A rapid and cost-effective gene expression assay for the diagnosis of well-differentiated and dedifferentiated liposarcomas

ABSTRACT Histology does not reliably distinguish benign lipomas from atypical lipomatous tumor / well differentiated liposarcoma, nor dedifferentiated liposarcoma from other pleomorphic sarcomas, entities with different prognosis and management. Molecular confirmation of pathognomonic 12q13-15 amplifications leading to overexpression of MDM2 is a diagnostic gold standard. Currently the most commonly used assay for this purpose is Fluorescent In-Situ Hybridization (FISH), but this is labor-intensive. We assessed if newer NanoString-based technology could allow for more rapid and cost-efficient diagnosis of liposarcomas on standard formalin-fixed tissues through gene expression. Leveraging large-scale transcriptome data from The Cancer Genome Atlas, we identified 20 genes, most derived from the 12q13-15 amplicon, that distinguish dedifferentiated liposarcoma from other sarcomas and can be measured within a single NanoString assay. Using 21 cases of histologically-ambiguous low grade adipocytic tumors with available MDM2 amplification status, we built a machine learning based analytical pipeline that assigns a given sample as negative or positive for liposarcoma based on quantitative gene expression. We validated the assay’s effectiveness on an independent set of 100 sarcoma samples (including 40 incident prospective cases), where histology was considered insufficient for clinical diagnosis. The NanoString assay had a 93% technical success rate, and an accuracy of 97.8% versus an MDM2 amplification FISH gold standard. Turnaround time at our institution with NanoString was considerably faster and less expensive than FISH.