Contents
36 — Core molecular methods Part 8 — Methods

36 — Core molecular methods

Before this: Ch 01 · Ch 02 · Ch 08 · Time: ~45 min

You will read papers whose methods sections assume this material. This chapter gives you the logic of each technique — the problem it solves and the physical fact it exploits — not the protocol. Protocols expire. The reason a method works does not.

What you'll be able to do

The core idea

Two properties make DNA and RNA unreasonably easy to work with, and proteins have neither.

A constant charge-to-mass ratio. Every nucleotide contributes one negative phosphate and roughly the same mass (Ch 01 §3). So electrophoretic mobility in a sieving matrix is a function of length alone, independent of sequence. One physical measurement, no per-molecule calibration, works for any DNA anyone has ever handled.

Programmable self-recognition. A sequence binds its complement and nothing else. That means you can address any target you like with a synthetic oligonucleotide — you specify the query as a string, order it, and it arrives. Detection, amplification, and manipulation all reduce to writing down a sequence.

Proteins have neither. Their charge depends on composition and pH; they have no complementary partner you can synthesise; and there is no protein . Every protein method therefore needs a bespoke physical reagent — an antibody raised in an animal — or a mass spectrometer. That one asymmetry explains why genomics scaled by six orders of magnitude and proteomics did not, and it is the thread running through everything below.


1. Getting the molecules out

The problem: a cell is a crowded mixture, and you need one class of molecule, intact and free of anything that inhibits enzymes downstream.

Break the cell open, denature and strip away protein, then exploit a chemical difference. The implementation binds nucleic acid to silica in high chaotropic salt (which strips the water shell) and elutes it in low salt; magnetic-bead versions of the same chemistry are what sits inside automated extractors. To get RNA rather than DNA, add DNase; the reverse needs RNase.

Three facts that propagate downstream:

2. Electrophoresis: the free gift of constant charge

Apply a field; DNA (polyanionic) migrates toward the anode. In free solution it does not separate at all: force scales with charge ∝ length, drag scales with length, so mobility is length-independent. Separation requires a sieving matrix, in which longer molecules are retarded more. Empirically, migration distance is approximately linear in −log(length) across the useful range of any given matrix.

      wells  ┌───┬───┬───┬───┐
             │ L │ A │ B │ C │        L = ladder, known lengths
   10 kb ────│ ▄ │   │   │ ▄ │
    3 kb ────│ ▄ │ ▄ │   │   │        A: single 3 kb product — clean PCR
    1 kb ────│ ▄ │   │ ▄ │ ▄ │        B: 1 kb only — wrong product or wrong template
   500 bp────│ ▄ │   │   │   │        C: two bands — non-specific amplification
   100 bp────│ ▄ │   │ ▄ │ ▄ │  ← primer-dimer: short, amplifies efficiently
             └───┴───┴───┴───┘
                    ▼ field
AgarosePolyacrylamide
Matrixpolysaccharide, large porescross-linked polymer, small tunable pores
Useful range~100 bp – 25 kb~5 – 500 bp
Resolutiona few percent of lengthsingle base
Typical useplasmid checks, products, digestsoligo QC, footprinting, protein (SDS-PAGE)

Two extensions worth knowing. Above roughly 20–50 kb, long molecules reptate end-on through the pores and mobility becomes length-independent again; pulsed-field electrophoresis restores separation by periodically switching field direction, forcing the molecule to reorient — a step whose cost scales with length. And capillary electrophoresis is the same physics in a polymer-filled capillary at single-base resolution: it is the readout of (§7) and of fragment-length assays generally.

Contrast with protein. Protein charge depends on amino acid composition and pH, so raw electrophoresis separates by an uninterpretable mixture of charge and shape. SDS-PAGE works by coating the protein in the detergent SDS at a roughly fixed mass ratio, manufacturing a constant charge-to-mass ratio. Protein methods spend effort recreating what DNA has for free.

3. PCR: exponential amplification of an addressed region

The problem: you hold one copy of a 500 bp locus embedded in 3.1 Gb of irrelevant sequence, and you need ~10¹² copies of that locus and nothing else.

