Contents
25A — Developmental genetics and the genetics of making a mouse Part 4 — Gene regulation

25A — Developmental genetics and the genetics of making a mouse

Before this: Ch 22 · Ch 23 · Ch 24 · Ch 25 · Time: ~55 min

Chapter 25 described development as a dynamical system: motifs, attractors, gradients, a network settling into spatially organised states. That is what development is. This chapter is how any of it was found out, and how you interrogate it in an animal you cannot .

Nobody deduced the Drosophila segmentation hierarchy from the network. It was found by poisoning flies, letting the embryos die, and sorting the corpses into piles by the shape of the damage — and the tiers of the hierarchy are the piles. When the same question was put to a mouse, where a screen costs three generations and a colony, the answer was to invent a way to break one chosen gene, in one chosen tissue, at one chosen hour. That invention is the most-used genetic tool in mammalian biology, and this is the chapter that names it.

What you'll be able to do

The core idea

You have a build pipeline. There is no source, no logs, no debugger, no way to attach to a running process. You have one instrument: break a single stage, run the build, and look at the artefact that falls out.

That is genetics applied to an embryo, and it is more powerful than it sounds, because the shape of the wreckage identifies the stage. A deleting a contiguous block of body segments broke something acting over a broad region. A mutation deleting every second segment broke something with two-segment periodicity — and you know that before you know what the gene is, what it encodes, or where it sits in the genome. Classifying the damage is discovering the architecture.

The second half runs the idea backwards. Once you can name a gene you want to break it deliberately, in a mammal, without killing the animal before the interesting part happens. That means separating three things a plain fuses together: which gene, which cells, when.

A developmental hierarchy is discovered backwards, from to tier, by classifying breakage. The classes come first; the molecules come later and have to fit.


1. Two logics of development: lineage and position

A cell has to become one thing rather than another, and there are exactly two kinds of information it can use.

Lineage. Fate follows from ancestry — which cell you descended from, through which divisions. Determinants are inherited: a dividing cell partitions cytoplasmic material unequally and the daughters differ because they got different stuff.

Position. Fate follows from location. Determinants are received: a cell reads a gradient or a contact signal and picks accordingly (Ch 25 §5).

Caenorhabditis elegans is the extreme of the first. Its somatic cell lineage is invariant — the same divisions in the same order producing the same cells in every animal, so each cell has a name and a fixed . The hermaphrodite soma generates 1,090 cells, of which 131 undergo programmed cell death at reproducible positions in the lineage, leaving the 959 somatic nuclei of the adult. Those 131 deaths are scheduled, not attrition, and the genes that schedule them were found by screening for mutants in which the extra cells survived.

A vertebrate limb bud is the extreme of the second: no invariant lineage, cells allocated to skeletal elements by position in a field, the field scaling with the size of the bud.

The experiment that distinguishes them is to kill one cell with a laser and watch the neighbours.

Lineage (mosaic)Positional (regulative)
Fate set byAncestry, via inherited determinantsLocation, via received signals
Ablate a precursorIts descendants are simply missingNeighbours re-pattern and fill the gap
Transplant elsewhereKeeps its original fateAdopts the fate of the new position
Scales with organ sizeNo — cell number is fixedYes — a bigger field gets more cells per band
Error correctionNone. A lost cell is a lost structureSubstantial. The field re-reads its coordinates
Pays for it withBrittleness, fixed body sizeImprecision, needing downstream sharpening

Neither organism is pure, and the exception is the useful part. The canonical lineage animal uses induction: six vulval precursor cells sit in the worm's ventral epidermis and the anchor cell in the overlying gonad signals to them. Kimble ablated the anchor cell before the L3 stage and all six precursors took the default, non-vulval fate. Conversely, vertebrate embryos contain lineage-restricted compartments whose cells never mix.

