Biology Atlas

Map / Scenarios

scenario · intermediate · partly written

A hit series that binds and then fails in cells

A worked decision: a screening hit series is potent against its purified target and then does nothing in cells, with the real STK33 campaign as the comparison.

The setup

You have a series that is potent on the purified enzyme and does nothing in the cells that are meant to need it. There are several defensible next moves, and each answers a different question: did the compound arrive, is the readout honest, does this cell context matter, or is the target the problem? 12

Reported, not yet replicated

Why no option is forced more

The evidence does not force one answer. The STK33 authors concluded that kinase inhibition was not a promising strategy, but their own probe report left off-target effects and high plasma protein binding open as alternative explanations, and the biochemical steps shared one luciferase readout with no non-luciferase check found. Engagement data would have separated these, and no engagement assay appears in the records opened. The sensible sequence is cheap checks of the readout and conditions first, then engagement, before concluding on the target. 2314

Our inference

Situation

A purified-protein kinase screen of a large library gave a few hundred primary actives, most of which held up on retest, and a handful that are potent in dose-response. A selective probe was made from one series. You then run it in cells that are supposed to depend on the target, and it does not reduce their viability. The real STK33 campaign (Broad, MLPCN) had this shape: 321,908 substances screened, 235 primary actives (0.073%), 174 of 224 retested confirmed, 13 submicromolar, and a probe, ML281, with an STK33 IC50 of 14 nM that did not meaningfully reduce cell viability up to 10 uM. 56217

Reported, not yet replicated

Ask

Your compounds bind the purified target and do nothing in cells. What do you do next?

Options

Accept the cell result and drop the series

Take the cell data at face value: the target is not a vulnerability in these cells, so stop chemistry on it. This is what the STK33 authors concluded for the related inhibitor BRD-8899, which failed to kill KRAS-dependent cells: kinase inhibition "does not represent a promising anti-KRAS therapeutic strategy". 321

Cost. Cheapest: no new assays, and it frees the team. Time and money are not priced in the sources.

Tells you. That, on the evidence you have, a potent selective inhibitor of the enzyme does not reproduce the hoped-for biology; the problem may be the target hypothesis rather than the chemistry.

Does not tell you. Why. The ML281 authors themselves noted the lack of effect could stem from off-target effects or from high plasma protein binding (99.6% human), and left this unresolved. Dropping the series does not distinguish "target is not needed" from "compound never engaged it".

See: Luminescent cell viability assay (CellTiter-Glo), Enzyme activity assay

Reported, not yet replicated

Run an orthogonal binding or engagement assay

Show that the compound binds the target by a different physical principle: SPR on the purified protein, and CETSA or NanoBRET in cells to see whether the compound reaches and sits on the target inside the cell. 41

Cost. Moderate. No public price was found for CETSA or NanoBRET; SPR cost is in the SPR record. NanoBRET needs a tagged target and a tracer; CETSA needs an antibody, mass spectrometry or a tag.

Tells you. Whether the compound arrives and binds in cells. A positive turns a null viability result into evidence that the target is not required; a negative points to permeability, ATP competition or exposure instead.

Does not tell you. Whether binding changes function. Both cell assays can also miss real binders (no thermal shift; ligands that do not compete with the tracer), so a negative is not proof of no binding. In the STK33 records no engagement step was found.

See: Surface plasmon resonance (SPR), Target engagement in cells (CETSA, NanoBRET)

Reported, not yet replicated

Check the readout and its artefacts

Ask what the signals actually measured. Every biochemical step in the STK33 funnel used the same luciferase-coupled readout (ADP-Glo), so a luciferase inhibitor would lower light like a kinase inhibitor; and the cell arm used CellTiter-Glo, where light tracks ATP, a proxy for cell number. Add a non-luciferase readout or a luciferase counter-screen, and a second viability method such as cell counting. 891021

Cost. Low to moderate: reruns of existing material with a different readout.

Tells you. Whether the enzyme potency was real or partly an artefact of one readout, and whether the cell result is missing an effect (or inventing one) because ATP is a poor proxy. In one study ATP and MTS proxies underestimated the potency of DNA-synthesis inhibitors compared with cell counting.

Does not tell you. Whether the target matters. No opened source covers ADP-Glo-specific interference, so how large the problem was in this campaign is unknown. Interference would explain false enzyme potency, but the retest and a selectivity panel (83 kinases at 1 uM) already argue against a pure artefact for the probe.

See: Enzyme activity assay, Luminescent cell viability assay (CellTiter-Glo), Reporter gene assay

Reported, not yet replicated

Test a different cell context

Repeat in other cells or a different phenotype: the cell screen used one KRAS-dependent leukaemia line (NOMO-1) and 72 h of CellTiter-Glo, which asks only whether cells stop producing ATP. 1191

Cost. Moderate: new cell lines and, if imaging is used, more cost per well.

