In Vitro Cytotoxicity: Purpose, Methods, and Meaning

In vitro cytotoxicity testing measures whether a substance kills or damages cells grown in a lab dish, and it has become one of the earliest checkpoints in drug discovery and chemical safety screening. The approach is faster and cheaper than animal testing, and decades of refinement have made it reasonably predictive of what a compound will do inside a living body.¹ But “cytotoxicity” is not one measurement with one meaning. The term covers a family of assays, each probing a different sign of cellular harm, and interpreting the results correctly requires understanding what each test actually detects and where the blind spots are.

Why Cytotoxicity Testing Exists

Before a new drug candidate, food additive, cosmetic ingredient, or industrial chemical reaches a living organism, researchers need a rough answer to a basic question: is this substance likely to poison cells? Running that test on cultured cells instead of animals is faster by orders of magnitude. A single 96-well plate can screen several compounds at multiple concentrations in a single day, using as few as a thousand cells per well.² That speed matters when pharmaceutical companies have thousands of candidate molecules to winnow down, or when regulators need to evaluate newly manufactured chemicals for which no animal data yet exist.

The goal is not to replace all animal testing in one stroke. In vitro cytotoxicity data is a filter: compounds that damage cells at low concentrations get flagged early, saving time and animal lives downstream. Compounds that pass get scrutinized further. In the regulatory world, this kind of screening has been formally integrated into safety evaluation frameworks. The International Organization for Standardization, for example, requires in vitro cytotoxicity assays as part of the biological evaluation of medical devices under ISO 10993-5, which specifies several accepted assay formats.³

What the Tests Actually Measure

No single assay captures every dimension of cell health. Instead, each major cytotoxicity assay looks at a different biological signal, and the choice of assay shapes what you can learn.

Metabolic Activity Assays

The most widely used approach relies on tetrazolium salts, commonly known by abbreviations like MTT, MTS, or XTT. These are dyes that healthy, metabolically active cells convert into a colored product. The more color you measure at the end of the experiment, the more living cells remain. The signal is a proxy for cell viability: if a compound kills cells or shuts down their metabolism, the color intensity drops in a dose-dependent way.

A related metabolic assay uses resazurin, sold under the brand name Alamar Blue. Living cells reduce this blue, nonfluorescent dye into a pink, highly fluorescent product called resorufin.⁴ Because fluorescence can be read without destroying the sample, this assay has a practical advantage: you can measure the same well at multiple time points and track toxicity as it develops. The reduced form of Alamar Blue has been found inside both the cytoplasm of living cells and the nuclei of dead cells, suggesting the chemistry is not entirely straightforward, but the assay remains popular because of its convenience and sensitivity.

Membrane Integrity Assays

When a cell dies by necrosis or late-stage damage, its outer membrane ruptures, spilling internal contents into the surrounding fluid. Lactate dehydrogenase, an enzyme normally confined to the inside of the cell, is one of those spilled contents. Measuring how much of this enzyme appears in the culture medium provides a direct readout of how many cells have lost membrane integrity.⁵ Because the enzyme is stable once released, the signal accumulates over time, making the LDH assay a solid indicator of cumulative damage. Research on pesticide exposure, for instance, showed that incubating nerve-like cells with various pesticides at low concentrations produced two- to three-fold increases in LDH leakage, confirming that even trace amounts caused measurable cellular harm.⁶

ATP Quantification

Adenosine triphosphate is the energy currency of living cells. When cells die, ATP production stops and existing stores are rapidly depleted. Measuring ATP via a light-producing reaction involving luciferase gives a fast, sensitive snapshot of how many cells remain metabolically active.⁷ The luminescent signal is proportional to ATP levels, and the assay requires very few cells, making it especially useful when working with scarce primary tissue samples. Across the field, ATP-based luminescence is often considered the most sensitive single-endpoint method for monitoring active cell metabolism.⁷

Beyond raw viability, the ratio of ADP to ATP within cells can help distinguish how cells are dying. In proliferating healthy cells, the ratio stays below about 0.11. In cells undergoing the orderly process of self-destruction known as apoptosis, the ratio rises into the range of 0.11 to 1.0. And in cells killed abruptly by harsh insults like heat shock, the ratio rockets above 15.⁸ That single measurement can tell you whether a compound is triggering a controlled shutdown or simply blowing cells apart, and those two outcomes have very different implications for safety.