One caution before the dispatch-table analogy takes hold: no cell knows its lineage. There is no address register and nothing reads a pedigree. "Invariant lineage" describes the output; the mechanism is entirely local — asymmetric partitioning at each division plus short-range signalling between the products. The determinism is emergent, not implemented — the source-code analogy from Ch 00 failing in its sharpest form.

2. The genetic dissection of a body plan

The Drosophila segmentation hierarchy is the canonical genetic pathway found by screening, and it is worth doing as genetics rather than as a molecular recap.

The screen

Christiane Nüsslein-Volhard and Eric Wieschaus ran it at EMBL Heidelberg over 1979–80. By their own later account they established about 27,000 inbred lines carrying an estimated 18,000 independently induced lethal , in three separate — X, second and third — because a lethal must be homozygosed to be seen and each chromosome needs its own balancer stock: a chromosome carrying that suppress recovery of crossovers, plus a recessive lethal and a marker, so a lethal mutation is maintained indefinitely as a heterozygous stock (Ch 37 §4, forward).

Two design decisions did most of the work. The readout was the cuticle of the dead embryo — a patterned external shell with denticle belts in a stereotyped arrangement, which survives the animal's death, mounts flat, and is scored in seconds. Durable, cheap, and — the part that matters — high-dimensional: it does not report "abnormal", it reports precisely which piece of the pattern is missing and where. And they screened for embryonic lethals deliberately, because anything patterning the body plan kills the embryo when broken; scoring viable adults would have discarded the whole class.

The 1980 Nature paper reported 15 loci whose mutation alters the segmental pattern, with an explicit saturation claim: these probably represented the majority of such genes in the genome.

The classification came before the molecules

They sorted the mutants by the geometry of the missing pattern, and got three classes:

ClassWhat the cuticle looks likeWhat the geometry impliesExamples named in 1980
GapA contiguous block of adjacent segments missingActs over a broad region — one domain, no periodicityKrüppel, knirps
Pair-ruleAlternate segments, or part of every alternate segment, missingMust be expressed with two-segment periodicityeven-skipped, odd-skipped, paired, runt
Segment polarityPart of every segment missing, remainder mirror-duplicatedOne-segment periodicity, and distinguishes front from back within each segmentgooseberry, patched

Read the middle column again. The periodicity of the defect predicts the periodicity of the gene's expression — a molecular prediction derived from the shape of a dead larva, years before any of these genes was cloned. When even-skipped was eventually stained it was in seven stripes. The screen had already said it would be.

graph TD S["EMS mutagenesis<br/>~27,000 balanced lines"] --> P["score cuticle of<br/>dead embryos"] P --> C{"geometry of the<br/>missing pattern"} C -->|"contiguous block"| G["<b>gap</b><br/>broad aperiodic domains"] C -->|"every 2nd segment"| PR["<b>pair-rule</b><br/>2-segment periodicity"] C -->|"part of every segment"| SP["<b>segment polarity</b><br/>1-segment periodicity<br/>+ within-segment axis"] G --> H["double mutants and<br/>cross-tier expression tests"] PR --> H SP --> H H --> R["<b>tiers, ordered</b><br/>each reads the one above,<br/>never the reverse"]

Ordering the tiers

Three classes are three piles, not a hierarchy. The order comes from asymmetric dependence: break tier k and everything downstream is corrupted while everything upstream is untouched. Remove bicoid and the gap-gene domains are lost or displaced; remove a gap gene and the pair-rule stripes shift; remove a pair-rule gene and segment-polarity stripes fail in alternate positions. None of it runs backwards. A staged build, where the direction of the arrow is the direction corruption propagates.

The maternal tier, and why it costs an extra generation

Above the gap genes sit the maternal-effect genes — bicoid, nanos and relatives — whose products the mother deposits in the egg. Their genetic signature is diagnostic: the embryo's phenotype is set by its mother's , not its own (Ch 11; Ch 25 §4). A homozygous mutant mother produces uniformly defective embryos whatever she is crossed to; her heterozygous sister's homozygous mutant offspring develop normally. The screen therefore shifts by a generation — you score the offspring of homozygous mutant mothers — and in a the inheritance looks lagged.

