The Drug That Should Have Worked
Glutaminase inhibition was the cleanest metabolic target in oncology. It failed in the most glutamine-dependent cancer there is. But the reason it failed tells us more about how tumors handle nitrogen than a success would have — and it says, fairly precisely, what to run instead.
If you had to pick one enzyme to shut off in a tumor, glutaminase would have been a defensible choice. Glutamine is the second-largest nutrient flux in most cancers. Glutaminase is the committed first step of its catabolism. The enzyme is druggable, the dependency is reproducible in culture, and an orally available, selective inhibitor existed.
It was tested in the most favourable setting anyone could nominate. It missed.
This is a follow-up to Ammonia in Disguise, which argued that a tumor's surplus nitrogen leaves largely as alanine rather than ammonia, and that the exits are redundant by design. If that is right, it predicts something specific about drugs aimed at this pathway. This piece checks the prediction against the trial record.
Part I / The Test Already Ran
Two Trials, Both Negative
Telaglenastat (CB-839) is the best-developed selective glutaminase-1 inhibitor. It went into two randomized settings chosen for maximum prior probability.
CANTATA was a global, randomized, placebo-controlled, double-blind trial of telaglenastat plus cabozantinib versus placebo plus cabozantinib in previously treated advanced or metastatic renal cell carcinoma. Clear-cell RCC is about as glutamine-addicted as solid tumors get — VHL-driven, pseudohypoxic, glycolytic. The trial did not meet its primary endpoint of progression-free survival.
KEAPSAKE went after a redox-defined population: non-small-cell lung cancer with KEAP1 or NRF2 pathway mutations. The rationale was good. KEAP1/NRF2 and LKB1/STK11 pathway lesions cooperate to produce enhanced glutamine dependence in KRAS-mutant lung adenocarcinoma — a genotype-matched vulnerability, which is exactly how modern targeted oncology is supposed to work. The trial was discontinued at interim analysis for lack of benefit.
Part II / The Bypasses
Glutamate Has Other Suppliers
Glutaminase makes glutamate. Blocking it matters only if glutamate becomes scarce — and in a tumor cell it does not, because several other routes arrive at the same molecule.
The transaminases. This is where the previous article becomes relevant to drug development. Alanine aminotransferase run in the reverse direction generates glutamate from alanine and α-ketoglutarate directly. So does aspartate aminotransferase, and the branched-chain transaminase, and glutamate dehydrogenase running reductively. A cell with abundant alanine has a glutamate supply that no glutaminase inhibitor touches. The transaminases are themselves targetable — in TAZ/YAP-activated breast cancer, GOT1 and PSAT1 are induced and the transaminase inhibitor aminooxyacetate represses growth in a TAZ/YAP-dependent manner — but they are a second target, not a consequence of hitting the first.
The stroma delivers. Pancreatic stellate cells secrete alanine that outcompetes glucose- and glutamine-derived carbon in the tumor's mitochondria. A tumor being fed amino acids by its neighbours is not meaningfully dependent on its own glutaminase. Cell-culture dependency assays, where the tumor cell is the only cell in the dish, systematically overstate the dependency they measure.
The cell eats its surroundings. Triple-negative breast cancer cells resist loss of glutamine transport by turning on macropinocytosis — engulfing extracellular protein and digesting it for amino acids. That bypasses the transporter and the enzyme together, and it is how a tumor consumes the stroma rather than merely being fed by it.
And the signalling reroutes. In pancreatic cancer, glutamine antagonism triggers compensatory ERK/MAPK activation through upregulation of Axl and ErbB-family receptor tyrosine kinases — a growth-signalling rescue layered on top of the metabolic one.
transaminases | stromal alanine | macropinocytosis | GLS2 | reductive GDH
There is also simple isoform redundancy. GLS1 and GLS2 are separate genes under different regulation, and telaglenastat targets GLS1. And metabolic plasticity defeats the drug outright in some contexts — IDH1-mutant glioma lines show low sensitivity to glutaminase inhibition for precisely this reason.
Part III / The Inversion
The Cell With No Bypass Is the One You Need
All of that explains a disappointing result. It does not explain a harmful one. For that you have to look at who else in the tumor is running glutaminase.
Best and colleagues combined CB-839 with anti-PD1 in STK11/Lkb1-deficient lung cancer — the genotype picked because those tumors are glutamine-dependent and their microenvironment accumulates glutamate, which on paper should have made the drug an immunological favour. The result went the other way. The combination blocked CD8 T cell clonal expansion despite anti-PD1 activation, reduced effector differentiation, and cut interferon-γ and granzyme B production. Their conclusion is blunt: glutaminase inhibition negatively impacts the CD8 T cells that anti-PD1 has just activated.