Dye Uptake Assays

The neutral red uptake assay takes a different approach entirely. Living cells actively absorb a weak dye called neutral red and store it in their lysosomes, the small internal compartments responsible for waste processing. When cells are damaged, their lysosomes lose the ability to retain the dye, so the amount of dye taken up decreases as toxicity increases.⁹ This assay has particular regulatory significance: it forms the basis of the OECD Test Guideline 432, which uses the 3T3 neutral red uptake phototoxicity assay to evaluate whether compounds become toxic when exposed to UV light. That guideline has been accepted by all OECD member states since 2004.¹⁰

How Cells Die Matters

A cytotoxicity number on its own, say an IC50 value showing that half the cells die at a certain concentration, does not tell you the full story. The way cells die carries information about the underlying mechanism of toxicity, which in turn affects how worried you should be about a compound’s safety.

Apoptosis is programmed cell death, characterized by specific biochemical events including activation of enzymes called caspases, release of a signaling molecule from mitochondria, and fragmentation of DNA into regular-sized pieces.¹¹ Necrosis, by contrast, is messier: the cell swells and bursts, and researchers often define it by the absence of those orderly apoptotic hallmarks. A compound that triggers apoptosis at high doses is a different safety concern from one that causes necrosis at low doses. Distinguishing between the two requires more than a single viability readout, which is one reason researchers often combine assays.

A related challenge is that many toxins damage cells by generating reactive oxygen species, essentially unstable molecules that attack cell structures from the inside. Mitochondria, the energy-producing compartments within cells, are frequent targets.¹² Research on the pesticide chlorpyrifos illustrates this pattern: at concentrations well below those needed to outright kill cells, the compound increased reactive oxygen species by about 83%, boosted a measure of fat damage within cell membranes, and depressed the electrical charge that mitochondria need to function.¹³ A standard viability assay at those sub-lethal concentrations might report “cells are alive,” missing the oxidative damage happening underneath.

Why a Viability Number Can Be Misleading

One of the most common misunderstandings about cytotoxicity data is treating a drop in viability signal as proof of cell death. Standard metabolic assays like MTT measure activity, not mortality. If a drug stops cells from dividing without actually killing them, the metabolic signal still drops because there are fewer cells at the end of the experiment than in the untreated control. That effect, called cytostasis, looks identical to genuine cell killing on a viability curve.

This distinction has real consequences. Researchers working with two cancer cell lines treated with cisplatin and topotecan found that while both drugs reduced viability, only a caspase inhibitor could partially rescue one of the cell lines, suggesting that in those cells the drug was causing genuine apoptotic death rather than just slowing growth. In the other cell line, none of the cell-death inhibitors meaningfully restored cell numbers, pointing toward growth arrest as the dominant explanation.¹⁴ Two drugs that look equally potent on an MTT plate may be doing entirely different things to cells, and those differences matter for how the compound will behave in a patient.

Impedance-based systems have emerged as one way to untangle these effects. Instead of adding a dye, these platforms measure the electrical resistance created by cells adhering to electrodes at the bottom of a well. As cells grow and spread, impedance rises; as they die and detach, it falls. Recording this continuously over hours or days produces a curve whose shape can distinguish cytostatic drugs, which cause impedance to plateau, from cytotoxic drugs, which cause it to drop. In one validation study, an algorithm applied to impedance curves successfully discriminated cytostatic, cytotoxic, and non-toxic compounds from a panel of 37 test substances, with results that correlated well with known animal data.¹⁵

The Nanoparticle Problem and Other Interference

Assay interference is a less glamorous but critical concern, especially in the rapidly growing field of nanomaterial safety. Nanoparticles have enormous surface areas and can interact directly with assay reagents in ways that produce false readings. Diesel particles, for example, reduced MTT dye in cell-free conditions at concentrations as low as 0.05 micrograms per milliliter, a level far below what many toxicity studies use. Other particle types, including activated carbon and soot, bound LDH enzyme and lowered its detectable concentration, making cells look healthier than they actually were.¹⁶