3. Hox genes as a genetic phenomenon

Chapter 25 §6 covers colinearity and the selector concept. Two points belong here because they are genetics rather than regulation.

Homeosis separates "build a structure" from "which structure". Ed Lewis's analysis of the bithorax complex (Nature 276:565, 1978) showed that removing its function transforms the third thoracic segment into a copy of the second — halteres become wings, giving a four-winged fly. Nothing novel was constructed; a segment that was going to be built anyway was built with the wrong identity. Identity is a separable variable, assigned by a gene on top of a structural programme that exists independently — a strong architectural claim, obtained entirely from mutants.

The sign of the transformation tells you what a selector does. For the pattern is consistent: transforms a posterior structure toward a more anterior identity (anterior is the default), while ectopic expression transforms anterior toward posterior. Where two Hox genes are co-expressed the more posterior usually prevails — visible genetically as between Hox genes, and one reason a single loss can be silent.

In mammals, redundancy blunts all of it. Single mouse Hox give partial, variably penetrant vertebral transformations; Hoxb4 nulls shift the second cervical vertebra part-way toward the first. Compound mutants for the paralogous Hoxa4, Hoxb4 and Hoxd4 transform more completely, with severity rising as mutant copies accumulate — the four vertebrate clusters came from two rounds of whole-genome duplication (Ch 35, forward), and the cover for each other. This is the first appearance of a problem §7 meets head on: a mammalian single-gene knockout systematically understates what the gene does.

4. Epistasis: ordering a pathway with a double mutant

Single mutants tell you a gene is required. They cannot tell you the order: two genes both needed for a structure, both mutants lacking it, identical information either way.

Double mutants can order them, and the reason is almost embarrassing once written down.

Derivation

Model a linear regulatory pathway as a composition of functions on a state:

   output  =  f_n ( … f_2 ( f_1 ( input ) ) … )

A null of gene i replaces f_i with the constant function OFF; a constitutive allele replaces it with the constant function ON. Both are the same operation: a clamp — replacing a function by a constant that discards its argument.

Compose two clamps at positions j < k, so j is upstream. Since f_k ignores its argument, nothing f_j produced can reach the output. The double mutant's is set by the clamp with the largest index.

   Switch regulatory pathway:  phenotype of double = clamp at argmax(index)
                               ⟹  the EPISTATIC mutation is the DOWNSTREAM one

That is exact, and it is nothing but "a constant function ignores its input". It also derives, for free, the condition everyone forgets: if both clamps have the same value, both orderings predict the same double-mutant phenotype and you learn nothing. The two single mutants must have opposite phenotypes.

The rule inverts

Take a substrate-dependent pathway — a biosynthetic chain through which material flows:

   precursor  ──(A)──>  intermediate  ──(B)──>  product

with the observable being which compound accumulates. Block A: precursor accumulates. Block B: intermediate accumulates. Block both: precursor accumulates, because nothing gets past the earliest block.

   Substrate-dependent pathway:  phenotype of double = block at argmin(index)
                                 ⟹  the EPISTATIC mutation is the UPSTREAM one

Same experiment, opposite reading. Avery and Wasserman set this out explicitly in 1992, and it remains the commonest error in reading a double mutant. Switch pathways are argmax; substrate pathways are argmin. Decide which you are in before converting "epistatic" into "upstream" or "downstream" — and note that the type is a claim about what your assay measures, not about the molecules. For a programmer: reading the value a pipeline returned, the last stage to write a constant wins; asking how far an item got down a conveyor, the earliest jam wins. Same pipeline, opposite answers.

The four conditions

  1. Both alleles must be null or fully constitutive. A hypomorph is a partial clamp, not a constant function, and the derivation collapses. Test independently — a true null is no worse opposite a chromosomal deletion of the locus than opposite itself.
  2. The single mutants must have opposite, distinguishable phenotypes. Derived above.
  3. The genes must act in the same pathway. Genes in parallel pathways give a double mutant merely worse than either — a genetic interaction, not an ordering.
  4. You must know the pathway type, because the rule inverts.