The mechanism is the asymmetry. An activated T cell becomes more glutaminase-dependent, not less — and unlike the tumor cell beside it, it has no stroma feeding it alanine, no macropinocytic habit, and little transaminase capacity. In the exchange described in the previous article, the T cell was already the price-taker. Here it is the only cell in the microenvironment for which glutaminase is genuinely rate-limiting.
Broad Blockade Does the Opposite
Which would be a dead end, except that widening the block inverts the outcome.
The glutamine antagonist DON and its prodrugs block every glutamine-utilizing reaction rather than one enzyme — nucleotide synthesis, amidotransfer, redox control, glutaminolysis. Leone and colleagues called the result divergent metabolic programs: under the same blockade, tumor cells fail to adapt and T cells do.
The immune effects are substantial. In a separate study, the DON prodrug JHU083 reduced myeloid-derived suppressor cell infiltration and drove their apoptosis, reprogrammed tumor-associated macrophages toward a proinflammatory phenotype with increased TLR4, CD80, CD86 and TNF and better cross-presentation to CD8 T cells, and raised tumor-infiltrating CD8 counts. Most of all, it rendered checkpoint-blockade-resistant 4T1 tumors responsive to combined anti-PD1/CTLA4. The tumor-targeted prodrug DRP-104 reproduces the pattern in KEAP1-mutant lung cancer, suppressing growth and enhancing checkpoint blockade — the same genotype KEAPSAKE tried to serve with a selective inhibitor.
Part IV / What to Run Instead
Five Combinations With Preclinical Support
If the problem is redundancy, the answer is not a better glutaminase inhibitor. It is either a wider block, a second block on the bypass, or a drug aimed at the consequence of the first block rather than at a parallel pathway.
| Partner | Pairing tested | What it closes off | Result |
|---|---|---|---|
| GPX4 | DON + RSL3 | The redox consequence of the block itself | Synergistic ferroptosis in murine and patient-derived pancreatic models |
| MEK/ERK | DRP-104 + trametinib | ERK rescue via Axl and ErbB | Significant survival extension over either alone in syngeneic PDAC |
| Glucose transport | Glutor + CB-839 | Glycolytic compensation | 40-fold improvement in growth inhibition of HCT116 colon cancer cells |
| Glycolysis, clinically available | Metformin + CB-839 | The same, with an existing drug | Synergy in KRAS-mutant ovarian cancer; little effect in wild-type |
| Transaminases | Aminooxyacetate | The alternate route to glutamate | Growth repression in TAZ/YAP-activated breast cancer |
Why the Ferroptosis Pairing Is Different
Four of those five add a second blockade. The GPX4 pairing does something else: it waits for the first block to create a vulnerability, then exploits it. That makes it the most interesting of the group, and the mechanism is not the one you would guess.
Restricting glutamine in pancreatic cancer cells raises PAXIP1-mediated H3K4me3, which upregulates both the ferroptosis driver HMOX1 and the defences GPX4 and SLC7A11. HMOX1 breaks heme down into labile iron, so the cell accumulates the substrate for lipid peroxidation at the same time as it reinforces the enzyme that holds peroxidation off. It survives by leaning on GPX4 — and becomes killable by inhibiting it. DON plus the GPX4 inhibitor RSL3 was synergistic in both murine and patient-derived xenograft models.
One detail matters for anyone reading across the ferroptosis literature: this route is independent of SLC7A11. It is not classical cysteine starvation, so xCT inhibitors are not interchangeable with GPX4 inhibitors here. The epigenetic step is doing the work.
And the Selection Problem Nobody Has Solved
Two genotypes enrich for glutamine dependence, and both come with a catch. KEAP1/NRF2 and LKB1/STK11 lesions cooperate to raise it — but STK11 deficiency is precisely the context in which glutaminase inhibition was shown to damage the anti-PD1 response. The biomarker that predicts tumor sensitivity also predicts immune harm. That overlap may be most of what went wrong in KEAPSAKE.
Two things predict failure and ought to exclude. IDH1-mutant glioma lines resist through plasticity. And asparagine is a known rescue from glutamine depletion — asparagine regulates cellular adaptation to glutamine deprivation — so tumors with high asparagine synthetase have a standing escape route before the first dose.
Part V / Clearance Instead of Blockade
The Route That Starves Nobody
Every intervention so far has been a blockade. There is another option, and it is the only one in this article with checkpoint-blockade data already attached: instead of stopping the tumor from making ammonia, help the host get rid of it.
The colorectal work tested three clearance strategies, and the results separate in a way that is more informative than a clean sweep would have been.
| Agent | How it clears nitrogen | What happened |
|---|---|---|
| Ornithine | Rate-limiting substrate of the urea cycle — adds disposal capacity | Lower serum ammonia, restored T cell proliferation and cytokine output, smaller MC38 and CT26 tumors; with anti-PD-L1, survival up 72% |
| Glycerol phenylbutyrate | Phenylacetate conjugates glutamine; excreted in urine | Significantly reduced MC38 tumor size as a single agent; not tested with checkpoint blockade |
| Lactulose | Acts on the gut microbiota | No significant effect on tumor size |
Lactulose failing is the useful negative. It works on microbial ammonia production in the gut, and in a colorectal model that should have been the obvious lever. That it did nothing points at the cancer epithelium rather than a dysbiotic microbiota as the source — which is consistent with the argument in the previous article, where the ammonia is what the tumor's own nitrogen handling leaves behind.
