Low-template DNA

When the Signal Becomes the Noise

When a sample contains very little starting material — a handful of cells on a door handle, a trace left by a fingertip on a knife — conventional amplification may produce a profile. Or it may produce fragments of one. The problem is that both outcomes can look, at first glance, like evidence.

Low-template work, sometimes called low-copy-number (LCN) analysis or touch DNA, pushes the polymerase chain reaction beyond the conditions under which it was validated for forensic use. Kary Mullis's original insight — that you could copy a target sequence exponentially through repeated heating-and-cooling cycles — was never designed with the extreme sensitivity that forensic laboratories now routinely attempt. Amplifying thirty or fewer copies of a DNA molecule produces results that are qualitatively different from amplifying the several hundred copies in a standard reference swab, and those differences carry direct consequences for interpretation.

What Goes Wrong, and Why

Three artefacts dominate the low-template landscape: drop-out, drop-in, and stutter.

Drop-out is the failure of a real allele to be detected. When only a few template molecules enter the reaction, the random variation inherent in early-cycle amplification can mean that one copy of a locus is copied efficiently and its partner is not. The result is a single peak where there should be two, or no peak at all. A profile that looks like that of a single contributor may actually carry the ghost of a second person whose alleles all dropped out. At the extreme, entire loci can fail, producing incomplete profiles that cannot be compared reliably against a database entry or a reference sample.

Drop-in is the complementary problem: a peak appears that belongs to nobody relevant. Environmental DNA — skin cells shed by casual contact with a surface, cells from a laboratory worker, a fragment that survived on the packaging — can be copied up to detectable levels alongside genuine sample. Because drop-in peaks tend to be low, some analysts apply a minimum peak-height threshold below which a signal is discarded, but that threshold is itself a judgment call, and setting it too high risks calling real alleles as drop-out.

Stutter — peaks one repeat unit shorter than a true allele, generated by polymerase slippage during amplification — appears in every STR profile, but at low template it becomes harder to distinguish from a genuine contributor. A stutter peak that would be unremarkable at 10 percent of its parent peak height becomes ambiguous when the parent is already marginal. The STR multiplex system was validated with robust template; its stutter ratios were characterised under those conditions and may not hold when the input drops by two orders of magnitude.

Stochastic effects is the collective term for all of this: when individual molecules rather than populations of molecules determine the outcome, results become variable run-to-run. The same extract, amplified twice, can yield different peak patterns. This is not contamination or error in the ordinary sense; it is a property of the physics. The term stochastic threshold describes the peak-height level below which these effects are considered to dominate the signal, below which a low peak carries no reliable genotype information. SWGDAM's published interpretation guidelines ↗ address how laboratories should set and apply this threshold, but the underlying validation is the laboratory's own responsibility, and practice has varied considerably.

CROSS-REF

Replication and Its Limits

One response to stochastic variability is to amplify the same extract multiple times and look for consistent peaks across runs — treating concordance as evidence of a true allele and treating peaks present in only one replicate as probable artefact. This approach, used in LCN regimes pioneered by the Forensic Science Service in the United Kingdom, increases sensitivity but introduces its own interpretive complexity: how many replicates are required, how many concordant peaks constitute confirmation, and what is done with the discordant ones. When the Forensic Science Service's LCN work came under judicial scrutiny in the case R v Hoey (Northern Ireland Crown Court, 2007), the judge found that the technique had not been adequately validated and that the profile evidence could not be relied upon — a documented instance of a court declining to accept a low-template result, as the judgment record reflects.

The United Kingdom's Forensic Science Regulator subsequently required that LCN laboratories demonstrate validation before continuing casework, and the Home Office commissioned an independent review. The wider lesson — that sensitivity is not inherently a virtue — has been slow to travel.

Transfer and Deposition

Low-template results are inseparable from the problem of how DNA arrives at a surface. Secondary transfer, the process by which DNA moves from person to object to different object without any direct contact between the depositor and the final surface, is measurable in laboratory studies. Shed cells and DNA-bearing material relocate easily. When a sample is so small that only thirty copies of a locus are available for amplification, the question of how those copies arrived becomes unanswerable by the chemistry alone.

This is the interpretive gap that low-template evidence opens. A full-template profile from a blood stain provides strong evidence of identity; the same numeric result derived from a trace on the outside of a bag does not establish that the contributor ever touched the bag. The result may be real — the chemistry may have worked — and the person's DNA may genuinely be there — but the route from deposition to detection carries too many unresolved steps for identity alone to carry the evidential weight it would carry in a higher-template context.

Probabilistic genotyping and Its Scope

The response to low-template complexity within the discipline has been to move interpretation toward probabilistic genotyping software, which models the probability of observing a given peak pattern given a particular set of genotype assumptions, including the probability of drop-out and drop-in at observed heights. Tools such as STRmix, developed in New Zealand with NIST validation support, explicitly incorporate stochastic parameters. They are a genuine advance over binary match-or-exclusion calls on incomplete profiles.

But their output is only as trustworthy as the model's assumptions about drop-out rates, stutter ratios, and mixture proportions — parameters that themselves shift as template quantity falls. The National Institute of Standards and Technology ↗ has published studies examining the performance of these tools across laboratories, and the variation in likelihood ratios assigned to identical profiles by different software implementations remains a live concern in the field. A ratio computed under optimistic parameters is not the same kind of claim as one computed under conservative ones, even if both appear as a single number in a report.

The honest characterisation of low-template evidence is this: the technique can detect DNA where older methods would return no result, and that is sometimes exactly what is needed. But the sensitivity that makes it useful also makes it the point at which a result becomes, in part, a property of the method rather than purely a property of the sample. Courts and examiners who treat a low-template likelihood ratio with the same confidence as a high-template one are comparing things that are not the same.

Three artefacts dominate the low-template landscape: drop-out, drop-in, and stutter.