Two standing cautions. orders genes in an information-flow sense and says nothing about physical contact (Ch 11). And in a chain containing a , a null of the repressor behaves like a gain-of-function of its target, so reason from the wiring rather than reciting the slogan — the worked example runs straight into this.

Ch 37 §11 (forward) generalises this from a qualitative ordering to a quantitative interaction measured across millions of double mutants, and to as a drug-target strategy. All of it rests on the logic derived here.

5. Mammalian development, in the amount you need

graph TD Z["zygote"] --> M["morula"] M --> B["<b>blastocyst</b> (mouse ~E3.5)"] B --> T["trophectoderm — outside<br/><i>→ placenta</i>"] B --> I["inner cell mass — inside<br/><i>→ embryo proper</i><br/><b>source of ES cells (§6)</b>"] I --> E["epiblast"] E --> PS["<b>primitive streak</b><br/>mouse ~E6.5 · human ~day 14–16<br/><i>gastrulation</i>"] PS --> EC["<b>ectoderm</b>"] PS --> ME["<b>mesoderm</b>"] PS --> EN["<b>endoderm</b>"] EC --> NP["neural plate → nervous system"] EC --> EP["epidermis"] NP --> NC["<b>neural crest</b><br/>border of plate and epidermis;<br/>delaminates and migrates"]

The first fate decision in a mammal is positional. Cells on the outside of the early embryo become trophectoderm and make placenta; cells on the inside become inner cell mass and make the animal. Inside-versus-outside is a coordinate, and the decision is read off it.

Gastrulation converts a single sheet of pluripotent cells into three layers. A furrow — the primitive streak — forms at the posterior of the epiblast (mouse around embryonic day 6.5, human around days 14–16). Epiblast cells move through it, lose their epithelial character, and are allocated to one of three germ layers:

Germ layerPrincipal derivatives
EctodermEpidermis and its appendages; the entire nervous system
MesodermHeart, blood, bone, skeletal muscle, kidney, connective tissue
EndodermThe gut tube and its outgrowths — lungs, liver, pancreas, thyroid

The neural crest, sometimes called a fourth germ layer, arises at the border between neural plate and epidermal ectoderm, detaches, and migrates throughout the embryo. Its derivatives read like a list assembled at random: neurons and glia of the peripheral nervous system, including the enteric nervous system; melanocytes, the pigment cells of skin, hair and inner ear; most of the craniofacial cartilage and bone; the adrenal medulla and the smooth muscle of the great vessels.

A syndrome is often a lineage, not an organ. Waardenburg syndrome combines congenital deafness with patchy depigmentation of skin, hair and iris; the type 4 form (Shah–Waardenburg) adds Hirschsprung disease — an aganglionic distal colon that obstructs because crest cells never arrived to build its nerve plexus. Ear, skin and colon share no anatomy. They share an ancestor, and once you know the derivative list the syndrome becomes a prediction (Ch 54, forward).

One precision point: germ layers label ancestry, not tissue type. Cranial neural crest makes bone and cartilage — tissues normally called mesodermal — while being ectodermal in origin.

6. Gene targeting: an experimental grammar

You have a candidate gene, from a fly or a human , and you want to know what it does in a mammal. Forward screening is unaffordable: recovering in mouse takes three generations and roughly thirty weeks per round (Ch 37 §4, forward). You must go the other way, from gene to , by breaking a chosen locus on purpose.

The obstacle. In budding yeast, is efficient enough that a product with 40 bp of at each end replaces an entire . Mammalian cells do not cooperate. When Smithies and colleagues first targeted a specific human chromosomal locus — the β-globin gene, in 1985 — the planned modification occurred in roughly one per thousand transformed cells; introduced DNA overwhelmingly integrates at random, intact, somewhere else. Two things had to be invented: a cell worth targeting, and a way to find one correct event among a thousand wrong ones.

