Non-Invasive NETseq Classifier Determines NET Diagnosis With High Accuracy, Sensitivity

Neuroendocrine tumor (NET) diagnosis is often delayed due to misdiagnosis or nonspecific symptoms, resulting in the occurrence of metastases in patients.

To enhance the detection of NETs in patients, a recent study sought to develop and validate a multi-gene native elongating transcript sequencing (NETseq) ensemble classifier using peripheral blood RNA sequencing to identify NET-related biomarkers and determine treatment response in patients receiving ¹⁷⁷Lu-DOTATATE peptide receptor radionuclide therapy (PRRT).

Blood samples were taken from a total of 178 patients with NETs and 73 healthy patients, which were then analyzed using RNA sequencing. A learning cohort of 59 PRRT-naïve patients with gastroenteropancreatic (GEP)-NETs and 38 healthy donors was formed to identify distinguishing gene features.

Five separate machine learning algorithms were trained and validated in the evaluation cohort (n=106), and the response to PRRT was measured in a PRRT cohort (n=46) and a PRRT response monitoring cohort (n=16).

The ensemble classifier determined NET from healthy patients samples with 100% accuracy in the learning cohort. In an evaluation cohort, the classifier demonstrated a 93% sensitivity rate (95% CI: 87.8%-98.03%) and a 91.4% specificity rate (95% CI: 82.1%-100%) for PRRT-naïve GEP-NETs (AUROC=95.4%).  The classifier also demonstrated an >87.5% sensitivity rate across different tumor characteristics, outperforming serum Chromogranin A sensitivity (χ²=21.89, P=4.161e−6).

In the PRRT cohort, significantly lower NETseq prediction scores were seen after treatment with ¹⁷⁷Lu-DOTATATE PRRT in comparison to non-responders. In an independent response monitoring cohort, paired samples (before PRRT and after the 2nd or 3rd cycle of PRRT) were analyzed.

The NETseq prediction score significantly decreased in partial responders (P=.002) and slightly reduced in those with stable disease (P=.068). The NETseq ensemble classifier identified PRRT-naïve GEP-NETs with ≥92% accuracy, and can play a potential role in early treatment response monitoring in the PRRT setting.

This blood-based, non-invasive classifier can serve as an optional tool for conventional methods in the detection and treatment response of patients with NETs.