Flat Biology's Blind Spot: What 2D Culture Still Gets Wrong About Drug Response
Team Dynamic Matrices | 26 January 2026
Two-dimensional culture built modern pharmacology. It is cheap, fast and reproducible in the narrow sense that a monolayer behaves consistently within one lab. It is also wrong in ways that matter for the two decisions it is most often used for: is this compound effective, and is it safe. The industry keeps it as the default screening substrate not because it best predicts human response but because the infrastructure was built around it. That inertia has a cost, and the cost shows up in trials that fail for reasons the preclinical data never surfaced.
The attrition numbers, honestly
Wong, Siah and Lo (Biostatistics, 2019) analysed more than 21,000 compounds and found an overall likelihood of approval from phase 1 of around 14 percent, and roughly 3 to 4 percent in oncology. Harrison (Nature Reviews Drug Discovery, 2016) attributed about half of phase 2 and 3 failures to lack of efficacy and a quarter to safety. Those are exactly the two properties 2D screening was supposed to have already characterised.
It would be an overstatement to blame culture format for all of that. It would be an understatement to say it plays no role. When a platform cannot capture diffusion-limited drug penetration, cell-matrix signalling, or the resistance mechanisms that emerge only in 3D architecture, it is not measuring a simplified version of the truth. It is measuring something structurally different.
Efficacy: the dose can be off by orders of magnitude
Comparisons of the same line in 2D and 3D find effective doses that differ by up to three orders of magnitude for the same compound (Edmondson et al., Assay and Drug Development Technologies, 2014). Karlsson et al. (Experimental Cell Research, 2012) showed colon cancer spheroids were markedly more resistant to 5-fluorouracil and oxaliplatin than the same cells grown flat, in a pattern that matches in vivo behaviour. A 2D IC50 against that line is a number that does not describe how the drug behaves against the tissue it is meant to treat.
Toxicity: the liver problem
Drug-induced liver injury remains a leading cause of post-approval withdrawal. Primary hepatocytes on plastic dedifferentiate within days and lose CYP450 activity, which is why 2D hepatotoxicity assays miss metabolism-dependent toxicants. Bell et al. (Scientific Reports, 2016) showed 3D primary hepatocyte spheroids retain metabolic competence for weeks and correctly flagged hepatotoxic compounds that 2D missed. Proctor et al. (Archives of Toxicology, 2017) tested 110 drugs in spheroids and achieved substantially better DILI prediction than monolayer assays. The metabolic machinery is different in 3D, and it is the 3D version that resembles a patient.
Where 2D is still fine, and where it is not
| Use | 2D acceptable? | Why |
|---|---|---|
| Triaging 100,000 compounds to 1,000 | Yes | Eliminating clearly inactive molecules; speed wins |
| Target engagement, biochemical mechanism | Often | Simplicity is an asset |
| Potency ranking of lead candidates | No | Dose shifts of 10 to 1000x in 3D |
| Hepatotoxicity, nephrotoxicity | No | Metabolic competence lost on plastic |
| Efficacy in solid tumours | No | No gradients, no matrix resistance |
| Go / no-go for candidate advancement | No | The blind spots sit exactly on the decision |
The problem is not that 2D exists. It is that it has become the default for decisions where its known blind spots undermine the conclusion being drawn.
What is changing on the regulatory side
The FDA Modernization Act 2.0 (2022) removed the statutory requirement for animal testing before human trials, and the FDA's April 2025 roadmap explicitly names organoids and microphysiological systems as acceptable evidence in preclinical safety. That implicitly acknowledges that better-than-2D in vitro data is both achievable and expected. Getting there requires 3D platforms as tractable as 2D: defined, reproducible matrices with mechanics matched to the tissue, in formats compatible with real throughput. That is an engineering problem the field is actively solving, and it is the one we work on.
A practical bridge for screening groups
We suggest a staged approach rather than a wholesale switch:
- Keep 2D for primary triage.
- Move secondary potency and all toxicity characterisation into a 3D format with a defined matrix. In our experience this is where the false positives and false negatives concentrate.
- Require matrix mechanics on the assay record (stiffness, and stress relaxation if viscoelastic) so results can be compared across sites.
- Choose a matrix that dissolves gently, so you can recover cells for the transcriptomic and proteomic follow-up that a hit deserves.
The uncomfortable conclusion is that a meaningful share of efficacy and toxicity data generated every year describes a biology that does not exist outside a dish. Flat culture was never wrong to build on. It is wrong to still be the default when better substrates exist and the cost of the discrepancy is measured in failed trials.
References
- Wong CH, Siah KW, Lo AW. Estimation of clinical trial success rates and related parameters. Biostatistics 20, 273–286 (2019).
- Harrison RK. Phase II and phase III failures: 2013–2015. Nat Rev Drug Discov 15, 817–818 (2016).
- Edmondson R et al. Three-dimensional cell culture systems and their applications in drug discovery and cell-based biosensors. Assay Drug Dev Technol 12, 207–218 (2014).
- Karlsson H et al. Loss of cancer drug activity in colon cancer HCT-116 cells during spheroid formation in a new 3-D spheroid cell culture system. Exp Cell Res 318, 1577–1585 (2012).
- Bell CC et al. Characterization of primary human hepatocyte spheroids as a model system for drug-induced liver injury, liver function and disease. Sci Rep 6, 25187 (2016).
- Proctor WR et al. Utility of spherical human liver microtissues for prediction of clinical drug-induced liver injury. Arch Toxicol 91, 2849–2863 (2017).
- U.S. FDA. Roadmap to reducing animal testing in preclinical safety studies (April 2025).
FAQs
For early, low-cost triage where speed matters more than translational accuracy, yes. The risk is treating those results as predictive of in vivo response.
More per well, in reagents and time. But the relevant comparison is the cost of a false signal carrying into a late-stage study. For any assay feeding a go/no-go decision, the trade-off favours 3D.
Embedded culture in a defined, pipettable matrix in 96- or 384-well plates. It runs on existing liquid handlers and imagers, and it keeps mechanics as a reportable variable.