Choosing a microfluidic material: PDMS, COP, COC, PMMA, PS and glass

If your device will contact a hydrophobic small molecule, do not make it from PDMS. That is the strongest and best-evidenced statement in microfluidic material selection, and it survives every qualification you can attach to it: identical microdevices fabricated from eight polymers showed compound absorption differing by more than 1000-fold, with PDMS the most absorptive material tested and more than 99% of chlorpromazine partitioning into the PDMS bulk within 24 hours.

For everything else — prototyping speed, feature fidelity, gas exchange, optical access — PDMS is frequently the right answer, which is why the material question cannot be settled by a single rule. The matrix below is the version we use.

The material matrix

The absorption problem, properly

This is the part of the material decision that is waved through most often and reversed least often — once a device is bonded and validated, the material is fixed — so it deserves the evidence in full.

The multi-polymer comparison. Kemas and colleagues fabricated identical microdevices from eight polymers and measured absorption of prototypic compounds spanning a wide range of physicochemical properties. Their findings, quoted directly from the paper:

• Absorption differed by more than 1000-fold between polymers, with PDMS the most absorptive and PTFE and thiol-ene epoxy the least.
• More than 99% of chlorpromazine partitioned into the PDMS bulk after 24 hours; absorption in PMMA and thiol-ene epoxy was around 1500 times lower.
• Absorption correlated strongly with compound hydrophobicity: logP alone explained the great majority of the variance (R² = 0.94), with additional contributions from hydrogen donor count and rotatable bond count.
• PDMS exhibited an almost binary absorption profile — negligible absorption for compounds with logP below 1, and absorption above 90% for logP above 1.25.
• No polymer absorbed compounds with logP below 1. For compounds with logP between 2 and 3.5, absorption was similar across all materials except PDMS, where it was significantly higher. Above logP 3.5, PTFE and thiol-ene epoxy were the least absorptive.
• Absorption was rapid: more than 50% of compound loss occurred within 10 minutes, with the materials then diverging by orders of magnitude at 2 and 24 hours.

The consequence they demonstrated is the reason this matters commercially. Using isogenic 3D human liver cultures, chlorpromazine toxicity was detected at therapeutically relevant concentrations in low-absorbing thiol-ene epoxy devices but not in PDMS (reported as 115.1% versus 5.9% viability in PDMS versus thiol-ene epoxy). Correcting the nominal in vitro TC50 for measured absorption brought the values into agreement with clinically reported toxic plasma concentrations: chlorpromazine nominal 10–50 µM, 93% absorbed, corrected 0.7–4.5 µM, against clinical 1.5–6 µM. Montelukast: nominal 10–50 µM, 96% absorbed, corrected 0.4–2 µM, against clinical 1 µM. Acetaminophen, at 0% absorption, needed no correction at all.

In other words, absorption does not add noise. It shifts your dose-response curve to the right by a compound-dependent amount, and it does so invisibly.

The complication. A second study is worth reading alongside it, because it cuts against a convenient assumption. van Meer and colleagues measured absorption of four cardioactive drugs by PDMS-coated culture wells against standard tissue-culture polystyrene. Verapamil and nifedipine were absorbed 20–50% more by PDMS than by TCPS after three hours; for Bay K 8644 there was no difference between the two. Bepridil absorption by PDMS exceeded 80% at three hours, reduced to roughly 50% by a commercial coating. Critically, they found no clear correlation between absorption and logP in their compound set, and none with molecular weight or a combination of the two, explicitly contradicting the proposal that all compounds above logP 2.67 show high absorption.

Both papers are correct and they are not in conflict. Across many compounds and eight materials, logP predicts absorption well. Within a small set of structurally particular compounds, it can fail. The practical rule that follows is: use logP to decide whether you have a problem, and measure your own compound to find out how large it is.

Mitigations, and their limits. Coatings work partially. A PTFE coating optimised on PDMS surfaces reduced absorbed Nile red fluorescence roughly two-fold going from 1–2% to 3–6% PTFE concentration. Sol-gel treatments, paraffin coatings and BSA passivation have all been published, with documented weaknesses — device swelling or dissolution, poor thermal stability, and short effective lifetimes respectively. A coating is a mitigation. Choosing a different polymer is a solution.

Autofluorescence: the imaging tax

If your readout is fluorescence, the substrate contributes background, and the size of that contribution depends on material and on how the device was made.

