Therapeutic nanoparticles are usually engineered to be large — roughly 20 to 200 nanometers — so the kidneys cannot filter them out (the renal cutoff sits near 5.5 nm) and they stay in circulation long enough to reach a tumor or a liver target. The price of that persistence is that they are eventually captured by scavenger cells in the liver and spleen, where they can sit for months. The field's standard safety check — bulk organ-level biodistribution over at most 28 days — cannot see which cell types hold the material, cannot distinguish one formulation from another tested in the same animal, and cannot tell whether the stress responses those cells mount are being measured correctly at the RNA level. The result is an acute-chronic gap: materials that look clean in short assays, like the nitrogen-doped graphene quantum dots that produced no overt illness in mice during a 28-day window while silently driving microglial ferroptosis in the hippocampus Nitrogen-doped graphene quantum dots induce ferroptosis through disrupting calcium homeostasis in microglia.
This proposal builds a single instrument that attacks three failures of that standard assay at once. We will tag hundreds of nanoparticle formulations with DNA barcodes assembled from a small stock of reusable, pre-synthesized modules — borrowing the cost-amortization logic of DNA movable-type printing Cost‐Effective DNA Storage System with DNA Movable Type — pool all formulations into one animal, and read out which cell type retains which formulation at 1, 7, 28, and 90+ days. We will run the screen in conditional knockout mice (Cre recombinase deleting a target gene only in macrophages, hepatocytes, or endothelial cells) to test which genes are necessary for clearance in which lineage. And in the same cells, we will measure RNA and protein jointly, because the Human Proteome Project's ten-year census established that mRNA abundance explains only 40% to 60% of protein abundance variance across human tissues A high-stringency blueprint of the human proteome — a gap that plausibly widens exactly in the stressed, particle-laden cells we care about.
Core Claim
The claim has two parts. First, that long-term nanomaterial clearance is a cell-type-specific, formulation-specific process that neither bulk biodistribution nor RNA-only profiling can resolve, and that a pooled, barcoded, single-cell screen in conditional knockout mice can map it at a cost no existing method approaches. Second, that the standard interpretation of a conditional knockout — phenotype equals gene absence — breaks in this regime, because residual protein outlives genomic excision while the retained particle outlives both, creating a measurable lag between gene deletion, protein depletion, and functional change in clearance.
Research Program
Barcode construction (synthesis cost as a constraint, not the subject). Screening hundreds of formulations in vivo is blocked in practice by barcode economics: one unique, chemically synthesized oligonucleotide per formulation, per batch, makes library costs grow linearly with the screen size, exactly the base-by-base overhead documented as prohibitive for large redundant libraries High-throughput DNA synthesis for data storage. We instead assemble each formulation's barcode from a fixed stock of modular oligos in the spirit of the BISHENG-1 movable-type platform, where each pre-synthesized DNA module was reused up to 10,000 times and drove storage cost to $122/MB [Cost‐Effective DNA Storage System with DNA Movable Type](/work/costeffective-dna-storage-system-with-dna-movable-type-ycfx). Concretely, each barcode is a codeword of $k$ motifs drawn from a library of $M$ modules, giving $\binom{M}{k}$ distinct codewords from only $M$ synthesized oligos; with $M = 8$ and $k = 4$ a single library of 8 oligos yields 70 symbols, and library-splitting across blocks scales this combinatorially. Reading a barcode back from a single cell is a coupon-collector problem — amplification recovers only a partial motif set — and we decode with the soft SC-LDPC approach of Sokolovskii et al., which kept frame error rates below $10^{-3}$ at 11 reads per symbol even under empirical cross-block interference of $\rho \approx 0.078$ on the HelixWorks nanopore dataset Coding Over Coupon Collector Channels for Combinatorial Motif-Based DNA Storage. Their decoder, validated on 21 synthesized DNA blocks with 3.7 million motif occurrences, is reused nearly unchanged; what is new is pointing it at single-cell amplicons from mouse tissue rather than archival DNA.
