Non-Invasive DNA Barcoding for Parallel LNP Delivery Screening: Design, Normalization, and Functional Validation

Pooled barcoding has changed the scale of lipid nanoparticle (LNP) discovery. Instead of administering one formulation to one group, investigators can associate each formulation with a unique sequence, combine the formulations, and use sequencing to compare their recovery. The approach saves experimental capacity, but it also compresses several biological events into a deceptively simple output: a table of barcode counts. Those counts may represent circulating material, tissue accumulation, cellular uptake, translated reporter, or a mixture of these states, depending on the barcode architecture and sample collected.

A 2026 study introduced snapCodes, benzylguanine-modified DNA barcodes that covalently bind an mRNA-encoded NanoLuc–SNAP-tag fusion product in vivo. The resulting DNA–protein complexes can be isolated from a small serum sample and sequenced, allowing several LNPs to be evaluated longitudinally in one nonhuman primate without terminal tissue collection at every time point [1]. This is an important shift from tissue-destructive barcode recovery toward a blood-readable functional reporter. It does not, however, make tissue localization or cargo-specific validation unnecessary.

The most useful way to design such a study is to treat non-invasive barcoding as one layer in an evidence chain. The library must be balanced. Reporter generation must be biologically linked to delivery. Serum recovery must be controlled. Sequencing counts must be corrected for input and technical background. Finally, shortlisted formulations must be tested in the tissue, cell population, and cargo context that defines success. This guide follows that chain and identifies where an apparently strong barcode can become a weak delivery claim.

This framework is intended for research-use study design and method evaluation; it does not establish clinical performance or regulatory suitability.

Define What the Barcode Is Expected to Measure

“Delivery” is not one event. An intravenously administered LNP first circulates and encounters proteins, blood cells, and vascular barriers. It may accumulate in an organ, bind a cell surface, enter a cell, escape an endosome, release its nucleic acid, support translation or target engagement, and finally produce a biological effect. A barcode recovered from bulk tissue can sit near the beginning of that sequence. A translated and secreted reporter sits later, but it still may not reproduce the behavior of another therapeutic or experimental payload.

Before selecting a barcode system, write the decision in operational terms. If the question is which formulation reaches the liver, tissue barcode enrichment may be sufficient for an initial screen. If the question is which formulation produces protein in hepatocytes, a translation-linked system is more relevant. If the question is which formulation edits a target locus in macrophages, neither bulk-organ barcode recovery nor a secreted reporter is the final endpoint. The decisive measurement is editing in the intended cell population.

It helps to classify readouts into five levels: physical presence in a sample, tissue accumulation, association with a target cell, cytosolic delivery and reporter expression, and intended cargo action. The screen should state which level it measures and which later level it is being used to predict. That statement determines the required controls and prevents a relative sequencing signal from being reported as direct proof of efficacy.

Figure 1: Evidence ladder separating LNP presence, tissue accumulation, cell entry, reporter translation, and cargo-specific action.

How Serum-Readable snapCodes Connect Translation to Sequencing

Conventional DNA barcoding assigns a short, amplifiable sequence to each nanoparticle formulation. After administration, tissues are collected and the relative abundance of each barcode is measured. The sequence is sensitive and scalable, but it is a separate cargo. It may remain detectable after the nanoparticle has failed to release or translate the principal mRNA. The snapCode design changes that relationship by requiring a translated SNAP-tag fusion product before the modified DNA barcode becomes part of the recoverable reporter complex.

In the reported workflow, each LNP formulation is associated with a distinguishable snapCode and an mRNA encoding a NanoLuc–SNAP-tag fusion protein. After functional mRNA delivery, the fusion protein is produced and binds its corresponding benzylguanine-modified DNA barcode. DNA–fusion protein complexes are exported, recovered from serum, affinity isolated, and decoded by sequencing. The original study compared six snapCoded LNPs in mice and nonhuman primates and used approximately 30 µl of serum for the non-invasive readout [1].

This architecture moves the readout closer to functional mRNA delivery because reporter translation is required. Yet the measured abundance remains a composite. It depends on mRNA release, translation, barcode–protein reaction efficiency, export from the producing cell, complex stability in blood, serum processing, affinity recovery, PCR, and sequencing. A high snapCode count therefore means that the complete reporter pathway produced an abundant circulating complex. It does not independently identify the organ that produced the complex.

