Virus-Like-Particle Delivery for Pooled CRISPR Screens in Primary Myeloid Cells: From Hybrid Cargo Delivery to Defensible Readouts

Primary human monocytes, macrophages, and dendritic cells are often the cells investigators most want to perturb and the cells least forgiving of a standard pooled CRISPR workflow. They are finite, donor variable, difficult to expand, sensitive to delivery stress, and ready to alter their state when exposed to nucleic acid or particle preparations. A pooled screen can therefore appear to work at the sequencing stage while failing much earlier: guide identity may reach the cell without sufficient editing, editing may occur only in a stressed subpopulation, or an apparent hit may reflect donor composition rather than perturbation biology.

Virus-like particles (VLPs) offer an important new option because they can deliver CRISPR proteins or ribonucleoprotein cargo transiently. But the most useful recent primary-myeloid implementation is a hybrid system, not a claim that a VLP carries an entire pooled guide library. In SLICeVLP, a VPX-lentiviral vector introduces and preserves the sgRNA identity, while engineered VLPs deliver Cas9 protein. That separation is the design principle for this article: guide identity, Cas9 exposure, functional editing, preserved cell state, phenotype, and sequencing readout are different checkpoints. Treating them as one variable is how many difficult-cell screens become uninterpretable.

This guide is written for research teams planning functional-genomics experiments in primary myeloid cells. It focuses on the evidence needed to decide whether a screen should proceed, how to prevent a low-information library from becoming a high-cost sequencing run, and how to validate a finding after the pooled discovery stage. It is for research use only and does not establish a clinical or therapeutic workflow.

Why Primary Myeloid Cells Break the Usual Pooled-Screen Assumptions

Many pooled-screen playbooks were optimized around proliferative cell lines. Those systems can supply enough cells to absorb modest transduction losses, maintain a library over several passages, and tolerate stable Cas9 expression. Primary myeloid cells change all three assumptions. They have a limited cell budget, their differentiation and activation states can shift during culture, and their innate sensing machinery can react to delivery components, residual nucleic acid, or prolonged expression of foreign proteins.

The relevant question is not simply whether a vector enters cells. A practical screen must preserve an interpretable relationship between each barcode and the edited cell state. Before ordering a large library, define the bottleneck with a small feasibility panel: measure viable recovery, one-guide-per-cell behavior, locus-level editing, target-protein or transcript reduction where applicable, and an activation-state panel. This is also the point to decide whether the endpoint can be read as guide abundance, fluorescence sorting, targeted molecular measurement, or single-cell transcriptional profiling.

For a bulk knockout screen, the limiting quantity can be expressed as a planning calculation:

Effective edited cells per guide = recovered cells × single-guide fraction × Cas9-exposed fraction × functional-edit fraction ÷ library size.

This is not a universal coverage threshold. It is a way to expose where coverage disappears. A team may begin with adequate donor cells but lose representation after guide delivery, then lose more cells after particle exposure or stimulation. Recording the numerator at each checkpoint makes the loss visible before a final guide-count table hides it.

Figure 1: The evidence chain for a pooled CRISPR screen in primary myeloid cells separates guide identity, transient Cas9 delivery, functional editing, cell-state preservation, phenotype, and sequencing readout.

The Hybrid Delivery Architecture: Keep the Barcode and Editor Questions Separate

In a hybrid VLP/lentiviral workflow, the sgRNA cassette and the editing protein do not have the same job. The guide vector creates a recoverable identity record. The engineered VLP or eVLP provides a timed editing pulse. SLICeVLP is a study-specific hybrid architecture, not a claim that every VLP delivery format has the same performance or is directly interchangeable. This architecture can be attractive in cells where maintaining stable Cas9 expression would be inefficient or would complicate state interpretation. It also creates a specific analytical obligation: validate each component independently before measuring the combined system.