Two primers — short synthetic oligonucleotides complementary to the flanks, pointing inward — plus a polymerase, plus thermal cycling:

95 °C  denature  ──►  strands separate
55–65 °C anneal  ──►  primers find their complements (and only those, if Tm is right)
72 °C  extend    ──►  polymerase copies from each primer's 3' end
                      repeat 25–40×

The enabling idea is thermostability: use a polymerase from a thermophile (canonically Taq, from Thermus aquaticus) whose activity survives the 95 °C denaturation step instead of being destroyed by it — Taq has a half-life of tens of minutes at 95 °C and extends at 72 °C. Without it, every cycle destroys the enzyme and you must add fresh enzyme by hand. With it, the process is a machine loop.

Why the product has defined ends

Track single strands. Let T = original (ends undefined), M = strands with a primer-defined 5' end but a run-off 3' end, S = strands defined at both ends. Each cycle, every strand present templates exactly one new strand: copying T yields M; copying M or S yields S.

T_n = 2                      (constant)
M_n = M_{n-1} + 2            →  M_n = 2n
S_n = 2·S_{n-1} + M_{n-1}    →  S_n = 2^{n+1} − 2n − 2

Check n = 3: 2 + 6 + 8 = 16 = 2⁴ strands ✓. The first double-defined strand appears at cycle 2 (S₂ = 2), and the first duplex made of two of them — the amplicon you would see on a gel — at cycle 3. By cycle 30 the undefined-end species are 62 strands out of 2³¹, i.e. a fraction of 3 × 10⁻⁸. The gel band is essentially pure S.

Efficiency and the plateau

With per-cycle efficiency E (0 ≤ E ≤ 1), copies after n cycles are

N_n = N_0 (1 + E)^n

Perfect doubling is E = 1. Real reactions run 0.9–1.0 early, then plateau. Causes, roughly in order of importance: at high product concentration the two product strands reanneal to each other faster than primers can find them; primers and dNTPs deplete; polymerase accumulates thermal damage.

The plateau has a consequence people repeatedly forget: endpoint yield converges regardless of starting amount. A reaction seeded with 10 copies and one seeded with 10⁶ end up at similar final concentration. Endpoint is therefore a presence/absence assay, never a quantitative one. Everything in §4 is a way around this.

Primer design as a constraint problem

4. Turning amplification into measurement

RT- prepends reverse transcriptase, which copies RNA into cDNA. No DNA-dependent accepts an RNA template — reverse transcriptase is itself a DNA polymerase, just an RNA-dependent one — so every amplification-based RNA method and most sequencing-based ones begin with this step, included. The exceptions are the methods that read RNA directly: Northern blotting and RNA-FISH (§6), and direct-RNA nanopore sequencing. Note the naming collision: RT-PCR is reverse- PCR; real-time PCR is . Papers use both.

qPCR: measure during the exponential phase

Read fluorescence every cycle, and take the cycle at which signal crosses a fixed threshold — the quantification cycle, Ct (or Cq). Two detection chemistries:

SYBR GreenTaqMan hydrolysis probe
Mechanismdye fluoresces when intercalated in any double-stranded DNAthird oligo inside the amplicon carries a fluorophore quenched by FRET; polymerase's 5'→3' exonuclease destroys it during extension, releasing signal
Specificityreports primer-dimers and off-target products toorequires a third independent sequence match
Cost / designcheap, nothing to designprobe synthesis per target
Control neededmelt curve, to confirm one productnone of that kind
Multiplexingnoyes — different dyes, different targets, one tube

Where Ct comes from. Fluorescence is proportional to copies: F_n = k N₀(1+E)ⁿ. Setting F = F_T at n = Ct and taking logs,

Ct = [ log(F_T / k) − log N_0 ] / log(1 + E)

Ct is affine in log N₀, with slope −1/log₁₀(1+E) per decade. That is the entire content of the standard curve: run a dilution series of known concentration, regress Ct on log₁₀(copies), and read efficiency off the slope m:

E = 10^(−1/m) − 1        perfect doubling ⇒ m = −1/log₁₀2 = −3.32

A slope of −3.32 is the canonical target; slopes far from it mean the assay is not doubling and none of what follows is valid.

