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.