The interference is not generic. Each nanoparticle type affects each assay differently. A separate study found that all tested nanoparticles interfered with optical measurements in the MTT, LDH, and DCF assays at higher concentrations, but the pattern varied: carbon black particles chemically oxidized a fluorescent probe in the absence of cells (creating a false signal for oxidative stress), while only zinc oxide nanoparticles significantly reduced LDH activity.¹⁷ Iron oxide nanoparticles posed yet another set of problems, interfering with both the MTT reagent itself and the fluorescence and luminescence readings used in multiple assay formats.¹⁸ The practical upshot: any study reporting cytotoxicity data for nanoparticles without demonstrating particle-free control experiments should be read with skepticism.

Cell Choice Shapes the Answer

Which cells you test a compound against is not a minor detail. Immortalized cell lines, the workhorse cultures kept in freezers at thousands of labs, behave differently from primary cells freshly isolated from tissue. A study comparing the responses of primary human keratinocytes with the immortalized HaCaT cell line found that the primary cells died at sodium dodecyl sulfate concentrations roughly three times lower than what was needed to kill the immortalized line.¹⁹ In other words, the immortalized cells were about three times more tolerant than the cells closer to what exists in your body.

This pattern does not always hold neatly. When dental resin compounds were tested across four different cell types, the ranking of which compound was most toxic stayed the same, but the actual concentration required to produce damage varied substantially. Primary cells and one transformed cell line were less sensitive, while the standard L929 mouse fibroblast line used widely in regulatory testing was the most sensitive.² For orthopaedic materials, an immortalized rat osteoblast line was consistently the most sensitive cell type tested, though its responses paralleled those of human cells closely enough that researchers suggested it could serve as a useful screening stand-in.²⁰

The takeaway is not that one cell type is “right” and another “wrong.” It is that a cytotoxicity number always has a cell type attached to it, and comparing results across studies that used different lines requires caution. A compound with an IC50 of 50 micromolar in L929 cells might look far less toxic in a primary cell system, or the reverse.

From Dish to Body

The hardest question in toxicology is translating what happens in a dish to what happens in a person. A compound might kill liver cells at 100 micromolar in a plate, but whether a realistic human exposure ever produces that concentration in the liver depends on absorption, metabolism, distribution, and elimination. Bridging that gap is the job of quantitative in vitro to in vivo extrapolation, or QIVIVE.

The idea is to build a computational model of how a chemical moves through a body, then use it to figure out what dose would produce the same concentration in a target organ that caused damage in the dish. A study testing anti-androgenic pesticides demonstrated this approach by first measuring their effects in vitro, then modeling fetal tissue concentrations in pregnant rats. Seven of nine pesticides for which no animal data existed were predicted to cause developmental effects, and follow-up animal experiments with three of them confirmed the predictions.²¹ The modeled fetal concentrations agreed well with what was later measured. That is the best-case scenario for QIVIVE: a genuine prediction that panned out.

A similar approach was used to predict toxicity thresholds for aloe-emodin, a naturally occurring compound found in aloe plants. Researchers used in vitro data from liver and kidney cells combined with pharmacokinetic modeling to estimate benchmark dose values in rats and humans. Their results suggested the liver is more sensitive than the kidney to aloe-emodin’s toxic effects, with the predicted benchmark dose for liver damage being roughly forty times lower than for kidney damage.²² These kinds of organ-specific predictions are exactly what simple cytotoxicity plates cannot provide on their own.

Three-Dimensional Models and Organs on Chips

Flat layers of cells in a dish behave differently from cells in a living tissue, where they exist in three dimensions, interact with neighboring cell types, and experience flowing fluids. Three-dimensional cell culture models, particularly spheroids (small clusters of cells that self-organize into ball-like structures), are one step closer to reality. Research comparing heavy metal toxicity in neuroblastoma cells grown in flat versus spheroid cultures found that the 3D arrangements were consistently less sensitive to cadmium and mercury than their flat counterparts: exposure to cadmium caused only a modest metabolic disturbance in spheroids, while flat-grown cells showed substantially greater damage to metabolic markers and energy stores.²³ Spheroids more closely mirror the relative resistance that cells in intact tissues show, which makes their toxicity data arguably more relevant to predicting what happens in a body.

