When researchers want to put large or charged molecules — DNA plasmids, messenger RNA, CRISPR editing complexes — inside a living cell, the most universal tool is electroporation: short electric pulses that push the cell membrane past a voltage threshold (roughly 0.2 to 1.0 volts across the bilayer) and open transient water-filled pores. It works on nearly any cell type, needs no carrier chemistry, and is a benchmark laboratory delivery standard. Yet the same technique that delivers cargo efficiently to cells in a dish routinely underperforms in living tissue, where lipid nanoparticles have been shown to overtake it for intramuscular DNA vaccination Lipid Nanoparticles Outperform Electroporation in Delivering Therapeutic HPV DNA Vaccines. That reversal is one concrete face of the in vitro–in vivo translation gap, and its causes are usually blamed on tissue field attenuation and electrode trauma. This proposal tests a different, unexamined suspect: the cell cycle.
The motivation comes from two facts that have not yet been connected. First, membranes are not static capacitors: their permeability changes in a stepwise, non-proportional way with lipid composition and ordering Data-driven equation for drug-membrane permeability across drugs and membranes, and cells remodel hundreds of structural and regulatory proteins as they move through growth and division phases Spatiotemporal dissection of the cell cycle with single-cell proteogenomics. Second, an asynchronous dish culture and a living tissue have very different distributions of cells across cycle phases — and every existing electroporation protocol treats that distribution as invisible. The core claim of this proposal is that the cell cycle phase of a cell at the moment of the pulse is a major, measurable, and currently unmodeled determinant of delivery efficiency, survival, and downstream gene and protein response, and that ignoring it contributes measurably to the in vitro–in vivo delivery gap.
What Exists and What Does Not
Three papers come closest to this intersection, and each leaves the specific hole this proposal fills. Lipid Nanoparticles Outperform Electroporation in Delivering Therapeutic HPV DNA Vaccines compares electroporation and three lipid nanoparticle formulations across in vitro expression and in vivo mouse immunity, but it treats the target tissue as a static recipient — it never asks what cell cycle state the electroporated cells were in, or whether that matters. Spatiotemporal dissection of the cell cycle with single-cell proteogenomics maps the cycle in extraordinary single-cell detail — 1,152 transcriptomes and 273,709 immunofluorescence-profiled cells — but it perturbs nothing: no membrane permeabilization, no delivery, no in vivo arm. And Microfluidic-Based Electrical Operation and Measurement Methods in Single-Cell Analysis reviews microfluidic devices that can trap single cells (dielectrophoretic capture above 90% efficiency) and electroporate them (up to 96.6% efficiency), and notes that impedance measurements can distinguish cell cycle stages — but as a technology review it never maps cycle state to delivery outcomes or downstream expression. On the computational side, transfer models that bridge cell lines to tumors — PRECISE: A domain adaptation approach to transfer predictors of drug response from pre-clinical models to tumors and DeepSADR: Deep Transfer Learning with Subsequence Interaction and Adaptive Readout for Cancer Drug Response Prediction — operate purely on chemical structures and static transcriptomes and contain no representation of physical delivery at all. Nobody, in short, has measured how pore formation, resealing, and macromolecular uptake vary with cell cycle phase, or tested whether that variation explains part of the delivery translation gap.
Core Claim
Electroporation outcomes — cargo uptake, membrane resealing, survival, and the downstream protein response — depend strongly and predictably on cell cycle phase, and this dependency can be measured label-free, modeled as a phase-conditioned delivery function, and used both to improve physical delivery and to explain a quantified fraction of the in vitro–in vivo expression drop. The claim is falsifiable: if delivery metrics are flat across phases once cell size and membrane area are controlled, the idea fails on its own terms.
Research Program
The program has four linked components, each producing an artifact the next one consumes. The first is a phase-resolved biophysical atlas. We will build on the microfluidic single-cell electroporation architecture reviewed in Microfluidic-Based Electrical Operation and Measurement Methods in Single-Cell Analysis: cells are trapped dielectrophoretically, staged into cycle phase by on-chip impedance spectroscopy (validated against a fluorescent FUCCI reporter, which changes color with cycle phase, in U-2 OS cells — the same line used in the proteogenomics study, so our measurements land directly on top of its expression map), and then electroporated with a swept range of field strengths and pulse durations. Per phase we measure uptake of fluorescently labeled dextran and mRNA, resealing kinetics, and survival, yielding a delivery function $D(\phi, E, \tau)$ where $\phi$ is cycle phase, $E$ field strength, and $\tau$ pulse duration. We will test whether adding phase as a predictor improves upon the best existing permeability model — the data-driven SISSO equation of Data-driven equation for drug-membrane permeability across drugs and membranes, which achieves a Pearson $r^2$ of 0.70 against 0.64 for a hydrophobicity-only baseline but contains no cell-state term.
