Controls and spike-ins for a sequencing lane

A lane that returns glossy sample reads and a dead control has not passed. The control exists so you can refuse the sample reads. This is a protocol overview at the level of decisions: which control class belongs on a DNA sequencing lane, which spike-in belongs in an RNA tube instead, and what you stop doing when the control fails. Volumes and cycle numbers stay in the manufacturer's instructions. How libraries become reads at all is next-generation sequencing from library to reads. The checks that sit beside controls before loading are what to check before a sequencing run.

What a lane control is measuring

A sequencing instrument can fail while still emitting files. Clusters can be too dense, too sparse, or dominated by one sequence so that base calling loses its balance. Registration can slip. A reagent can underperform. Sample reads from a complex genome sometimes disguise that, because the genome is diverse enough to look "normal" in a summary plot.

A PhiX-style control is a library of a small, well-known bacteriophage genome, or an equivalent characterised control genome, spiked into the lane in a fraction the instrument vendor specifies. You know what the alignment should look like. You can compute an error rate that is about the machine and the base caller, rather than about your polymorphism. On instruments that need base diversity, the same spike also balances a low-diversity amplicon pool. The fraction is a specification, not a guess. A control that was added at an unrecorded volume cannot be interpreted when it comes out too strong or too weak.

The idea of a reference material is older than sequencing. NIST's reference material programmes are a useful reminder that a control is a defined substance with a defined use. PhiX-style DNA is that idea applied to a lane. It is not a universal positive control for your biology.

A known variant controls the sample and the pipeline

The lane control does not know who your samples are. A second control does: a sample, processed with the others, that carries a variant you have already seen by another method. After you demultiplex and call, that variant should reappear with a sane allele balance. If it does, this pipeline on this batch can recover that class of variant. If it does not, you have a swapped plate, a broken BED file, a wrong reference build, or a filter that ate the call. Those are different failures, and the missing variant does not by itself say which one. It does say that you stop.

Choose the known variant to resemble the claim you will make. A heterozygous single-base change does not vouch for a large deletion. A clonal Sanger-confirmed mutation does not vouch for a variant present in one percent of molecules. State the resemblance in the record.

A no-template library, carried through adapters and indexing, belongs in the same batch when contamination is plausible. Reads in that well mean the adapters, the indexes, or the bench introduced DNA. Interpret the real samples only after you understand those reads. This is the sequencing cousin of the no-template control described for amplification in general laboratory practice.

ERCC-style spike-ins are an RNA reagent

The External RNA Controls Consortium mixtures are sets of synthetic transcripts, added to RNA at known relative amounts. They let you see whether an RNA-seq experiment preserved abundance relationships, within the limits of that mixture. They are not a DNA sequencing control. They are not a substitute for PhiX. They do not measure cluster error on a genome instrument, and adding the RNA to a finished DNA library does not recreate the test they were designed for.

If your project is RNA, add the spike-in to the RNA, following the spike-in protocol, and still decide separately how the lane itself will be controlled. Protecting the RNA until that moment is its own craft, outlined in protecting RNA during extraction. If your project is DNA, leave the ERCC tube in the freezer. A data column labelled spike that contains the wrong class of molecule will be plotted, and the plot will be meaningless.

Branching when a control fails

Write the stop rule before the run, in the same spirit as a sequencing statement of work.

If the PhiX-style library produces no reads, or produces an error rate outside the range you set from the instrument guidance, do not call sample variants. Check loading, reagents and instrument metrics. A repeat of the sequencing, from the same pool, is a different decision from a repeat of the library. If the control looks right and a sample's known variant is missing, stop that batch's genotypes and audit identity, reference build and filters. If the no-template well has indexed reads, stop and treat the batch as contaminated until the source is found. If only one sample looks thin and the controls are healthy, the failure is that sample's library, which is the territory of troubleshooting a failed library quantification.

Do not average a failed control into a pass because three samples look exciting.

Equipment and reagents, as classes

You need the control library the instrument family recommends, a way to quantify it with a named method, and a sample sheet that reserves its index or its demultiplexing path so control reads are not dumped into a sample. You need the previously genotyped sample, with the locus written down. You need, for RNA projects only, the spike-in mixture specified by that RNA protocol. Quantification of the control uses the same discipline as any other library: the number must name its method.

An instrument-class overview is the Illumina next-generation sequencing introduction. Other instrument families specify their own control genomes. Use the one that belongs to the machine you are loading. Mixing the noun PhiX onto a platform that expects a different control is how a "failed control" is manufactured from a paperwork error.

Quiet ways a control fails to be a control

The control library was made with a different adapter, so it never clustered, and someone concluded the instrument was down. The known variant was recorded on a different genome build, so the caller looked in the wrong place and "failed". The spike-in was ERCC, the samples were genomic DNA, and a beautiful standard curve was fitted to noise. The PhiX fraction was so high that it starved the samples, which is a design error, not a successful control. The known-variant sample was prepared on a different day with a different kit, so it no longer shares the batch's failure modes.

A control that cannot fail is décor. If you will explain away every bad control metric, you do not have a control. Write the metric you will obey.

Deposited control reads can help a later reader, in an archive such as the European Nucleotide Archive, when policy allows the deposit. A sentence in the methods that says "a control was used", with no identity and no result, does not.

Research use and the organism in the tube

PhiX-style DNA is nucleic acid, and your samples may be much more sensitive than the phage genome. The control does not change the biosafety level of the sample, the ethics of a human sequence, or the containment of a pathogen genome. Institutional rules decide those, with documents such as the WHO laboratory biosafety manual as a public reference rather than as a permit. A passed control is not a diagnostic validation. It is evidence about this run, in this research project.

A control that thawed during a power cut

Control libraries and spike-in aliquots are small and easy to ruin. A freezer that warmed overnight can nick a control you have used for a year. The next lane then "fails" for a reason that lives in the control tube. After an outage, quarantine the aliquot and open a new one, or re-quantify and re-check length if the manufacturer allows that aliquot to continue. In a hot season, do not leave the control strip on the bench while the sample plate is assembled. Record the lot. A changed lot is a new control until you have seen it pass once.

What to specify when you ask for a controlled run

State the instrument class, the control genome you expect on the lane, the fraction policy you want followed, and the known variant you will use as a sample control, including its build. State that RNA spike-ins are in or out, and why. State the stop rule. Catalogue classes for library reagents sit in the genomics and sequencing catalogue. The research setting is the genomics research pathway.

Discuss a genome run through the whole-genome sequencing enquiry reference, an amplicon pool through the amplicon sequencing enquiry reference, and a single-locus confirmation of that known variant through the Sanger DNA sequencing enquiry reference. Put the control plan in the quote request. These are enquiry references so the method can be discussed with the same nouns you used here.