Guide delivery should be evaluated by barcode recovery, single-guide distribution, and retention after the intended recovery interval. Cas9-VLP delivery should be evaluated by an orthogonal locus-editing assay and, when possible, by a functional control phenotype. A fluorescent reporter is useful only if it answers the same biological question as the intended cargo. A reporter-positive cell can still carry a weakly active guide, receive insufficient editor, or change state because of delivery stress.

Start with a compact panel that includes non-targeting guides, guides against a well-characterized positive-control locus, and multiple guides per test gene. Use a locus where editing can be confirmed with CRISPR sequencing, then verify the amplicon design and depth before scaling. For large libraries, an early test of plasmid and packaged-guide composition using pre-made library sequencing can reveal representation loss that cannot be repaired after cells have been exposed.

MOI and VLP dose should not be optimized as though they are a single lever. Guide-vector input affects identity complexity and multi-guide risk. VLP input affects editor exposure, viability, and potentially cell-state disturbance. Keep a two-dimensional pilot matrix: guide-vector level on one axis and VLP dose on the other. Select the region that preserves guide representation and viable cells while achieving the functional edit needed for the specific endpoint. The best condition is rarely the condition with the strongest isolated reporter signal.

Design Library Scope From the Cells You Can Actually Recover

A primary-cell screen begins with a cell budget, not a gene list. Work backwards from the minimum edited, interpretable cells needed at the final readout. Then add losses for guide delivery, sorting if used, VLP exposure, recovery, stimulation, and library preparation. If the calculation produces an unrealistic donor requirement, shrink the library or change the screen format before starting.

Focused libraries are often more defensible than genome-scale libraries in donor-derived myeloid cells. A pathway library, a prioritized candidate set, or a transcription-factor panel gives more guides per hypothesis and permits donor replication. In contrast, a broad library that reaches the sequencing instrument with uneven representation may produce a nominally genome-wide result without reliable negative evidence for most genes.

Build the guide set with redundancy rather than placing the full claim on one sequence. Independent guides that converge in direction are more informative than one extreme guide count. Include non-targeting controls to define the technical distribution, safe-targeting controls that target a preselected neutral genomic location to identify cutting-related background effects, and pathway-positive controls that can demonstrate whether the stimulation and readout are responsive. Use CRISPR screen sequencing at the pilot stage to compare guide distributions before and after each major step, rather than treating the final endpoint as the first library-quality assessment.

For a FACS-compatible phenotype, plan the gate before the library is made. Specify the marker, stimulation window, compensation controls, live-cell gate, and how many cells are expected in each selected fraction. For a transcriptional phenotype, decide whether a targeted panel can answer the question or whether unbiased profiling is necessary. A small targeted response panel may be more robust for a known inflammatory axis. A broader state-discovery question can justify single-cell RNA sequencing or an integrated single-cell RNA-seq data analysis workflow, provided that guide capture and donor metadata are built into the analysis plan.

Figure 2: A cell-budget decision tree shows how guide-library size, recovery losses, functional editing, and readout type determine whether a primary myeloid pooled screen is feasible.

Build Representation Checkpoints Into the Physical Workflow

Library representation is not a single QC number. It is a chain of distributions, and each distribution should have a sample retained for sequencing. The practical checkpoints are: the plasmid library; the packaged VPX-lentiviral guide pool; post-transduction cells before VLP exposure; post-VLP cells after recovery; and the final sorted or sequenced population. Those samples allow a team to locate the first step at which guides became skewed.

At each checkpoint, evaluate guide dropout, relative-abundance skew, read depth per guide, and correlation with the preceding stage. A library can retain a respectable median read count while losing a critical subset of guides. Plotting the cumulative fraction of guides above a defined minimum and reviewing the bottom tail are more informative than reporting only a mean. If a guide class systematically drops out, examine sequence features, vector production, and cell-state effects before interpreting biological depletion.