Ingredient one — a cell that can still build an animal. Evans and Kaufman (Nature 292:154, 1981) and, independently, Martin (PNAS 78:7634, 1981) derived permanent embryonic stem (ES) cell lines from the inner cell mass of mouse blastocysts. What matters is not that they grow in culture but that injected back into a host blastocyst they contribute to every tissue of the animal, including the germ line. A change made in a dish can be transmitted to offspring.

Ingredient two — positive–negative selection. Mansour, Thomas and Capecchi (Nature 336:348, 1988) solved the needle-in-a-haystack problem with a construct whose geometry does the work.

   TARGETING VECTOR (linear)

   ┌──────────────────┬────────┬──────────────────┬──────────┐
   │  5′ homology arm │  neo   │  3′ homology arm │  HSV-tk  │
   └──────────────────┴────────┴──────────────────┴──────────┘
    └──── copied in by homologous recombination ────┘   ↑ outside the arms

   CORRECT TARGETING (homologous recombination)
       gains neo, does NOT gain HSV-tk   → G418 resistant, ganciclovir resistant  ✓ survives
   RANDOM INTEGRATION (end joining)
       whole linear molecule inserts     → G418 resistant, ganciclovir SENSITIVE  ✗ killed
   NO UPTAKE
                                         → G418 sensitive                          ✗ killed

The logic is a set difference implemented in chemistry — {took up DNA} \ {integrated randomly} — and each half follows from where the marker sits:

Reported enrichment: about 2,000-fold, and independent of whether the target gene is expressed in ES cells — the clause that made the method general rather than a trick for selectable loci. Enrichment is not proof: confirm by Southern blot — size-fractionate digested genomic DNA on a gel, transfer it to a membrane and probe it — or by long-range PCR with one primer outside each arm. That assay, and the electroporation step in the diagram below (a pulsed electric field that makes the membrane transiently permeable to DNA), are laboratory hardware from Ch 36 (forward).

graph LR V["targeting vector<br/>arms + neo + tk"] --> EP["electroporate<br/>into ES cells"] EP --> SEL["G418 + ganciclovir<br/><i>~2,000× enrichment</i>"] SEL --> CONF["Southern / long-range PCR<br/>primer outside each arm"] CONF --> INJ["inject into host<br/>blastocyst"] INJ --> CHI["<b>chimera</b><br/>patchwork coat"] CHI --> CROSS["cross to wild type"] CROSS --> GERM["<b>germline transmission</b><br/>coat colour of pups = assay"] GERM --> HOM["heterozygotes → intercross<br/>→ homozygous null"]

The coat-colour step is the elegant part. Use ES cells from an agouti strain and host blastocysts from a black strain: a chimeric pup has a patchwork coat whose agouti fraction estimates the ES contribution, and because agouti is , an agouti pup from a chimera × black cross can only have come from an ES-cell-derived gamete. A free -transmission assay, readable at weaning. Three groups reported germline transmission of targeted in 1989 — Thompson et al. in Cell, Koller et al. in PNAS, Zijlstra et al. in Nature; Capecchi, Evans and Smithies shared the 2007 Nobel Prize for the principles behind it.

Knockout, knock-in: same grammar, different payload

The homology arms are an address; what sits between them is a payload.

Payload between the armsWhat you get
Selection cassette replacing essential exons — a null allele
A single altered Knock-in point mutation, modelling a human variant in its own locus
A fluorescent protein in frame at the start codonReporter under the gene's own regulation — not a transgene's guess at it
A humanised exon or whole locusHuman protein sequence in mouse physiology
A recombinase, e.g. CreA driver line expressing Cre in exactly that gene's cells (§7)
An essential exon flanked by loxP sitesA conditional allele — silent until triggered (§7)

write(address, payload). Knockout was never a separate technique from knock-in.