“Non-invasive” should also be interpreted precisely. The method avoids euthanasia and terminal tissue processing for each observation, but the experiment still involves in vivo dosing and a blood draw. “Low-volume longitudinal blood readout” is the more informative description when comparing study designs.

Figure 2: Serum-readable snapCode workflow from pooled LNP administration to reporter translation, DNA–protein capture, and barcode sequencing.

Design a Barcode Library That Can Survive Amplification and Error Correction

Barcode diversity is useful only when the sequences can be distinguished after synthesis, amplification, and sequencing. Build the formulation barcode set as a controlled whitelist rather than using arbitrary random strings. Avoid extreme GC content, long homopolymers, repetitive motifs, strong predicted secondary structures, and complementarity to universal primers. Maintain enough Hamming or edit distance between valid sequences so that a single sequencing error does not convert one formulation ID into another. Barcode-design reviews generally recommend checking sequence composition and neighborhood distance before oligonucleotide synthesis [2].

Separate the formulation identifier from the unique molecular identifier (UMI). The formulation barcode answers which LNP generated the molecule. A UMI can help estimate how many original template molecules entered amplification and expose PCR duplication. It does not automatically remove every amplification bias, especially when the number of template molecules is small or UMI diversity is insufficient. Retain both raw reads and unique barcode–UMI molecule counts so that an apparently dominant formulation can be checked for PCR jackpot behavior.

Use common flanking sequences for amplification and sequencing, but test the complete oligonucleotides individually before pooling. A valid in silico design can still show sequence-specific synthesis yield, capture efficiency, or PCR recovery. An equivalence panel should expose every barcode to the same amount of reporter complex or synthetic template and verify that counts remain acceptably similar after the planned extraction and library workflow. Barcodes with persistent under-recovery should be redesigned rather than corrected by a large post hoc factor.

For the sequencing stage, targeted amplicon sequencing can provide focused depth across short barcode loci, while sequencing of pre-made libraries can support projects in which the investigator supplies indexed barcode libraries. The project brief should specify the barcode whitelist, UMI structure, sample indices, expected insert size, and desired output table rather than describing the material simply as “DNA for NGS.”

Control the LNP Pool Before Interpreting Any Biological Ranking

Pooling should not hide formulation failures. Characterize each LNP individually for concentration, encapsulation, particle size, polydispersity, and RNA integrity before combining it with other candidates. A formulation that precipitates, leaks cargo, or contributes far less material than intended can generate a low barcode signal for reasons unrelated to in vivo delivery. Repeat key measurements after pooling to detect interactions or aggregation that appear only in the combined preparation.

Equal volume is rarely equal input. Build the pool from calibrated molar or mass measurements and preserve the calculation that connects stock concentration to final dose. Then sequence an aliquot of the final pre-injection pool. This input measurement captures deviations introduced by oligonucleotide synthesis, conjugation, formulation, recovery, mixing, and early amplification. Nominal recipes alone cannot provide that correction.

Cargo composition is another experimental variable. Barcoded mRNA studies found that replacing barcoded mRNA with short barcoded DNA changed LNP size, polydispersity, and in vivo ranking. In one 16-formulation comparison, b-DNA and b-mRNA delivery correlated weakly in liver and spleen, whereas barcoded mRNA better predicted delivery of functional mRNA [3]. A screen intended to select carriers for a large mRNA, siRNA, or editing system should therefore use a cargo-matched proxy where feasible and explicitly validate the final cargo later.

Pool size should follow the weakest measurement layer, not the theoretical number of possible barcode sequences. As more formulations enter the pool, the dose per member, independent molecule count, and separation between high and low performers can fall. Estimate whether the lowest expected candidate will still produce enough molecules after serum recovery and UMI deduplication. Include reference formulations at known performance levels to show that the study can distinguish strong, intermediate, and weak delivery under the chosen pool size.

Plan Serum Sampling as a Kinetic Experiment

A single time point can favor one part of the reporter pathway. Early sampling may emphasize rapid translation and release. Later sampling may reflect persistence, redistribution, or slow clearance of the DNA–protein complex. Select time points around a pilot kinetic profile and preserve repeated measurements from the same animal when the design allows. The result should show rank stability and time-dependent changes, not only a pooled average that erases them.