Formulation panel. Roughly 100 formulations spanning the clearance spectrum: redox-cleavable polycations modeled on the disulfide-cross-linked cLPEI carrier Accelerated Endosomal Escape of Splice-Switching Oligonucleotides Enables Efficient Hepatic Splice Correction, whose polycationic fragment fate after disulfide cleavage remains uncharacterized; and persistent inorganic cores of the Fe3O4@Au Preparation and characterization of Fe3O4@Au-C225 composite targeted nanoparticles for MRI of human glioma and Bi@ZIF-8 Rapid synthesis of a Bi@ZIF-8 composite nanomaterial as a near-infrared-II (NIR-II) photothermal agent for the low-temperature photothermal therapy of hepatocellular carcinoma families, which lack any enzymatic degradation route and represent the worst-case chronic retention.
Conditional knockout arms. Candidate clearance genes — scavenger receptors, autophagy components, and the ferroptosis regulators GPx4 and SLC7A11 implicated by the quantum-dot study Nitrogen-doped graphene quantum dots induce ferroptosis through disrupting calcium homeostasis in microglia — will be floxed and crossed to LysM-Cre/CD11b-Cre (Kupffer cells and macrophages), Alb-Cre (hepatocytes), and Tie2-Cre (sinusoidal endothelium). This follows the logic that made conditional ablation decisive in complex niches: Tian et al. used CD11b-DTR depletion and myeloid-specific H19 knockout to isolate the macrophage-autonomous driver of liver fibrosis Conditional depletion of macrophages ameliorates cholestatic liver injury and fibrosis via lncRNA-H19, Zhang et al. isolated macrophage MED1 in vascular injury Zhang et al., and Liu et al. showed that sensory-neuron MyD88, not immune-cell MyD88, governs the acute-to-chronic pain transition Liu et al.. None of these lineage-dissection tools has ever been aimed at nanomaterial clearance; the closest work, the cLPEI study, explicitly left both clearance kinetics and genetic mechanism untouched.
Discordance measurement. In each sorted cell population we collect paired single-cell RNA and protein measurements (antibody-derived tags where surface-accessible, targeted mass spectrometry to the Human Proteome Project's PE1 stringency of at least two unique peptides of at least nine amino acids A high-stringency blueprint of the human proteome). The primary statistic, computed per gene in particle-burdened versus particle-free cells of the same type, is
$$\Delta r = r_{\text{particle-high}} - r_{\text{particle-low}},$$
where $r$ is the Pearson correlation between log mRNA and log protein abundance for a gene across cells within a burden bin. The working hypothesis, grounded in the 40–60% transcript-protein ceiling A high-stringency blueprint of the human proteome and in the demonstration that persistent protein activity outlives transient transcript signaling in long-term memory Persistently Increased Expression of PKMzeta and Unbiased Gene Expression Profiles Identify Hippocampal Molecular Traces of a Long‐Term Active Place Avoidance Memory and “Shadow” Proteins, is that $\Delta r$ is substantially negative for clearance-machinery and ferroptosis genes — that chronic particle stress decouples transcript counts from the protein state that actually governs the cell's fate.
Proposed Experiments
Experiment 1 validates barcode survival and decoding in vitro: BV2 microglia and primary hepatocyte–Kupffer co-cultures dosed with barcoded carriers (the 50 µg/mL viability threshold from the quantum-dot study Nitrogen-doped graphene quantum dots induce ferroptosis through disrupting calcium homeostasis in microglia as the stress reference), with single-cell amplicon decoding benchmarked against bulk readout. Success is frame error rate at or below the Sokolovskii soft-decoding result of $10^{-3}$ under comparable interference; their own comparison showed soft decoding beating hard-erasure decoding at materially lower read coverage Coding Over Coupon Collector Channels for Combinatorial Motif-Based DNA Storage.
Experiment 2 is the pooled wild-type screen: one intravenous pool of roughly 100 formulations, harvests at 1, 7, 28, and 90 days, sorting liver and spleen into Kupffer cells, sinusoidal endothelial cells, hepatocytes, and splenic macrophages, with per-cell-type barcode counts yielding a retention half-life per formulation per lineage. The 90-day arm deliberately extends past the 28-day ceiling that the file shows bounding every cited clearance study.