Relative quantification (ΔΔCt). Since N₀ ∝ 2^(−Ct) when E = 1:

within a sample:   target/reference  = 2^−(Ct_target − Ct_ref)  = 2^−ΔCt
across samples:    fold change       = 2^−(ΔCt_treated − ΔCt_control) = 2^−ΔΔCt

The inner ratio cancels input amount and extraction efficiency; the outer ratio cancels anything common to both genes. Two assumptions do all the work and both routinely fail: equal efficiency for target and reference (relax it with the Pfaffl formula, which uses each assay's measured E), and a reference gene that genuinely does not change. The second is the compositional-data problem in miniature — you are measuring a ratio and calling it an abundance — and it returns at full scale in Ch 47.

Digital PCR: absolute counting

qPCR is relative. Converting Ct to copies needs a standard curve, and the answer inherits the standard's calibration error.

Digital PCR removes the calibration entirely. Partition the reaction into 10⁴–10⁷ droplets or microwells so that most contain zero or one template molecule. Amplify to endpoint — the plateau that ruined quantitative endpoint PCR is now a feature, because it drives every occupied partition to the same saturated signal, so the readout is binary. Count positives.

Partitioning is Poisson with mean λ templates per partition, so with n partitions and k positive:

P(empty) = e^−λ = (n − k)/n     ⇒   λ = −ln(1 − k/n)      copies = λn

The correction matters: at k/n = 0.21 the naive count undercounts by ~11%, because some positives held two molecules. The estimate is absolute — its units are counted partitions and a known partition volume, with no reference material. Precision is Poisson, Var(λ̂) = (e^λ − 1)/n, so it improves as √n and beats qPCR decisively for small fold changes and low copy numbers: rare- detection in circulating tumour DNA, copy-number calls, viral load, and library quantification.

5. Cloning: cut, join, propagate

The problem: obtain unlimited identical copies of a defined construct, and get a cell to express it.

Restriction enzymes are a bacterial immune system: they cleave incoming phage DNA at short recognition sites, while the host methylates its own copies of those sites to spare them — self versus non-self by covalent marking. The sites are palindromic in the reverse-complement sense, because the enzyme is a homodimer and a two-fold symmetric protein binds a two-fold symmetric site:

EcoRI, GAATTC                      SmaI, CCCGGG
5'-G     A A T T C-3'              5'-C C C   G G G-3'
3'-C T T A A     G-5'              3'-G G G   C C C-5'
   staggered cut → 5' overhangs       flush cut → blunt ends
   "sticky": AATT anneals to any      joins anything to anything;
   fragment cut with EcoRI            no directionality, lower efficiency

A 6-bp cutter cuts on average every 4⁶ = 4,096 bp, a 4-bp cutter every 256 bp — modulated by real genome composition (NotI, GCGGCCGC, is rare in vertebrate DNA because CpG is depleted). Sticky ends only hydrogen-bond; DNA ligase forms the phosphodiester bonds that make the join covalent.

Vectors. A plasmid is a small circular replicon with three mandatory parts: an so the host copies it (the origin also sets copy number), a selectable marker, usually antibiotic resistance, and a multiple cloning site. Expression vectors add a , a binding site or Kozak sequence, and a terminator.

Transformation — chemical competence plus heat shock, or electroporation — succeeds in a tiny fraction of cells. This is why the marker is selection, not screening: plate on antibiotic and only transformants survive. Blue-white screening adds a second layer: the cloning site sits inside lacZα, so an empty vector complements the host's β-galactosidase, cleaves X-gal, and gives a blue colony; an insert disrupts the reading frame and gives white. Note precisely what that establishes — something was inserted. Not the right thing, not in the right orientation, not free of errors. You still sequence.

Why modern assembly displaced it

Restriction cloning has a structural flaw: it requires recognition sites that are absent from the insert, present in a usable arrangement in the vector, compatible in buffer and temperature, and it leaves scars. Finding a workable enzyme pair is a -satisfaction problem that often has no solution, and it does not scale past two or three fragments.