The second component asks whether the pulse itself corrupts the expression machinery we hope to study. Spatiotemporal dissection of the cell cycle with single-cell proteogenomics established that 85% of cycling proteins do not cycle at the transcript level, and that even the concordant 15% show a median 7.7-hour delay between peak transcript and peak protein. We will measure single-cell RNA and immunofluorescence protein levels at 0, 4, 8, and 24 hours after electroporation, per phase, tracking cell cycle gatekeeper proteins (CDK1, Cyclin B1, p27, PLK1) whose degradation-driven regulation is well characterized in the arrest literature, e.g. p27 destabilization through SKP2 in Curcumol Induces G1 Phase Arrest in SK-Hep-1 Cells by Targeting SKP2-Mediated p27 Degradation and PLK1 stabilization in USP7 is a novel Deubiquitinase sustaining PLK1 protein stability and regulating chromosome alignment in mitosis. The question is whether ion influx and membrane stress accelerate degradation or phosphorylation of these labile kinases — which would mean post-electroporation transcript readouts are systematically misleading in a way nobody has quantified.
The third component is the translation ladder: the same delivery perturbation applied in a flat 2D culture, in a 3D extracellular-matrix hydrogel (the architecture whose loss in standard culture is itself a documented driver of the translation gap, as in Development and characterization of a novel injectable thyroid extracellular matrix hydrogel for enhanced thyroid tissue engineering applications), and in mouse muscle in vivo. In the in vivo arm we replicate the design of Lipid Nanoparticles Outperform Electroporation in Delivering Therapeutic HPV DNA Vaccines — intramuscular reporter plasmid delivered by electroporation versus SM-102 lipid nanoparticles, read out by antigen expression and IFN-γ ELISpot — with one addition: we measure the cell cycle distribution of the electroporated tissue (Ki-67 and phospho-histone H3 staining, plus a FUCCI reporter line). We can then decompose the in vitro–in vivo expression drop into a phase-distribution component (predicted by scaling our in vitro $D(\phi, E, \tau)$ by the tissue's measured phase distribution) and a residual component attributable to field attenuation and tissue damage. This decomposition is the proposal's central deliverable: a number saying how much of the gap is cell cycle and how much is physics.
The fourth component is a compact phase-conditioned delivery predictor, trained on the atlas, and a conceptual bridge to the computational translation literature: it is exactly the physical-delivery feature that PRECISE and DeepSADR lack, and we will test in the in vivo arm whether it predicts measured expression better than a phase-blind delivery model at equal parameter count.
Baselines
The naive reference point is what nearly every lab does today: fixed-parameter electroporation on an asynchronous culture, with no knowledge of cycle state. The best-case in vitro number for this approach comes from the microfluidic literature itself, which reports electroporation efficiencies up to 96.6% with dielectrophoretic capture above 90% Microfluidic-Based Electrical Operation and Measurement Methods in Single-Cell Analysis — but that headline efficiency is measured on delivery alone, before accounting for survival differences across phases, and our baseline will score it on viable-delivered fraction (cells that both receive cargo and remain proliferative at 24 hours) so the comparison is apples-to-apples. We will run this baseline on our own chip with the same cells and cargo so the only difference is the absence of phase information.
The strongest existing method on the delivery-translation question is the lipid nanoparticle, specifically the SM-102 formulation that produced the highest IFN-γ splenocyte responses among all regimens including electroporation in Lipid Nanoparticles Outperform Electroporation in Delivering Therapeutic HPV DNA Vaccines (the paper reports the ranking but no exact spot count, so no reported result found on Lacuna for the absolute ELISpot value). We will run this head-to-head in the in vivo arm with the same plasmid, mice, timepoints, and readouts as the electroporation arms. Beating LNPs outright on vaccination-grade immunity is not our claim; the pre-specified success criteria are instead that (a) cycle-adaptive electroporation closes at least half of the in vitro–in vivo expression drop of fixed-parameter electroporation, and (b) our phase-conditioned predictor explains in vivo expression with materially higher correlation than a phase-blind predictor fit with the same data and degrees of freedom.
Two further comparators sharpen the picture. For cycle staging, the strongest published label-free classifier is DeepFlow, a 42-layer convolutional network that classifies seven cell cycle stages from imaging flow cytometry at 98.73% accuracy versus 92.35% for a feature-boosting baseline Reconstructing cell cycle and disease progression using deep learning; our impedance-based staging must approach this accuracy against FUCCI ground truth or we will adopt imaging-based staging instead. Finally, the key ablation isolates the proposal's central ingredient: identical chip, cells, cargo, and total pulse energy, but with parameters selected phase-blind (a single fixed protocol) versus phase-adaptive. If the phase-adaptive arm does not raise the viable-delivered fraction, the atlas may be real but the adaptive payoff is not.
Why not just…?
Why not just use lipid nanoparticles in vivo and abandon electroporation? For the specific application of intramuscular DNA vaccination, the LNP paper already shows this is the pragmatic winner, and we do not dispute it. But carrier-free physical delivery remains the only species-agnostic route for cargoes that nanoparticles cannot carry well or at all — Cas9 ribonucleoprotein delivery into pluripotent stem cells Highly efficient genome editing via CRISPR–Cas9 in human pluripotent stem cells is achieved by transient BCL-XL overexpression and genetic manipulation of non-model organisms with robust walls Genetic tool development in marine protists: emerging model organisms for experimental cell biology — and, more basically, the LNP paper's own design treats the target tissue as a static recipient and never identifies why in vitro electroporation benchmarks failed in vivo. Our decomposition experiment answers that question in a way that substituting LNPs cannot. If cycle-adaptive pulsing fails to close any of the gap, the field-mechanics explanation wins and this proposal will say so.
