Science1 distinct publisher3 min readPublished
A repurposed retroviral shell protein lets a culture report its own gene activity again and again instead of once, so a single well becomes its own before-and-after control, at population-average resolution.
The Scientist · Science desk

Compiled by The ScientistSomething wrong?How this is made
The readout leaves the sample alive, and that changes the unit of comparison. In a destructive design, every timepoint is a different plate, because reading the transcriptome means killing the cells to get at their RNA [1]. The difference you measure between hour zero and hour 48 therefore carries whatever separated those two plates before anyone touched them. When the same population reports repeatedly into its medium, baseline and post-perturbation samples come from one well, and the contrast is paired [3].
The bookkeeping follows from that. Take a six-timepoint, four-condition time course run in triplicate. Destructive sampling needs 6 x 4 x 3 = 72 cultures, since each reading spends one. Self-reporting needs 4 x 3 = 12, and the ratio is just the number of timepoints [14]. The specific numbers are an illustration; the ratio is structural.
The mechanism is lifted from retroviruses, which package their RNA genomes in protein shells to move between cells [5]. The team engineered mammalian cells to express a retroviral structural protein that sits in the membrane, recruits cellular RNA as well as viral RNA, wraps it in a shell and buds off into the surrounding liquid [4]. Sample the medium, isolate the RNA, sequence [3]. What matters operationally is that the apparatus is encoded in the cell rather than bolted to the outside of it. Co-first author Mohamad Najia argues that this molecularly encoded route could be more broadly enabling for an average life science lab than robotics or mechanical biopsies, particularly for time-dynamic questions [12]. Senior author Paul Blainey and first author Jacob Borrajo say they chose the harder molecular path deliberately, taking the ease of adoption of CRISPR tools as the model [13]; Blainey's description of the older options is stabbing cells or cutting pieces off them [7].
The thing this account does not tell you is how faithfully the RNA inside the particles tracks the RNA inside the cell, gene by gene. No capture-efficiency or bias figures appear in the phys.org write-up, nor any report of whether cells constitutively budding virus-like particles behave like cells that are not [16]. That second one is the control question a reviewer would ask first: the sensor is built into the specimen. Resolution is the other boundary. The medium sample yields the transcriptome of a cell population [15], so this does not do what single-cell sequencing exists to do. The one heterogeneity result described is coarser than that: in a co-culture of two human cell types, tags on the particles let the two signals be told apart during analysis [9].
Where the design argument is strongest is the sample you cannot afford to take apart. On spheroids of human endothelial cells, the team captured short-term transcriptional changes after biochemical stimulation without dissociating the spheroid [10], and with Linda Griffith's lab at MIT they applied it to organ-on-a-chip devices built to mimic organ physiology [11]. For a screen whose question is trajectory rather than variance, the paired design is available to any lab that can get the construct into its cells [8]. For a question about which cells in the dish are diverging, dissociation is still the assay.
Ranked by verification strength, evidence, and original report placement.
Existing methods for measuring a cell's transcriptome rely on killing the cell to access the RNA inside it, and offer only a one-time snapshot.
Researchers at the Broad Institute and MIT invented a 'cellular self-reporting' approach that lets living cells share their own transcriptomes without being killed; the live-cell transcriptomic method is described in the journal Cell and relies on virus-like particles that cells use to package and deliver RNA to their culture medium.
Scientists can sample the medium to isolate the RNA and sequence it, and can do this repeatedly to reveal how gene activity in the same cell population changes as the cells mature or respond to perturbations.
The team engineered mammalian cells to express a retroviral structural protein that can encapsulate viral RNA and also the cell's own RNA; integrated into the cell membrane, the protein recruits cellular RNA, forms a shell around it to create a virus-like particle, and buds off from the membrane into the liquid medium.
The team took inspiration from retroviruses, which evolved the ability to package their RNA genomes in protein shells in order to spread from one infected cell to another.
Study senior author Paul Blainey is a core member of the Broad Institute and a professor of biological engineering at MIT, and said the concept was 'complete science fiction' when the work started.
Distinct publishers with included, body-backed reporting in this cluster.
phys.org
1 article · September 4, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
science
Cells hoard polyamines to keep iron quiet, Whitehead team reports in Cell1 distinct publisher
science
A gene-ranking tool that disagrees with GWAS, and 21 asthma genes to argue about1 distinct publisher
product
MIT's magnet trick makes correlated microwave signals without the cryostat1 distinct publisher
build
Proptech AI pilots stall at the read path out of the system of record1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One peer-reviewed anchor, read only through its own press office
There is a real checkable object at the bottom of this story: a Cell paper with a DOI, named first and senior authors, and a specific list of systems it was tried in. What reaches us is the institution's summary of that paper, and the two measurements a method like this stands or falls on, fidelity of the captured RNA and the behaviour of cells making the shell protein, are not in it.
The inventing lab, plus one collaborator down the corridor
Two sets of hands appear in this reporting: the Blainey lab, and Griffith's group at MIT whose chips the method was run inside. The closing line hopes other labs will try it, which is a fair reading of where uptake stands. No plasmid deposit, reagent source, or licensing path is named, so a lab persuaded by the story has nothing to act on yet.
Range foregrounded, accuracy unaddressed
Nothing in the copy is overclaimed outright, and the release is candid that resolution is population-level and single cells are still future work. The overstatement is one of proportion: paragraph after paragraph on how many systems it worked in, and not a line on how closely the RNA in the supernatant matches the RNA in the cell, which is the number that decides whether any of the applications hold.
The institution reporting on itself, quoting itself
This is MIT News copy about MIT and Broad work, republished by phys.org with the credit line attached. Every quote comes from an author; two of them make the case that ordinary labs will be able to run it, and one makes the case for long-horizon, high-risk funding. No methods researcher outside the paper is asked whether a population-average live readout answers the questions they would want it for.
Firm on who and what, unfixed on how well
Who did the work, where it was published, and which cell systems were tried are all pinned down by named people and a DOI, so we treat those as settled. Performance is a different matter, and a single institutional account cannot settle it; our reading is deliberately unresolved on quality until the paper's controls are read or another lab runs the method.