In cancer research we like our drug targets to behave consistently. Block a protein that a tumour depends on, and the tumour should suffer. One enzyme has stubbornly refused to follow that script.

The enzyme is adipose triglyceride lipase, or ATGL, encoded by the PNPLA2 gene. It catalyses the rate-limiting first step of neutral lipolysis, cleaving stored triacylglycerol into diacylglycerol and free fatty acids. In plain terms, it opens the cell's fat reserves and releases fuel that a tumour can oxidise for energy or divert into membranes and signalling lipids. On paper it looks like an appealing target but the literature is contradictory. Published reports describe effects ranging from tumour-suppressive to tumour-promoting, so that blocking ATGL either starves cancer cells, does nothing at all, actively helps the tumour along, or even initiates one in the first place. Our paper offers a mechanistic reason why all of these can be true at once.

We study prostate cancer, and our starting point was prostate-specific membrane antigen, or PSMA, the protein behind prostate cancer PET imaging and radioligand therapy. One feature of those scans is that a single tumour often does not light up evenly. Bright and dim regions sit next to each other. That patchiness matters clinically, because it complicates patient stratification for radioligand therapy. We wanted to know what it was telling us about the biology underneath.

We used the PET signal as a map and sampled paired PSMA-high and PSMA-low regions from within the same tumours and profiled them by label-free LC-MS/MS proteomics and reduced representation bisulfite sequencing, reading out protein abundance and DNA methylation side by side. Of the 4473 proteins we identified, the differential signal in PSMA-high regions pointed towards lipolytic reprogramming, and one enzyme stood out: ATGL. We then tested the association by immunohistochemistry in a tissue microarray cohort of 90 treatment-naïve patients, where ATGL and PSMA expression was highly correlated (r=0.74, P<0.0001) as predicted from the proteomics approach. Thus, PSMA was not simply marking cells on their surface, but pointing us towards a coordinated metabolic program.

The obvious next step was to target ATGL. If these tumour cells lean on fat, then inhibiting the enzyme that liberates that fat should be a therapeutic vulnerability. Using NG-497, an inhibitor selective for human ATGL, we confirmed suppression of triacylglycerol hydrolase activity across a panel of prostate cancer cell lines and assessed its effect on cell proliferation. LNCaP cells were markedly sensitive, with an IC50 of 2.2 µM, while the others were far more resistant, with IC50 values of 27 µM in 22RV1, 51 µM in PC3, and not reached below 80 µM in DU145. Whatever determined the outcome, it was not how much PSMA or ATGL target the cell had.

The determining factor was metabolic flexibility. Using Seahorse XF Mito Fuel Flex Test assays, we measured two distinct properties. First, the fuel a cell relies on at baseline, and second, its flexibility, defined operationally as the capacity to increase oxidation of one fuel when the pathways supplying the others are blocked. LNCaP and 22RV1 shared a fatty acid dependency, but they differed sharply in flexibility. LNCaP cells were intrinsically inflexible. Under ATGL inhibition they showed only modest shifts in fuel use and failed to expand their flexibility at all, so once we cut off their lipolytic supply they had nowhere to turn. 22RV1 cells did the opposite, executing coordinated rewiring towards anaerobic glycolysis and glutamine, and sustaining lipolytic flux through constitutively elevated hormone-sensitive lipase. That compensation was visible morphologically while NG-497 caused marked lipid droplet accumulation in LNCaP, PC3 and DU145 cells on BODIPY™ staining, but barely any in 22RV1. The vulnerability in LNCaP had never been the enzyme itself but the absence of an alternative.

That gives us a general principle. Targeting a metabolic enzyme only works when the tumour cannot compensate. Where escape routes exist, the cell reroutes its flux and survives, and reduced lipolysis can even be protective by limiting the lipotoxic burden of fatty acid overload, which is precisely how the same intervention ends up looking beneficial to the tumour. Seen through this lens, the contradictions in the literature dissolve. ATGL behaves like a tumour promoter in cells that cannot escape and like a tumour suppressor in cells that can. The long-running disagreement was never really about the enzyme. It was about the metabolic wiring of whichever model system each laboratory happened to be studying.

None of this logic is specific to lipolysis, or to prostate cancer. Every tumour runs on a metabolic network, and in every network some routes carry the load while others are redundant. Cut a redundant route and flux reroutes. Cut a load-bearing one in a cell with no alternative, and the tumour falls. What counts as load-bearing changes from one tumour to the next, so a tumour's metabolism reads almost like a fingerprint. It need not even be uniform within a single tumour. Think back to where we started, with a prostate that lit up unevenly on its scan. Because PSMA expression tracks this lipolytic programme, that patchy image is plausibly a map of metabolic patchiness, with different regions wired differently and therefore carrying different vulnerabilities.

The practical message is that measuring the target is not enough. Before turning any metabolic enzyme into a therapy, we need to know whether the tumour can route around it, which means profiling the network alongside target expression rather than counting a single protein. Prostate cancer offers a convenient entry point, because PSMA is already imaged routinely and flags these lipolytic, potentially vulnerable tumours. Our data also hint at how that vulnerability might be deepened deliberately. Enzalutamide, a mainstay of treatment in advanced disease, increased both ATGL and PSMA in the cell lines we tested, revealing reciprocal regulation between AR signalling and the PSMA/ATGL axis and raising the possibility that standard hormone therapy primes a tumour for a follow-up strike at its lipolytic machinery. We would stress that this is a hypothesis rather than a result. By our own central argument, higher target expression does not by itself predict sensitivity, so whether enzalutamide also narrows a tumour's metabolic options is the experiment that still has to be done.

Those limits are worth stating plainly. Our functional data come from cell lines, our patient data are correlative, and NG-497 awaits in vivo efficacy and safety testing. What we think travels beyond prostate cancer is the principle rather than the particular enzyme. Find the tumour that has painted itself into a metabolic corner, and you have found something you can aim at.