Delta-delta Cq and what it assumes

The doubling formula is short. The list of things it assumes is longer. Delta-delta Cq, written ΔΔCq, is a way to turn threshold cycles into a relative fold change between a sample and a calibrator, after accounting for a reference transcript. It is the calculation behind a great many RT-qPCR bar charts. This troubleshooting page is about what has to be true before that bar means what the label says. The assay context is RT-qPCR for relative expression, and the study around it is from cells to a gene expression result.

A fold change that arrived too quickly

The people who need this page already have a spreadsheet. Target Cq, reference Cq, a control column, a treated column, and a formula that subtracts twice and raises 2 to a power. The decision is whether to keep that formula, replace it with an efficiency-corrected one, or stop talking about fold change until a broken assumption is repaired. The MIQE guidelines ask you to describe the analysis method, including normalisation and how Cq was determined. "We used delta-delta Cq" is the start of that description, not the end.

The arithmetic, said once

For each tube, subtract the reference Cq from the target Cq. That difference is ΔCq. It adjusts the target for the amount of material that the reference says was there. Then subtract the calibrator's ΔCq from the sample's ΔCq. That second difference is ΔΔCq. If both assays double every cycle, the relative quantity of the sample compared with the calibrator is 2 raised to the power of minus ΔΔCq. A ΔΔCq of minus 1 is a twofold higher relative abundance. A ΔΔCq of plus 1 is a twofold lower one. The calibrator becomes 1 by construction, because you subtracted it from itself.

The unit of that number is "relative to this calibrator, given this reference". It is a ratio of ratios. It becomes copies per cell only if you built a different experiment: a calibrated standard curve in copies, and an independent measure of cell number. Most expression plates never did that, and they should not be labelled as if they had.

Three assumptions that carry the formula

The first assumption is matched amplification efficiency. The simple powers-of-two formula treats the target assay and the reference assay as equally efficient, and treats that shared efficiency as a clean doubling. You earn the assumption with standard curves, as set out in standard curves and primer efficiency. If the target sits at 85 percent and the reference at 105 percent, each cycle is a different multiplication. Subtracting Cq values then mixes two rulers. The distortion is small when all samples have nearly the same Cq, and it grows when you claim a large fold change or when the samples span a wide range. An efficiency-corrected relative calculation, of the kind associated with Pfaffl's approach, inserts the measured efficiencies instead of assuming 2. It is still a model. It still needs an honest slope.

A widely used planning window of 90 to 110 percent does not mean two assays inside that window are interchangeable. The window is where many people start. The formula you publish is a separate choice, and it should match the slopes you measured.

The second assumption is a real calibrator. The calibrator is the sample you call the baseline: untreated cells, time zero, a reference tissue, a pooled control. Every fold change is a comparison with that tube. If the calibrator was degraded, mislabelled, or a different cell type, every other sample inherits the error. Biological replicates of the calibrator group are wiser than a single favourite well. Averaging Cq values needs care, because Cq is on a log scale. Many workflows average the relative quantities, or they state clearly that they averaged Cq inside a tight technical pair. Say which you did.

The third assumption is reference-gene stability in this experiment. ΔCq removes whatever the reference did. If the reference rose with the treatment, the target appears to fall. Stability is a measurement, described in choosing reference genes instead of hoping. When the claim is quantitative, more than one stable reference is the safer denominator. A geometric mean of validated references is a common way to combine them. An untested GAPDH is a hope.

Further assumptions hide in the plate. Cq values should sit inside the linear range of each assay. The same threshold rule should be used throughout. The amplicon should be the transcript you named, which you can check against a gene model on Ensembl and a specificity search on NCBI Primer-BLAST. A no-template control and, where genomic DNA can amplify, a no-reverse-transcription control, keep the Cq attached to cDNA.

Where to stop when an assumption fails

Lay the three assumptions next to the notebook. Missing curves mean you do not know the efficiencies. Either run the dilutions or change the sentence to a qualitative pattern of Cq values with the caveat visible. Curves that disagree by a wide margin mean the simple formula is the wrong tool. Switch to an efficiency-corrected calculation or redesign the worse primer pair. Do not "pick the fold change that looks familiar".

A calibrator with a failed reference amplification cannot anchor the plate. Repeat it. Do not borrow a calibrator Cq from a different day unless you have shown the runs are comparable. A reference gene that tracks the treatment is retired. If the only reference you have is the moving one, the fold changes already in the figure are provisional.

Inhibition in a subset of samples looks like high Cq and a dramatic fold change. Dilute those cDNAs or spike a control RNA. If the ΔCq returns toward the rest of the group, you were measuring dirt, not biology. Technical triplicates that disagree by a large Cq gap are a pipetting or sealing problem. The well that makes the story better is not automatically the true one. Set a rule for outliers before you look at the groups.

Symptoms worth trusting as symptoms

A fold change that explodes when you swap reference genes is a reference problem, not a software problem. A fold change that shrinks when you apply the measured efficiencies was being helped by the assumption of perfect doubling. A target that looks induced in Cq and uninduced once RNA integrity is balanced across groups was a degradation story. None of these patterns tell you the biological direction by themselves. They tell you which assumption to test next.

Comparing results to an RNA-seq table adds another ruler. Sequencing counts are normalised across thousands of features and depend on the library type. A qPCR fold change can confirm the direction of a named transcript and still disagree on magnitude. Isoforms are a frequent reason: the primer pair sees one exon junction, and the count sums several transcripts. Look up the amplicon on the gene model before you call one method wrong.

Research claims, not a clinical unit

Delta-delta Cq supports a research comparison under the assumptions you checked. It does not diagnose a patient, and it does not replace an ethics decision for human samples. Report the calibrator, the references, the efficiencies and the replicate structure in the same place as the fold change. Readers can then see the unit. Institutional biosafety rules still govern the nucleic acid you started from, including any infectious template.

A calibrator that thawed

A power cut, a warm courier pouch, or an afternoon on a humid bench can damage the calibrator more than the samples that stayed frozen. Because every ΔΔCq subtracts that calibrator, one bad baseline tilts the whole figure. If the freezer alarmed, measure integrity or rerun a reference panel before you reuse yesterday's Cq. The arithmetic will happily produce a fold change from a compromised tube. The troubleshooting step is to refuse that convenience.

Asking for help without skipping the assumptions

If you are still choosing mixes and reference assays, the molecular biology catalogue is the reagent side, and the nucleic acid analysis pathway is the wider sample path. Say in the enquiry which formula you plan to use and whether efficiencies are already known.

When the trouble is that a handful of qPCR assays cannot answer a transcriptome question, the mRNA sequencing enquiry reference and the differential expression analysis enquiry reference are the pages to read before you write. They are enquiry references. Put the contrast, the species, the calibrator definition and the replicate plan on the quote request, and ask whether a quotation is possible. A method can be discussed from that specification. The fold-change assumptions stay your scientific responsibility either way.