A direct comparison of thermoplastic fabrication routes found that at 350 nm excitation there was significant autofluorescence across all materials tested, with all COP sheets — treated or untreated — significantly lower than polystyrene. At 647 nm, autofluorescence was negligible across the board. Between those extremes, at 492 nm and 572 nm, material and treatment mattered critically: all COP sheets showed negligible autofluorescence at both wavelengths, comparable to 647 nm, while thermally treated polystyrene showed significantly increased autofluorescence above 400 nm, appearing not as uniform background but as large distinct circular patches comparable in size and brightness to fluorescently labelled cells.

Three things follow:

1. Fabrication process, not just material, sets autofluorescence. Thermal bonding of PS made it worse; solvent bonding did not. If you specify a material without specifying the bonding method, you have not specified the optical behaviour.
2. COP is the strong default for fluorescence work in the green and red channels.
3. UV-cured 3D-printed resins are the worst case. uFluidix notes that any 3D print material exhibits enormous autofluorescence when UV light is used to cure the resin, alongside poor optical quality and residual uncured material in closed channels.

Note also what the commercial platforms have converged on. MIMETAS builds the OrganoPlate on 150 µm thin microscope-grade glass with explicitly stated non-absorbent materials. Emulate sells the Chip-R1 Rigid Chip specifically marketed on a low-drug-absorption profile for ADME and toxicology, alongside its elastomeric Chip-S1. Both are commercial acknowledgements of the same two problems.

Gas permeability cuts both ways

PDMS is highly gas-permeable. In a thick-walled culture device with a shallow chamber that is a genuine advantage: oxygen reaches the tissue without active perfusion, and CO₂ equilibrates with the incubator. It is also why open-top PDMS culture devices work at all for dense 3D constructs.

The same property causes three problems:

• Evaporation. Water vapour leaves through the bulk. In small-volume chambers over multi-day culture, medium osmolality drifts. This is why vendors sell evaporation-control accessories — eNUVIO lists an evaporation minimiser as a stocked accessory for its OMEGA line.
• Volatile compound loss. Anything that partitions to the gas phase leaves through the walls, compounding absorption into the bulk.
• Bubble behaviour. Gas permeability helps clear bubbles in some geometries and helps generate them in others, particularly with temperature swings.

Thermoplastics and glass have essentially no gas permeability, which removes all three problems and introduces a fourth: you must supply oxygen deliberately, by perfusion, headspace or a membrane.

Cost per part, by volume

These bands are reasoned estimates from published process descriptions, not quotations. Treat the shape of the curves as informative and the absolute numbers as illustrative only.

The crossover to watch is PDMS versus embossed thermoplastic. Below roughly a hundred parts, PDMS usually wins on total cost and always wins on schedule. Above roughly a thousand, thermoplastic wins decisively. Between those figures the answer depends almost entirely on how many post-processing operations touch each part — which is the same conclusion the quote anatomy page reaches from the other direction.

Decision shortcuts

Choose PDMS when: you are prototyping and the geometry will change; you need elastomeric function such as valves, stretchable membranes or deflecting force-readout posts; you need gas permeability for a dense 3D construct in an open chamber; and either no small molecules are involved or every compound of interest has logP below about 1.

Choose COP or COC when: fluorescence imaging matters, hydrophobic compounds are involved, and you expect to scale beyond a few hundred parts. This is the default for commercial organ-on-chip consumables for exactly those reasons.

Choose PMMA when: you need low absorption and rigid parts at moderate cost with easy machining, and you do not need COP’s optical performance or solvent resistance.

Choose glass when: absorption must be effectively zero, high-NA imaging through the substrate is required, or organic solvents are present. Accept the fabrication cost and the fragility.

Choose PTFE or thiol-ene epoxy when: the compounds are highly hydrophobic — logP above 3.5 — and absorption is the dominant constraint. These were the least absorptive materials in the eight-polymer comparison.

Choose polystyrene when: you want to match the surface chemistry of standard tissue-culture plastic and your imaging avoids the 350–572 nm autofluorescence trap — and if you do choose it, specify solvent bonding rather than thermal bonding.

What to write in the RFQ

State the material and the constraint that drove it in the same sentence: “COP, because the assay uses a logP 4.2 compound and PDMS absorption would invalidate the dose-response.” A fabricator who knows the constraint can propose an alternative that meets it. A fabricator who only knows the material will quote what you asked for, including when what you asked for is wrong.

The full RFQ field list is on the custom fabrication page.