assay · intermediate · partly written
Luminescent cell viability assay (CellTiter-Glo)
A one-reagent plate assay that breaks open the cells and uses a luciferase enzyme to turn the ATP in the well into light, read as a proxy for how many healthy cells are left.
Also called: CellTiter-Glo, Cell Titer-Glo, ATP luminescence viability
In his words: what it tells you
He said: "I've seen high-throughput screening done for Cell Titer-Glo, which just tells you if a cell is alive or not." 1
Amit's hypothesis
What it actually tells you
Your format is right: cells, compound, add reagent, read. The gap is what 'alive' means to the readout. CellTiter-Glo measures ATP, not life, and 'ATP is proportional to cell number' is the vendor's premise, not a law. In one study gemcitabine cut cell counts about 74% while ATP fell under 20%. So it tells you how much ATP is in the well. 23
Reported, not yet replicated
How it works more
The reagent contains detergent, which bursts the cells and releases their ATP, and it must stop the cell's own ATP-eating enzymes (ATPases) from destroying the ATP before it is read. A thermostable luciferase called Ultra-Glo, built from a different firefly than the classic enzyme because the classic one is not stable in detergent, consumes ATP and a luciferin substrate and emits light. More ATP means more photons, which a plate reader counts. The signal half-life is over five hours, so plates can be read in batches.
The vendor bulletin (TB288, revised 1/23) describes a single reagent added to cells in serum medium, with no washing or medium removal. These are vendor claims. 2
Reported, not yet replicated
Where it breaks and how people check more
Arrested cells keep ATP. In HT29 cells at 48 h, gemcitabine cut cell count about 74% at maximum effect (log EC50 -8.7) but ATP fell under 20% with no valid fit; etoposide's ATP EC50 was about -4.5 versus -6.8 by count. The authors' reason: cells arrested in S or G2 phase keep growing and gain mitochondria. Longer exposure narrows the gap. This is one paper in one cell line. Checks: count nuclei by imaging, use a DNA-content readout, or use growth-rate metrics.
Glycolysis masks mitochondrial toxins. Highly proliferative lines use glycolysis in glucose despite working mitochondria (the Crabtree effect) and tolerate respiration poisons; the vendor advises re-testing ATP-depleting hits in galactose and reading a membrane-integrity marker beside ATP.
Luciferase inhibitors. Profiling more than 70,000 samples found about 3% active against luciferase, 681 (0.9%) with IC50 below 10 uM, and they also blocked several commercial luciferase detection reagents. Less light would look like toxicity (our inference). That study was not of CellTiter-Glo screens and no independent measurement of the false-hit rate in them was found. Check with a cell-free well (medium, compound, reagent) or a non-luciferase readout.
Temperature, edge effects and mixing give uneven signal; the Sanger protocol rejects plates whose control coefficient of variation exceeds 0.18. 34526
Reported, not yet replicated
Does a hit mean anything? more
Not alone. A drop in light means less ATP. The next steps of the cascade (orthogonal viability, imaging, dose response) exist to say which story is true: fewer cells, arrested cells, poisoned mitochondria or an inhibited reporter enzyme. In the STK33 screen the 7.5% active rate at 80% inhibition is the thing to distrust first, because it includes everything that stops growth, poisons mitochondria or inhibits luciferase. 735
Our inference
Protocol numbers and screen formats deep
Vendor bulletin: reagent volume equal to the medium volume (100 uL in 96-well, 25 uL in 384-well); equilibrate about 30 min, shake 2 min, wait 10 min, read. Linear (r2 0.99) from 0 to 50,000 HEK293 cells per well in 96-well plates; 50 cells gave signal above three standard deviations over background. A 10 mL kit covers 100 assays at 96-well volumes or 400 at 384-well; 100 mL covers 1,000 or 4,000.