Experiment 3 repeats the screen in the three conditional knockout lines for two or three lead genes selected from Experiment 2's retention outliers, with Cre induction staggered at fixed offsets before and after particle dosing to measure the lag among genomic excision, residual protein depletion (assayed directly by targeted mass spectrometry), and the functional change in barcode retention — the empirical test of the second core claim.
Experiment 4 measures $\Delta r$ in particle-high versus particle-low cells at days 28 and 90, and asks whether the strongest published RNA-to-protein prediction tool, scChord — which achieved the lowest RMSE and highest Pearson correlation on GSE100866 (~8,600 cord blood cells, 13 proteins) and GSE164378 (>161,000 peripheral blood cells, 228 proteins), training on 100,000 cells in about 27 minutes with a 229 MB GPU footprint scChord: A Probabilistic Manifold Rectification Framework for RNA-to-Protein Translation — retains that accuracy inside particle-laden cells. scChord itself reports increasing difficulty with correlation-structure reconstruction as protein panels scale to the hundreds; no exact CMD value is reported on Lacuna for scChord. No reported result found on Lacuna for scChord's per-protein RMSE or PCC decimals, so we will establish our own in-house reference on GSE100866 before applying it to our data.
Baselines
The naive reference point is the field's default: bulk organ biodistribution of fluorescently or radiolabeled particles at a 28-day endpoint, as practiced throughout the cited nanotoxicology literature, including the quantum-dot study's bulk microarray and cytokine readouts Nitrogen-doped graphene quantum dots induce ferroptosis through disrupting calcium homeostasis in microglia. We will run it in parallel at the same doses and time points, and it is the strongest competitor on simplicity and cost per animal. It will be beaten if the pooled screen resolves retention differences between cell types or between formulations that bulk assay reads as a single organ-level number — concretely, if per-lineage half-lives differ by more than the bulk assay's measurement spread, or if two formulations with identical bulk liver content differ in which lineage holds them. A second known weakness we will quantify directly is label dissociation: bulk fluorescent or radiolabeled readings can report tag clearance rather than core clearance, an artifact the file flags as endemic.
The strongest existing method for the protein-prediction question is scChord scChord: A Probabilistic Manifold Rectification Framework for RNA-to-Protein Translation, with totalVI's joint negative-binomial model of paired RNA and surface protein Joint probabilistic modeling of paired transcriptome and proteome measurements in single cells as the second reference. We will run both on our paired data with the same cells, the same train/test splits, and a matched compute budget on our own hardware, using their published configurations and our tuning budget applied equally to each. Beating them is not the point — they are oracles for when RNA-only measurement suffices. The decisive comparison is conditional: if scChord's held-out error on clearance-machinery and ferroptosis proteins in particle-high cells stays at its particle-low level, RNA-only profiling is adequate under particle stress and our discordance claim fails; if its error grows by a margin exceeding split-level noise, direct proteomics is necessary and the atlas justifies its cost.
Two ablations isolate the proposal's key ingredients. First, a barcode ablation: an identically sized sub-library using one commercial, uniquely synthesized oligo per formulation, at the same information content and the same decoding budget, isolates the modular-combinatorial ingredient; we will report realized cost per formulation for both arms. Second, a measurement ablation: the full pipeline run with RNA-only profiling (imputing protein with scChord) versus paired measurement, isolating what the discordance arm contributes to the clearance mechanism we infer.
Why not just…?
Why not just buy commercial DNA barcodes? At pilot scale this is a real threat, and we treat it as the primary ablation rather than argue it away. The honest accounting: at roughly 100 formulations, unique commercial barcodes may cost no more than the labor of standing up modular assembly, since BISHENG-1's $122/MB figure includes automated hardware Cost‐Effective DNA Storage System with DNA Movable Type. The modular approach wins in two regimes the ablation will delimit: as the library grows into the thousands (cost per distinct codeword falls toward zero because each module is reused across codewords, the same amortization that gave 10,000 reuses per module), and under single-cell amplicon dropout, where partial motif recovery is exactly the coupon-collector channel that soft decoding was built for Coding Over Coupon Collector Channels for Combinatorial Motif-Based DNA Storage. If the commercial arm matches our decoding error rate and total cost at every library size we test, we will report that the cost argument does not hold at screen scale and the proposal's burden shifts entirely to the biological claims — which stand independently of the barcode economics.
