HSP90 inhibition blocks ERBB3 and RET phosphorylation in myxoid/round cell liposarcoma and causes massive cell death in vitro and in vivo

Myxoid sarcoma (MLS) is one of the most common types of malignant soft tissue tumors. MLS is characterized by the FUS-DDIT3 or EWSR1-DDIT3 fusion oncogenes that encode abnormal transcription factors. The receptor tyrosine kinase (RTK) encoding RET was previously identified as a putative downstream target gene to FUS-DDIT3 and here we show that cultured MLS cells expressed phosphorylated RET together with its ligand Persephin. Treatment with RET specific kinase inhibitor Vandetanib failed to reduce RET phosphorylation and inhibit cell growth, suggesting that other RTKs may phosphorylate RET. A screening pointed out EGFR and ERBB3 as the strongest expressed phosphorylated RTKs in MLS cells. We show that ERBB3 formed nuclear and cytoplasmic complexes with RET and both RTKs were previously reported to form complexes with EGFR. The formation of RTK hetero complexes could explain the observed Vandetanib resistence in MLS. EGFR and ERBB3 are clients of HSP90 that help complex formation and RTK activation. Treatment of cultured MLS cells with HSP90 inhibitor 17-DMAG, caused loss of RET and ERBB3 phosphorylation and lead to rapid cell death. Treatment of MLS xenograft carrying Nude mice resulted in massive necrosis, rupture of capillaries and hemorrhages in tumor tissues. We conclude that complex formation between RET and other RTKs may cause RTK inhibitor resistance. HSP90 inhibitors can overcome this resistance and are thus promising drugs for treatment of MLS/RCLS.

The hypothesis addressed here is that if a metasignature existed, the genes in the signature would reflect the essential transcriptional features of myxoid liposarcoma (MLA). There are 26 signatures analyzed for this case. At the significance threshold of Q-value less than 0.05, 87 genes were present in at least 22 of the 26 studies with minimum false discovery rate of 0.0714. All of the genes existing in this meta-signature are listed in Table S1.
We have used the limma package in R/Bioconductor platform to calculate moderated t-statistics and p-values (1). Multiple testing issue was addressed by calculating q-values (2). To identify the meta-signature, we have modified and implemented the procedure described in (3) as follows:

1.
A set S of differential expression analyses were selected and a significance threshold T was chosen to define differential expression signatures from the selected analyses (T default = 0.10). Genes with a q-value below the threshold T were selected.

2.
Genes were sorted by the number of signatures they were present in.

3.
The number of genes present in each possible number of signatures was tallied (N 1 , N 2 , ..., N 26 ).

4.
Random permutations were performed in which the actual q-values were assigned randomly to genes per signature, so that the genes in each signature changed randomly, but the number of genes in each signature remained the same. Randomization pattern was the same between signatures ensuring the dependence of genes across signatures during the randomization process. This simulation generated a tally of number of genes present in each possible number of random signatures (E 1 , E 2 , ..., E 26 ).

5.
The significance of intersection among the true signatures was assessed by the minimum meta-false discovery rate (mFDRmin) as: 6. If mFDR min was less than 0.10, a meta-signature was defined as those genes that were significantly differentially expressed (q ≤ T) in at least j of S analyses, where j was equal to i when mFDR min was defined.

7.
If no meta-signature was defined by using T default , steps 2 through 6 were repeated where T was lowered by 50% at each iteration until either a metasignature was defined or the number of genes in two or more signatures reached zero.
The expression rate in Table S1 was defined as: Nummer om signature in which the is gene was significantly regulated / Number of signatures in which the gene was upregulated For example, AGT displayed an expression rate of 26/26, which means that AGT was upregulated in all signatures. GYG2 on the other hand showed an expression rate of 22/19, which means that the gene was differently expressed in 22 of 26 signatures. Nineteen of these 22 regulates were upregulations.
In the original method, the random signatures were generated without considering the dependency of genes across different comparisons. In our method, the same randomization pattern was applied to signatures that were generated from the same dataset, enabling a more realistic simulation. Another modification was to use moderated t-statistic instead of standard t-statistic when calculating p-values, thus enhancing the robustness of the test statistic.
Two public microarray datasets were analyzed to identify a meta-signature for MLS (Table S2). These datasets contained a range of sarcomas, including MLS, and we aimed to find a meta-signature of expressed genes for MLS by performing a two class differential expression analysis with MLS versus each of the different tumor types. In total, we have performed 26 comparisons, thus generated 26 signatures (Table S3).