Broad STK33 screen: reagent diluted 1:3 with PBS, 30 uL added to 30 uL of cells and compound after 72 h, 30 min equilibration, 10 min, read on an EnVision. GDSC: 2.5 uL of CellTiter-Glo 2.0 per 1536-well, with separate day-1 plates for baseline. 2876
Reported, not yet replicated
Question answered
Readout physical
Luminescence (photons counted by a plate reader, reported as relative light units) from a luciferase reaction that consumes ATP and a luciferin substrate in a lysed well. More ATP gives more light. One number per well; compound wells are divided by vehicle-only wells to give percent of control, and dose series give curves. 2
Reported, not yet replicated
Not measured
It cannot separate dead cells from cells that have stopped dividing, cannot say which mechanism acted, cannot show whether a protein is gone, and says nothing about single cells. Cells that grow larger or gain mitochondria while arrested keep their ATP; glucose-fed cancer cells can shrug off mitochondrial poisons; and a compound that inhibits the luciferase itself lowers the light without touching the cells. 345
Reported, not yet replicated
Setting
Where in cascade
Sample prep
No washing or medium removal. Add reagent equal to the medium volume (100 uL for a 96-well plate, 25 uL for 384-well), equilibrate the plate about 30 min to room temperature, shake 2 min, wait 10 min, then read. Run no-cell control wells (medium, compound and reagent) alongside. 2
Reported, not yet replicated
Conditions
Temperature gradients, uneven seeding and edge effects give uneven signal; tall plate stacks need longer equilibration and adherent cells need more mixing. ATP per cell varies by cell type, and contact inhibition at high density breaks linearity, so titrate cells per well first. DMSO up to 2% only minimally affects light. Linear (r2 0.99) from 0 to 50,000 HEK293 cells per well in 96-well plates; 50 cells gave signal above three standard deviations over background. Exposure time matters: longer exposure narrows the gap between ATP and cell count for arrested cells. 23
Reported, not yet replicated
Throughput
No per-day figure was found. The format is one reagent and no wash, so robots handle it: the Broad STK33 cell screen tested 37,420 substances in 384-well plates, and Sanger's GDSC2 runs in 1536-well plates (297 compounds across 969 cell lines). Class set from those scales (our reading). 79
Our inference
Time per run
Engineerable variants
Reagent and plate formats
CellTiter-Glo 2.0 (used by Sanger) and a 3D version exist as Promega products (names only, not read in detail). The assay runs in 96-, 384- (TB288) and 1536-well formats (Sanger). Kits: 10 mL covers 100 assays at 96-well volumes or 400 at 384-well; 100 mL covers 1,000 or 4,000. 286
Reported, not yet replicated
Day-1 baseline plates
GDSC reads separate day-1 plates so growth can be computed rather than raw ATP. The growth-rate-inhibition literature argues raw cell-count scoring is distorted by division rate. 610
Reported, not yet replicated
Galactose instead of glucose
Swapping glucose for galactose forces cells onto mitochondria; Promega advises re-testing ATP-depleting hits in galactose (vendor). 4
Reported, not yet replicated
Multiplex with a membrane-integrity readout
Promega's Mitochondrial ToxGlo reads a dead-cell protease (membrane integrity) first, then ATP, in one well (vendor). 4
Reported, not yet replicated
Use cases
Broad STK33 cell screen (PubChem AID 2330)
NOMO-1 cells (a KRAS-dependent leukaemia line), 750 cells in 30 uL per well in 384-well plates, 25 nL compound by pin tool, 72 h, then reagent. 37,420 substances were tested and 2,817 were active at 80% inhibition (7.5%, our arithmetic). The record itself describes an inhibitor-of-STK33 cell screen, so a hit means less ATP-light, not a confirmed STK33 hit. Deposited 2010-02-03. 7
Reported, not yet replicated
GDSC2 drug-sensitivity dataset (Sanger)
Switched to CellTiter-Glo in 2015: 1536-well, 72 h, 297 compounds, 969 cell lines. The older GDSC1 used Syto60 (a DNA dye) or resazurin (a colour-change metabolic dye): 403 compounds, 970 lines. Protocol: 7.5 uL cells, 24 h, acoustic dispensing, 0.1% DMSO, 72 h, 2.5 uL CellTiter-Glo 2.0, 10 min, read; a plate passes only if the negative-control coefficient of variation is below 0.18. 96
Reported, not yet replicated
Before after
Status quo before
Earlier ATP assays needed several steps to extract the ATP before it could be measured (the report's account of the vendor's own framing). 2
Reported, not yet replicated
What forced it
Screening needs a readout that a robot can run on thousands of wells without washing or medium removal. That this was the driver is the report's reading of the vendor's aim, not a statement in the bulletin. 2
Our inference
The trick
One reagent contains detergent to lyse the cells and an engineered thermostable luciferase (Ultra-Glo, from a different firefly, Photuris pennsylvanica, because the classic Photinus enzyme is not stable in detergent). Its properties allow reaction conditions that inhibit the cell's own ATP-eating enzymes, so the ATP survives until it is read. 2
Reported, not yet replicated
Why it worked
A glow signal with a half-life over 5 hours lets plates be read in batches without injectors. The same cheap number then served a 37,000-compound primary screen and a 969-cell-line pharmacology atlas, which also means every dataset built on it inherits its blind spots. 279
Reported, not yet replicated
Data availability
GDSC2 (Genomics of Drug Sensitivity in Cancer, Sanger) public
Drug-sensitivity data measured with CellTiter-Glo: 297 compounds across 969 cell lines, with the protocol published. 96
Caveats. Inherits the readout's blind spots (arrested cells keep ATP, luciferase interference). GDSC1 used different readouts, so the two are not the same measurement.
Reported, not yet replicated
PubChem AID 2330 (Broad STK33 cell screen) public
Primary-screen results for 37,420 substances in NOMO-1 cells, with the full assay protocol. 7
Caveats. A hit means less ATP-light, not an STK33 hit; the planned THP-1 toxicity counter-screen result was not opened.
Reported, not yet replicated
Promega CellTiter-Glo kits purchase
The reagent kit sold by the vendor. 8
Caveats. The vendor's own price is behind a login and was not read; reseller prices seen in search summaries were not opened and are not recorded.
