RT-qPCR remains the gold standard for nucleic acid quantification. Yet, studies consistently show that improper normalization and weak statistical practices are among the top causes of irreproducible results (Bustin et al., 2009).
This mini playbook distills over two decades of qPCR best practice (Vandesompele et al., 2002; Hellemans et al., 2007) into 11 actionable tips you can apply immediately to strengthen your experiments and publications.
Part I. Biostatistical analysis
1. Always log-transform your data. Gene expression data are log-normally distributed: most values cluster below 1 (downregulation), while a few stretch to infinity (upregulation). On a linear scale, twofold down and twofold up are not symmetric. Log transformation (base 10 or 2) restores symmetry and ensures parametric tests treat fold-changes appropriately (Motulsky, Intuitive Biostatistics). Without log transformation, P-values may be misleadingly biased toward upregulated genes.
2. Respect pairing in your study design. Many qPCR designs compare matched samples: before vs. after treatment, mutant vs. wild-type littermates, or two conditions applied to the same cell line. If data are paired, tests like the paired t-test or Wilcoxon signed-rank test exploit this structure, reducing background variability. Ignoring pairing treats related data as independent and can inflate error, masking real differences (Hellemans & Vandesompele, 2011). Always encode pairing in your analysis plan.
3. Choose your statistical test upfront. Small sample sizes are common in qPCR. With n < 24, you cannot rely on the central limit theorem to "rescue" parametric methods. In such cases, parametric tests (t-test, ANOVA) may give overly optimistic P-values, while non-parametric tests (Mann–Whitney, Kruskal–Wallis) are more conservative but safer. Decide on the test before looking at your data, to avoid "p-hacking." For borderline sample sizes, report both methods transparently, but interpret cautiously (Nature Methods, Points of Significance series, 2013).
4. Report confidence intervals, not just error bars. Error bars communicate uncertainty — but only if you choose the right metric. For group means, the 95% confidence interval (CI) is superior to standard deviation (SD) or standard error of the mean (SEM). A CI shows the range in which the true population mean lies with 95% certainty, making it directly interpretable (Hellemans et al., 2007). For normalized expression, propagate technical and biological sources of error, including reference gene variance and PCR efficiency.
5. Adjust for multiple testing. Testing many genes inflates false positives. At α = 0.05, analyzing 10 genes gives a ~40% chance of at least one spurious "significant" hit. Apply multiple-testing corrections such as Šidák (conservative) or Benjamini–Hochberg false discovery rate (FDR) (less conservative, more power) (Benjamini & Hochberg, 1995). Report both raw and adjusted P-values.
6. Biological replicates are non-negotiable. PCR replicates only measure technical variability — they cannot substitute for biological replication. At least 3 biological replicates per group are needed for CI-based inference; 6 pairs are required for Wilcoxon signed-rank, and ≥8 combined samples for Mann–Whitney (Hellemans & Vandesompele, 2011). Underpowered experiments waste resources and risk irreproducibility. Design replication strategy before data collection.
7. When in doubt, consult a statistician. Molecular biologists must know the basics — but no one is an expert in everything. A consultation with a biostatistician, especially during the design phase, can prevent wasted experiments. They can advise on power analysis, test selection, and randomization. But remember: statisticians cannot fix poor experimental design. Garbage in, garbage out.
Part II. Reference gene normalization
8. Use multiple validated reference genes. Normalization removes technical variation from qPCR data. Using a single unvalidated "housekeeping" gene is risky: it can introduce errors up to 6.4-fold in 10% of cases, and >20-fold in rare cases (Vandesompele et al., Genome Biology 2002). Instead, use 2–5 validated, stably expressed reference genes identified via geNorm or equivalent algorithms. Once established under specific conditions, this panel can be reused for future studies with similar samples.
9. Verify stability in your experiment. Reference gene stability can drift under new experimental conditions. Always check stability using metrics like the geNorm M value and coefficient of variation (CV) (Hellemans et al., Genome Biology 2007). Recommended thresholds: homogeneous samples (cell culture) M < 0.5, CV < 25%; heterogeneous samples (tissues, biopsies) M < 1, CV < 50%. If genes exceed thresholds, expand your candidate panel or improve sample processing. Stability assessment is impossible with only one reference gene — another reason multiple are essential.
10. Reference genes don’t need to be in the same plate. Relative quantification only requires equal treatment of all samples. Follow the "sample maximization" strategy: run all samples for a given gene in the same qPCR run. Reference genes can be measured separately — even with different chemistries or platforms — and still provide valid normalization (Hellemans et al., 2007). Inter-run calibration is only required if samples are split across multiple runs.
11. Reference gene ≠ equal expression. A common misconception is that reference genes must be expressed at the same level as the target gene. This stems from Northern blot days, when band intensity dictated detectability. qPCR’s wide dynamic range makes expression level irrelevant - only stability across conditions matters (Bustin et al., Clinical Chemistry 2009). Choose reference genes based on stability, not abundance.
Key takeaways
- Log-transform data to restore symmetry before statistical testing.
- Respect pairing and select tests before analysis, not after.
- Use confidence intervals and correct for multiple comparisons.
- Replicate biologically, not just technically.
- Normalize with multiple validated reference genes and check stability.
Jan Hellemans
Co-founder, Clarida
Creator of the qbase algorithm, co-author of the MIQE guidelines, and co-founder of Biogazelle (acquired 2022). Translated qbase+ into a globally adopted qPCR analysis platform serving pharma, biotech, and academic labs worldwide.