Standardize the blood matrix. Serum and plasma are not interchangeable when clotting, anticoagulants, protein binding, affinity capture, or nuclease activity affect recovery. Use one matrix throughout a study and document tube type, blood volume, clotting interval where applicable, processing delay, centrifugation, storage, and freeze–thaw history. Record hemolysis or lipemia, because an unusually high or low recovery may be a sample-quality problem rather than a delivery phenotype.

Collect a pre-dose sample from every animal. It can reveal barcode carryover, cross-contamination, index leakage, or nonspecific capture before treatment. A post-dose blank time point that is too early for plausible translation can also help distinguish circulating input material from the reporter-dependent complex, although its interpretation depends on the reporter kinetics. Technical replicate extractions from selected samples are useful for separating serum-processing variance from biological variance.

Serum sampling reduces the need for serial tissue collection, but it cannot localize the source. A formulation may generate the same serum reporter level through high translation in a small compartment or modest translation across a large organ. A target-organ project still needs a tissue-resolved confirmation cohort. Cell-type questions may require sorting, spatial assays, or single-cell RNA sequencing alongside barcode or reporter measurements.

Figure 3: Longitudinal serum sampling paired with a smaller tissue-confirmation cohort for LNP localization and cell-type resolution.

Build Background Correction From Independent Control Layers

No single negative control explains every source of background. An extraction blank monitors reagent and handling contamination. An untreated or pre-dose serum sample monitors animal-specific background. A naked-barcode control measures the fate of unencapsulated sequence. A reporter-negative control tests whether DNA can enter the capture fraction without translated fusion protein. A synthetic spike-in introduced at a defined processing step monitors extraction, affinity capture, amplification, and sequencing recovery.

The input pool is the denominator for formulation-specific enrichment. For barcode i at time t, a practical relative metric is the sample fraction of UMI-deduplicated molecules divided by that barcode’s fraction in the pre-injection pool. In words:

Input-corrected abundance = sample barcode fraction ÷ input-pool barcode fraction.

This ratio corrects unequal starting representation but does not produce an absolute delivery efficiency. It is compositional: one barcode’s fraction can rise because another formulation falls. It also depends on the recoverable serum reporter pool. Include total recovered molecules, spike-in recovery, and the input-corrected value in the analysis table so that relative enrichment is not mistaken for absolute exposure.

Do not subtract low counts into negative values or use an arbitrary pseudocount without reporting it. Establish a minimum independent-molecule threshold from blanks, negative controls, and replicate variability. Values below that threshold should be reported as detected below the reliable quantification range or not reliably quantified, depending on the evidence. A zero after filtering is not the same as biological absence.

UMI-aware deduplication should precede relative-abundance calculation. Demultiplex with unique dual indices where possible, compare unexpected barcode–sample combinations, and investigate index hopping or cross-talk. The final analysis should retain a sample-level QC table containing total reads, valid-whitelist reads, unique molecules, spike-in recovery, blank contamination, and the fraction of library members above the reporting threshold.

Figure 4: Background-correction and normalization pipeline using pre-dose serum, extraction blanks, spike-ins, UMIs, and the sequenced input pool.

Separate True Reporter Complexes From Technical Leakage

A serum barcode signal can be created before the intended biology occurs. Free barcode may remain after formulation or purification. Reporter-independent DNA may bind nonspecifically to capture reagents. Aerosolized PCR product can contaminate low-template libraries. Template switching or index hopping can distribute abundant sequences into other samples. Each failure mode requires a different check, which is why untreated serum alone is insufficient.

Run purified barcode and purified reporter controls through the capture chemistry. Verify that barcode recovery requires the intended covalent complex and that the complex remains stable over the sample-handling interval. Challenge the assay with large concentration differences between barcodes to test whether the highest member generates false signal in neighboring indices. Maintain physically separated pre-amplification and post-amplification areas when possible.

Sequence-level QC should compare each observed read to the approved whitelist and apply a documented edit-distance rule. An error-correction algorithm must not merge a genuine low-frequency barcode into a nearby abundant sequence. This is one reason to design the whitelist with adequate sequence separation rather than trying to solve crowded sequence space after the experiment.

