Cell biology assays — standardizing migration, invasion and reporter readouts

Cell biology assays are the workhorses of mechanistic research — migration, invasion, adhesion, reporter expression, protein interaction — but they are also among the least standardized. Two labs running the same Boyden chamber invasion assay on the same cell line routinely report invasion indices that differ by two- to three-fold, not because the biology is different but because the protocol has ten undocumented degrees of freedom. This piece addresses what to lock down and how to analyze the output so that the results survive inter-lab scrutiny.

Scratch / wound-healing migration assay

The scratch assay is ubiquitous and reproducible when controlled; uncontrolled, it is one of the highest-variance assays in cell biology. The hidden variables:

• Scratch width at T=0: different operators, different pipette tips, different seeding densities produce scratch widths ranging from 300 to 900 µm. Normalize all migration data to initial scratch width — not to absolute closure in microns.
• Cell density at time of scratch: cells must be confluent but not over-confluent; quantify confluence by image analysis before scratching and report it.
• Serum concentration in migration medium: serum drives migration. Many labs "reduce serum to 0.5%" without validating that this is the minimum chemotactic gradient. A serum concentration series in a Boyden chamber determines the effective gradient for your cell line.
• Image acquisition timing: automated imaging at fixed intervals is the only way to get a smooth migration curve. Manual endpoint snapshots miss kinetics and introduce operator-selection bias.

Analysis: measure wound area (not width — area is rotation-invariant) at every time point with ImageJ/CellProfiler, normalize to T=0, fit a linear model on the first half of closure (where migration is not inhibited by confluence), and report the migration rate (µm²/h) with 95% CI. Report at least n=6 independent experiments, each with ≥ 3 technical replicates (wells per condition).

Boyden chamber invasion assay

The transwell invasion assay measures invasion through a Matrigel layer and is one of the most protocol-dependent assays in oncology research. Variables that must be locked in the SOP:

• Matrigel lot and coating thickness: Matrigel is a biological extract with batch-to-batch variation. Use a single aliquot for each experiment; record lot number; verify gel thickness by confocal z-stack on a sacrificed insert.
• Matrigel concentration: 0.5 mg/mL vs. 2 mg/mL produces dramatically different invasion indices for the same cell line. Pick one concentration and never deviate.
• Serum gradient across the membrane: document upper- and lower-compartment serum concentration explicitly. The gradient is the chemoattractant.
• Cell number and starvation duration: seed at a density validated to produce sub-confluence at the membrane at assay endpoint. Starvation prior to seeding (0.5–1% serum, 16 h) synchronizes cell cycle and reduces proliferation contribution to apparent invasion.
• Fixation and staining timing: crystal violet, calcein-AM, DAPI — pick one and fix the timing. Staining intensity drifts with fixation time.

QC inclusion criterion: non-invaded side must be free of cells (validate with a no-Matrigel control insert counted in parallel — if cells reach the lower face without Matrigel, the seeding density or culture conditions are wrong).

Reporter gene assays (luciferase, GFP/RFP)

Dual-reporter systems (Firefly + Renilla or Firefly + β-galactosidase) are the gold standard because the Renilla/β-gal control normalizes for transfection efficiency, cell number, and non-specific lysis. Single-reporter assays without a transfection control are a common source of inflated effect sizes.

Critical standardization points:

• Transfection efficiency monitoring: cotransfect a GFP or mCherry construct and measure % positive cells by imaging or flow cytometry before lysis. Experiments with < 40% efficiency should be discarded — the surviving signal comes from a non-representative cell population.
• Passage number: reporter expression changes with passage number in many cell lines. Validate the assay window in early, middle and late passage and document acceptable passage range.
• Lysis and reading timing: luciferase activity decays after lysis. Process all wells within 30 min of adding lysis buffer; use a plate stacker if scale demands it, not endpoint lysis of the whole plate at once.
• Normalization: report Firefly/Renilla ratio, not raw luminescence. Do not report fold-changes over vehicle without reporting the absolute ratios — reviewers need to judge the absolute assay window.

Proximity ligation assay (PLA)

PLA (Duolink or equivalent) detects protein–protein interactions or post-translational modifications in situ with single-molecule sensitivity. The assay is powerful but the quantification is frequently mishandled:

• PLA dot count per cell vs. per image: report dots per cell (normalize to nuclear count or cell mask area), never dots per image — cell density varies between conditions.
• Background control: run the assay with one primary antibody omitted. A PLA with > 3 background dots per cell per field for the single-antibody control has non-specific amplification that will inflate the positive signal.
• Confocal Z-stack vs. widefield: widefield images integrate signal from the full cell volume and overestimate dot counts in thick cells. Use a single confocal slice at the nuclear midplane for counting unless three-dimensional distribution is specifically studied.

Statistical framework for cell biology data

Three persistent errors in cell biology statistics:

1. Treating technical replicates as independent experiments: wells on the same plate on the same day share plate effects, reagent batch, and experimenter. They are pseudoreplicates. Use the well mean as a single data point; n for the t-test is the number of independent experiments (biological replicates), not the number of wells.
2. Reporting only mean ± SEM from three experiments: n=3 is almost never enough to power a two-sample test to 80% — you need ~6 per group. Report the data points, not just the error bar.
3. Multiple comparisons without correction: a figure with six treatment groups and fifteen pairwise comparisons with unadjusted p-values will show two to three "significant" results by chance alone. Use Tukey or Dunnett correction for comparisons to a single control.

How AiLabrix fits

Drop the image analysis export (CellProfiler, Fiji, Operetta, IncuCyte) plus the experimental metadata CSV. The pipeline imports wound area over time, invasion counts per insert, reporter ratios, or PLA dot counts; applies the correct normalization; runs mixed-effects models accounting for the plate/experiment hierarchy; performs multiple-comparison-corrected hypothesis tests; generates migration curves, bar plots with individual data points, and SAR-style perturbation matrices. Signed PDF with reproducibility metrics per assay type, power analysis for current n, and suggestions for minimum n to reach 80% power at the observed effect size. [email protected].