
(b) x4 is itself a deletion, spanning A2 through A4.
The proof needs no extra data. x4 fails to recombine with D2 and with D3. But
D2 = {A1, A2} D3 = {A3, A4} D2 ∩ D3 = ∅
D2 and D3 do not overlap. A single point cannot lie inside two disjoint intervals. The containment model is therefore violated, and since the assay is sound, the object being localised is not a point.
Now fit the alternative. A deletion fails to recombine with another deletion exactly when the two overlap (their union covers no wild-type sequence in the shared region), and recombines when they are disjoint (each parent supplies what the other lacks). Take x4 = deletion of {A2, A3, A4} and predict the whole row:
| D1 | D2 | D3 | D4 | D5 | D6 | D7 | D8 | D9 | D10 | |
|---|---|---|---|---|---|---|---|---|---|---|
| deletion of A2–A4 overlaps? | yes | yes (A2) | yes | no | no | no | yes (A2) | yes (A4) | no | no |
| predicted | − | − | − | + | + | + | − | − | + | + |
| observed | − | − | − | + | + | + | − | − | + | + |
Every cell matches. Be careful about how much the table actually pins down. It proves x4 removes at least A2, A3 and A4. It cannot decide whether x4 also removes A1, because both deletions containing A1 — D1 and D2 — already overlap x4 at A2, so they would read "−" either way. The right-hand end is pinned: x4 recombines with D4 = {A5…A8}, so it does not reach A5.
The confirming experiment: test for reversion. Plate a large number of x4 alone on K-12(λ) at high density. A point mutation reverts at some low but non-zero rate; a deletion never reverts, because there is no sequence left to restore. That is exactly the criterion Benzer used to identify his deletions in the first place, and it is a completely independent line of evidence from the recombination table.
(c) The adaptive tree.
cross vs D1 (A1–A4)
/ \
− / \ +
cross vs D2 (A1–A2) cross vs D5 (A5–A6)
/ \ / \
− / \ + − / \ +
D7 (A2) D8 (A4) D9 (A6) D10 (A8)
/ \ / \ / \ / \
− + − + − + − +
A2 A1 A4 A3 A6 A5 A8 A7
Three crosses reach any of the eight intervals — log₂ 8 = 3 — using only seven of the ten deletions. D3, D4 and D6 are never needed. That is the chapter's point about avoiding quadratic work: interval containment queried by binary search.
Now run x4 through it. D1 → −, so go left. D2 → −, so go left. D7 → −. The tree returns A2, confidently and wrongly.
This is the part worth carrying away. Binary search is correct only under its precondition — here, that the object is a point inside exactly one interval. The adaptive tree never asks a question whose answer could contradict that precondition, so it cannot detect its own failure; it just returns a plausible-looking interval. The three "redundant" deletions D3, D4 and D6 are what turn an unverifiable answer into a checkable one, and D3 is the specific cross that exposes x4.
Redundancy is not waste in a search whose model might be wrong. The extra queries buy you error detection, and a mapping panel with no redundancy will silently mis-assign every deletion in your collection.
(d) The counting.
Pairwise, as the chapter frames it:
n(n − 1)/2 = 2000 × 1999 / 2 = 1,999,000 crosses
Deletion mapping first:
assignment: 2,000 mutants × 3 crosses = 6,000
within-interval pairwise, 250 per interval:
8 × (250 × 249 / 2) = 8 × 31,125 = 249,000
─────────
total 255,000
255,000 versus 1,999,000 — a 7.8-fold saving, 1.74 million crosses avoided.
Notice where the remaining cost sits: 249,000 of the 255,000 crosses are the within-interval pairwise work, not the assignment. The assignment step is almost free. So the way to improve the scheme is not to optimise the search — it is to subdivide further, because that shrinks the quadratic term:
| Intervals | Crosses each | Assignment | Within-interval | Total | Saving |
|---|---|---|---|---|---|
| 8 | 3 | 6,000 | 249,000 | 255,000 | 7.8× |
| 16 | 4 | 8,000 | 124,000 | 132,000 | 15.1× |
| 47 | 6 | 12,000 | 41,553 | 53,553 | 37.3× |
The assignment cost grows as n log k while the residual quadratic falls as n²/k. That is the whole argument for collecting more deletions, and it is why Benzer's map got finished at all.
(e) Eight equal intervals across rIIA's 2,178 bp:
2,178 / 8 = 272 bp per interval = 272 / 106 ≈ 2.57 map units
So a single deletion assignment locates a mutation to ~272 bp — already far better than a classical gene map, and obtained with three yes/no crosses and no counting whatever.
To go finer you must switch back from the binary predicate to the frequency: cross the mutant against other point mutants in the same interval and measure recombination. The floor on that measurement is ~0.02 map units, which is
0.02 × 106 ≈ 2.1 bp
a further ~130-fold improvement, and about the base pair.
Why the floor is so low, and it is not because phage recombine unusually well. It is the selection. Plating the burst on K-12(λ) gives zero background — only r⁺ grows — so 10⁹ progeny can be screened on one dish and a single recombinant is a visible plaque. That pushes the detectable frequency to ~10⁻⁸. A Drosophila geneticist scoring 10⁴ flies bottoms out near 10⁻⁴. Benzer bought four orders of magnitude by choosing an organism where selection does the counting, and that is the same trick as the auxotroph plate in problem 4 and the transductant plate in problem 5.
Treat the bp conversion as approximate — a phage genome pairs several times per infection, so a phage map unit is not a meiotic and the 106 bp/unit factor is soft. The order of magnitude is not soft, and the conclusion stands: the gene is an interval, resolvable to about the base pair, a decade before anyone sequenced one.
Where you went wrong
| Error pattern | What to re-read |
|---|---|
| Read an absolute time of entry as a map position, or extrapolated to t = 0 and called the result the origin | Problem 1(d) — the data contain Λ − O, one number for two unknowns |
| Added cotransduction frequencies, or added (1 − C) values, as if they were distances | Problem 2(a) — C is cubic in d; only L(1 − C^(1/3)) adds |
| Reported C = 0 as a distance instead of a lower bound, or trusted a C near 1 to the base pair | Problem 2(d) — both ends of the range are dead for different reasons |
| Concluded from a complementation failure that the are at the same site | Problem 3(d) — function, recombination tests position |
| Concluded from a small that the mutations are in the same gene | Problem 3(c) — a 7.4 bp pair straddled the cistron boundary; a 116.6 bp pair did not |
| Scored a complementation test by counting colonies, and called 10⁻⁶ "weak complementation" | Problem 4(c) — complementation is a lawn; rare colonies are revertants plus recombinants |
| Omitted the reversion controls, or the cis control, and so could not exclude dominance | Problem 4(b), 4(d) |
| Told generalized from specialized transduction by comparing frequencies rather than which markers moved | Problem 5(a) — both are ~10⁻⁶; only the marker distribution separates them |
| Got the direction of the bound wrong when a model omits a lossy step | Problem 5(d) — 9.3 × 10⁻⁵ is a lower bound on transducing particles |
| Predicted plasmid incompatibility from sequence similarity, or copy number from plasmid size | Problem 6(c) — the test is functional; the controller sets both |
| Treated "never exposed to the drug" as evidence against resistance | Problem 6(d) — the determinant travels on a replicon, not in a lineage |
| Treated the λ switch as merely switchable, and so could not say why a lysogen is stable | Problem 7(b), 7(e) — feedback is what makes a decision into a memory |
| Trusted a binary search that returned a clean answer, with no redundant query to check its precondition | Problem 8(c) — the adaptive tree confidently returns A2 for a deletion |