Why not just profile RNA and impute the proteins? Because the imputation tools are the very evidence that this fails under stress. The 40–60% transcript-protein ceiling is a global average across healthy tissues A high-stringency blueprint of the human proteome; scChord showed that deterministic regression collapses multimodal protein distributions to conditional means, and that even its own flow-matching model degrades on correlation structure at high protein dimension scChord: A Probabilistic Manifold Rectification Framework for RNA-to-Protein Translation; totalVI's background-decoupling applies only to surface proteins Joint probabilistic modeling of paired transcriptome and proteome measurements in single cells. Whether that failure worsens specifically in particle-laden cells is an open empirical question — no reported result found on Lacuna for transcript-protein correlation under chronic nanoparticle stress — and Experiment 4 is designed to settle it either way.
Why not just use global knockouts or pharmacological inhibitors? Global deletion of essential clearance or ferroptosis genes risks embryonic lethality or organism-wide compensation, the classic sledgehammer problem that conditional ablation exists to solve — the file's knockout literature is a catalog of mechanisms that only became visible under lineage restriction, from macrophage H19 in fibrosis Conditional depletion of macrophages ameliorates cholestatic liver injury and fibrosis via lncRNA-H19 to neuronal MyD88 in chronic pain Liu et al.. Inhibitors have the opposite failure: they bathe whole tissues, so rescue by Fer-1 or calcium blockers in the quantum-dot study Nitrogen-doped graphene quantum dots induce ferroptosis through disrupting calcium homeostasis in microglia cannot establish microglial-autonomous necessity in vivo, since the same drugs alter vascular tone and neurotransmission everywhere. That said, inhibitor rescue is cheaper and faster, so we will use it as a screening filter for candidate genes before committing to floxed lines — an alternative we adopt where it works rather than resist.
Decisive Checks
The program fails informatively in three specific ways. If pooled barcode retention matches bulk biodistribution in every lineage and formulation (Experiment 2), cell-type resolution adds nothing and the screen's cost advantage alone cannot justify it. If scChord's accuracy holds in particle-high cells (Experiment 4), RNA-only screening of clearance phenotypes is adequate and the paired-proteomics arm is unnecessary. If barcode retention kinetics are identical across all conditional knockout lines (Experiment 3), clearance is dominated by passive physical capture rather than genetically regulated metabolism — a negative result, but one that redirects the field away from a mechanism it currently assumes. The timing experiment stands on its own either way: the stagger between Cre excision, protein depletion, and functional clearance change is, to our knowledge, the first direct measurement of how knockout interpretation must be corrected in the presence of bioaccumulated material.
References
- Nitrogen-doped graphene quantum dots induce ferroptosis through disrupting calcium homeostasis in microglia
- Cost‐Effective DNA Storage System with DNA Movable Type
- A high-stringency blueprint of the human proteome
- High-throughput DNA synthesis for data storage
- Coding Over Coupon Collector Channels for Combinatorial Motif-Based DNA Storage
- Accelerated Endosomal Escape of Splice-Switching Oligonucleotides Enables Efficient Hepatic Splice Correction
- Preparation and characterization of Fe3O4@Au-C225 composite targeted nanoparticles for MRI of human glioma
- Rapid synthesis of a Bi@ZIF-8 composite nanomaterial as a near-infrared-II (NIR-II) photothermal agent for the low-temperature photothermal therapy of hepatocellular carcinoma
- Conditional depletion of macrophages ameliorates cholestatic liver injury and fibrosis via lncRNA-H19
- MED1 Deficiency in Macrophages Accelerates Intimal Hyperplasia via ROS Generation and Inflammation
- TLR signaling adaptor protein MyD88 in primary sensory neurons contributes to persistent inflammatory and neuropathic pain and neuroinflammation
- Persistently Increased Expression of PKMzeta and Unbiased Gene Expression Profiles Identify Hippocampal Molecular Traces of a Long‐Term Active Place Avoidance Memory and “Shadow” Proteins
- scChord: A Probabilistic Manifold Rectification Framework for RNA-to-Protein Translation
- Joint probabilistic modeling of paired transcriptome and proteome measurements in single cells