Abstract Background: Gene expression signatures are becoming increasingly common in clinical settings, for diagnosis, risk stratification, and treatment response prediction. However, deployment in clinical practice often entails a molecular platform different from the one used for development. Transporting molecular signatures across platforms (“bridging”) cannot be achieved directly, due to differences in scales and distributions in gene expression and requires bespoke methods. Here, we focused on the bridging of a clinicopathological and gene expression model (CP-GEP) across two reverse transcription-quantitative polymerase chain reaction (RT-qPCR) platforms: a single-plex SYBR Green qPCR, and a multiplex TaqMan probe-based qPCR (Idylla TM platform). CP-GEP is a regression model, combining eight genes (plus two housekeeping genes) and two clinicopathologic variables (age and Breslow thickness), able to stratify patients with melanoma into two groups based on their risk of sentinel lymph node metastasis. Methods: For the bridging, we selected 150 samples of primary melanoma patients who underwent a sentinel lymph node biopsy from a cohort of stage I-III melanoma patients. We measured the expression of the eight genes on both qPCR platforms and quantified them as ∆Ct. Within a bootstrapping scheme (500 repeats), for each gene, we learned a mapping (“bridging function”) between ∆Ct spaces, using all eight genes as regressors. Exploiting the linearity of per-gene bridging functions, we reparametrized the CP-GEP model so that we could directly use it in the target platform. We assessed the performance of the bridging, in terms of Percent Agreement (PA), defined as the percentage of the concordant binary outputs between the original and the bridged models; both in the bootstrapped samples and in 15 samples from an independent cohort. Results: In the bootstrapped samples, we obtained a PA of 100%, in the intervals for which we had pre-defined acceptance criteria, based on the distance from the cut point; and we had a mean PA of 89.2% [95% CI, 85.23%, 94.6%] when we considered all samples. In the external validation, we achieved a PA of 100%; all performance metrics were identical. The continuous scores of the original and bridged models were in excellent agreement, with an R2 equal to 0.95 both in the bridging and external validation cohorts. Conclusions: We showed that a model, combining gene expression and clinicopathologic variables, originally developed on a single-plex SYBR Green RT-qPCR, often used in diagnostic labs, could be robustly reparametrized to be compatible with a multiplex TaqMan probe-based qPCR (Idylla TM platform), more suitable for point-of-care-diagnostics. The ability to transfer models across different molecular platforms can greatly accelerate the translation of fundamental research to clinical practice. Citation Format: Lara Pozza, Sena Zumrutcu, Lizanne Bosman, Ashley van der Spek, Melanie Moreau, Dennie Tempel, Jvalini Dwarkasing, Domenico Bellomo. From diagnostic labs to rapid near-patient testing: Bridging across RT-qPCR platforms of a clinicopathological and gene expression model for cutaneous melanoma patients [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 3780.
Read full abstract