Tells you. Whether the null is specific to one line, one timescale or one readout; imaging could ask a question you did not have to pick in advance.

Does not tell you. Which cells to choose: the reports do not say which other lines were tested for STK33. A second null across lines would strengthen the case that the target is not a vulnerability; a hit would need the engagement and artefact checks above before it could be trusted.

See: High-content imaging and morphological profiling, Luminescent cell viability assay (CellTiter-Glo), Western blot

Our inference

Re-examine the enzyme potency and conditions

Look at how the potency number was made. The STK33 retest called a compound active if its IC50 fell within one log above the top concentration, so some potencies were extrapolations, and enzyme IC50 depends on ATP and substrate through Cheng-Prusoff. 786

Cost. Low: reanalysis, plus enzyme kinetics at several ATP levels.

Tells you. How much of the biochemical potency would survive realistic cellular ATP, and whether the mode of inhibition (not reported for ML281) predicts a drop in cells.

Does not tell you. Anything about the target biology. For ML281 the kinetic mode and Km(ATP) are not reported, and cellular ATP effects on STK33 engagement were not measured in the sources.

See: Dose-response assay, Enzyme activity assay

Reported, not yet replicated

What really happened

The STK33 funnel (Broad MLPCN)

Primary screen (PubChem AID 2661): ADP-Glo luminescent kinase assay, 1536-well, single concentration, mostly singlicate, hit cutoff 25% inhibition; 321,908 substances, 235 active. Retest (AID 2821): 224 tested, 174 active, at eight concentrations from 0.1 to 12 uM. Counter-screens named in the paper were PKA and Aurora B; 13 hits were submicromolar and one quinoxalinone was chosen on tractability and selectivity. ML281 reached STK33 IC50 14 nM, about 550-fold over Aurora B and over 700-fold over PKA; at 1 uM among 83 kinases only FLT3 and KDR were inhibited by 25% or more besides STK33. 56287

Reported, not yet replicated

What the cells said

ML281 did not meaningfully reduce viability up to 10 uM, and the authors wrote that synthetic lethality with KRAS "cannot be recapitulated with selective STK33 small-molecule inhibitors". A related inhibitor, BRD-8899, the most potent and selective in its study (not ML281 itself), likewise failed to kill KRAS-dependent cells. A separate cell screen (AID 2330, NOMO-1 cells, CellTiter-Glo, 72 h) ran 37,420 substances with 2,817 active at 80% inhibition (7.5%), a rate about a hundred times the biochemical one; a THP-1 counter-screen was planned and no result was found. 231191

Reported, not yet replicated

The lesson drawn

Cells asked a different question: does losing this kinase matter? In this case the target, not the chemistry, was the likely problem, but the authors left off-target effects and plasma protein binding open. The enzyme assay delivered exactly what it measures: a potent selective inhibitor of the enzyme. 281

Reported, not yet replicated

Still open here

Linked from

Sources

  1. The screening cascade: concept breakdown (our own earlier write-up, unpublished). Kept in the atlas repository; not published.
  2. Weiwer et al.. Weiwer et al. 2012: the STK33 probe ML281 (PMC full text). PubMed Central (2012). paper; opened 2026-11-17 all records citing this
  3. Luo, T., et al.. Luo et al. 2012: STK33 kinase inhibitor BRD-8899 in KRAS-dependent cells, PNAS. PubMed Central (2012). paper; opened 2026-11-17 all records citing this
  4. HowStuffWorks: target engagement in cells (our own earlier write-up, unpublished). Kept in the atlas repository; not published.
  5. PubChem BioAssay AID 2661: STK33 primary screen (ADP-Glo). PubChem. database, not peer reviewed; opened 2026-11-17 all records citing this
  6. PubChem BioAssay AID 2821: STK33 inhibitor retest, 8-point dose response. PubChem (2010). database, not peer reviewed; opened 2026-11-17 all records citing this
  7. HowStuffWorks: dose-response curve fitting (our own earlier write-up, unpublished). Kept in the atlas repository; not published.
  8. HowStuffWorks: enzyme kinetics assays (our own earlier write-up, unpublished). Kept in the atlas repository; not published.
  9. HowStuffWorks: luminescent viability, CellTiter-Glo (our own earlier write-up, unpublished). Kept in the atlas repository; not published.
  10. 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
  11. PubChem AID 2330: Broad STK33 cell-based screen. PubChem (2010). database, not peer reviewed; opened 2026-11-17 all records citing this
  12. Amit's dictation of 2026-11-17: what the atlas must contain. Kept in the atlas repository; not published.
Where it sits on the map
History of this record (6)

This record has been changed 6 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.

Status
partly written
Last updated
2026-11-19
Record
content/scenarios/hit-series-that-fails-in-cells.json