3D Cell Culture 101: Choosing Your System

Team Dynamic Matrices | 14 November 2025

When people ask us how to "move to 3D", the honest answer is that there is no single 3D. There are four architectures that cover almost everything we see in the field, and each one answers a different kind of question. Most of the frustration we hear about comes from picking the architecture first and the question second. So let us start with the question.

First, decide what you are actually measuring

Before comparing systems, write down the one readout that will make or break the project. In our experience it falls into one of these buckets:

• Viability, growth or a simple drug response where you need throughput more than fidelity
• Morphogenesis and organisation (lumens, polarity, layering, branching) where the surrounding matrix is part of the experiment
• Mechanobiology where you need to set and change stiffness or stress relaxation deliberately
• Transport, flow or tissue-tissue interfaces where perfusion is the point

Once you know which bucket you are in, the choice of system is usually obvious. Below is how we think about each.

Embedded culture

Cells are fully surrounded by a hydrogel, so mechanical and biochemical cues arrive from every direction rather than from a single surface. This is the default for organoids and for any study where cell-matrix interaction is part of the biology. The landmark intestinal organoid protocol from Sato and colleagues (Nature, 2009) was an embedded culture, and almost every organoid lineage since has followed the same logic.

What we like: it supports the widest range of downstream work, from live imaging to sequencing, and it runs in standard plates.

What to watch: the matrix is now an experimental variable. If it is undefined and varies by lot, that variability lands directly on your organoids. We wrote a separate post on why that matters for reproducibility.

Scaffold-based culture

Cells are seeded onto a pre-formed structure, often porous and often fabricated (electrospun fibres, decellularised tissue, printed lattices). This is the right tool when macro-geometry is the experiment: engineered cartilage, bone constructs, or anything destined for a defined shape.

What we like: architecture you can specify at the millimetre scale.

What to watch: signalling is dominated by the scaffold surface, so mechanical cues are less uniform than in an embedded gel, and cell recovery for analysis is usually harder.

Spheroid culture

Cells aggregate in low-attachment plates or hanging drops with no external matrix. It is the cheapest entry point into 3D and it scales well. For a first-pass cytotoxicity screen or a simple hypoxia gradient, it is often all you need.

What we like: fast, cheap, robot-friendly.

What to watch: there is no matrix to tune, so anything involving stiffness, ligand presentation or ECM remodelling is out of scope. Spheroids also develop necrotic cores past a few hundred micrometres, which can confound drug penetration studies.

Microfluidic and organ-on-chip systems

Perfused channels, often with two or more compartments separated by a membrane, that let you model flow, shear, barrier function and tissue-tissue interfaces. Ingber's review in Nature Reviews Genetics (2022) is the best single overview of what these systems can and cannot do.

What we like: dynamic conditions no static plate can reproduce.

What to watch: hardware cost, a real learning curve, low throughput and (in PDMS devices) absorption of small hydrophobic drugs into the chip material, which can silently lower effective dose.

How the four compare

Our recommendation for a lab starting out

Start with embedded culture in a well-characterised matrix. It gives you the largest option space later: you can add flow, you can scale down to spheroid-like aggregates inside the gel, and you can move the same cells into a chip once you know what they need. The reverse path, from chip back to plates, usually means starting over.

Two practical tips from our own bench:

1. Run your first embedded experiment alongside your existing 2D assay for the same endpoint. The difference in drug response between 2D and 3D is often large (we cover this in Flat Biology Is Lying to You), and seeing it in your own hands is more convincing than any review.
2. Pick a matrix you can specify. If you cannot state the stiffness and stress relaxation of your gel, you cannot report them, and you cannot troubleshoot them when results drift.