Understand When Barcode Abundance Is Not Functional Cargo Delivery

For conventional b-DNA screens, the gap is direct: DNA recovery can occur when an LNP is attached to a cell, internalized but trapped in an endosome, or degraded after uptake. Detection says that amplifiable sequence reached the analyzed material. It does not prove that a therapeutic RNA entered the cytosol. High-throughput LNP reviews therefore recommend evaluating biodistribution and functional output as separate, complementary measurements [4].

Translation-linked systems narrow this gap but do not remove it. snapCode abundance reflects translation of the reporter construct and subsequent appearance of the complex in serum. A therapeutic mRNA may differ in length, structure, modification, untranslated regions, translation efficiency, or intracellular localization. siRNA requires loading into RISC and target knockdown rather than translation. A genome editor requires expression of several components, access to the nucleus or another compartment, target recognition, and a measurable edit. The best serum reporter LNP can therefore be a rational candidate without being the confirmed winner for another cargo.

Reporter biology can also distort comparisons. If one tissue secretes the fusion product efficiently and another retains or degrades it, equivalent intracellular translation may produce unequal serum signals. Differences in vascular access, protein clearance, or immune recognition can further reshape the circulating profile. These effects do not invalidate the assay; they define what it measures. The screen ranks the complete serum-reporter pathway.

Pooled attribution creates an additional problem in systems where a functional reporter gate and a separate barcode are merely co-delivered. Multiple LNPs can enter one cell. The barcode found in a reporter-positive cell may not be the formulation that generated the functional event. SANDS reduces some model constraints by using a translated surface reporter and then sequencing associated DNA barcodes, but it still depends on co-localization and tissue sampling [5] QuART links mRNA translation to reverse-transcribed barcode production in specific tissues or cells, offering another route to a function-related nucleic acid signal [6].

Protein or peptide barcodes move even closer to translation because the decoded species is produced from the mRNA, but they sacrifice the amplification advantage of DNA and can require more sample for LC–MS/MS detection [7]. The correct platform is not the one with the most functional-sounding label. It is the one whose measurement layer matches the current decision.

Figure 5: Four situations in which high barcode abundance can overstate target-cell or cargo-specific LNP performance.

Match Orthogonal Validation to the Intended Cargo

Validation should test the next claim, not repeat the same barcode measurement with a second PCR. For an mRNA payload, quantify cargo-specific RNA and protein in the target tissue and compare the leading formulations individually. IVT mRNA sequencing can help confirm reporter or payload sequence integrity before formulation, while synthetic RNA sequencing and modification analysis can support characterization when engineered RNA identity and modifications are project variables.

For siRNA, the decisive endpoint is target knockdown in the intended cell population, ideally at both RNA and protein levels. For a nuclease or base editor, measure on-target products and by-products by CRISPR mutation sequencing. If the locus or edit architecture requires broader coverage, targeted region sequencing can examine a defined interval. Broader genome-editing sequencing workflows may be appropriate when the validation question extends beyond one short amplicon.

For tissue targeting, pair the serum trajectory with tissue barcode or reporter measurements. Bulk tissue can confirm organ-level localization but may still be dominated by abundant non-target cells. Sorting, single-cell analysis, or spatial localization may be required when the goal is a rare immune, stromal, or endothelial population. A multi-omics strategy can be justified when delivery, transcriptional response, and cell-state changes must be interpreted together, but it should not be added merely to make the project appear comprehensive.

Rank Candidates With Uncertainty, Not a Single Winner Column

A screen produces a ranking, but the ranking is not equally certain at every position. Require concordance across biological replicates, examine whether candidates remain above the reporting threshold at relevant time points, and calculate intervals around input-corrected abundance. A formulation that ranks first because of one extreme animal should not automatically advance over a consistently strong second candidate.

Retain absolute QC quantities beside normalized values. Two formulations can have similar enrichment ratios while differing substantially in recovered molecules. Low-count ratios are especially unstable. Flag candidates whose rank changes when one sample is removed, when alternative reasonable pseudocounts are used, or when technical replicate extraction is considered. These sensitivity checks are more informative than reporting extra decimal places.

Do not validate only the top hit. Choose a small panel that spans strong, intermediate, weak, and mechanistically interesting profiles. Include at least one formulation whose serum ranking and prior tissue behavior disagree, because such a comparator tests the central assumption of the non-invasive readout. Retest the candidates individually and with the intended cargo. This limits winner’s-curse effects and reveals whether pooling changed relative performance.