Reported, not yet replicated
Teach this
In one sentence
CellTiter-Glo lyses the cells and uses a luciferase to turn their ATP into light, so it tells you how much ATP is in the well, a proxy for healthy cell number and not a yes-or-no on life. 2
Reported, not yet replicated
The five-minute version
Cells in wells get compound; you add one reagent that bursts them and makes light in proportion to their ATP; a plate reader gives one number per well. It is the cheapest way to get a number from thousands of wells, so it is the default for 'did the compound hurt or slow the cells'. But ATP is a proxy: arrested cells can keep their ATP while the count falls (gemcitabine in one study), glucose-fed cancer cells shrug off mitochondrial poisons, and compounds that inhibit the luciferase look like toxicity. A hit is a drop in light, and the cascade's next assays say why. 2345
Reported, not yet replicated
What a listener will ask
So does it count living cells?
No: it measures ATP in the lysed well, which the vendor treats as proportional to cell number; that holds less well when cells are arrested rather than dead.
Why is it used in screens?
One reagent, no washing, a glow signal that lasts for hours, so robots can run and read many plates.
How do you know a hit is real?
Test a cell-free well for luciferase inhibition, re-test in galactose, count nuclei by imaging, and run a dose-response curve.
The common wrong idea
It tells you whether a cell is alive or dead.
It tells you how much ATP is in the well, and ATP is only a proxy for healthy cell number: arrested cells can keep their ATP.
The one citation to hand over: 3
Still open here
- Price: no opened page showed a price (Promega hides it behind login; a distributor page timed out). Cost per well is therefore not given; the academic core recharge figure for a cell-based screen was not re-opened at source for this record and is not relied on.
- The Promega claim that its engineered luciferase resists inhibitors better than the native enzyme could not be checked: the source page returned a 404, so no figure is given.
- How often luciferase interference actually produces false hits in CellTiter-Glo screens: no independent measurement found. The 3% luciferase-inhibitor rate comes from a general library profile, not from CellTiter-Glo screens.
- Whether the Broad CTRP dataset used CellTiter-Glo with 16-point dilution series rests on a search-result summary of a page that could not be opened (unconfirmed); the CTRP paper's abstract does not name the readout. CTRP assay time not confirmed.
- The underlying paper on glycolysis masking mitochondrial toxins (Marroquin 2007) was seen only in search summaries: unconfirmed here.
- The claim that evaporation is the main cause of edge effects in 1536-well plates rested on a lid-seller note that was not re-opened: unconfirmed and not used.
- The luciferase reaction equation (TB288 Figure 3) could not be read; the stoichiometry is not given here.
- A THP-1 counter-screen was planned in the Broad STK33 cell assay to flag general toxicity; no result was opened.
- Machine-learning use at this step (for example modelling drug sensitivity from the GDSC data) is not covered by the reports read; ml_role is left unwritten.
- Cost per campaign: no basis found for a range; left unwritten.
Not yet written: Cost, Capital, ML role.
Linked from
Sources
- Amit's dictation of 2026-11-17: what the atlas must contain. Kept in the atlas repository; not published.
- CellTiter-Glo Luminescent Cell Viability Assay technical bulletin TB288 (rev. 1/23). Promega (2023). vendor-page, not peer reviewed; opened 2026-11-17 all records citing this
- Chan et al.. Comparison of ATP readout and cell counting for drug potency (Chan et al., PLoS ONE 2013). PLoS ONE (2013). paper; opened 2026-11-17 all records citing this
- Arduengo. Differentiating mitochondrial toxicity from other types of mechanistic toxicity (Arduengo, Promega, May 2012). Promega (2012). vendor-page, not peer reviewed; opened 2026-11-17 all records citing this
- Auld et al.. Characterization of luciferase inhibitors in a 70,000-sample library (Auld et al., J Med Chem 2008), Europe PMC listing. J Med Chem (via Europe PMC) (2008). paper; opened 2026-11-17 all records citing this
- GDSC screen information (2019-02-11). Wellcome Sanger Institute (2019). report, not peer reviewed; opened 2026-11-17 all records citing this
- PubChem AID 2330: Broad STK33 cell-based screen. PubChem (2010). database, not peer reviewed; opened 2026-11-17 all records citing this
- CellTiter-Glo Luminescent Cell Viability Assay, product page G7570. Promega (2026). vendor-page, not peer reviewed; opened 2026-11-17 all records citing this
- Sanger DepMap: drug sensitivity datasets (GDSC1 and GDSC2). Wellcome Sanger Institute (2026). database, not peer reviewed; opened 2026-11-17 all records citing this
- Hafner et al.. Growth-rate inhibition metrics for breast cancer cell lines (Hafner et al., Sci Data 2017), Europe PMC listing. Scientific Data (via Europe PMC) (2017). paper; opened 2026-11-17 all records citing this
Where it sits on the map
History of this record (3)
This record has been changed 3 times, on the dates below. The note kept against each change is written for the people who maintain the atlas, in their own shorthand, so it is not shown here; it is in the record itself.
- Throughput class
- high
- Emerging
- no
- Status
- partly written
- Last updated
- 2026-11-17
- Record
- content/assays/luminescent-cell-viability-assay.json