Adapters indexes and barcode hopping

A multiplexed run is a promise that each read can be walked back to one tube. Adapters make the fragment readable. Indexes, and the hopping that moves them, decide whether that walk is honest. This is a troubleshooting page for the failure where the FASTQ looks excellent and the sample identity does not. The chemistry that attaches those adapters is in library prep is where most runs are won. The run that consumes them is in next-generation sequencing from library to reads.

If you sequence one library alone, hopping has nowhere biologically wrong to send a read, though the index read can still fail. The trouble starts when many libraries share a flow cell and a wrong index looks like a real sample.

What the adapter is doing on the molecule

On an Illumina-style library the adapter ends are the sequences the flow cell oligo binds, often discussed as P5 and P7. Inside the adapter sit the binding sites for the read primers. The instrument does not "know" your sample name. It knows those sequences. A fragment without them does not form a proper cluster. A fragment made of two adapters and no insert forms a cluster anyway, and the read is adapter sequence. That dimer problem is a prep problem. It becomes an index problem only because dimers can carry indexes too, and they hop and mis-assign with enthusiasm.

The index is a barcode of a handful of bases, read in a separate index read rather than as part of the genomic insert. Dual indexing uses two barcodes. In the usual vocabulary i7 is the index next to the P7 end, read as index 1, and i5 is the index next to the P5 end, read as index 2. Both must be known to the demultiplexer. A unique molecular identifier, when a kit includes one, is a molecule-level tag and is not a sample index. Mixing those two ideas in a sample sheet assigns molecules to people, or people to molecules.

Indexes should differ from each other at enough positions that a single substitution in the index read does not turn sample A into sample B. That edit distance is a design property of the index set. A homemade set of barcodes that differ by one base will leak reads between samples even when no hop occurs. Prefer a set built for that purpose, and record the sequences you actually used.

How a hop becomes another sample's read

Barcode hopping, also called index hopping, is the assignment of a read to an index pair it was not ligated with. The mode that has drawn the most attention sits on patterned flow cells that amplify by exclusion amplification. Free index primers left over from library PCR can anneal to a library molecule and be extended, stamping a new index onto a fragment that already had another identity. The new combination then forms a cluster. Non-patterned flow cells are less famous for this path and are not immune to index error from misreads or from mixed primers. The practical consequence is the same: a read lands in the wrong pile.

Free adapters and free primers are the fuel. Under-cleaned libraries, too much adapter at ligation, and PCR conditions that leave primer behind all raise the fuel. A cleanup that the size trace says still contains a dimer peak is a hopping risk as well as a cluster thief. The manufacturer's cleanup for that kit is the instruction to follow. This page will not invent a bead ratio.

Unique dual indexes change the consequence. Each sample is the only owner of its i7 and the only owner of its i5. A hop that swaps one end produces a pair that no sample owns. Demultiplexing can park those reads as undetermined. You lose a fraction of data, which is visible, and you do not silently donate reads to the neighbouring tube.

Combinatorial dual indexes are weaker because the sheet contains many crossings of a shared i7 list and a shared i5 list. Swap one end and you often invent a crossing that is another real sample. The demultiplexer is then doing what you asked. It has no way to notice that this pair was never ligated in one tube. Single indexes are weaker still. There is only one barcode to get wrong, and every barcode on the sheet is a real sample. A substitution or a hop is enough.

A blank library, prepared and indexed beside the samples, is the control that makes hopping visible. Reads in the blank after demultiplexing are not a microbiome you discovered in water. They are misassignment, contamination, or both. Judge the rate against the blank before you interpret a low-level taxon or a low allele fraction in a real sample. Low-level claims are exactly where hopped reads disguise themselves as biology.

Failures that survive a perfect-looking read quality

Read quality can be high on a misassigned read. The insert was sequenced well. The barcode was the part that lied. Filtering on Q30 will not remove hopping. It may enrich for it if the hopped clusters were bright.

Forcing index mismatches is a second way to launder the problem. Allowing two mismatches in a short index will assign near-misses, including some hops and some sequencing errors, to a real sample. Tighten the mismatch allowance and watch the undetermined fraction move. A sudden drop in assigned reads after tightening is information. It means the previous assignment was generous.

The i5 orientation trap is not hopping, and it looks similar from across the room because the sample files are empty. Instrument workflows differ in whether index 2 is read in the orientation printed on the tube or as its reverse complement. A sample sheet transcribed from a different instrument family sends the whole lane to undetermined. Reverse-complement the i5 column for the instrument that ran, demultiplex again, and only then conclude the prep failed. Write the instrument family on the sheet so the next person does not undo the fix.

A contaminated index primer stock stamps the same wrong barcode onto every library in a batch. Unique dual indexes do not save you if the primer you added was already mixed. Aliquot index primers. A shared working stock that has been opened through a dusty week is a common source of a barcode that appears where it was never assigned.

Swaps before indexing are invisible to every sheet. If the tubes were exchanged at extraction, each library is internally consistent and biologically wrong. Indexes prove that this library stayed with this barcode after the index PCR. They do not prove the nucleic acid was the donor named on the tube. Sample tracking is a lab process. The barcode is one column of it.

Infectious samples and a shared flow cell

Indexing does not reduce the hazard of the organism. Pooling happens after you have decided the libraries may share a bench. A sample that may be infectious stays under the institutional biosafety decision through extraction and library prep. This page is not that decision, and demultiplexed reads are not a diagnostic identification. Do not treat a hopped pathogen-like sequence in a blank as a clinical finding. Treat it as a library-assignment problem until the controls say otherwise.

Write the index scheme into the specification

When two laboratories share a run, the specification has to name the index layout in words a stranger can implement: unique dual index, the sequence of each index in the orientation the instrument will read, the mismatch allowance, and who keeps the sheet. A spreadsheet of sample names without sequences cannot be demultiplexed. A spreadsheet of sequences without the instrument family can be demultiplexed wrongly. Heat and courier delays matter here mainly because a paper sheet and a tube rack that travel separately get reconciled from memory. Send the sheet with the tubes, and keep a copy that does not depend on one inbox.

What to ask before the libraries are pooled

Ask which index design will be used, whether a blank is indexed beside the samples, what fraction of reads you should expect in undetermined, and which instrument workflow sets the i5 orientation. Ask for the sequences, not only the kit's well numbers, because well numbers from two kit lots are not a sequence. The genomics research pathway is the surrounding design. Reagent classes are in the genomics and sequencing catalogue.

A multiplexed genome project can be discussed against the whole-genome sequencing enquiry reference. If the biological question is one amplicon and one trace, the Sanger DNA sequencing enquiry reference avoids a multiplex entirely. Put the index table in the quote request. The habit of specifying assignment rules before the run, including what a blank is allowed to contain, sits with commissioning a sequencing or proteomics study.