Tumor type prediction via tissue- and liquid-based comprehensive genomic profiling: High-specificity tobacco signature detection to support lung cancer diagnosis.

Abstract

8036 Background: Cigarette smoking exposes the lungs to tobacco mutagens, producing a distinct mutational pattern with elevated tumor mutational burden and strand bias for C>A mutations, aiding identification of lung origin in cancer of unknown primary (CUP). We evaluated a tobacco signature (TSig) caller for diagnosing lung cancer on tissue (TBx) and liquid biopsies (LBx) tested via the FoundationOne CDx (F1CDx) and FoundationOne Liquid CDx (F1LCDx) comprehensive genomic profiling (CGP) assays. Methods: We analyzed 351,611 TBx and 68,888 LBx samples, assessing TSigs (COSMIC v2 signatures 4 and 29) in a research use only capacity in cases with ≥10 somatic non-driver variants. For LBx, ctDNA tumor fraction (TF) was estimated via aneuploidy, fragment length, and variant features. TSig caller performance and co-occurring genomic alterations were evaluated against submitted diagnoses. Concordance was assessed in paired TBx and LBx samples (TF ≥1%) collected within 90 days of one another. Results: In all, 20.3% (71,211/351,611) of TBx and 13.6% (9,385/68,888) of LBx specimens had sufficient somatic variants for TSig analysis, with TSigs detected in 13.1% (9,302/71,211) of TBx and 10.2% (954/9,385) of LBx cases. Of TSig+ TBx cases, 87.5% (8,140/9,302) were submitted with a primary lung cancer diagnosis, 6.2% (579/9,302) as CUP, and 6.3% (583/9,302) as non-lung cancer. For TSig+ LBx cases, 81.9% (781/954) were submitted as primary lung cancers, 6.4% (61/954) as CUP, and 11.7% (112/954) as non-lung cancer. TSig+ cases were enriched for TP53 , KRAS , STK11 , KEAP1 , SMARCA4 , and MET alterations ( P < 0.001), consistent with their association with smoking-related lung cancer, regardless of submitted diagnosis, indicating that many cases submitted as CUP or non-lung cancer represented misdiagnosed lung cancers. Conversely, TSig− lung cancers were enriched for EGFR , ALK , ROS1 , and RET alterations ( P < 0.001), common in non-smokers. For cases with sufficient variants for TSig analysis, the TBx TSig caller had a high specificity of 97.1% but a lower sensitivity of 26.3%, with an accuracy of 66.3%, positive predictive value (PPV) of 87.5%, and negative predictive value (NPV) of 63.1%. LBx performance was comparable, with 96.7% specificity, 18.6% sensitivity, 61.8% accuracy, 81.9% PPV, and 59.5% NPV. In 272 paired TBx and LBx lung cancer samples, the positive percent agreement for TSig detection was 61.0%. Conclusions: TSig analysis identified misdiagnoses in 6.3% of TBx and 11.7% of LBx cases and supported lung origin in 6.3% of CUP cases. High specificity and PPV established TSig+ results as strong indicators of lung cancer, while lower sensitivity reflected the intrinsic limitation of the biomarker in detecting non-smoking-related cancers. These data highlight the utility of F1CDx and F1LCDx TSig analysis in refining lung cancer diagnosis and treatment.