The sequencing assay should match the question. Short, well-defined guide cassettes are suited to targeted amplification and amplicon sequencing services. Where multiple fixed loci, guide cassettes, or donor-specific markers need to be evaluated in one run, a carefully designed multiplex PCR sequencing strategy can keep the QC panel compact. For candidate confirmation, targeted region sequencing provides a way to examine the edited locus rather than inferring editing from phenotype alone.

Use DNA-level and RNA-level checkpoints for different purposes. DNA sequencing answers whether the intended edit and guide identity are present. RNA measurements answer whether the perturbation changed the expected biological program. For a bulk endpoint, conventional RNA-seq can help distinguish a genuine pathway shift from a generic stress response. Do not substitute transcript-level evidence for direct editing evidence, or the reverse.

Choose the Readout That Preserves the Causal Link You Need

A pooled screen only earns its conclusion when the assay retains a credible guide-to-phenotype relationship. Different readouts answer different questions:

• Bulk guide abundance is suited to survival, proliferation, or enrichment/depletion under a defined condition. It is efficient but cannot resolve heterogeneous cell states.
• FACS-sorted guide enrichment works when a surface marker, reporter, or intracellular signal can be gated reproducibly. Its strongest use is a pre-specified, robust phenotype with enough recovered cells in each fraction.
• Perturb-seq couples guide identity to transcriptome state at single-cell resolution. It is powerful for heterogeneous responses but has stricter requirements for guide capture, ambient RNA control, doublet management, and donor-aware modeling.
• Optical or imaging-based screens are valuable when localization, morphology, or organelle state is the phenotype, but they require a separate mapping method and image-QC plan.

Do not select the readout because it is the most elaborate option. Select it because it can falsify the hypothesis. If a candidate is expected to alter a single induced surface marker, FACS enrichment plus guide counting may be stronger than a small single-cell data set with poor guide capture. If the hypothesis concerns alternative inflammatory states, single-cell profiling may be needed to separate a true state transition from differential survival.

Figure 3: A readout-selection map compares bulk guide counting, FACS enrichment, Perturb-seq, and imaging screens by phenotype type, retained guide-to-cell linkage, and primary failure mode.

Protect the Myeloid State While You Measure Editing

Primary myeloid cells can respond to delivery, culture density, serum changes, cytokine handling, and plate effects. A screen that uses only an unstimulated, untreated control cannot distinguish a gene-specific effect from a delivery-associated state shift. Include controls that isolate the contribution of each component: untreated cells, guide-vector-only cells, VLP-only cells, mock-particle controls where feasible, and the full hybrid delivery condition. If the biological endpoint includes stimulation, apply the same stimulation schedule across the comparison set and include donor as a first-class design variable.

State control requires more than a viability measurement. Track morphology or imaging features, live-cell recovery, baseline marker expression, and at least a concise activation or interferon-response signature. Predefine the signatures that would trigger a stop decision. For example, a condition with acceptable editing but a widespread particle-associated response that overwhelms the intended signal should be redesigned, not advanced because the guide counts are technically clean.

Randomize donor samples across plates and process them in balanced batches. Keep operator, particle lot, guide-vector lot, stimulation reagent lot, and sequencing batch in the metadata table. These fields are not administrative extras. They are needed to determine whether an apparent hit aligns with the intended perturbation or with a hidden workflow variable. When final editing calls are generated from targeted DNA data, a separate variant-calling analysis can provide a consistent framework for documenting edit spectra and indel profiles across donors and conditions.

Analyze Donors as Biological Replicates, Not as Technical Repeats

Primary-cell screens frequently fail through over-aggregation. Pooling donors may increase a raw cell count but can conceal opposite guide effects, donor-specific guide recovery, or a batch that dominates a comparison. Maintain donor labels from collection through sequencing. If cells must be combined before one assay, preserve donor identity with a compatible tagging design or run sufficient donor-resolved replicates to model donor effects explicitly.

