The investigators designed the recordanelle as a searchable intracellular archive. Its mitochondrial anchor consists of the first 70 amino acids of human TOM20, which tethers the structure to the outer mitochondrial membrane. An assembler peptide inspired by proteins that form biomolecular condensates creates a local compartment, while a final RNA-binding domain determines which molecules are collected. By swapping that domain, the researchers could target engineered RNAs, messenger RNAs with poly(A) tails or other RNA classes. Because the recorded material remains associated with a fraction that can be isolated and sequenced, a cell can be analyzed at a later time without requiring continuous observation of every transcriptional event.

Initial tests showed that the components must be matched carefully. Constructs carrying the MS2 and PP7 coat proteins recruited reporter RNAs bearing the corresponding stem loops, while swapping the coat protein redirected capture to the other reporter. Depending on the assembler, enrichment ranged from threefold to 11-fold, and the ratio between targeted and untargeted reporter RNA ranged from 1.2-fold to sixfold. Some full-length FUS and EWSR1 variants also attracted RNA without a specific binder, whereas the disordered-region assemblers TAF15N and FUSN showed less obvious nonspecific capture. In HEK293T cells expressing a FUS-based design, RNA sequencing detected no differentially expressed genes across the whole cell relative to a control lacking the RNA-binding domain, suggesting that the construct itself did not measurably disrupt the cellular transcriptome under the tested conditions.

The researchers next replaced stem-loop-specific proteins with binders that recognize features common to eukaryotic messenger RNAs. These included proteins that attach to the poly(A) tail, a 5′ cap-binding protein and several export-associated or nonspecific RNA binders. PABPC1- and PABPC4-based recordanelles produced consistent recruitment of reporter and endogenous RNA, reaching as much as 15-fold enrichment in mitochondrial fractions. A larger bulk RNA-sequencing experiment with synthetic spike-ins then assessed global capture. At a threshold of more than 1.5-fold enrichment and a false discovery rate below 0.01, recordanelles represented 64% of expressed genes when paired with FUSQG and 90% when paired with TAF15N. Nonbinding controls enriched fewer than 0.1%. These percentages refer to the fraction of transcript species represented in the record, not the percentage of all RNA molecules physically bound; the authors interpret the results as sampling of a substantial part of the cytoplasmic RNA pool.

Protection, rather than recruitment alone, is the central feature that distinguishes the optimized design. In reporter cells carrying a doxycycline-inducible mCherry::ms2 transgene, messenger RNA rose during a 20-hour treatment and then returned toward baseline after the stimulus was removed. Recordanelles based on MCP, PCP, LARP1 or LARP4A captured the engineered RNA but did not protect it better than a nonbinding control, with whole-cell transcript recovery near 0.07. PABPC1- and PABPC4-based designs raised one-day recovery to roughly two- to fourfold, and the reporter remained detectable for at least 48 hours. The difference suggests that direct attachment to the poly(A) tail may help shield RNA from degradation, although the study did not establish the precise protective mechanism.

Two TAF15N-based versions were carried forward. RecordanelleV1 used PABPC4 lacking its C-terminal region, whereas recordanelleV2 used a point-mutated version, PABPC4ΔC161A, that could record the transient reporter without changing its initial expression level. By two days after doxycycline removal, both versions retained significant mCherry::ms2 signal in whole cells and in the mitochondrial fraction, while the MCP control did not. The investigators selected V2 for subsequent work because V1 sometimes produced an apparent increase in recorded interferon-stimulated genes between the stimulation and recovery measurements, possibly reflecting translation of stored transcripts. That behavior was not observed with V2, making it the more consistent recording architecture in these assays.

The strongest test used endogenous immune signaling. HEK293T cells were exposed to 60 units per milliliter of interferon beta for 20 hours, after which the cytokine was removed. At the time of stimulation, interferon-stimulated genes such as IFIT1, IFI44L, OAS3 and IFIT3 showed the expected response in every construct condition. Two days later, the signal had decayed in control cells and in cells carrying the nonbinding MCP recordanelle, but remained detectable in cells carrying recordanelleV2. Bulk RNA sequencing found that the V2 cells retained all differentially expressed genes detected at the peak of stimulation, corresponding to 100% recall in that experiment, and the recovered pattern captured biological processes associated with the earlier interferon response. In nonpassaged cells, the signal persisted for at least five days; when cells were passaged, it remained detectable for three to four days but was lost from day five onward, consistent with dilution or turnover during proliferation.

Single-cell RNA sequencing showed that the same information could be recovered without first purifying the mitochondrial fraction. At the peak of interferon exposure, the analysis identified a distinct cluster characterized by interferon-stimulated genes, which the researchers called the IFNB response cluster. Two days after cytokine removal, that cluster disappeared from control cells but remained substantially represented among interferon-treated cells carrying recordanelleV2. A module score based on interferon-stimulated genes similarly fell to baseline in stimulated control cells and remained elevated in V2 cells. This is significant because it demonstrates that the recorder preserves cell-specific transcriptional history at single-cell resolution, allowing researchers to identify which cells experienced a transient signal even after the conventional cellular RNA profile has returned to baseline.

Functional measurements indicated that preserving RNA did not necessarily prolong the inflammatory response at the protein level. Quantitative proteomics showed robust induction of several interferon-stimulated proteins at the peak of treatment, but almost all returned to baseline two days after cytokine removal, with recordanelleV2 and the MCP control indistinguishable except for two proteins. The recordanelles also did not significantly alter ATP production, substantially increase apoptosis, measurably disrupt mitochondrial membrane potential or significantly change aerobic metabolism in the assays conducted. Proteomic analysis detected no distinct alteration in respiratory-chain proteins, and the overall proteome of V2-expressing cells showed no significantly affected biological process relative to mock transfection. Nevertheless, the authors emphasize that any transcriptional recorder is itself a perturbation, potentially adding transgene burden, occupying mitochondrial space, binding RNA unnaturally, depleting transcripts from the cellular pool and imposing energetic costs for storage.

The study therefore presents recordanelles as a promising bridge between high-resolution snapshots and continuously maintained cellular histories. The strongest results came from a single fusion protein in cultured human cells, and the work was published in Nature Biotechnology as an open-access report. Future versions may add enzymes that encode lineage or temporal order, while the current method could also support live-cell profiling of RNA-binding specificity or deliberately sustain selected signaling pathways. For now, the central advance is that a living cell can retain a broad, sequence-readable account of earlier RNA activity, including transient endogenous responses, without requiring continuous imaging or repeated destructive sampling. Whether the approach can be adapted for primary cells, tissues or organisms, and how its perturbations behave outside engineered HEK293T cultures, remains to be tested.

Subject of Research: Development and laboratory testing of a synthetic organelle for transcriptome-wide RNA recording in human cells

Article Title: Recordanelles for transcriptome-wide single-cell RNA recording in human cells

Article References: Reinkemeier, C. D., Manero-Carranza, M., Farouni, R., Filice, F., Klotz, A., Schneider-Heieck, K., Andermann, T. L., & Platt, R. J. (2026). Recordanelles for transcriptome-wide single-cell RNA recording in human cells. Nature Biotechnology. https://doi.org/10.1038/s41587-026-03327-6

Image Credits: AI Generated

DOI: 10.1038/s41587-026-03327-6

Keywords: recordanelles, single-cell RNA sequencing, transcriptome recording, synthetic organelles, RNA preservation, interferon signaling, mitochondrial condensates, HEK293T cells, transcriptomics, transcriptome-wide, single-cell, recording