Skip to content
rnaprog
This methodology article is published in English; the interface and SEO metadata are localised. Read the original English version.

siRNA off-targets live in the seed: why position 2–8 decides your specificity

Most siRNA off-target effects come from a 7-nucleotide seed match, not from the guide as a whole. What that means for candidate ranking, why internal seed counts are a weak proxy, and when you need a real transcriptome search.

rnaprog editorial 7 min read

A 21-mer siRNA has roughly four million possible 7-mers in a human transcriptome. If your guide pairs perfectly at positions 2–8, it can silence genes you never intended to touch — and the effect will not show up in your primary knockdown assay.

Why the seed matters more than the rest

Argonaute-loaded guide strand pairs with mRNA in two phases. Positions 2–8 (the seed) must pair for the initial recognition step; the rest of the guide stabilises the interaction but tolerates mismatches. That asymmetry is why a guide with perfect complementarity to your target can still bind hundreds of other transcripts at their 3′ UTRs, where seed matches are enriched.

The practical consequence: two guides with identical GC and identical predicted on-target activity can differ tenfold in off-target burden, and the difference is almost entirely in the seed.

What we count, and what we do not

Our candidate table reports an off-target count. Be clear about what it is: the number of times the guide’s 7-mer seed appears elsewhere within the target region you supplied. That is an internal-repeat check, useful for spotting guides that will hit multiple sites in your own amplicon or transcript, and useless as a specificity estimate.

A real specificity assessment requires searching the guide’s seed against a reference transcriptome (or genome, for nuclear targets) and counting 3′ UTR matches. That needs an aligner and a reference index — tens of gigabytes, minutes of compute. It is not something a design run can do in two seconds, and any tool claiming otherwise is doing something weaker than it implies.

So: use the internal count to avoid obvious repeats. Do not read it as “this guide has zero off-targets”.

Ranking signals that do carry weight

When you cannot run a transcriptome search, these composition rules correlate with real behaviour well enough to be useful:

  • Seed GC between 30% and 60%. Very GC-rich seeds bind too many unintended 3′ UTRs; very AT-rich seeds are weak on-target.
  • A or U at position 19 of the guide (position 1 of the antisense 5′ end). The classic RISC asymmetry rule: the strand whose 5′ end is less stably paired becomes the guide. U or A at the 5′ antisense end biases loading correctly.
  • G or C at position 19 of the sense strand — the mirror of the same rule.
  • No immune-stimulatory motifs. GUCCUUCAA, UGUGU, UUCG and similar motifs trigger innate sensing in some cell types. If you see them, expect an interferon response that may confound your readout.
  • No poly(T) ≥ 4 if the guide is expressed from a U6 cassette rather than delivered as an oligo.

Why a pool, not a single duplex

Twelve candidates at 3 credits each is 36 credits — cheaper than one failed knockdown experiment. The pool gives you a distribution: some candidates will sit at 45% GC with a clean seed and no immune motifs, and those are the three worth ordering. Ordering three duplexes instead of one costs little more and removes the “did it fail because the design was bad or because the biology was” question from your first experiment.

Practical note on the target region

Supply the transcript region you actually intend to silence, not the whole mRNA. If you paste 8,000 nt, the tool will return 12 candidates scattered across the transcript — some in the 5′ UTR, some in the CDS, some in the 3′ UTR — and their behaviour will differ for reasons unrelated to sequence quality. Pick a region (typically 100–500 nt around the site you care about), then design.

If you do not know where to start, the CDS is a reasonable default: it is usually well covered by transcriptome annotations, and silencing there avoids the 3′ UTR’s microRNA-binding competition.

Первый запуск начинается с 300 бесплатных кредитов

Хватит на 6 mRNA-конструкций или пул из 12 siRNA, включая QC и экспорт. Без карты.

300 credits · No card required · You keep the rights to your sequences