The Sequential Design
Ornithine and phenylbutyrate are usually discussed as interchangeable scavengers. They are not, and the difference suggests they should be run together rather than chosen between.
Ornithine acts fast and acts twice: it feeds the urea cycle, and it supports fixation of ammonia onto glutamate to make glutamine. But glutamine is a holding pen rather than an exit — the nitrogen is bound, not gone, and it can come back out. Phenylbutyrate closes that loop. Converted to phenylacetate, it conjugates glutamine into phenylacetylglutamine, which leaves in the urine carrying both of glutamine's nitrogens with it.
phenylbutyrate: glutamine → phenylacetylglutamine → urine (gone)
One agent captures the nitrogen, the other removes the container. On paper that is a more durable reduction than either alone, and I have set the case out separately.
And the Objection to It
Which runs into this article's own argument.
The sequential design works by depleting glutamine. That is the mechanism, not a side effect. But glutamine depletion is precisely what disabled the CD8 T cells in Part III — activated T cells need glutamine, have no bypass for it, and become more dependent on it, not less. An intervention whose stated purpose is to clear the metabolic brake on those T cells, and which does so by removing a nutrient they cannot replace, is working against itself somewhere along the dose curve.
How much it matters is an empirical question nobody has answered. The conjugation happens in liver and kidney, so what counts is how far glutamine falls inside the tumor, not in plasma — and that may be negligible. But the experiment is obvious and undone: the pair plus anti-PD-L1, reading out tumor-infiltrating T cell function rather than tumor size. Note that the combination with checkpoint blockade was run with ornithine alone, which adds disposal capacity without consuming anything the T cell needs. Ornithine is the safer half of the pair on exactly this reasoning.
Where the Polyamine Problem Goes
Ornithine has its own liability, and it is a real one: ornithine decarboxylase converts it into putrescine, the entry point to polyamine synthesis, and polyamines support tumor growth. Supplying ornithine supplies that pathway too.
The obvious fix is to block ODC with difluoromethylornithine, and the usual framing treats this as damage control. The data suggest it is considerably more than that. Polyamine blockade with DFMO plus a polyamine transport inhibitor blocked tumor growth in immunocompetent mice but not in athymic nude mice lacking T cells — the effect requires adaptive immunity. It reduced Gr-1+CD11b+ myeloid suppressor cells, raised intratumoral CD3+ T cells, and blocked IL-4 induction of arginase activity, shifting macrophages away from the protumor phenotype.
Those are the same suppressive myeloid populations the glutamine antagonists act on in Part III. So DFMO is not a patch for ornithine's side reaction. It is an immune-enabling agent in its own right that happens to also redirect ornithine toward the urea cycle, which is where you wanted it.
Two caveats on that regimen, both worth stating plainly. Phenylbutyrate is also a pan-HDAC inhibitor across classes I, IIa and IIb — so any benefit it shows is ambiguous between nitrogen scavenging and chromatin effects, and demonstrating the ammonia mechanism requires ornithine or benzoate as a control. And the three-drug combination has never been tested together in a tumor model. The individual pieces have support; the assembly is an argument.
Part VI / What the Failure Means
Redundancy Is Evidence, Not Absence
The easy reading of CANTATA and KEAPSAKE is that glutamine metabolism was overrated. I think that reading is exactly backwards.
A pathway with one route through it is a pathway the cell can afford to have interrupted. A pathway with five routes through it is one the cell cannot. Nitrogen handling has transaminases, isoform duplicates, stromal supply lines, scavenging by engulfment, and a reductive route through glutamate dehydrogenase — and the cell maintains all of them. That is not what unimportant looks like. It is what load-bearing looks like.
And the clearance route from Part V sidesteps the problem entirely. Promoting ammonia detoxification reactivated T cells, decreased tumor growth and improved anti-PD-L1 efficacy without inhibiting a single tumor enzyme. It works with immunotherapy for the same reason the selective inhibitor works against it — it does not take a nutrient away from the cell that cannot replace it.
Which returns the question the previous article ended on. If a tumor's nitrogen economy is a set of redundant exits, then the interventions that fail are the ones that plug a single exit, and the ones that work either close all of them at once or help the host handle what comes out. Both of the results that look most promising here are of the second kind.
That is a strange place for cancer metabolism to have arrived at. The field spent fifteen years looking for the enzyme to switch off. The two interventions with the cleanest immunological data switch nothing off at all.
%20(3).png)
No comments:
Post a Comment