Reporter gene assays guide — Part 7: Assay normalisation
Normalisation Strategies
A raw reporter signal is not data. It is a number that depends on cell number, transfection efficiency, cell viability, substrate concentration, instrument settings, and a dozen other variables that have nothing to do with the biology you are trying to measure. Normalisation is the process of removing these variables so that what remains reflects the actual experimental effect. It is also the part of reporter assay design that gets the least attention in published methods sections and is the most common reason published data cannot be reproduced.
What normalisation actually corrects for
A well-normalised reporter assay removes three classes of nuisance variation:
Plate-level variation: well-to-well differences in cell number, transfection efficiency, lysis efficiency, pipetting volume, and substrate addition. This is the largest source of variability in most plate-based assays and the easiest to address.
Sample-to-sample variation: differences in cell health, confluence, passage number, and any treatment that affects viability or general cellular activity. This matters more in long assays and in primary cells.
Treatment-specific confounders: effects of your experimental treatment on the normalisation control itself. This is the most insidious source of error and the hardest to detect.
A normalisation strategy that handles the first two but not the third will give you precise, reproducible, and completely wrong data.
The gold standard: dual-luciferase
Dual-luciferase is the most-used normalisation in reporter assays for good reason. Two different reporters (typically firefly and Renilla) are assayed in the same well, and the ratio of their activities controls for everything except changes that affect both reporters in the same way. Firefly is the experimental reporter driven by your promoter of interest. Renilla is driven by a constitutive promoter (usually TK or CMV) on a co-transfected plasmid.
What dual-luciferase controls for: cell number, transfection efficiency, lysis efficiency, sample handling, and most pipetting errors. The ratio of two reporters in the same well cancels out these effects.
What dual-luciferase does not control for: changes in the constitutive promoter's activity caused by the experimental treatment. If your drug inhibits CMV (and many do, since CMV has NF-κB sites that respond to inflammation), your Renilla signal drops, the ratio shifts, and you will measure an apparent induction of your experimental promoter that is actually a repression of the control.
This is the classic "renilla gets stressed too" problem. The fix is to choose a normalisation promoter that is minimally affected by your experimental conditions. EF1α is generally a safer choice than CMV. TK-Renilla is also a reasonable choice. PGK is widely used in lentiviral systems.
Other normalisation strategies
Co-transfected fluorescent protein. A fluorescent protein (often GFP or mCherry) driven by a constitutive promoter on a separate plasmid, measured alongside the luciferase. This is the dual-fluorescent equivalent of dual-luciferase but with lower dynamic range. It is useful when your primary reporter is itself a fluorescent protein, but otherwise the dual-luciferase approach gives better signal-to-noise.
Bicistronic reporters (IRES or 2A). The two reporters are on the same plasmid, translated from the same mRNA. This guarantees that every cell that expresses one reporter also expresses the other, eliminating cell-to-cell variation in transfection. The trade-off is that the second reporter is often expressed at lower levels, and the ratio between the two can be sensitive to sequence context.
Cell number normalisation. Dyes that measure total DNA (Hoechst, DAPI), total protein (Bradford, BCA), or cell viability (CellTiter-Glo, resazurin) can be used to normalise to cell number rather than transfection efficiency. This is a weaker form of normalisation than dual-reporter approaches but is useful when transfection control is impractical.
Imaging-based cell counting. High-content imaging systems can count cells and measure reporter fluorescence per cell, giving single-cell normalisation. This is increasingly common in HTS and removes most of the population-averaging artefacts that plague well-level assays.
RNA-based normalisation. qPCR of the reporter mRNA normalises to transcript level, removing any effects of translation efficiency, protein stability, and post-translational regulation. This is more labour-intensive than dual-luciferase but gives information about whether changes in reporter activity are transcriptional or post-transcriptional.
Endogenous gene normalisation. Measure the expression of an endogenous gene (by qPCR or Western) that is unaffected by your treatment, and normalise to it. This is the most biologically meaningful form of normalisation for some applications but is rarely used in standard reporter assays because of the labour and cost.
Choosing a normalisation strategy
| Strategy | Controls for | Best for | Limitations |
|---|---|---|---|
| Dual-luciferase (firefly/Renilla) | Cell number, transfection, lysis | Standard promoter/enhancer work | Constitutive control can be affected by treatment |
| Bicistronic (IRES or 2A) | Cell-to-cell variation in transfection | Single-cell analysis, viral vectors | Lower expression of second reporter |
| Co-transfected fluorescent protein | Cell number, transfection | When primary reporter is fluorescent | Lower dynamic range |
| Cell number dye | Cell number only | When transfection control is impractical | Misses transfection variation |
| qPCR of reporter mRNA | Transcriptional regulation | Mechanism studies | Labour-intensive, requires RNA prep |
| Imaging-based cell count | Cell number, single-cell variation | HTS, single-cell analysis | Requires imaging platform |
Common normalisation mistakes
Using the same promoter for the experimental and the control. This is surprisingly common in published work and is the single most damaging normalisation mistake. If your experimental and control promoters are the same, the ratio will not change even if the underlying activity of the promoter changes: you have built a null normalisation. Always use a different promoter for the control.
Trusting the constitutive control without validating. Every constitutive promoter is affected by some treatment. CMV is affected by inflammation and cell cycle. EF1α is affected by cellular stress. PGK is affected by glucose. Validate that your specific drug or treatment does not affect the control promoter before relying on dual-luciferase data.
Not normalising at all. For a one-off experiment, the temptation is to skip normalisation and report raw firefly luciferase activity. This is rarely publishable, and even for pilot data it makes the experiment very hard to interpret. The minimum acceptable normalisation is to control for cell number (by protein, DNA, or viability measurement) and to express the result as fold change over a control condition.
Using the wrong plate-reader settings. The signal-to-noise ratio of your assay is determined in part by the gain and integration time on the plate reader. If your signal is at the bottom of the dynamic range, the noise dominates. If it is at the top, the signal is saturating. Optimise plate-reader settings for each new construct, and consider running a dilution series to confirm linearity.
Statistical normalisation
Once you have collected the data, the right statistical treatment matters. Reporter assays are usually analysed as ratios (treatment/control) or as the ratio of the two reporters (firefly/Renilla). The ratio data is not normally distributed, so non-parametric tests (Mann-Whitney, Wilcoxon) are often more appropriate than t-tests. For multi-condition experiments, ANOVA on the log-transformed ratios is the standard approach.
For screens, the Z-factor is the most useful single quality metric. It compares the dynamic range of the assay to the variability of the positive and negative controls:
Z' = 1 − (3σ_positive + 3σ_negative) / |μ_positive − μ_negative|
A Z' above 0.5 is the standard threshold for a good screening assay. Below 0, the assay is worse than random. Most well-optimised reporter assays achieve Z' between 0.6 and 0.8. Anything below 0.5 needs further optimisation.