Organ-on-a-chip technology pushes this further. These microfluidic devices contain tiny channels lined with living cells from specific organs, with fluid flowing through to simulate blood supply. By connecting multiple organ compartments on a single chip, researchers can study how a drug metabolized in a liver compartment produces toxic byproducts that then damage a kidney compartment. A liver-kidney-on-chip system demonstrated this with aflatoxin B1, a fungal toxin: the liver cells metabolized the compound while kidney cells downstream showed flow-dependent toxic responses, something impossible to capture in a standard static plate.²⁴ These platforms remain expensive and technically demanding, but they are increasingly seen as the bridge between simple cell-plate screening and whole-animal studies.²⁵

Multiparametric Assays and High-Content Screening

Rather than running one assay at a time, modern platforms increasingly measure multiple cell-health indicators simultaneously from the same well. A high-content screening approach validated for human hepatotoxicity prediction measures eight indicators at once, covering nuclear shape, membrane integrity, mitochondrial function, and cell proliferation.²⁶ This kind of multiparametric read provides a richer picture than any single measurement and helps flag compounds that damage cells through mechanisms a lone assay would miss.

Technology transfer studies have confirmed that this approach is reproducible across different labs. In validation work, a multiparametric live-cell screening assay was found to be highly concordant with human liver toxicity, including rare toxicities that standard assays tend to miss. The protocol’s success was attributed to using human cells that can metabolize drugs, exposing cells for 72 hours to capture slower-acting toxicants, and testing across a wide concentration range of 30- to 100-fold above the therapeutically active concentration.²⁷

Machine Learning and the Tox21 Library

The U.S. Tox21 consortium has screened roughly 10,000 drugs and environmental chemicals through dozens of in vitro assays, generating an enormous dataset of bioactivity profiles. Researchers have begun feeding that data into machine learning algorithms alongside chemical structure information to predict toxicity without running any new experiments at all. Models combining both in vitro bioactivity data and chemical structure descriptors achieved strong predictive accuracy for acute systemic toxicity, with performance metrics in the range of 0.73 to 0.93 depending on the algorithm and input type.²⁸

A parallel effort focused specifically on organ-level toxicity found that mixing bioactivity descriptors from high-throughput screening with chemical structure or chemotype descriptors produced the most accurate predictions of which organs would be damaged in repeat-dose animal studies.²⁹ These models are not yet ready to replace all animal testing, but they represent a future where the first pass of toxicity evaluation for a new chemical could be entirely computational, informed by a massive library of prior in vitro results. The integration of such computational tools with 3D organoids and organ-on-chip systems is already under way, with the goal of cross-validating simulations against experimental evidence from increasingly realistic cell models.³⁰

Real-Time Monitoring Without Labels

Conventional assays are endpoint measurements: you add a dye, wait, then read the result. That gives you a snapshot but not a movie. Real-time cell analysis systems, which track impedance continuously via electrodes embedded in the culture surface, fill that gap by offering label-free, uninterrupted monitoring of cell adhesion, shape changes, and growth rate.³¹ Because no reagent is added and cells are never disturbed, the data captures the full time course of a toxic response: the lag before damage begins, the speed at which it progresses, and whether cells attempt to recover.

The technology has proven especially useful for studying immune cell killing, a context where timing matters enormously. When peptide-specific immune cells were co-cultured with target cells on impedance-sensing plates, the resulting drop in impedance tracked the kinetics of immune-mediated cell death in real time, detecting apoptosis and detachment as they happened.³² That kind of dynamic data cannot be captured by any endpoint assay, no matter how sensitive, and it is increasingly important for evaluating immunotherapies where the therapeutic mechanism is itself a form of controlled cytotoxicity.