Index Hopping: How Reads End Up in the Wrong Sample, and How to Detect It

Index hopping is a sequencing artifact in which a small fraction of reads from a multiplexed run get assigned to the wrong sample’s index (barcode) during demultiplexing. On the platforms most exposed to it, the effect is usually low-level — often well under 2% of reads — but on a run where every sample matters (rare-variant calling, low-input single-cell work, forensic or clinical samples pooled together), even that level of cross-sample contamination can produce a false-positive variant call or an artificially “detected” transcript that never belonged in that sample.

What index hopping actually is

In a multiplexed sequencing run, many samples are pooled into one lane or flow cell, each tagged with a short index sequence (or a pair of indexes, i5 and i7) so that reads can be sorted back to their sample of origin after sequencing — a step called demultiplexing. Index hopping happens when a read carries the correct insert sequence for Sample A but gets tagged, during cluster generation, with an index that belongs to Sample B. The read is then demultiplexed into the wrong sample’s FASTQ file. See FASTQ Format Explained for what that output actually looks like and how index information is encoded in the read header.

This is distinct from ordinary sequencing or index-read errors, which produce reads that fail to match any index cleanly and get discarded as “undetermined.” Index hopping is worse in one specific way: the hopped read matches a real index belonging to a different real sample, so it looks like valid data and is silently included in that sample’s results unless something specifically screens for it.

Why it happens: ExAmp chemistry and patterned flow cells

Index hopping is a chemistry-driven artifact, not a software bug or a labeling mistake. It’s most pronounced on Illumina platforms that use patterned flow cells with exclusion amplification (ExAmp) chemistry — the HiSeq 4000, NovaSeq 6000, and NextSeq 2000 families are the commonly cited examples. In ExAmp chemistry, the DNA library fragments and the amplification primers are mixed together in free solution before cluster generation, rather than the primers being exclusively surface-bound on the flow cell as in older bridge-amplification chemistry (used on the MiSeq, for instance).

That free-solution mixing is the mechanism: if library preparation leaves any free-floating, unligated index primers or adapters in the pooled library — which is common to some degree with any prep, but worse with incomplete cleanup — those free adapters can anneal to and prime the wrong template fragment during cluster generation. The result is a cluster that reads out the correct insert sequence but the wrong index. Because MiSeq’s bridge-amplification chemistry doesn’t mix primers and templates in solution the same way, it shows measurably lower index-hopping rates (commonly cited below 0.05%) than patterned, ExAmp-chemistry instruments (commonly cited in the 0.1%–2% range, with PCR-free library preps running toward the higher end of that range since PCR-free workflows tend to carry more residual free adapter into pooling).

How to detect it

Index hopping is detected at the demultiplexing stage, not by inspecting individual reads. The diagnostic signature to look for is unused index hopping: unexpectedly low but nonzero read counts assigned to index combinations that were never actually used in your sample pool. If your pool used indexes A, B, and C on samples 1, 2, and 3, a well-behaved run should show essentially zero reads for combinations like “A paired with C’s partner index” if that pairing was never assigned to a real sample. A small but consistent number of reads showing up under those unused combinations — rather than zero — is exactly what hopped reads look like, because a hopped read effectively pairs one sample’s index with a fragment or partner index from another sample.

Practically, this means:

• Generate a demultiplexing report that includes read counts for all index combinations, not just the ones you assigned to samples — most demultiplexing tools (e.g., Illumina’s bcl2fastq/bcl-convert) can report this if configured to do so.
• Look specifically at the unused/unassigned combinations. A near-zero count across the board is reassuring; a low but clearly nonzero and roughly consistent count across many unused combinations points to index hopping rather than random noise.
• Estimate a rough hopping rate by comparing reads in unused combinations to reads in the used, expected combinations — this gives you a sense of what fraction of your “real” sample data may also be contaminated by hopped reads from other samples, even though those hopped-in reads are harder to isolate directly.

This check costs nothing beyond reading a report you likely already generate, and it’s the only reliable way to know whether a given run’s index-hopping rate is negligible or high enough to affect downstream calls — sequencer model and chemistry tell you the risk profile, not the actual rate on your specific run.

Mitigation: unique dual indexing, not combinatorial indexing

The most effective mitigation is switching from combinatorial (or single) indexing to unique dual indexing (UDI). The distinction matters:

• Combinatorial dual indexing reuses a limited set of i5 and i7 indexes across many samples — e.g., 12 i5 indexes and 8 i7 indexes combined to uniquely tag 96 samples. Any given i5 or i7 sequence is shared by multiple samples in the pool.
• Unique dual indexing assigns every sample a combination where both the i5 and the i7 index are unique to that sample and used nowhere else in the pool.

Because combinatorial indexing reuses individual indexes across samples, a hopped read only needs to acquire the wrong index on one end to be misassigned to another real sample that shares that index — and that misassignment is invisible, because the resulting combination is still one of the “expected” ones in the pool. With unique dual indexing, a hopped read would need to independently acquire the wrong index on both ends to land on another real sample’s combination — a much lower-probability event. Any read with a hopped index on only one end produces a combination that isn’t among the pool’s assigned pairs, so it shows up as exactly the “unused combination” signature described above and can be identified and discarded during demultiplexing rather than silently contaminating a sample’s data.

Additional, complementary steps: minimize residual free adapter and unincorporated index primer carried into pooling (tighter bead-cleanup or size-selection steps in library prep — see the NGS library prep kit guide for what to evaluate in a kit’s cleanup chemistry), pool at appropriate dilution rather than over-concentrating libraries before loading, and treat any nonzero unused-combination count in your demultiplexing report as a standing QC check for every multiplexed run, not a one-time investigation.

Frequently asked questions

Is index hopping the same as an index read error?

No. An index read error produces a read that doesn’t match any assigned index cleanly and is discarded as undetermined. Index hopping produces a read that matches a real index belonging to a different real sample, so it’s included in that sample’s data unless it’s specifically filtered out.

Does index hopping affect all Illumina sequencers equally?

No. Platforms using patterned flow cells with ExAmp chemistry (e.g., HiSeq 4000, NovaSeq 6000, NextSeq 2000) show measurably higher index-hopping rates than platforms using non-patterned flow cells with bridge amplification (e.g., MiSeq), because ExAmp chemistry mixes library fragments and amplification primers in free solution during cluster generation rather than keeping primers surface-bound.

Does unique dual indexing eliminate index hopping entirely?

No — it doesn’t stop the underlying chemistry from producing hopped reads. What it does is make hopped reads identifiable: a read that hops on only one index no longer matches any of the pool’s assigned combinations, so it can be flagged and removed at demultiplexing instead of silently landing in the wrong sample’s data.

Does index hopping matter for a low-plex, high-input run?

It matters least there. Index hopping is most consequential when sample input is low, sensitivity requirements are high (rare-variant calling, minimal residual disease, single-cell work), or when samples with very different expected signal are pooled together, since a small fraction of cross-contaminating reads is more likely to change a result.

Related concepts

Index hopping is one of several run-level artifacts worth checking alongside general sequencing QC — see PhiX control for the standard run-quality spike-in, and Unique Molecular Identifiers (UMIs) for a related but distinct tagging strategy used to distinguish true molecular duplicates from PCR duplicates rather than to prevent index misassignment. For broader multiplexed-experiment design considerations, see RNA-seq: Experimental Design Through Analysis.