7. Conditional and inducible alleles: the null is embryonic lethal, now what?

How often this happens. Not occasionally. From its first 1,751 unique gene , the International Mouse Phenotyping Consortium reported 410 lethal lines and 198 subviable ones — about 23% and 11%, so roughly a third of null never produce a healthy adult homozygote (Dickinson et al., Nature 537:508, 2016). For those genes a straight knockout answers "is it essential?" and then stops.

Two distinct failure modes need different fixes. If the animal dies before the tissue of interest exists, you must restrict in space — delete in one lineage only. If the animal survives but has compensated during development, giving a null with a suspiciously mild , you must restrict in time — delete acutely, in an animal that developed normally. The second is subtler and commoner than expected: a viable stable mutant has had all of development to reroute around the loss, and has been selected for tolerating it. Acute removal has not.

Cre-lox

Cre is a 38 kDa site-specific recombinase from bacteriophage P1 (Sternberg and Hamilton, 1981). Its target, loxP, is a 34 bp site: two 13 bp inverted repeats flanking an 8 bp asymmetric spacer. The asymmetry is the whole design — it gives the site a direction, and the relative orientation of two sites decides the outcome.

   ──▶─────────────▶──      same orientation      →  EXCISION   (intervening DNA deleted)
   ──▶─────────────◀──      opposite orientation  →  INVERSION  (intervening DNA flipped)

The architectural move is to separate the allele from the trigger and put them in different animals.

Cross them, and the exon is deleted only in cells that have expressed Cre — permanently, and in every descendant.

The founding experiment is exactly the "now what" case. Gu, Marth, Orban, Mossmann and Rajewsky (Science 265:103, 1994) showed that deletion of the β promoter and first exon is lethal, then used a Cre transgene to make the identical deletion only in T cells. Those mice lived, and the requirement could finally be examined.

Inducibility adds the time axis. Fuse Cre to a mutated ligand-binding domain of the oestrogen receptor (CreER, and the more sensitive CreERT2, carrying G400V/M543A/L544A). The fusion sits in the cytoplasm and cannot reach DNA until ligand binds, and the make it respond to the drug 4-hydroxytamoxifen while ignoring endogenous oestrogen. Space from the promoter, time from the drug.

FLP-FRT, from the yeast 2-micron plasmid, is a second orthogonal system with the same architecture (FRT is likewise 13 bp repeats around an 8 bp spacer), so two independent operations can run in one animal — classically FLP to remove the selection cassette, Cre to delete the exon later.

The feature-flag analogy, and the four places it breaks

A floxed allele really is a feature : the code path exists, and a separate trigger decides whether it is taken, scoped by lineage and time. The divergences are exactly where experiments go wrong.

The same trick, read forwards: lineage tracing

Change what sits downstream of the loxP sites and the tool changes character. Put a transcriptional stop cassette flanked by loxP sites in front of a reporter at a ubiquitously expressed locus — conventionally Rosa26 (Soriano, Nature Genetics 21:70, 1999). The reporter is silent; Cre excises the stop and it switches on permanently, in that cell and every descendant, whether or not Cre is ever active again. Cre is no longer a deletion tool but an irreversible, heritable label applied to a chosen population at a chosen moment.

Barker and colleagues (Nature 449:1003, 2007) knocked EGFP-IRES-CreERT2 into the Lgr5 locus and crossed it to a Rosa26-lacZ reporter. One pulse of tamoxifen labels scattered single cells at the base of intestinal crypts; weeks later the label runs in continuous ribbons from crypt base to villus tip, contains every differentiated cell type, and persists for life. Self-renewal and multipotency — the definition of a stem cell — demonstrated rather than asserted, from a marker gene and two alleles. The multicolour version (R26R-Confetti, whose Brainbow cassette picks one of four fluorescent proteins at random) makes neighbouring clones distinguishable, which is what showed crypt maintenance to be neutral competition between equivalent stem cells rather than an ordered hierarchy (Snippert et al., Cell 143:134, 2010).

