TRAF7 Mutated Subgroups Differ in Sphenoid Wing Meningiomas with Hyperostosis

Introduction: Sphenoid wing meningiomas (SWMs) can involve the sphenoid bone to varying degrees. Our previous study found that the degree of bony involvement can predict genomic subgroups in SWMs such that tumors with bone invasion were nearly 30 times more likely to harbor somatic NF2 mutations. SWMs with only hyperostosis were over four times more likely to have a TRAF7 somatic mutation. However, TRAF7 mutations can cooccur with mutations in KLF4 and PI3K signaling pathway molecules, resulting in differences in transcriptome and potentially varied clinical relevance. We sought to investigate the tumor and patient characteristic associated with different TRAF7 SWM subgroups, and whether certain features can discriminate these subgroups using machine learning.

Methods: All patients who underwent surgery for SWMs and whose tumors underwent whole exome sequencing were reviewed. The extent of bony involvement was radiographically classified as (1) none, (2) frank tumor invasion, (3) only hyperostosis, or (4) both. Sphenoorbital meningiomas (SOMs) are further classified based on periorbital involvement and exhibit impressive hyperostosis. We assessed correlations between clinical features and TRAF7 subgroups, which include TRAF7 mutations alone (“TRAF7”), comutation of TRAF7 with KLF4^K409Q (“KLF4”) or PI3K signaling molecule mutations (“PI3K”). For discriminating the subgroups, we used four types of machine-learning classifiers: logistic regression, random forest, k-nearest neighbor, and the native Bayes. The best models were selected based on the accuracy in 10-fold cross validation.

Results: Among 64 SWMs, 11 were TRAF7 alone subgroup, while 11 and 7 were classified as KLF4 and PI3K, respectively. No TRAF7 subgroups showed frank tumor invasion of bone. PI3K was more likely to be higher grade (43%), compared with TRAF7 alone and KLF4 combined (5%; p = 0.031). Similarly, PI3K was correlated with atypical histology (p = 0.031). PI3K was more likely to occur in male patients, compared with TRAF7 alone or KLF4 combined (57 vs. 14%, p = 0.033). Among patients with hyperostosis only, presenting diplopia was associated with higher odds of PI3K (OR = 14; p = 0.033). The prevalence of SOMs in TRAF7 alone (46%) was over four times higher than that among KLF4 and PI3K combined (11%; p = 0.07). Patients with TRAF7 alone were on average 10 years younger than those in PI3K or KLF4 groups (50 vs. 61, p < 0.01). Random forest outperformed other classifiers and among all SWMs, the accuracy was 0.94 (95% CI: 0.85, 0.98) for predicting TRAF7 alone, and 0.91 (95% CI: 0.81, 0.96) for predicting PI3K. Positive predictive values calculated by applying models on hold-out samples (30% of all SWMs) were 0.74 and 0.95 for TRAF7 alone and PI3K, respectively, which are consistent with the parameters reported in our previous study on meningiomas.

Conclusion: Although hyperostosis bony involvement is a significant predictor of underlying TRAF7 SWM mutations, the presence and type of comutation can be associated with different clinical features. PI3K subgroup tends to be higher grade. Diplopia and male gender can discriminate PI3K from other TRAF7 mutated subgroups among patients with SWMs. Future studies with larger sample sizes are needed.

Publikationsverlauf

Artikel online veröffentlicht:
15. Februar 2022