Reference stability: are your qPCR reference genes steady at all?

Test whether the genes you normalise expression to are steadier across human tissues than genes of the same abundance, with the false discovery rate controlled across candidates.

What it doesCopy Anchor Link

Name the candidate reference genes. Every protein-coding gene GTEx v8 measures gets the standard deviation of its log2 expression across the tissues you choose (all 54 by default), each tissue being GTEx's median over its donors. Each candidate is then ranked among the genes nearest it in mean expression, because less abundant genes vary more for reasons that have nothing to do with regulation, and a gene is only steady if it is steadier than genes of its own level. The result lists the candidates steadiest first, each with its mean, its standard deviation, the share of its peers that vary less, its q for steadier and its q for more variable, and the range of expression its peers span. Beside it, the four rankers the field uses order the candidates against each other, reading each tissue as one sample.

When you'd use thisCopy Anchor Link

• Before choosing a reference gene. A ranker always names a most stable gene, even when every candidate varies as much as any gene of its abundance. Ranking each candidate among its peers asks whether the best of them is steady at all.
• Checking a published normaliser. Classic reference genes were chosen in a handful of cell types. Testing them against all of GTEx shows which of them still hold across tissues.
• Narrowing to one organ. Choose only the tissues your experiment spans, such as the thirteen brain regions, and the peer ranking is recomputed on those.

What this computesCopy Anchor Link

For every gene in GTEx v8 (19,107 protein-coding genes):

• the mean and the sample standard deviation of log2(TPM + pseudocount) across the chosen tissues, the pseudocount 1 by default

For each candidate:

• its expression-matched peers: the genes nearest it in mean expression, 1,000 by default, ties broken by table order, and the range of mean expression they span
• its percentile: the share of those peers less variable than it
• a p for steadier, (1 + peers at least as steady) / (peers + 1), and the mirror p for more variable; the smallest p the test can give is 1 / (peers + 1)
• q-values for each direction, adjusted across the candidates by Benjamini–Hochberg (independent or positively dependent tests) or Benjamini–Yekutieli (any dependence, less power), and the steadier and more variable calls at the false discovery rate you set

Across the candidates, with each tissue as a sample and −log2(TPM + pseudocount) in place of Cq:

• geNorm M and its stepwise exclusion order (Vandesompele 2002)
• NormFinder stability (Andersen 2004)
• BestKeeper: the SD and coefficient of variation of each candidate and its correlation with the BestKeeper index (Pfaffl 2004)
• comparative ΔCt: the mean SD of each candidate's differences from every other (Silver 2006)
• the RefFinder consensus: the geometric mean of the four ranks

Every result states, in words generated from the run, where geNorm's two most stable genes sit among their own peers, the smallest q the settings allow when no call is possible, which candidates are so abundant that their peers span more than a 16-fold range, which symbols GTEx does not carry, and that tissue medians measure variation between tissues, not between people.

Worked exampleCopy Anchor Link

The ten reference genes of Vandesompele et al. (Genome Biology 2002) across all 54 GTEx tissues, 1,000 peers, Benjamini–Hochberg at q ≤ 0.05:

• None of the ten is steadier or more variable than its peers.
• HMBS comes closest to steady: only 16 of its 1,000 peers vary less (p 0.017, q 0.17).
• B2M and ACTB vary more than 85% and 84% of their peers.
• TBP, which geNorm pairs with UBC as most stable, sits at the 42nd percentile, an ordinary gene of its abundance. geNorm orders TBP, UBC, HMBS, SDHA, YWHAZ, RPL13A, GAPDH, HPRT1, ACTB, B2M; every ranker puts B2M last.
• Across the thirteen brain tissues alone, GAPDH becomes the steadiest (11th percentile), still without a call.

ControlsCopy Anchor Link

Show decides how much of the result the table shows: Against their peers or Against each other.

LimitationsCopy Anchor Link

• Tissue medians, not samples. Each tissue is GTEx's median over its donors, so this measures variation between tissues. A gene steady across tissue medians can still vary between people, treatments or disease states within one tissue, which is what a qPCR normaliser must survive. Validate a shortlist on your own samples.
• Peers are matched on mean only. At the top of the expression range few genes are as abundant, so the peers of the most abundant candidates span a wide range and include noisier, less abundant genes, which makes a variable gene look ordinary.
• Human, protein-coding, GTEx v8. A symbol GTEx does not carry under its current HGNC name is left out and named (GAPD is now GAPDH).
• The rankers order the candidates you name against each other; none of them says whether any candidate is steady in absolute terms.

Where the numbers come fromCopy Anchor Link

Every gene's mean, standard deviation, peers, percentile and p is checked against base R on all 19,107 genes, across all 54 tissues and across the thirteen brain tissues with different settings, and the q-values against R's p.adjust. The rankers are the ones the qPCR tool runs, checked against eleven (geNorm), ctrlGene and NormqPCR (NormFinder, BestKeeper, comparative ΔCt) and the RefFinder consensus on Vandesompele's own 85-sample plate. Expression is GTEx v8's gene median TPM per tissue.

AutomationCopy Anchor Link

The same test is a workflow node and a tool the assistant can call, with the same settings and defaults as the panel.