A PET tracer aimed at the PD-L1 protein has managed to map that target’s expression across every lesion in patients with non-small cell lung cancer (NSCLC), not just in the fragment a biopsy needle removes. The study, published in August 2026 in Radiology by researchers at the First Affiliated Hospital of Xiamen University in China, showed that an SUVmax above 5.6 identifies high PD-L1 expression, and that immunotherapy responders had far greater uptake than non-responders — median 5.9 versus 1.9.
The problem biopsy does not solve
In current practice, deciding to treat NSCLC with a checkpoint inhibitor rests on a single number: the PD-L1 tumor proportion score (TPS) measured by immunohistochemistry on a tissue sample. The usual strata separate negative (<1%), intermediate (1–49%) and high (≥50%) expression. The flaw is well known and poorly solved: PD-L1 shows pronounced heterogeneity both within a single tumor and between different lesions in the same patient.
That means a needle crossing one plane of a nodule may find 60% positive cells while an adrenal metastasis in the same patient sits at 2%. Immunohistochemistry is excellent at what it measures — it just measures a piece. Molecular imaging offers the opposite: whole-body visualization and quantification across every visible lesion, at the cost of far lower spatial resolution and no direct cellular information.
What the study measured and found
The tracer is 68Ga-NK224. Thirty-seven patients with newly diagnosed, recurrent or metastatic NSCLC underwent PET/CT with the tracer, with PD-L1 expression assessed by immunohistochemistry for comparison.
Three findings carry the conclusion. First, the threshold: an SUVmax of 5.6 best discriminated high PD-L1 expression — the cutoff that matters for a treatment decision. Second, the association with outcome: among patients who received immunotherapy, responders showed a median SUVmax of 5.9 against 1.9 in non-responders, a separation of more than threefold. Third, heterogeneity itself: uptake variability was greater when measured across the whole lesion than when restricted to the biopsy plane, which is a quantitative demonstration that the fragment underestimates how scattered the target really is.
It helps to recall what SUV actually is, since the number only means something normalized. It is the ratio of tissue activity concentration to injected activity corrected for body mass:
$$\mathrm{SUV} = \frac{C_{\text{tissue}}\ [\mathrm{kBq/mL}]}{A_{\text{inj}}\ [\mathrm{kBq}]\, /\, m\ [\mathrm{g}]}$$
where $C_{\text{tissue}}$ is the activity concentration measured in the voxel or region of interest, $A_{\text{inj}}$ is the administered activity and $m$ is patient mass. That is why an SUVmax of 5.9 is comparable across patients of different sizes — and why any error in weight, injection timing or scanner calibration contaminates the number directly.
Where the tracer comes from and what was already known
NK224 is not an antibody, and that choice explains much of the exam’s performance. It is a cyclic peptide derived from DK221, coupled to a NOTA chelator — an arrangement that allows labeling with either gallium-68 or fluorine-18. Small peptides clear the blood quickly, producing high tumor-to-background contrast in a short window; radiolabeled antibodies, by comparison, require 24 to 96 hours between injection and acquisition, which is unworkable in routine practice.
The tracer’s first-in-human characterization was published in 2025 in Clinical Cancer Research, a single-center prospective trial run by the same institution between May 2023 and November 2024, with 31 analyzable patients. That work established the operating parameters: acquisition 60 minutes after injection, an effective dose of $2.06 \times 10^{-2}$ mSv/MBq — in line with other gallium-68 agents — and no adverse events during injection or across four hours of observation.
The TPS-stratified numbers from that study help interpret the 5.6 threshold: median SUVmax of 2.6 in PD-L1 negative lesions (<1%), 4.3 in intermediate lesions (1–49%) and 7.8 in high expressers (≥50%), with a statistically robust association (H = 24.994; P < 0.001). And one result deserves attention: 18F-FDG showed no equivalent correlation with TPS (P = 0.138). Glycolytic metabolism is not a proxy for PD-L1 expression — the targeted tracer is measuring something FDG cannot see. Dispersion between lesions in the same patient had a median coefficient of variation of 27.5%, reaching 53.2% in some cases.
Implications for practice and for the imaging service
If these findings hold up in multicenter work, the effect is not to replace immunohistochemistry — it is to decide where to biopsy and when to distrust the result. A patient with 50% uptake dispersion across lesions is a patient whose single-sample TPS does not represent the disease. In that scenario PET would steer the biopsy toward the highest-uptake lesion, cutting the risk of a sampling false negative.
The parallel with breast imaging is direct: the same target-imaging logic applied to HER2 PET/CT for predicting treatment response shows molecular imaging migrating from staging into treatment selection. And the move is not isolated within nuclear medicine: the recent FDA approval of a tau PET agent confirms that target-specific tracers are moving out of research and into the market.
For the service there are three operational consequences. First, protocol: an exam whose decision threshold is 5.6 demands tight control of calibration, post-injection timing and region-of-interest delineation — because SUV variability across scanners can run 10% to 20% when protocols are not harmonized. Second, reporting: quoting only the index lesion’s SUVmax throws away the newest piece of information the exam provides, which is the dispersion across lesions. Third, integration with pathology — this is an exam that only generates value in multidisciplinary discussion, read against the TPS.
Limitations and what is missing
Sobriety about the evidence stage is warranted. This is a single-center study with a small sample in an Asian population, with the 5.6 threshold derived from the cohort itself — which almost always over-optimizes a cutoff. There is no external validation yet, and no demonstration that choosing immunotherapy by PET improves survival compared with choosing by TPS. The 5.9-versus-1.9 separation between responders and non-responders is striking, but it was measured in a treated subgroup inside an already small cohort.
There is also a problem the imaging community knows well: interpretation variability. Even in well-established exams, chest CT reading varies considerably between radiologists — and a quantitative exam with a single threshold inherits that problem at the lesion-delineation step. Before any clinical adoption, the mandatory next step is a multicenter trial with a pre-specified threshold and a survival endpoint, not a correlation endpoint.
Source: AuntMinnie — New immuno-PET tracer shows promise in lung cancer