At the guide-count stage, compare each donor and condition against its matched control before seeking a common effect. At the single-cell stage, examine guide recovery by donor, library complexity, mitochondrial and stress features, doublet rate, and the number of cells contributing to each guide-by-state conclusion. A statistically striking guide is not automatically a credible hit if it is supported primarily by one donor or one poorly recovered library.

Use three layers of evidence when prioritizing a hit: guide-level concordance, locus-level edit confirmation, and phenotype reproducibility in donor-resolved experiments. This framework narrows the difference between a sequencing association and a functional-genomics result. It also produces a clearer handoff to downstream validation than a ranked list alone.

Figure 4: A donor-aware QC dashboard separates library representation, editing evidence, viability, activation-state markers, and phenotype consistency before hit ranking.

Validate Hits With an Orthogonal, Staged Plan

A pooled result should start validation rather than end it. Re-test each candidate with at least two independent guides, preferably in an arrayed format that removes competition among library members. Confirm the on-target locus with targeted sequencing, then measure the phenotype with the same primary assay that defined the screen. If the phenotype is transcriptional, repeat the relevant state signature in a donor-resolved experiment instead of relying only on a global differential-expression score.

Use rescue or pathway-logic experiments only when the biology and assay support them. A rescue construct that changes baseline cell state can create a second confounder. A better sequence is often: confirm edit spectrum, verify target reduction, reproduce the phenotype with independent guides, test a mechanistically related perturbation, and then consider rescue. The goal is not to make a candidate look compelling by adding assays. The goal is to make the causal interpretation harder to refute.

Off-target assessment belongs in the validation plan when a candidate will support detailed mechanistic follow-up. The appropriate depth depends on the editing modality, guide architecture, and intended interpretation. A focused CRISPR off-target validation design can be chosen after the primary hit set is narrowed, rather than consuming the cell budget of the discovery screen.

Figure 5: A staged validation workflow moves from pooled guide evidence to independent-guide confirmation, targeted edit measurement, donor-resolved phenotype testing, and mechanism-focused follow-up.

A Decision Matrix for Choosing the Perturbation Format

Hybrid VLP/lentiviral delivery is not automatically the right choice. It is most useful when the project needs a pooled, recoverable guide identity and benefits from transient Cas9 exposure in difficult primary cells. A guide-vector plus Cas9 nucleofection workflow may be acceptable for smaller studies where electroporation stress has been characterized. Stable Cas9 expression may fit expandable myeloid-like models but changes the relevance of a primary-cell study. Arrayed RNP delivery can be the best answer for a short candidate list, because it removes library-representation constraints and makes donor replication more practical.

Figure 6: A delivery-format decision matrix positions hybrid VLP and guide-vector delivery, nucleofection, stable Cas9 models, and arrayed RNP experiments by scale, state preservation, and validation burden.

What a Defensible Mini-Library Pilot Should Prove

A mini-library pilot is not a reduced version of the final screen. It is a diagnostic experiment designed to remove the largest uncertainties before valuable primary-cell material is committed. Its library should contain enough guides to reveal representation behavior, but remain small enough that every guide can be supported by the available donor cells. Include the same guide-vector backbone, particle production process, stimulation schedule, and planned barcode-recovery assay that will be used later. Changing all of those elements at once in the full screen prevents a useful failure analysis.

First, prove that the guide cassette remains measurable from the biological material that will be analyzed. Sample the plasmid pool, guide-vector preparation, cells after guide delivery, and cells after the recovery period. If the final readout will use sorted fractions, include a mock sort in the pilot. This establishes whether cell handling, gating, or low DNA input is the primary source of guide loss. The aim is not to demand perfectly flat abundance. The aim is to know the baseline distribution well enough to recognize selective dropout later.

Second, separate a molecular editing result from a useful functional perturbation. A target may show indels yet retain enough protein to leave the planned phenotype unchanged. Conversely, a strong phenotype may emerge from delivery-associated stress rather than target disruption. Select one control whose editing can be confirmed at DNA level, one control whose effect can be read in the chosen biological assay, and one non-targeting guide set that experiences the same delivery sequence. If those three controls disagree, a full library will usually magnify the uncertainty rather than solve it.