Cross-species stability is another decision axis. SANDS showed that LNP performance and cellular responses can differ across murine, primate, and human hepatocyte contexts [5]. A mouse-to-primate rank change should be investigated as a biological result, not removed as a batch effect. The report should keep species, cell context, dose, sampling matrix, and time point attached to every ranking.

Figure 6: Candidate-ranking dashboard combining input-corrected abundance, unique molecules, time-point stability, replicate uncertainty, and tissue confirmation.

A Three-Stage Study Design for Defensible LNP Selection

Stage 1: non-invasive pooled discovery. Qualify individual LNPs, sequence the input pool, administer the mixed library, and collect longitudinal serum samples. Recover reporter-linked complexes, apply whitelist and UMI processing, and rank formulations using input-corrected abundance with replicate uncertainty. The output is a shortlist, not a final delivery claim.

Stage 2: tissue-resolved confirmation. Test a smaller candidate panel in a design that collects the organs and cell populations relevant to the project. Compare serum signal with tissue barcode, reporter RNA, reporter protein, and cell-type localization. This stage identifies whether circulating reporter output is generated by the intended biological compartment.

Stage 3: cargo-matched functional validation. Reformulate the selected LNPs with the intended mRNA, siRNA, guide/editor combination, or other payload. Test each candidate individually under the intended dose and route. Measure the endpoint that defines success, such as protein expression, target knockdown, or editing outcome. Confirm that the final cargo does not materially alter LNP quality attributes relative to the screening proxy.

These stages can share samples and controls, but they should not be collapsed into a single score. A useful final report displays how each candidate moved through the evidence layers and why it advanced or stopped. That structure makes a negative validation result informative: it can reveal a tissue-source problem, reporter-specific bias, cargo incompatibility, or insufficient sequencing evidence.

Figure 7: Three-stage LNP selection funnel from non-invasive pooled screening to tissue-resolved and cargo-matched validation.

Frequently Asked Questions

Conclusion

Non-invasive DNA barcoding can make parallel LNP screening more longitudinal and less dependent on terminal tissue collection. Its strongest use is candidate ranking within a defined reporter system. The method becomes misleading when serum abundance is treated as a direct map of organ delivery or as proof that a different therapeutic cargo will function.

A defensible project preserves four evidence layers: the composition of the administered pool, the quality of the serum reporter signal, the location of delivery in tissues and cells, and the action of the intended cargo. Barcode design, input correction, UMI-aware processing, and negative controls protect the first two layers. Tissue and cargo-matched validation complete the chain. Used this way, non-invasive barcoding does not replace biological confirmation; it makes that confirmation smaller, more focused, and easier to interpret.

Reference:

1. Jang B, et al. Non-invasive nanoparticle barcoding in nonhuman primates. Nature Biotechnology. 2026.
2. Johnson MS, Venkataram S, Kryazhimskiy S. Best Practices in Designing, Sequencing, and Identifying Random DNA Barcodes. Journal of Molecular Evolution. 2023;91:263–280.
3. Guimarães PPG, et al. Ionizable lipid nanoparticles encapsulating barcoded mRNA for accelerated in vivo delivery screening. Journal of Controlled Release. 2019;316:404–417.
4. Lam K, et al. High-Throughput Strategies for Streamlining Lipid Nanoparticle Development Pipeline. Advanced Science. 2025.
5. Hatit MZC, et al. Species-dependent in vivo mRNA delivery and cellular responses to nanoparticles. Nature Nanotechnology. 2022;17:310–318.
6. Wang KC, et al. A Reverse Transcription Nucleic-Acid-Based Barcoding System for In Vivo Measurement of Lipid Nanoparticle mRNA Delivery. ACS Bio & Med Chem Au. 2025;5(1):35. DOI: 10.1021/acsbiomedchemau.4c00081.
7. Odunze U, et al. RNA encoded peptide barcodes enable efficient in vivo screening of RNA delivery systems. Nucleic Acids Research. 2024;52:9384–9396.
8. Sago CD, et al. High-throughput in vivo screen of functional mRNA delivery identifies nanoparticles for endothelial cell gene editing. Proceedings of the National Academy of Sciences. 2018;115:E3351–E3360.