For a programmer: git blame for cells — an irreversible tag written at a chosen commit and inherited by everything downstream. Ch 48 (forward) does the same reconstruction from endogenous , with no engineering at all.

8. Targeting versus editing (forward-looking)

It is easy to conclude that made everything above obsolete, and that conclusion is wrong in a specific and useful way.

Gene targeting (this chapter)Genome editing (Ch 38, forward)
How the address is specifiedKilobase-scale arms20-nt guide RNA plus a PAM
What determines the final sequenceYour vector, copied in by The cell's choice of repair pathway
Where it worksSpecies with -competent ES lines — mouse, and with effort ratAny species, any cell you can deliver to
Time to a founder animalRoughly a year: clones, screening, chimeras, breedingWeeks, by direct injection into zygotes
Precision of payloadArbitrary, kilobase-scale, exact easy; precise edits hard, and very hard in non-dividing cells
What it buysControl over what, and via Cre-lox over when and whereSpeed, and reach into any organism

Editing replaced the addressing step, not the grammar. Floxed exons, tissue-specific Cre drivers, lox-stop-lox reporters and drug-inducible recombinases are unchanged, remain the standard way to ask a spatially and temporally restricted question in a mammal, and are now routinely built with CRISPR — because CRISPR makes inserting the loxP sites cheap. The conditional outlived the technology that produced it, which is the usual fate of a good abstraction.


Common misconceptions

What people thinkWhat's actually true
tells you which gene acts firstIt tells you which single mutant the double resembles. In a switch pathway that gene is downstream (the last clamp wins); in a substrate pathway it is upstream (the earliest block wins). Classify the assay before applying the rule
Any two mutants can be ordered by building the doubleOnly with opposite, distinguishable , null-or-constitutive , and genes in the same pathway. Two Vulvaless mutants give a Vulvaless double and exactly zero information — which follows from the derivation, not from experience
C. elegans proves fate is set by lineage; vertebrates prove it is set by positionBoth use both. Ablate the worm's anchor cell and all six vulval precursors take the default fate — induction, in the canonical lineage animal. And no cell anywhere reads its own : invariance is the output of purely local asymmetric divisions
The segmentation gene classes were defined by what the genes do molecularlyThey were defined by the geometry of a cuticle defect in dead embryos, years before any was cloned. The two-segment periodicity of the pair-rule phenotype predicted the seven stripes; it was not derived from them
A mouse tells you what the gene doesAbout a third of null lines never yield a healthy adult homozygote (410 lethal, 198 subviable of the first 1,751 IMPC lines), and survivors had all of development to compensate. You measure an animal built without the gene, which is not the consequence of losing it
Cre-lox makes a tissue null for the geneRecombination is incomplete and varies by locus, cell type and driver. The tissue is a mosaic, and unrecombined cells frequently have a growth advantage and repopulate it. "Conditional knockout" names an intention, not a
A floxed allele is a knockout alleleIt is engineered to be functionally wild type until Cre arrives — the loxP sites sit in . Homozygous floxed mice without Cre are a required control, and a phenotype in them invalidates the line
made gene targeting obsoleteIt replaced the addressing step. Conditional alleles, Cre drivers, lox-stop-lox reporters and inducible recombinases are unchanged, and are now built with CRISPR. Editing buys speed; targeting's grammar buys control over when and where

Worked example: ordering the vulval induction pathway from double mutants

The goal. Given a set of C. elegans mutants and no molecular information at all, produce a linear order of gene action.

The system. Six vulval precursor cells, P3.p–P8.p, sit in a row in the ventral epidermis. The anchor cell signals from the overlying gonad; the three nearest generate the vulva and the other three take the default fate and fuse with the epidermis. Two , and they are opposite — which is what makes the analysis possible. Vulvaless (Vul): no vulva, so the animal cannot lay eggs, progeny hatch internally and it becomes a "bag of worms" — scoreable by eye on a plate. Multivulva (Muv): extra ventral protrusions, where precursors that should have taken the default fate made vulval tissue instead.