Why not just synchronize the cells with thymidine or nocodazole blocks and electroporate the whole population? Chemical synchronization forces the confound we are trying to measure directly into the system: it is, as the proteogenomics study puts it, inherently stressful and distorts natural expression patterns Spatiotemporal dissection of the cell cycle with single-cell proteogenomics. A thymidine-blocked G1/S population is not a physiological G1/S population, and delivery measured on it would conflate drug stress with cycle state. We instead stage cells label-free (impedance, validated against FUCCI) and perturb nothing before the pulse. As a check rather than an argument, we will run one chemically synchronized arm and quantify how far its delivery metrics sit from the label-free phase-matched measurements — turning the objection itself into a result.
Why not just read out transcripts after delivery, since RNA methods are cheaper and more sensitive? The transcript-protein decoupling is the reason: 85% of cycling proteins show no transcript-level cycling, the concordant minority lags by a median 7.7 hours, and 71% of single-cell protein variability remains unexplained even after cycle modeling Spatiotemporal dissection of the cell cycle with single-cell proteogenomics. Since our second component specifically tests whether permeabilization stress perturbs protein stability and phosphorylation of cycle gatekeepers — effects invisible to RNA counts, as the phosphorylation-dependent AKT/STAT3 signaling measurements in Dasatinib regulates LPS-induced microglial and astrocytic neuroinflammatory responses by inhibiting AKT/STAT3 signaling illustrate — a transcript-only design could not distinguish "the pulse changed expression" from "the pulse changed protein half-life." The dual-level design is not gold-plating; it is the only readout that can answer the question.
Why not just bridge the gap computationally, as PRECISE and DeepSADR already do? These models are genuinely strong at what they model — DeepSADR reaches an AUC of 0.957 for sorafenib response prediction versus 0.811 for the next best baseline on clinical data DeepSADR — but they contain no physical delivery dimension at all, and even within their own domain, domain adaptation costs accuracy (PRECISE reports a median 3.9% reduction in Pearson correlation on pre-clinical prediction as the price of transfer PRECISE). A model trained on static transcriptomes cannot tell you that a therapy whose efficacy depends on physically opening pores will behave differently in a tissue whose cells sit in different cycle phases. Our program does not compete with these models; it produces the measured, phase-resolved delivery term they are missing, and tests in the in vivo arm whether adding it improves prediction. The honest threat is that the phase effect exists but is too small to matter at tissue scale; the decomposition experiment is built precisely to detect that outcome cheaply.
Decisive Checks
Three outcomes would settle the proposal's fate. First, if the per-phase delivery atlas shows no variation beyond what cell size and membrane capacitance predict, the core claim is dead, and we will report the null result — it would still be the first controlled measurement of its kind and would exonerate cycle state in the delivery gap. Second, if phase-adaptive pulsing does not raise the viable-delivered fraction over the fixed protocol at matched total energy, the adaptive application fails even if the atlas is informative, and the work remains a measurement contribution rather than a method. Third, if the phase-distribution component explains little of the in vitro–in vivo expression drop relative to the residual field-attenuation component, then the LNP paper's implicit story — that tissue mechanics, not cell state, drive the reversal — is confirmed, and the phase-conditioned predictor should not be grafted onto transfer models like PRECISE and DeepSADR. Each of these is a pre-registered, quantitative criterion, not a judgment call, and any one of them converts the proposal into a clear positive or negative contribution to the systematic-validation literature.
References
- Lipid Nanoparticles Outperform Electroporation in Delivering Therapeutic HPV DNA Vaccines
- Data-driven equation for drug-membrane permeability across drugs and membranes
- Spatiotemporal dissection of the cell cycle with single-cell proteogenomics
- Microfluidic-Based Electrical Operation and Measurement Methods in Single-Cell Analysis
- PRECISE: a domain adaptation approach to transfer predictors of drug response from pre-clinical models to tumors
- DeepSADR: Deep Transfer Learning with Subsequence Interaction and Adaptive Readout for Cancer Drug Response Prediction
- Curcumol Induces G1 Phase Arrest in SK-Hep-1 Cells by Targeting SKP2-Mediated p27 Degradation
- USP7 is a novel Deubiquitinase sustaining PLK1 protein stability and regulating chromosome alignment in mitosis
- Development and characterization of a novel injectable thyroid extracellular matrix hydrogel for enhanced thyroid tissue engineering applications
- Reconstructing cell cycle and disease progression using deep learning
- Highly efficient genome editing via CRISPR–Cas9 in human pluripotent stem cells is achieved by transient BCL-XL overexpression
- Genetic tool development in marine protists: emerging model organisms for experimental cell biology
- Dasatinib regulates LPS-induced microglial and astrocytic neuroinflammatory responses by inhibiting AKT/STAT3 signaling