The review, led by Shouqing Hou, Hongru Ma, Xuye Lang and Yuan Yao, examines transcription termination from bacteria through archaea to human cells. Its central insight is deceptively simple: a terminator is not one thing. It can mean a predicted hairpin, a mapped RNA end, a polyadenylation site, a region of readthrough, or a place where polymerase actually lets go of the DNA. These are different measurements of different events, and treating them as interchangeable has muddied decades of mechanistic comparison. By separating the signal from the assay that detects it, the authors hope to make cross-species comparisons meaningful again.

To do this, the team introduces the concept of a signal grammar, a system-specific set of rules that connects sequence elements, RNA structures, molecular readers, conditional gates, and kinetic windows to the final outcome of polymerase release. A bacterial hairpin followed by a U-tract, for instance, is a compact grammar: the hairpin must fold before the U-tract is fully transcribed, and the weak RNA-DNA hybrid that follows destabilizes the elongation complex just enough for the transcript to slip free. In contrast, the human RNA polymerase II system uses a polyadenylation signal, auxiliary sequence elements, and a downstream window where an exonuclease called Xrn2 chases the polymerase to force release. Same problem, radically different solutions.

The authors formalize this idea with five operational dimensions: elements, syntax, readers, gates, and kinetics. Elements include sequence motifs, RNA structures, RNA accessibility, and processing sites. Syntax describes their order, orientation, and spacing. Readers are the molecular machines that interpret the signal, from RNA polymerase itself to factors like Rho, aCPSF1, or the cleavage and polyadenylation machinery. Gates are conditional switches such as RNA folding, translation status, or chromatin state. Kinetics describe the race between signal recognition and polymerase escape. Crucially, the authors stress that a list of features is not a grammar; only the rules connecting them, including co-requirement, spacing dependence, and alternative routes, give the system predictive power.

Bacterial intrinsic termination, the best-studied case, illustrates how a grammar works in practice. Cryo-electron microscopy structures of Escherichia coli termination complexes show a sequence-induced pause, hairpin nucleation after the nascent RNA emerges from the polymerase exit channel, and later intermediates associated with transcript release. Local re-pairing of template and non-template DNA from the upstream edge of the transcription bubble can shorten the RNA-DNA hybrid, facilitating hairpin completion and RNA release. If the hairpin does not form before the polymerase escapes its kinetic window, transcription simply continues. Quantitative studies using synthetic terminator libraries have shown that efficiency spans a continuum rather than a binary on-off switch, shaped by hairpin stability, U-tract length, loop geometry, and downstream sequence composition.

But even the supposedly factor-independent intrinsic system turns out to be context-dependent. In Bacillus subtilis, the elongation factor NusA stimulates many weak or non-canonical intrinsic terminators, and NusG acts both alone and cooperatively with NusA at a large fraction of them. These findings blur the old distinction between intrinsic and factor-dependent termination, suggesting that accessory proteins can extend the functional range of imperfect hairpin-U-tract modules by slowing elongation or stabilizing hairpin formation. Cross-host studies in Halomonas bluephagenesis, Corynebacterium glutamicum, and E. coli further show that terminator performance and portability must be tested in the intended biological context, not assumed from sequence alone.

Rho-dependent termination, the other major bacterial route, operates on an entirely different grammar. Rather than a compact stop element, Rho requires an accessible, often C-rich RNA region called a rut site, productive loading and ATP-dependent translocation, and a downstream polymerase pause or slow zone. Translation status acts as a major conditional gate: actively translating ribosomes can shield nascent RNA from Rho, whereas stalled or delayed translation exposes it. Structural work on E. coli Rho-NusG complexes has visualized a closed Rho ring with RNA threaded through its central channel, bridged to polymerase by NusG, poised for translocation. Genome-wide mapping in E. coli and Mycobacterium tuberculosis reveals that Rho suppresses pervasive antisense transcription and enforces transcription-translation coupling across broad genomic territories.

Archaeal termination adds yet another layer of complexity. U-rich tracts can promote intrinsic-like release without requiring an upstream hairpin, but they can also recruit cleavage factors. Recent work on the CPSF73-family ribonuclease aCPSF1 in Methanococcus maripaludis showed that its depletion causes genome-wide termination defects, and that its N-terminal KH domains recognize terminator U-tracts while its nuclease activity is required for high efficiency. Cryo-EM structures of the Thermococcus kodakarensis FttA-Spt4/5 complex reveal how a ribonuclease can be physically coupled to the elongation complex, with Spt5 bridging the nuclease to polymerase. These findings suggest that archaea use both intrinsic-like and cleavage-linked routes in parallel, rather than occupying a simple evolutionary midpoint between bacteria and eukaryotes.

In eukaryotes, the polyadenylation signal is only the beginning of the story. PAS recognition and cleavage establish RNA processing, but polymerase release typically occurs further downstream, within a broad termination window shaped by Xrn2/Rat1 exonuclease activity, allosteric changes in the elongation complex, polymerase kinetics, and chromatin state. Alternative polyadenylation, readthrough transcription, and non-canonical routes such as Sen1-mediated and DNA-directed termination add further complexity. The authors emphasize that PAS strength reflects hexamer identity, flanking motifs, spacing, and cellular factor availability, not the AAUAAA sequence alone. Chromatin can even shift polyadenylation site usage by altering polymerase elongation rate, linking nuclear organization directly to transcript architecture.

The review concludes with a call for claim-matched validation in the age of machine learning. Computational tools for terminator prediction, from RhoTermPredict to deep learning models of polyadenylation, inherit the limitations of their training labels. A poly(A)-site model is not an RNAPII-release model, and an intrinsic-terminator classifier is not a Rho predictor. The authors propose an iterative design-build-test-learn cycle in which training datasets retain experimentally supported positives, weak effects, discordant assays, hard negatives, and failed host transfers. Generative models coupled with terminator-strength predictors have already produced synthetic terminators with efficiencies above 90 percent in E. coli, but cross-host portability remains an empirical outcome, not a computational guarantee. As the authors put it, the central bottleneck is not a shortage of candidate motifs or models, but the comparison of results generated by different measurements, and progress will depend on claims that remain matched to the observable and biological context that support them.

Subject of Research: Cross-domain signal grammars of transcription termination

Article Title: Cross-domain signal grammars of transcription termination

Article References: Hou, S., Ma, H., Lang, X., & Yao, Y. (2026). Cross-domain signal grammars of transcription termination. Advanced Biotechnology, 4(4), Article 42. https://doi.org/10.1007/s44307-026-00134-7

Image Credits: AI Generated

DOI: 10.1007/s44307-026-00134-7

Keywords: transcription termination, RNA polymerase, intrinsic terminator, Rho-dependent termination, archaea, polyadenylation signal, RNAPII, signal grammar, machine learning, terminator engineering, gene expression, synthetic biology