The proposal is to regroup the reporting requirements into seven stages that follow the order in which an experiment is actually carried out, and to attach to each stage a short list of things to write down at the time. Table 1 gives the stages, the core information to report for each and the reason a reader needs it. Figure 1 shows the stages in sequence. The checklist is deliberately shorter than MIQE. It is meant as a floor for laboratories that currently report much less, and nothing in it conflicts with the full guideline.
The seven stages of the proposed reporting checklist, with the core information for each
| Stage | Report at minimum | Why the reader needs it |
|---|---|---|
| 1. Design and replication | Groups, number of biological replicates, what a replicate is, controls included | Sets the population to which any inference applies |
| 2. Sample and nucleic acid | Source, storage, extraction method, quantity, purity and integrity assessment | Degraded or impure template biases Cq values |
| 3. Reverse transcription | Input amount, priming strategy, enzyme, whether a no-enzyme control was run | This step is the least reproducible in the workflow |
| 4. Assay and validation | Target accession, primer sequences, amplicon length, specificity check, efficiency | Allows the assay to be checked and repeated |
| 5. Run and raw data | Instrument, chemistry, reaction volume, cycling conditions, Cq method, no-template controls | Cq depends on instrument and software settings |
| 6. Normalisation | Reference genes used and the evidence that they are stable in these samples | An unstable reference creates or hides differences |
| 7. Analysis and statistics | Quantification model, handling of replicates and outliers, test used, software | Shows how Cq values became the reported result |
The seven reporting stages in the order an experiment proceeds, with gaps found at write-up referred back to the record
The first thing to record is what was compared with what and how many independent biological units stood behind each group. The distinction between biological and technical replication is the one most often blurred. Three wells pipetted from one cDNA preparation measure pipetting precision and nothing more, and a difference between two samples each run in triplicate has a sample size of one per group however small the error bars look (Vaux et al., 2012). The report should say plainly what the unit of replication was, an animal, a plant, a culture started on a different day, and which controls accompanied the experimental samples.
A reader needs to know where the material came from, how it was preserved, how nucleic acid was extracted and how its amount and quality were judged. Integrity deserves a sentence of its own. Partly degraded RNA still amplifies, but it does so unequally across transcripts and amplicon positions, so comparisons between samples of different integrity are unsafe (Fleige and Pfaffl, 2006). Laboratories with access to capillary electrophoresis can report an RNA integrity number (Schroeder et al., 2006). Those without it should state what they did instead, for example inspection of ribosomal bands on a gel, and should not leave the matter unmentioned. Whether and how genomic DNA was removed belongs here too.
Conversion of RNA to cDNA is the step at which most of the technical variance in a two-step experiment enters. Yield depends on the enzyme, on whether priming is by oligo(dT), random oligomers or gene-specific primers, and on the amount of RNA put in (Stahlberg et al., 2004). The report should give all three, state that the same input amount was used across samples being compared or explain how unequal input was dealt with, and say whether reactions lacking reverse transcriptase were run to detect signal from contaminating DNA. Where a one-step format was used, in which reverse transcription and amplification share a tube, that should be said, since the two formats are not interchangeable when results from different studies are compared. It also helps the reader to know whether all samples in a comparison were reverse transcribed in the same batch, because batch-to-batch differences at this step are easily mistaken for biology.
An assay that cannot be identified cannot be repeated. The database accession of the target, the primer sequences, any probe sequence and the amplicon length should be given, or for a commercial assay the supplier's identifier and whatever context sequence the supplier discloses. Two properties need evidence. Specificity can be supported in silico, with a tool such as Primer-BLAST (Ye et al., 2012), and empirically by a single melting peak or a product of the expected size. Amplification efficiency should be stated together with how it was estimated, because the estimate itself is imprecise when derived from a short dilution series (Svec et al., 2015).
The instrument, the detection chemistry, the reaction volume and the cycling conditions are facts that cost nothing to report. So is the method by which Cq values were obtained. A Cq is not a property of the sample alone: it changes with the baseline correction and with where the threshold is placed, and different software makes these choices differently (Ruijter et al., 2009). The report should state the software and settings, give the results of the no-template controls, and wherever possible deposit the raw Cq values as a supplementary file so that the calculation can be repeated.
Normalisation to a reference gene corrects for differences in input and reverse transcription yield, and it is where published experiments most often go wrong. No gene is constitutively stable in every tissue and treatment. Vandesompele et al. (2002) showed that normalising to a single unvalidated reference can introduce large errors and proposed using the geometric mean of several references chosen for stable expression in the samples at hand. A model-based alternative for ranking candidates was given by Andersen et al. (2004). The checklist therefore asks for the reference genes by name, their number, and the evidence that they were stable under the conditions of the study. A statement that a gene is commonly used is not such evidence.
Finally the report should say how Cq values were turned into the quantities shown in the figures. The comparative method of Livak and Schmittgen (2001) assumes that target and reference amplify with equal and near-perfect efficiency, and a paper that uses it should say that the assumption was checked. Where efficiencies differ, an efficiency-corrected model is appropriate (Pfaffl, 2001). The handling of technical replicates, any exclusion of wells and the rule applied, the scale on which statistics were computed and the test used all belong in the methods. Fold changes are ratios, and analysis on the logarithmic scale is usually the defensible choice (Schmittgen and Livak, 2008).