Both replacements move specificity from a fixed enzyme alphabet to arbitrary designed sequence.

graph LR D["designed sequence"] --> S["synthesis or PCR<br/>with overlap/BsaI tails"] S --> A["one-pot assembly<br/><i>Gibson / Golden Gate</i>"] A --> T["transformation"] T --> SEL["antibiotic selection"] SEL --> V["colony screen<br/>+ full-plasmid sequencing"] V --> U["expression, editing,<br/>reporter, library"]

The third option is increasingly to skip assembly: synthetic DNA is now cheap enough that ordering a gene-length fragment can cost less than the labour of building it.

6. Hybridisation: addressing a sequence with its complement

One idea, many instruments: a labelled nucleic acid probe finds its complement in a complex mixture, and stringency — temperature, salt, formamide — sets how many mismatches you tolerate. Raising stringency trades sensitivity for specificity along a continuous dial.

MethodTargetDetected withQuestion it answers
SouthernDNAlabelled complementary probeis this sequence present, in what size fragment, in how many copies
NorthernRNAlabelled complementary probeis this transcript present, at what size (isoforms are visible), how abundant
Westernproteinantibody — not a probeis this protein present, at what apparent size, how abundant
FISHDNA/RNA in situlabelled probe, imagedwhere, and how many copies, per cell
MicroarrayDNA or RNAmillions of immobilised probeshow much of each of a predefined set of sequences

Western blotting is not . Ed Southern's surname gave the Southern blot its name; Northern and Western are puns on it. The joke misleads a large fraction of students into believing Westerns work by base pairing. They cannot — proteins do not base-pair. A Western works because someone immunised an animal against your protein and purified the antibodies. That is the core idea of this chapter made concrete: the DNA and RNA methods take a string as input, the protein method takes a reagent that must be produced, validated, and re-produced.

FISH deserves separate emphasis because it answers a question sequencing answers badly. Bulk sequencing averages over cells; FISH reports per nucleus. When the question is "what fraction of cells carry this amplification" or "is this fusion present in a minor subclone", single-cell spatial information is the point, and FISH remains standard clinical practice for exactly that.

Microarrays invert the blot: instead of one probe against a whole sample, immobilise millions of known probes at known positions and wash the labelled sample over them. Each spot is one hybridisation assay; the array runs them in parallel. Expression arrays lost to for a reason that generalises — a hybridisation assay can only measure what you designed onto it, and its dynamic range is squeezed between background and saturation; a sequencing assay counts, and is discovery-capable. Genotyping arrays survive precisely because their query set is fixed by design: a known panel of common SNPs, cheap per sample, which is what makes population-scale association studies affordable (Ch 51).

7. Sanger sequencing

Chain termination converts a sequence-reading problem into a length-separation problem — which, by §2, is the one problem DNA hands you for free.

Run a single-primer extension containing normal dNTPs plus a small fraction (~1:100 to 1:1000) of dideoxynucleotides, which lack the 3'-OH. Any ddNTP incorporated terminates that chain, because the next phosphodiester bond has nothing to attach to. Across millions of molecules you obtain a population of fragments terminating at every position, each carrying a fluorescent dye identifying which base terminated it. Separate by capillary electrophoresis at single-base resolution and read the colour order.

primer ──►  ...GATTACAG
            ──────────────────────────► fragment length
   A   ▁▁▁█▁▁▁▁▁▁█▁▁▁▁▁▁▁█▁
   C   ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█▁▁▁▁     read peak colours in length order
   G   █▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█      → G A T T A C A G
   T   ▁▁▁▁▁▁█▁█▁▁▁▁▁▁▁▁▁▁▁

Usable read length is roughly 500–1,000 bases: the first tens of bases are lost to primer and dye artefacts, and the far end fails when the relative length difference between consecutive fragments falls below the resolving power of the capillary.

It survives in the era of massively parallel sequencing for a specific reason. Confirming a variant is not about accuracy alone — it is about error independence. Sanger uses different chemistry, a different sample path, and different analysis software, so it does not share the systematic failure modes that matter most: mismapping in a paralogous region, index hopping between samples, a sample swap. Universal Sanger confirmation of every next-generation call has been retired in many laboratories in favour of confirmation triggered by quality metrics, but orthogonal confirmation of reportable findings and of sample identity remains normal practice (Ch 55). Know its limits too: it reads a mixture as superimposed traces, so a heterozygous renders everything downstream unreadable without , and it cannot see much below ~15–20%.