GenePhenotype
lin-3Vul
let-23loss of functionVul
let-60loss of functionVul
let-60Muv
lin-45loss of functionVul
lin-1loss of functionMuv
lin-15loss of functionMuv

Step 1 — establish this is induction at all. Kimble's laser ablation (1981): destroy the anchor cell before the L3 stage and all six precursors take the default fate. The instruction comes from outside the lineage that receives it.

Step 2 — the wrong turn, and it costs a generation. The obvious first cross is let-23(lf) × lin-45(lf): two genes both required, order unknown. Build the double. It is Vul, and you have learned nothing. Both clamp the output to the same value, so both orderings predict the same phenotype and the experiment cannot discriminate. Not bad luck — §4 derived it. analysis requires opposite phenotypes, and finding that out by building the cross costs weeks.

Step 3 — the informative crosses. Pair each Vul mutant with a Muv mutant.

   let-23(lf) ; let-60(gf)   →  Muv    let-60 epistatic  →  let-60 DOWNSTREAM of let-23
   lin-3(lf)  ; let-60(gf)   →  Muv    let-60 epistatic  →  let-60 DOWNSTREAM of lin-3
   let-60(lf) ; lin-1(lf)    →  Muv    lin-1  epistatic  →  lin-1  DOWNSTREAM of let-60
   lin-45(lf) ; lin-1(lf)    →  Muv    lin-1  epistatic  →  lin-1  DOWNSTREAM of lin-45
   let-23(lf) ; lin-15(lf)   →  Vul    let-23 epistatic  →  let-23 DOWNSTREAM of lin-15
   lin-15  ──⊣  [ lin-3  →  let-23  →  let-60  →  lin-45 ]  ──⊣  lin-1  ──→  vulval fate

Step 4 — the rule gives position, not sign. lin-1 loss of function is Muv: removing lin-1 produces more vulva, so wild-type LIN-1 prevents vulval fate. It is a , and the pathway works by relieving its repression. Nothing in "the epistatic mutant is downstream" told you that. Position comes from the double mutants; the sign of each edge comes from the single-mutant phenotype taken alone. Fuse the two inferences and you draw a pathway of activators that is topologically right and mechanistically backwards. Same at the other end: lin-15(lf) is Muv, so LIN-15 is a negative regulator too, acting above the receptor.

Step 5 — verify the pathway type. Everything above assumed a switch regulatory pathway, where the observable is the state of the final decision. Run the same experiment on a two-step biosynthesis scored by which compound accumulates and a⁻ b⁻ accumulates the precursor, so the upstream mutation is epistatic — opposite answer, same experiment (§4). Classify the assay before converting "epistatic" into a direction.

Step 6 — audit the alleles. Every inference assumed let-60(gf) is genuinely constitutive and let-23(lf) genuinely null. A hypomorphic let-23 leaks signal, the double comes out partially Muv, and the ordering becomes uninterpretable rather than merely wrong. Check by placing each allele opposite a chromosomal deletion of its locus.

What this bought. The chain reconstructed above is, molecularly, EGF → EGF receptor → Ras → Raf → MEK → ERK → an Ets-family repressor; let-60 encodes a Ras protein. It is conserved intact in humans and among the most frequently mutated pathways in cancer (Ch 56, forward) — ordered correctly, edge by edge, by counting bumps on the underside of a worm.

What generalises. (i) Find mutants with opposite phenotypes; same-phenotype pairs are uninformative by construction. (ii) Build the double. (iii) The epistatic mutation is the one the double resembles. (iv) Decide switch-versus-substrate before converting "epistatic" into "upstream" or "downstream". (v) Take signs from single mutants, never from the ordering rule. (vi) Confirm the alleles are null or constitutive. It is bisection over a pipeline — with the that you cannot step through it, only clamp one stage and read the final output.

Connections