Designing a Cell Biology Experiment

A student images one stretched dish and one unstretched dish, scores 600 cells in each, and reports a difference with a p-value below 0.0001. How many experiments did the student do?

One. Every cell in the stretched dish shared the same dish, the same pipetting, the same afternoon on the stretcher. If anything about that dish was odd, all 600 cells carry the oddity together. This lesson is about designing an experiment that can actually tell you something, using one running case: does stretching an epithelium make its cells proliferate? The cells are MDCK cells, a line derived from dog kidney epithelium (Madin-Darby canine kidney), grown in culture as a sheet. Everything described here was measured in those cultured cells.

Start from two models, not one hypothesis

“Stretch makes cells proliferate” sounds like a hypothesis, but it is hard to fail. Almost any rise in almost any measurement can be read as support. A sharper start is two models that predict different things.

• Model A: stretch is sufficient. Stretched quiescent cells re-enter the cycle and go on to divide.
• Model B: stretch lets cells start the cycle, but a second signal, from the Wnt pathway, is needed to finish it.

Wnt proteins are secreted signals that stabilize beta-catenin, a protein that then switches on target genes; the lesson on Wnt signaling covers the pathway. The two models agree that stretch should push cells into S phase. They disagree about division. That is where the experiment has to look.

Think of two suspects with the same alibi for the morning and different alibis for the afternoon. Questioning them about the morning gets you nowhere. Before you collect a single image, write down what each model predicts, as a number, on a measurement you choose in advance. Also write down what result would favour neither. An experiment that can only confirm is not a test. The lesson on experiments that distinguish made the same point about perturbation and rescue.

What counts as N

The experimental unit is the thing independently assigned to a treatment: a dish, a stretch device, a mouse. The observational unit is the thing you measure: a cell, a field of view. N, the sample size, is the number of experimental units (Lazic and colleagues, 2018). Measuring more cells adds observational units and no experimental units. Treating cells as independent replicates when they share a dish is called pseudoreplication. Repeating the whole procedure on separate days is what supplies independent information; Lazic and colleagues call the choice of days rather than wells a scientific judgement, made because it tests whether an effect survives a fresh run.

Worked problem

Here is a made-up data set, invented for teaching. Three independent runs, on different days. Each run has one control dish and one treated dish, and 200 cells are scored per dish for EdU, a thymidine analogue that cells build into DNA as they replicate it.

Run	Control (% EdU+)	Treated (% EdU+)	Difference
1	20	34	14
2	25	27	2
3	15	28	13

The wrong analysis: pool the cells. Control: 40 + 50 + 30 = 120 of 600 cells, 20.0%. Treated: 68 + 54 + 56 = 178 of 600, 29.7%. The difference is 9.7 points. Treating each cell as independent, the standard error of the difference is

SE = sqrt(0.20 × 0.80 / 600 + 0.297 × 0.703 / 600) = 0.0248, about 2.5 points

so z = 9.7 / 2.5, about 3.9. It looks decisive.

The right analysis: the dish is the unit, and each run pairs its two dishes.

1. Paired differences: 14, 2, 13. Mean = 29 / 3 = 9.67 points.
2. Deviations from the mean: 4.33, -7.67, 3.33. Sum of squares = 18.8 + 58.8 + 11.1 = 88.7. Divide by n - 1 = 2: variance 44.3, so SD = 6.66.
3. SE = 6.66 / sqrt(3) = 3.84.
4. t = 9.67 / 3.84 = 2.51, with 2 degrees of freedom. The two-sided p is about 0.13.
5. 95% interval: 9.67 ± 4.303 × 3.84 = 9.67 ± 16.5, about -6.9 to 26.2 points.

Conclusion. The effect is suggestive, not shown. The interval runs from a small decrease to a large increase. Notice what p = 0.13 does not say: it does not show “no effect”. Scoring 2,000 cells per dish would not help, because the uncertainty comes from run-to-run variation. More independent runs would.

A SuperPlot (Lord and colleagues, 2020) shows this in one figure. Each cell is a small dot, colored by run; each run’s mean is a large symbol of the same color; the comparison uses the large symbols. Here run 2 would stand out at once as the run where treatment barely moved the mean.

Analysis

pool all cells one value per run

Resample cells Back to the lesson’s table

Quantity	Lesson’s table (expected)	This draw (observed)

N is the number of independent runs, not the number of cells. The data are invented for teaching; Resample redraws each dish’s cells from its percentage, so observed counts wobble around the expected ones. This is a sketch of the logic, not a result.

Measuring the right thing: entry versus completion, area versus cell number

The student’s evidence for proliferation was more green cells. In MDCK cells carrying Fucci reporters (fluorescent ubiquitination-based cell-cycle indicator), red mKO2-Cdt1 marks G0 and G1, and green mAG-Geminin marks cells from S phase until mitosis (see the lesson on cell-cycle control). A green cell has entered S. It has not necessarily divided.

