AI-Designed PARP-1 PET Tracer Evaluated in Mice and 4 Patients

Key Takeaways

• Among people with breast or ovarian cancer undergoing exploratory PET/CT, representative tracer uptake was descriptively consistent with high versus low lesion PARP-1 staining.
• Cellular uptake was greater in higher-PARP-1-expression lines and fell significantly after blocking with unlabeled compound.
• In 22Rv1 tumor-bearing mice, DOTA-FZPF showed significantly greater tumor uptake and lower liver activity than NOTA-FZPF.
• Mouse tumor tracer uptake declined during fuzuloparib administration despite similar subsequent PARP-1 staining in treated and control animals.
• No post-injection adverse events or vital-sign abnormalities were observed in the small exploratory human cohort.

Poly(ADP-ribose) polymerase 1 (PARP-1) expression can vary across cancer lesions, while a biopsy samples only a selected site. That mismatch complicates efforts to characterize PARP-1 across a patient's disease. An investigational gallium-68 tracer derived from the PARP inhibitor fuzuloparib could visualize uptake at multiple sites with positron emission tomography (PET), rather than relying entirely on tissue from one location. Whether such a tracer can reliably characterize lesion biology across sites remains an open clinical question.

Biopsy remains the clinical gold standard for assessing PARP protein expression in the investigators' framing, but a sampled lesion need not represent other metastatic sites. At Fudan University Shanghai Cancer Center in China, Ji and colleagues’ PARP-1 tracer study in Materials Today Bio connected artificial intelligence (AI)-assisted compound prioritization to chemical synthesis, cell testing, mouse imaging, and preliminary human positron emission tomography/computed tomography (PET/CT). This sequence allowed investigators to assess whether a candidate probe could be made and examined alongside tissue staining, without establishing a clinical test.

Scaffold-constrained recurrent neural network (RNN) generation produced seven fuzuloparib analogs. Convolutional neural network (CNN)-guided docking, a synthetic-complexity score, building-block review, deep neural network (DNN) target prediction, and molecular dynamics informed prioritization. Investigators selected NH2-FZPF partly because its amine permitted derivatization and a building block was available, then prepared DOTA-FZPF and NOTA-FZPF gallium-68 tracers. Predicted target rankings and docking poses were selection tools, not experimental demonstrations of selective binding. Testing included radiochemical purity and stability assays, uptake and blocking experiments in higher-PARP-1 22Rv1 and A2780 versus lower-expression AsPC-1 cells, and tumor-bearing mouse imaging. Comparative DOTA-versus-NOTA micro-PET/CT and separate tissue biodistribution measurements used three animals per group.

A separate six-mouse 22Rv1 experiment used serial scans during fuzuloparib administration through day 21. Exploratory DOTA-FZPF PET/CT involved three patients with breast cancer and one with ovarian cancer. Investigators also examined immunohistochemistry from representative biopsied lesions and monitored participants after injection.

Both radiotracers showed high radiochemical purity and remained more than 95% intact after three hours in saline and human and mouse serum, as well as in mouse urine collected at three hours. In cell experiments, uptake was greater in higher-expression 22Rv1 and A2780 cells than in AsPC-1 cells and fell significantly with excess unlabeled lead compound. These measured uptake and blocking findings are distinct from the computational predictions used to choose the candidate molecules.

In 22Rv1 mouse micro-PET/CT at 60 minutes, tumor uptake was 5.03 ± 0.77 percentage of injected dose per gram (%ID/g) with [68Ga]Ga-DOTA-FZPF versus 2.77 ± 0.51 %ID/g with [68Ga]Ga-NOTA-FZPF (P = 0.014). PET-derived liver uptake was 2.43 ± 0.47 versus 7.27 ± 2 %ID/g, respectively; these were imaging measurements, not ex vivo biodistribution values. DOTA tracer activity was also higher in high-expression mouse tumors than in lower-expression AsPC-1 tumors, and blockade reduced activity in 22Rv1 and A2780 tumors.

During fuzuloparib administration in 22Rv1 mice, DOTA tracer uptake in tumors declined, although PARP-1 staining at the final assessment was similar in treated and control animals. The authors suggested that drug occupancy could account for reduced tracer availability despite persistent target expression. This experiment evaluated a possible pharmacodynamic imaging signal in mice, not treatment efficacy or response prediction in patients.

In exploratory human scans, a breast lesion with high PARP-1 staining had a maximum standardized uptake value (SUVmax) of 3.7 versus 1.8 in the contralateral breast. A hepatic metastasis with low staining had an SUVmax of 1.3 versus 1.1 in normal liver, placing its activity close to background. These separate representative cases provide descriptive imaging–pathology concordance, not a quantitative correlation across patients or an estimate of diagnostic accuracy.

Investigators observed no adverse events or vital-sign abnormalities after injection among the four participants, an observation that does not establish general safety. The few clinical cases, restricted tumor types, and limited biopsy staining samples prevented a quantitative comparison of SUVmax with PARP-1 expression. The authors also noted modest mouse tumor-to-liver contrast, consistent with the difficulty of distinguishing the human hepatic lesion from liver background. Diagnostic performance, patient benefit, response prediction, and human drug-occupancy measurement remain unestablished; the Shanghai experience alone does not establish applicability to U.S. clinical imaging.

Further clinical evaluation would need to examine more lesions and tumor types and address hepatic tumor-to-background contrast. Whether PET can measure PARP inhibitor target engagement in humans is another question left open by the mouse dosing experiment.

Clinician Questions

Did AI target prediction establish selective PARP-1 binding for the fuzuloparib-derived PET probes?

The DNN ranked PARP-1 as the top predicted target for fuzuloparib, NH2-FZPF, DOTA-FZPF, and NOTA-FZPF. That ranking applies to the parent compound and both chelator-bearing candidates but does not directly demonstrate selective binding.

Did gallium-68 DOTA-FZPF or NOTA-FZPF produce organ abnormalities in short-term mouse testing?

In normal mice injected with [68Ga]Ga-DOTA-FZPF or [68Ga]Ga-NOTA-FZPF, investigators examined tissues after seven days and reported no apparent histopathological abnormalities in the heart, liver, spleen, lungs, or kidneys. That finding concerns short-term animal histology, not established human organ safety.