Introduction: Variable genomic breakpoints have been identified through the application of target-capture DNA next-generation sequencing (NGS) for ALK, ROS1, and RET fusion detection in NSCLC. We investigated whether ALK, ROS1, and RET genomic breakpoint location can predict matched targeted therapy efficacy.

Methods: NSCLCs were analyzed by DNA NGS, target-specific RNA NGS, whole-transcriptome sequencing, and immunohistochemistry.

Results: In total, 3787 NSCLC samples were analyzed. DNA NGS detected ALK, ROS1, and RET fusions in 241, 59, and 76 cases, respectively. These fusions were divided into canonical (single EML4-ALK, CD74/EZR/TPM3/SDC4-ROS1, and KIF5B/CCDC6-RET fusions), noncanonical (single non-EML4-ALK, non-CD74/EZR/TPM3/SDC4-ROS1, and non-KIF5B/CCDC6-RET fusions), and primary/reciprocal (both primary and reciprocal rearrangements were detected) subtypes on the basis of genomic breakpoint position, and noncanonical and primary/reciprocal subtypes were defined as uncommon fusions. Further RNA sequencing and immunohistochemistry revealed that six of 47 (12.8%) uncommon fusions were actually nonproductive rearrangements that generated no aberrant transcripts or proteins. Moreover, genomic breakpoints of canonical ALK and RET, but not ROS1, fusions always predicted breakpoints at the transcript level, whereas 85.4% (35 of 41) of uncommon fusions actually produced canonical fusion transcripts. Patients with uncommon ALK fusion (n = 31) who received first-line crizotinib exhibited shorter median progression-free survival than those with canonical ALK fusion (n = 53, 8.4 mo versus 12.0 mo, p = 0.004). However, no difference in progression-free survival was observed when only ALK RNA or protein-positive cases were analyzed (p = 0.185).

Conclusions: Uncommon ALK, ROS1, and RET genomic breakpoint is an unreliable predictor of matched targeted therapy efficacy. Functional validation by RNA or protein assay may add value for the accurate detection and interpretation of rare fusions.