8. Antibodies, and what you can build from them

An antibody is a protein that binds one epitope — a small patch of surface — tightly. Raised by immunisation, they come in two forms: polyclonal (a serum containing many binders to many epitopes on the target; robust to epitope damage, variable between batches, finite in supply) and monoclonal (one B-cell clone, one epitope; renewable and reproducible, but fails completely if that epitope is masked or altered). Sequencing the clone and expressing it recombinantly gives the reproducibility of a monoclonal with an inexhaustible supply.

A systemic caveat: a substantial fraction of commercial antibodies do not detect what the label claims. Validation means a or knockdown control showing the signal disappears — not a band of the expected size.

Immunoprecipitation couples the antibody to a bead and pulls the target out of a lysate, along with whatever is attached to it. Wash stringency is the control knob and it is a precision/recall curve: harsh washes remove background and real partners together.

ChIP (chromatin immunoprecipitation) is the move that made the technique genomic. Crosslink protein to DNA in living cells with formaldehyde, fragment the chromatin, immunoprecipitate the protein of interest, reverse the crosslinks, and sequence the DNA that came along. The output is a genome-wide map of where that protein was — the basis of Ch 49. Two things to hold onto: the essential controls are input chromatin and a non-specific antibody, and the result is a population average of a stochastic occupancy (Ch 01 §5) — a "bound site" is one bound often enough, in enough cells, to survive the wash.

Epitope tagging sidesteps the reagent problem: engineer a short peptide tag (, HA, His) onto the gene, and use one heavily validated anti-tag antibody for every target you will ever study. This converts an unsolved biochemistry problem into a cloning problem — a move worth recognising, because the field makes it repeatedly.

9. Reporters: making expression observable

Measuring expression usually destroys the sample. A reporter fuses the regulatory sequence you care about to a gene whose product you can measure non-destructively.

Fluorescent protein (GFP and relatives)Luciferase
Signalintrinsic fluorescence, needs only oxygenenzymatic light from an added substrate
Resolutionsingle cell, live, spatial, over timepopulation, usually endpoint after lysis
Sensitivity / rangemoderate; cellular autofluorescence sets a floorvery high; near-zero background, wide dynamic range
Main artefactslow maturation and high stability — poor at reporting decreases; the fusion may perturb the proteinneeds substrate; kinetics vary with substrate access

Beyond expression level, a fluorescent fusion reports localisation — where in the cell the protein is — which no lysate-based method can.

What a classical reporter assay actually establishes is sufficiency: this isolated fragment, on a plasmid, in this cell type, can drive expression. It does not establish necessity at the endogenous locus, in native chromatin, at native copy number. That gap is the whole reason validation moved to perturbing the native sequence (Ch 38, Ch 52). The scaled version of the sufficiency assay is the massively parallel reporter assay: give each candidate element a unique barcode inside the transcript, transfect the whole library at once, and read expression by sequencing barcodes in RNA versus DNA — a one-at-a-time assay converted into a counting problem over 10⁵ elements. by input DNA counts is the same ratio logic as dual-luciferase and as ΔΔCt.

10. Protein–protein interactions, and their error directions

splits a into its DNA-binding domain and its activation domain. Fuse "bait" to one and "prey" to the other. If bait and prey interact, they reconstitute a functional activator and switch on a reporter gene, which the yeast reads out as growth or colour. Because the whole assay is genetically encoded, it scales to library-against-library .

Co-immunoprecipitation pulls down endogenous protein A and blots (or mass-spectrometers) for B.

Each method fails in a characteristic direction, and knowing the direction is more useful than knowing the protocol:

MethodOver-reportsUnder-reports
Y2Hpairs forced into the same compartment at high concentration that never co-occur in vivo; self-activating and sticky baitsinteractions needing absent in yeast; membrane proteins; anything the fusion misfolds
Co-IPabundant sticky proteins; complexes that form after lysis destroys compartmentalisationtransient or low-affinity interactions that do not survive washing
Proximity labelling (an enzyme fused to the bait tags neighbours within nanometres, in living cells)neighbourhoods rather than binding partnersgenuinely binding partners with no accessible tagging residues

Interaction networks are unions of assays with these biases. Any topological statistic computed on an interactome — degree distribution, hub identity, modularity — inherits them, which is worth remembering before drawing conclusions from network structure (Ch 25).