Third, evaluate the temporal relationship between editing, recovery, and stimulation. Some primary myeloid phenotypes are only interpretable after cells return to a stable baseline following delivery. Others are transient and require a tightly defined stimulation window. Record the time at which guide identity is stable, the time at which the edit becomes measurable, the time at which viability stops declining, and the time at which the phenotype is assayed. This timeline determines whether a null result means a nonessential gene or a readout collected before the perturbation could act.

Fourth, test whether donor variation changes the operating window. A condition that appears acceptable in one donor can be too stressful, too weakly edited, or too skewed in another. The pilot should therefore use donors as independent biological observations. The objective is not to claim population-level effect sizes from a small study. It is to learn whether the workflow has a reproducible technical operating range and whether the eventual analysis must be stratified by donor, maturation state, or baseline response.

A practical go or no-go summary should document four evidence blocks: library retention, target editing, myeloid-state preservation, and endpoint usability. Library retention asks whether enough of the planned guide set persists through the final sampling step. Target editing asks whether barcode-positive cells show an appropriate on-target edit spectrum. State preservation asks whether delivery controls create broad activation, toxicity, or compositional shifts. Endpoint usability asks whether the assay has sufficient dynamic range and recovered cells to compare perturbations. A weak result in any block is a reason to change one variable and repeat the pilot, not a reason to apply stronger statistics to fragile data.

This staged approach also sharpens collaboration between vector production, cell biology, sequencing, and analysis teams. The vector group can own guide-pool composition and delivery consistency. The cell team can own recovery and state controls. The sequencing group can own barcode and amplicon quality. The analysis team can set dropout metrics, guide aggregation rules, and donor-aware models before results arrive. When responsibilities are named at the pilot stage, the final screen has a clear evidence trail rather than a collection of disconnected QC files.

A Stage-Gated Implementation Plan

Gate 1: define the claim. State whether the screen seeks survival, a marker shift, a transcriptional state, or a morphology phenotype. Define the falsifying observation and the appropriate readout.

Gate 2: prove the hybrid components separately. Confirm guide representation and single-guide behavior. In parallel, confirm VLP-mediated editing and a functional positive-control phenotype.

Gate 3: run a mini-library pilot. Include the planned controls, at least two donors where feasible, and sequencing checkpoints before and after each major biological transition.

Gate 4: launch only with a preserved evidence chain. Advance when viability, activation-state QC, guide coverage, edit confirmation, and endpoint recovery support the intended inference. Otherwise, change library scope, delivery dose, or readout before committing the larger cohort.

Frequently Asked Questions

References:

1. Jung H, Devant P, Ching C, et al. Virus-like particles enable targeted gene engineering and pooled CRISPR screening in primary human myeloid cells. Nature Biotechnology. 2026. doi:10.1038/s41587-026-03258-2.
2. Banskota S, Raguram A, Suh S, et al. Engineered virus-like particles for efficient in vivo delivery of therapeutic proteins. Nature Biotechnology. 2022. doi:10.1038/s41587-021-01219-7.
3. Mangeot PE, Risson V, Fusil F, et al. Genome editing in primary cells and in vivo using viral-derived nanoblades loaded with Cas9-sgRNA ribonucleoproteins. Nature Communications. 2019. doi:10.1038/s41467-018-07845-z.
4. Parnas O, Jovanovic M, Eisenhaure TM, et al. A genome-wide CRISPR screen in primary immune cells to dissect regulatory networks. Cell. 2015. doi:10.1016/j.cell.2015.08.059.
5. Jung H, Devant P, Ching C, et al. GSE327644: Virus-like particle-enabled gene engineering and pooled CRISPR screening in primary human myeloid cells. Gene Expression Omnibus. 2026. Dataset accession; no DOI assigned.