Organ-on-chip vs Transwell vs static culture: a decision matrix

The short answer: use a Transwell unless you can name the specific physiological variable that only perfusion provides, and say how you will measure it. Shear stress, apical-basal flow, recirculating multi-organ linkage and mechanical strain are real capabilities that a static insert cannot deliver. “More physiological” is not one of them.

That framing is not a criticism of organ-on-chip technology. It is how you avoid buying a throughput-costly, consumable-locked platform to answer a question a $12 insert already answers, and it is also how you recognise the cases where the chip is genuinely the only option.

The decision matrix

When Transwell is the right answer

This section is the reason to trust the rest of the page. There is a strong commercial incentive across this industry to present the chip as the default and the insert as the legacy option, and it is not true. Here are the cases where a permeable support is not merely acceptable but correct.

1. Your readout is transport across a barrier with a clean area term

Apparent permeability calculations need a defined surface area, a defined donor and receiver volume, and a well-mixed compartment. A Transwell gives you all three by construction. On a chip you have a flow field, a residence time and a geometry-dependent effective area, all of which have to be characterised before the number means anything.

If the question is “what is the Papp of this compound across this epithelium”, the insert is not the inferior tool. It is the tool the calculation was designed around.

2. The tissue does not need flow to differentiate properly

Some epithelia mature perfectly well static. Altis Biosystems’ RepliGut Planar is instructive here: it is a primary human intestinal stem cell model grown submerged on a semi-permeable membrane, producing absorptive enterocytes, goblet cells and enteroendocrine cells, with both apical and basal surfaces accessible for dosing and sampling — and it is deliberately not perfused. A company whose entire business is human intestinal models chose the static architecture, because for the gut questions they answer it is the right one.

The general principle: match architecture to the biology, not to the marketing. Ask whether flow changes the phenotype you care about, and demand a citation rather than an assertion.

3. You need throughput and replicates more than you need realism

Statistical power is a physiological consideration too. A 96-insert format at four replicates gives you 24 conditions. A 12-chip culture module at four replicates gives you three. If your compound set is large, the insert answers the question this quarter and the chip answers it next year.

4. Regulatory or historical comparability matters

Transwell-based transport and barrier assays have decades of accumulated data behind them. If your result needs to sit alongside a body of existing measurements — internal historical data, a literature comparison, or a validated test guideline — a novel architecture creates a comparability burden that has to be discharged before anyone believes the number.

5. Your compound is hydrophobic and the available chip is elastomeric

This one is decisive and frequently missed. Kemas and colleagues fabricated identical microdevices from eight polymers and measured compound absorption differing by more than 1000-fold, with PDMS the most absorptive; more than 99% of chlorpromazine partitioned into the PDMS bulk within 24 hours, while absorption in PMMA and thiol-ene epoxy was around 1500 times lower.

A polystyrene or polycarbonate Transwell with a low-absorption compound is a more accurate experiment than a PDMS chip with a compound that has vanished into the walls. If you do go to a chip with such a compound, use a rigid, low-absorption chip — Emulate sells the Chip-R1 explicitly on that basis, and MIMETAS states non-absorbent materials as a platform property. The material guide sets out the full data.

6. You have no dedicated culture staff

Pumped chip platforms have a real operational burden: priming, bubble management, tubing connections, and per-chip fluidic handling. Every one of those is a skill and a failure mode. If the work will be done by rotating students or by an engineering group without daily culture experience, an insert plate is a system that tolerates inexperience and a tubed chip is not.

The mitigation, if you want a chip anyway, is to choose a pumpless architecture. MIMETAS’s gravity-driven OrganoFlow rocker perfusion removes pumps, tubing, bubbles and fluidic connections entirely, at the cost of shear control. That is often the right trade.

When the chip genuinely wins

Equally honestly, here is when nothing else will do.

• Shear stress is a variable in your biology. Endothelial alignment, glycocalyx integrity, renal proximal tubule transporter expression and hepatocyte zonation all respond to flow. If the phenotype you need requires shear, no insert produces it. Note also that orbital shaking generates meaningful shear at the periphery of wide wells and can align and activate endothelial cells — so test whether a shaker gets you there before buying a platform.
• You need mechanical strain. Breathing lung, peristaltic gut. Only stretchable chips do this.
• You need recirculating multi-organ linkage. Pro-drug activation in liver acting on a second tissue, or PK-like exposure profiles across compartments. This is the strongest structural case for the category and it is not approximable with inserts.
• Continuous sampling of a small volume over time. Longitudinal media sampling from a perfused circuit is a fundamentally different dataset from endpoint sampling.
• Long-term culture where static media exhausts. Perfusion sustains metabolically demanding tissue for weeks where static culture cannot.
• The tissue geometry is the point. Perfusable lumens, angiogenic sprouting into matrix, migration across a gradient. Membrane-free chip architectures were built for these.

Cost per data point: how to do the arithmetic

Only one number in this comparison is public. The Corning HTS Transwell-24 system lists at $556.00 per case of two units through Fisher Scientific, captured 2026-09-01 — approximately $11.60 per insert at that pack size, and individual-format inserts in smaller packs are cheaper still per unit. No organ-on-chip vendor publishes a consumable price at all.

So the comparison has to be built from a quote. Use this structure and fill the chip column from the vendor.

The bottom row is the only number that compares across architectures. Cost per chip is not comparable to cost per insert, because a chip that needs six replicates where an insert needs three is twice as expensive at the same unit price.

Ask every chip vendor the five questions set out on the platform comparison page: instrument cost including required accessories, consumable cost at three annual volumes, cell cost where the platform requires vendor cells, the ready-to-use tissue premium, and a three-year total on your actual experiment count.

A short decision path

1. Does your readout need a defined apical and basolateral compartment? If no, static may be sufficient — start there.
2. Does the biology require flow, strain, or organ-to-organ linkage, and can you name the variable and the measurement? If no, Transwell. If you cannot name it, the honest answer is no.
3. Is your compound hydrophobic? If yes, either a Transwell or a rigid low-absorption chip. Never an elastomeric chip without an absorption correction.
4. How many conditions per year do you need? Divide by replicates and controls before comparing with chip throughput figures. Organ-chips are not a throughput technology.
5. Who runs it? No dedicated culture staff means either an insert plate or a pumpless plate-format chip.
6. Can someone else run it for you? Several platform vendors offer studies as a service. Buying the answer is frequently cheaper than buying the capability, particularly for a first study — see custom organ-on-chip development.

The honest summary

Organ-on-chip is a physiological-relevance technology with a throughput cost and an operational cost. When the variable it adds is one your biology actually depends on, nothing substitutes for it and the cost is worth paying. When it is not, you have bought a more expensive, more fragile, lower-throughput version of an experiment a permeable support does better.

The field’s own most successful products acknowledge this. A dedicated intestinal model company ships a static planar architecture. A leading chip vendor removed the pumps. Those are not concessions; they are the correct engineering answers to specific questions, and choosing well means being willing to arrive at the unglamorous one.

How we can help

We hold no reseller relationship with any platform vendor, which means we can run the cost-per-usable-condition comparison honestly and tell you when the answer is a rocker and a box of inserts. Where a chip is genuinely required, we run the same five-question pricing enquiry across vendors and normalise the responses. Send us the readout, the compound properties, the conditions per year and who will be running the work.