This is exactly what Benham-Pyle and colleagues found in 2016. In quiescent MDCK-II monolayers, about 10% of unstretched cells were Geminin-positive. Strain alone, at a maximum of about 8.5% in their live-imaging device, increased the number of cells in S and G2 without increasing the number tracked from G1 through S to division. Wnt3A alone did not move quiescent cells into S phase in this preparation. Strain plus Wnt3A raised mitotic entry above either input alone. More green, then, was a pile-up, not more division.

Two more limits of the reporter matter. Red does not distinguish G0 from G1, so red alone cannot prove a cell is quiescent. And the construct has a dark period after mitosis, when neither color is bright, so a snapshot undercounts cells that have just divided.

Area has its own trap. Suppose two circular MDCK tissues, 1.7 and 3.4 mm across (the sizes Heinrich and colleagues used in 2020), each with an edge moving outward at a constant 30 micrometres per hour for 10 hours, an idealized interval. The radii go from 850 to 1,150 micrometres and from 1,700 to 2,000.

area ratio = (R + vt)² / R²: (1150/850)² = 1.83 and (2000/1700)² = 1.38

The small tissue grows 83% in area and the large one 38%, with identical edges and no division rate in the calculation at all. Percent area growth is not proliferation. Count nuclei, as the lesson on microscopy and images describes, and track divisions.

Controls that rule out the boring explanations

Each control answers one dull explanation.

• Handling. Unstretched devices assembled, mounted and imaged identically. Otherwise “stretch” includes everything done to the stretched device.
• The medium. Wnt3A-conditioned medium comes from mouse L cells engineered to make Wnt3A, so the control is medium from parental L cells, which carries everything except the Wnt3A.
• The drug. Vehicle (usually DMSO) at the same concentration. Benham-Pyle and colleagues used SU6656 at 10 µM to inhibit Src-family kinases and found it blocked strain-induced EdU incorporation. A concentration used in a paper is a condition, not a measure of potency.
• Off-target effects. A casein kinase I inhibitor, which allows beta-catenin to build up, raised mitotic entry when combined with strain, and iCRT3, which blocks beta-catenin-dependent transcription, prevented that rise. The authors themselves flagged possible off-target effects of the casein kinase I inhibitor. One inhibitor never proves its nominal target. Add an independent genetic perturbation with rescue (CRISPR and functional genomics; drugs against a part).
• The observer. Positivity thresholds kept the same across conditions, as the 2016 authors did, ideally fixed before analysis, and scoring blind to condition.
• The stretch itself. Measure the strain the cells actually feel and analyze only the strained region. The two devices in the 2016 study reached different maxima, about 8.5% and 15%, so their results should not be merged.

A pathway reporter checks engagement: TOPdGFP, a destabilized GFP driven by beta-catenin-responsive sites, asks whether Wnt3A did anything at the level of transcription (hacking the cell covers reporters).

Putting it together

Design problem

A student claims “stretch makes epithelial cells proliferate” from more green Fucci cells after stretching an MDCK monolayer. Redesign the experiment.

One way to do it

Design. A 2 × 2 factorial: no stretch or stretch, crossed with control (parental L-cell) or Wnt3A-conditioned medium. Run all four arms on each day, assigning devices to arms at random within the day. N is the number of independent days, chosen from a pilot’s run-to-run variation.

Primary endpoint, fixed in advance. The fraction of cells red at the start that are tracked through green (S/G2) to an observed division within a set window. Call it Q.

Predictions, with invented thresholds (they are teaching choices, not values from any paper).

• Model A: Q(stretch) - Q(control) is at least 10 percentage points.
• Model B: Q(stretch) - Q(control) is at most 2 points, and Q(stretch + Wnt3A) - Q(stretch) is at least 10 points.
• Inconclusive: a stretch-alone difference between 2 and 10 points, or confidence intervals wide enough to fit both models.

Secondary endpoints. Nuclear counts and density, EdU, TOPdGFP, and cell loss, each reported separately.

Controls. Identically handled unstretched devices; L-cell medium; vehicle for any drug; fixed thresholds; blinded scoring; measured strain. For molecular claims, a genetic perturbation of beta-catenin signaling with rescue alongside any inhibitor.

Analysis. Day-paired device means, shown as a SuperPlot.

Compare with the published result. In Benham-Pyle and colleagues’ 2016 study, strain alone raised S/G2 without more completed cycles, and strain plus Wnt3A raised mitotic entry: the pattern Model B predicts, in quiescent canine MDCK-II monolayers in culture. Whether the same holds in other epithelia, or in tissue, is a separate experiment.

These habits carry into the rest of the unit. The lessons on epithelia, stem cells and cell state and disease all rest on claims about proliferation and identity, and each claim is only as good as its unit, its endpoint and its controls. For the probability behind the statistics, see the lesson on probability and crosses.