The Promoter You Did Not Put There
7 min read · Updated October 30, 2026
A sigma-70 promoter is two short boxes about seventeen bases apart. That is a weak enough pattern that ordinary coding sequence contains it, which means your construct almost certainly contains promoters you did not design — some of them pointing backwards.
Here is what a scan of a perfectly ordinary cassette returns, and the one input mistake that makes the answer useless.
Finding a promoter is not a finding
Scan a 409 nt cassette — a strong constitutive promoter, a 5' UTR, a short coding sequence and a terminator — and 235 candidate promoters are scored, from 664 start-position states. In a 529 nt version the count is 319.
So the output of a promoter scan is a ranking, never a yes or no. The question it answers is which of these is strongest and which way does it point, and a tool that reported found a promoter would be reporting a property of the sigma-70 consensus rather than of your construct.
The strongest promoter in the cassette was antisense
In that 409 nt scan, the highest-rate candidate sits on the reverse strand with its transcription start at position 82, with a −35 of TTGAGA and a −10 of TAGAAT separated by 17 bases. Four of the eight strongest were antisense.
A promoter pointing backwards inside a cassette transcribes antisense RNA across everything upstream of it. The symptom is reduced expression of the gene you care about, and it looks like a cloning problem, a plasmid-loss problem or bad luck — anything except what it is. No forward-only scan will ever show it to you.
That is the argument for scanning both strands by default rather than as an option somebody has to know to switch on.
A promoter at the first base of your input cannot be scored
The same cassette begins with BBa_J23119, the consensus Anderson promoter, and in that scan it does not appear among the top 100 of 235 candidates. Prepend 120 bases of unrelated sequence, changing nothing else, and it is suddenly rank 1 — transcription start at 156, −35 TTGACA, −10 TATAAT, spacer 17, with a predicted rate of 39,305 against 4,015 for the best antisense candidate in the same molecule.
The reason is that the model reads sequence upstream of the −35, where the UP element binds, and a promoter sitting at position 1 has none. The scan does not refuse; it scores what it can see, and what it can see is not your promoter.
So give a scan context on both sides. A cassette pasted exactly from its first base is the commonest way to get an answer that is confidently about the wrong thing.
The energy terms are the actionable part
Each candidate comes with the free-energy decomposition the model is built from: the −35 and −10 boxes, the spacer, the discriminator, the extended −10, the UP element and the initial transcribed region. In the worked J23119 case the total is −2.79, with the −10 contributing −1.45, the −35 −0.99 and the spacer −0.09, while the UP element and discriminator work against it at +0.44 and +0.20.
That is more useful than a rate on its own, because it says which part to change. A weak promoter whose penalty is in the discriminator is a different edit from one whose −35 is two bases off consensus.
What the rate is and is not
The number is a model's estimate, fitted to in vitro transcription of thousands of sigma-70 promoters. Against the two independent in vivo datasets the authors tested, it explains R² = 0.45 and 0.60 — so the ranking between two promoters is usable and the absolute rate is not something to plan a yield around.
The widely quoted R² = 0.80 is a held-out tenth of the authors' own in vitro data, and a promoter in a cell is the in vivo case. Between a third and a half of the variance there is unexplained.
It is also sigma-70 only, and E. coli only. Another sigma factor, another organism, or a promoter under activator or repressor control is outside what the model knows — and it says nothing at all about mRNA stability or translation, which are the next two things that decide whether a transcript becomes protein.
Frequently asked questions
What is a cryptic promoter and why does it matter?
Sequence that matches the promoter consensus well enough to initiate transcription without having been designed to. Inside a construct it produces transcripts nobody intended, and when it points backwards it produces antisense RNA across the genes upstream of it — which reads as poor expression rather than as a sequence problem.
Why scan the reverse strand for promoters?
Because a promoter pointing backwards is still a promoter. In a worked 409 nt cassette the highest-scoring candidate of 235 was antisense, and four of the top eight were. A forward-only scan cannot report any of them, and the symptom they cause looks like a cloning problem.
My promoter does not show up in the scan. Why?
Most often because it sits at the very start of the input. The model reads sequence upstream of the −35 box, so a promoter at position 1 is scored without the context it needs — in a worked example, adding 120 bases of upstream padding moved the construct's own promoter from outside the top 100 to rank 1. Always include flanking sequence.
How accurate is a predicted transcription rate?
Good enough to rank, not to plan with. Against the two independent in vivo datasets its authors tested, the model explains R² = 0.45 and 0.60; the commonly quoted 0.80 is from their own in vitro data. Use it to compare promoters in the same construct, and measure anything the yield depends on.
Related references
Related tools
Find every sigma-70 promoter in a region on both strands, with the free-energy terms behind each transcription rate — then build a library of promoters spanning a range.
Build a multi-gene operon from a promoter, coding sequences and a terminator, then scan the assembled molecule for the internal promoters, ribosome binding sites and terminators that the joins created.
Predict the translation initiation rate at every start codon, and design a 5' UTR to hit a target expression level.
Scan a coding sequence for premature stops, cryptic RBS/polyA signals, unwanted restriction sites, GC extremes and repeats.
Auto-detect promoters, tags, origins and resistance markers in a plasmid, then run a deep scan against pLannotate's full feature databases.