11. Mass spectrometry: the discovery-capable protein assay

Antibodies can only find proteins you already suspect. Mass spectrometry is the way to identify and quantify proteins without a per-target reagent. The measurement is simple — mass-to-charge ratio, m/z — and everything else is inference.

Bottom-up logic. Whole proteins are too large to measure informatively, so digest with a sequence-specific protease. Trypsin cleaves after lysine and arginine — a restriction enzyme for proteins — yielding peptides of tractable length that carry a basic residue at the C-terminus and therefore ionise well. Separate peptides by liquid chromatography, ionise by electrospray, measure m/z (MS1), select a peptide, fragment it by collision, and measure the fragments (MS2). Adjacent fragments in the resulting ladder differ by the mass of one amino acid.

Identification is a database search, not reading. You compute theoretical fragment spectra for every peptide in an in silico digest of the proteome — which means mass spectrometry is downstream of genome (Ch 44) and blind to anything not annotated. Observed spectra are scored against theoretical ones, and false discovery is controlled by target–decoy search: search a reversed or shuffled database alongside the real one and count decoy hits above threshold as a direct empirical estimate of the false-positive count. It is an empirical null, and it is what makes proteomics FDR statements meaningful.

Acquisition strategy determines the missing-data structure of your matrix:

Quantification is either label-free (peak intensity or spectral counts, requiring careful and inheriting the compositional problem) or isobaric labelling, where samples are tagged with chemically identical, equal-mass labels whose fragments differ, mixed, and run once — eliminating run-to-run variance at the cost of ratio compression when contaminating peptides are co-isolated.

Three hard limits explain why the proteome is not "solved" the way the genome is. Abundance dynamic range spans ten or more orders of magnitude in plasma while a single run covers perhaps four to five, so abundant proteins mask everything else. Peptides are not uniquely assignable to proteins, giving a protein inference problem structurally identical to reads in . And modifications are substoichiometric, requiring dedicated enrichment. Underneath all three sits the fact from the top of this chapter: there is no amplification step for protein. Nucleic acids can be copied; proteins must be detected as they are.

12. Question → method

QuestionMethodThe assumption it rests on
Is this DNA fragment the expected length?Agarose gelLength is the only thing mobility reflects
Is this specific sequence present, and at what copy number?, or dPCR for absolute countsPrimers are specific; efficiency is known
How many copies exactly, in a rare-variant background?Digital Partitions are independent and Poisson-distributed
Has expression of gene X changed between conditions?RT-qPCR with ΔΔCtReference gene is genuinely invariant; efficiencies match
What is the sequence of this one amplicon?SangerSample is clonal or a resolvable
Is this variant real, or a mapping artefact?Orthogonal confirmation (Sanger or a second chemistry)The two methods fail independently
Which cells carry this rearrangement, and how many?FISHProbe specificity; adequate cell counts
a known panel across 500,000 peopleGenotyping arrayEverything you need is on the array
Where in the genome does protein P bind?ChIP + sequencingAntibody is specific; crosslinking is unbiased
Is this protein present, and at what size?The antibody detects what the label says
Which proteins are present, without a prior list?LC-MS/MSPeptides are in the searched database and above the noise
Does this regulatory element drive expression?Reporter assaySufficiency on a plasmid implies relevance
Do proteins A and B interact?Y2H to generate, co-IP or proximity labelling to testEach assay's error direction is understood
Build a defined multi-part constructGibson / Golden GateOverlaps or overhangs are orthogonal; sequence-verify anyway

