Cluster analysis of immunohistochemical profiles in synovial sarcoma, malignant peripheral nerve sheath tumor, and Ewing sarcoma

As a result of overlapping morphologic and immunohistochemical features, it can be difficult to distinguish synovial sarcoma, malignant peripheral nerve sheath tumor, and Ewing sarcoma/primitive neuroectodermal tumor in core biopsies. To analyze and compare immunohistochemical profiles, we stained tissue microarrays of 23 synovial sarcomas, 23 malignant peripheral nerve sheath tumors, and 27 Ewing sarcomas with 22 antibodies potentially useful in the differential diagnosis, and analyzed the data with cluster analysis. Stain intensity was scored as none, weak, or strong. For CD99, tumors with membranous accentuation were independently categorized. Cluster analysis sorted five groups, with like tumors clustering together. Synovial sarcoma clustered into two groups: one cytokeratin and EMA positive ( n = 11), the other mostly cytokeratin negative, EMA positive, bcl-2 positive and mostly CD56 positive ( n 9). Malignant peripheral nerve sheath tumor clustered into two groups: one S100 positive, with nestin and NGFR positivity in most ( n 10), the other mostly S100 negative, and variably but mostly weakly positive for nestin and NGFR ( n 11). Ewing sarcomas clustered into a single group driven by membranous CD99 staining. Thirteen cases failed to cluster (outliers), while three Ewing sarcomas clustered into groups of other tumor types. Paired antibodies for each tumor type determined by visual assessment of cluster analysis data and statistical calculations of specificity, sensitivity, and predictive values showed that EMA/CK7 for synovial sarcoma, nestin/S100 for malignant peripheral nerve sheath tumor, and membranous CD99/Fli-1 for Ewing sarcoma yielded high specificity and positive predictive values. Cluster analysis also highlighted aberrant staining reactions and diagnostic pitfalls in these tumors. Hierarchical cluster analysis is an effective method for analyzing high-volume immunohistochemical data.