Choosing Reference Genes for RT-qPCR: Stability Testing and Common Pitfalls

Choosing the wrong reference gene is the single most common way to invalidate a qPCR experiment. GAPDH and beta-actin are used by default in thousands of studies, but their expression varies significantly across tissues, treatments, and cell types. If your normalizer is not stable, every fold-change value you calculate is wrong—and no amount of statistical testing downstream can fix it.

This guide covers how to select and validate reference genes for qPCR normalization, including the three major stability algorithms, practical criteria for choosing candidates, and mistakes that lead to retracted conclusions.

Why Reference Gene Stability Cannot Be Assumed

A reference gene (also called a housekeeping gene or endogenous control) should have constant expression across all samples in your experiment. The problem is that no gene is universally stable. GAPDH expression changes under hypoxic conditions. Beta-actin (ACTB) varies with cell proliferation. 18S rRNA can be affected by total RNA input differences because it is not polyadenylated and behaves differently during reverse transcription.

The MIQE guidelines (Bustin et al., 2009) require that reference gene stability be validated for each experimental system. This is not a suggestion—journals increasingly reject manuscripts that use unvalidated reference genes.

Common Reference Gene Candidates by Category

Start with a panel of 8–12 candidate genes drawn from different functional categories. Using genes from different pathways reduces the risk that a shared biological process affects all your normalizers simultaneously.

• Cytoskeletal: ACTB (beta-actin), TUBB (beta-tubulin)
• Metabolic: GAPDH, PPIA (cyclophilin A), PGK1
• Ribosomal: 18S rRNA, RPL13A, RPS18, RPLP0
• Transcriptional: TBP (TATA-binding protein), POLR2A
• Other: HPRT1, B2M, YWHAZ, SDHA, HMBS, UBC

Which candidates are most likely to be stable depends on your tissue type, organism, and experimental treatment. There are no shortcuts here—you must test them empirically.

How to Run a Reference Gene Validation Experiment

The validation experiment is straightforward but often skipped due to perceived cost or time. Here is the minimum protocol:

1. Select 8–12 candidate genes from the categories above, biased toward genes reported as stable in your tissue type in the literature.
2. Collect representative samples spanning all experimental conditions in your study (treatment vs. control, all time points, all tissue types).
3. Run qPCR for all candidates across all samples in triplicate.
4. Analyze stability using at least two of the three major algorithms: GeNorm, NormFinder, and BestKeeper.
5. Select 2–3 stable genes and normalize using their geometric mean.

Using the geometric mean of multiple reference genes (rather than a single gene) is recommended by Vandesompele et al. (2002) and is now considered standard practice.

The Three Stability Algorithms Explained

GeNorm

GeNorm calculates a stability measure called the M-value for each candidate gene. It works by comparing pairwise variation between genes—genes that vary together across samples are considered co-regulated and therefore stable relative to each other. The algorithm iteratively removes the least stable gene until two remain.

• Threshold: M-value <0.5 for homogeneous tissues, <1.0 for heterogeneous samples
• Strength: Also calculates the optimal number of reference genes via the V-value (pairwise variation). V <0.15 means adding another gene does not improve normalization.
• Limitation: Can rank co-regulated genes as stable even if both vary together with treatment.

NormFinder

NormFinder uses a model-based approach that accounts for both intragroup and intergroup variation. It identifies genes with minimal combined variation across your experimental groups, making it better at detecting genes that appear stable overall but vary between conditions.

• Threshold: Stability value <0.15 is considered excellent
• Strength: Less susceptible to co-regulation artifacts than GeNorm
• Limitation: Requires group assignment (treatment vs. control), which may not apply to all experimental designs

BestKeeper

BestKeeper uses raw Ct values to calculate the standard deviation and coefficient of variation for each candidate gene. Genes with the lowest SD across samples are ranked as most stable.

• Threshold: SD <1.0 Ct across all samples
• Strength: Simple, uses raw Ct values without conversion
• Limitation: Does not account for group structure or co-regulation

Interpreting Conflicting Rankings

It is normal for GeNorm, NormFinder, and BestKeeper to produce different rankings for the same dataset. This is expected because they use different mathematical models. The practical approach:

• If a gene ranks in the top 3 across all three algorithms, it is a strong candidate.
• If rankings diverge significantly, prioritize NormFinder for experiments with distinct treatment groups, and GeNorm for determining how many reference genes to use.
• Never rely on a single algorithm. The consensus recommendation is to use at least two methods and look for agreement.

Common Mistakes in Reference Gene Selection

• Using GAPDH or ACTB without validation: These are the most popular reference genes and also the most frequently shown to be unstable. Popularity is not evidence of stability.
• Validating in one tissue and applying to another: Reference gene stability is tissue-specific, treatment-specific, and even species-specific. Validation must be repeated for each new experimental context.
• Using a single reference gene: A single gene provides no internal check on stability. The geometric mean of 2–3 validated genes is significantly more robust.
• Skipping validation for time-course experiments: Gene expression of housekeeping genes can drift over developmental stages or extended treatment periods.
• Ignoring primer efficiency differences: If your reference gene primers have substantially different efficiency from your target gene primers, normalization will introduce systematic error. Validate primer efficiency for all genes in your panel.

Reporting Reference Gene Validation in Publications

MIQE requires that you report the following in your methods section:

• Which candidate reference genes were evaluated
• Which stability algorithm(s) were used and the resulting stability values
• Which genes were selected and why
• Whether normalization used a single gene or geometric mean of multiple genes
• The primer efficiency for each reference gene

Reviewers increasingly check for this information. Providing stability rankings in supplementary data strengthens your manuscript and preempts reviewer requests for additional validation.

Reference gene selection is not a preliminary chore—it is a critical experimental decision that determines the validity of every fold-change value in your study. Spend the time to validate properly, use multiple algorithms, normalize with the geometric mean, and report what you did.