Common misconceptions

What people thinkWhat's actually true
Western blotting works by , like Southern and NorthernIt works by antibody binding. Proteins do not base-pair. The name is a pun on Ed Southern's surname, and it misleads more students than any other naming accident in biology
How much product you get tells you how much you started withOnly during the exponential . Reactions plateau to similar endpoint yields regardless of input, which is exactly why measures a cycle, not a quantity
qPCR gives absolute quantitiesIt gives relative ones, unless calibrated against a standard curve whose accuracy it then inherits. Digital PCR is the absolute method, and it gets there by counting partitions rather than calibrating
Restriction sites are palindromes the way "racecar" isThey read identically on the reverse complement, not backwards along one strand. The symmetry exists because the enzyme is a homodimer
A band of the right size confirms you have the right moleculeIt confirms a length. Identity requires a sequence-specific probe, a diagnostic digest, or sequencing. Correctly sized wrong products are common
is obsoleteIts value is error independence, not accuracy. It fails differently from massively parallel sequencing, which is precisely what orthogonal confirmation requires
A hit means the two proteins interact in a cellIt means they can interact when both are forced into a yeast nucleus at high concentration. They may live in different compartments, cell types, or decades of developmental time
More cycles means more sensitivityPast plateau, no. And at very low input, stochastic sampling in early cycles produces and jackpot effects — the answer becomes noisy, not merely small
A GFP fusion faithfully reports protein levelFluorescent proteins mature slowly and degrade slowly, so they track increases poorly and decreases very poorly, and the fusion itself can perturb folding, localisation and turnover
An antibody sold as "anti-X" detects XA substantial fraction do not. The only real validation is loss of signal in a or knockdown

Worked example: quantifying one gene three ways

A treatment is claimed to induce gene TGT1. You measure it by RT- against reference gene ACTB, and confirm by digital .

Step 1 — validate the assay with a standard curve. A tenfold dilution series of a plasmid standard, 10⁶ down to 10² copies, gives Ct = 16.20, 19.55, 22.87, 26.19, 29.53.

slope m = (29.53 − 16.20) / (2 − 6) = 13.33 / (−4) = −3.333
E = 10^(−1/m) − 1 = 10^(1/3.333) − 1 = 10^0.3000 − 1 = 1.995 − 1 = 0.995

Efficiency 99.5%, slope essentially the ideal −3.32. The E = 1 assumption behind ΔΔCt is justified for this assay — and this step is not optional, because ΔΔCt without it is a number with no defined meaning.

Step 2 — relative quantification.

control:  Ct(TGT1) = 24.60,  Ct(ACTB) = 20.10   →  ΔCt = 24.60 − 20.10 = 4.50
treated:  Ct(TGT1) = 22.15,  Ct(ACTB) = 20.05   →  ΔCt = 22.15 − 20.05 = 2.10

ΔΔCt = 2.10 − 4.50 = −2.40
fold change = 2^(−ΔΔCt) = 2^2.40 = 5.28

TGT1 is up ~5.3-fold. Note that ACTB moved 0.05 cycles between conditions — reassuring, though it demonstrates only stability of ACTB, not that it is unaffected by the treatment.

Step 3 — absolute confirmation by digital PCR. The treated cDNA is partitioned into n = 20,000 droplets of 0.85 nL each; k = 4,200 are positive at endpoint.

k/n = 0.2100
λ   = −ln(1 − 0.2100) = −ln(0.7900) = 0.2357 templates per partition
copies in reaction = λn = 0.2357 × 20,000 = 4,714

The naive count of 4,200 would understate by (4714 − 4200)/4714 ≈ 11%, because ~476 positive partitions — n[1 − e^−λ(1 + λ)] — held more than one molecule, accounting for the 514 templates the raw count never sees. Analysed volume = 20,000 × 0.85 nL = 17 µL, so the concentration is 4,714 / 17 ≈ 277 copies µL⁻¹ — absolute, with no standard curve anywhere in the derivation.

Step 4 — why bother, when qPCR already gave an answer. Compare precision.

dPCR:  Var(λ̂) = (e^λ − 1)/n = (1.2658 − 1)/20,000 = 1.329 × 10⁻⁵
       SD(λ̂) = 0.00365   →   CV = 0.00365 / 0.2357 = 1.5%

qPCR:  typical technical SD ≈ 0.15 cycles
       CV in copy number = 2^0.15 − 1 = 11.0%

A 5.3-fold change is far outside qPCR's noise, so here the two agree and qPCR was sufficient. But had the claimed effect been 1.3-fold — the size of most real regulatory changes — an 11% CV would have made it unresolvable, while a 1.5% CV would not. The method you need is set by the effect size you are chasing, not by the biology of the gene.

Connections