Comprehensive Molecular Analysis of Inflammatory Myofibroblastic Tumors Reveals Diverse Genomic Landscape and Potential Predictive Markers for Response to Crizotinib

Inflammatory myofibroblastic tumor (IMFT) is an ultra-rare neoplasm composed of myofibroblastic and fibroblastic spindle cells with an inflammatory infiltrate consisting of plasma cells, lymphocytes, and/or eosinophils (1). This orphan malignancy can affect people at any age, with a predilection for children, adolescents, and young adults. It most commonly arises in lung, retroperitoneum, and abdominal cavity (2). Metastases are rare and the mainstay of treatment of IMFTs is radical surgery, but the disease has a tendency to recur locally after initial resection. Although a recent retrospective series suggested that some cases can respond to cytotoxic treatment, the role of chemotherapy in IMFT is still poorly established (3). Approximately 50% of IMFTs harbor a gene rearrangement activating anaplastic lymphoma kinase (ALK), which is known to be the most common driver in this rare disease. A number of other genetic abnormalities such as rearrangements of ROS1, RET, or neurotrophic tyrosine kinase 3 (NTRK3) have been described previously in individual patients (4–6), making IMFTs potentially sensitive to specific kinase inhibitors.

The European Organization for Research and Treatment of Cancer (EORTC) conducted the 90101 “CREATE” study, a very first prospective phase II trial exploring the use of crizotinib, an oral ALK/ROS1/MET inhibitor in this rare disease (7). The study assessed the activity of this agent in ALK-positive and ALK-negative cases. ALK positivity was defined by ALK immunopositivity and/or ALK rearrangement on FISH present in at least 15% of tumor cells. High antitumor activity of crizotinib was observed in ALK-positive IMFT (50% achieved objective response). Therefore, ALK was considered as an important predictive biomarker in this setting. However, the observation of tumor shrinkage in patients with ALK-negative disease prompted the hypothesis that other molecular events may also have contributed to the response to crizotinib in some cases, which led to the current exploratory analysis.

To date, most studies have focused on the identification of specific fusions and fusion partners in IMFT (8–11). Other genomic alterations such as copy number changes (e.g., loss of 19q13.32) or gene mutations affecting, for example, TP53, MDM2, KIT, or PIK3CA, have been identified in published case reports (12–14). Because of the rarity of IMFT, the molecular profile remains poorly understood. The aim of the current study was to perform a comprehensive molecular analysis of archival tumor material from patients with IMFT enrolled in EORTC trial 90101 “CREATE” to describe the molecular landscape of this ultra-rare sarcoma. Furthermore, we aimed to correlate molecular findings with clinical outcome in the trial participants.

The collection of archival tumor tissue blocks at study entry and central biobanking in a commercial biorepository (BioRep) was a mandatory component of the EORTC 90101 study protocol (http://www.eortc.be/services/doc/protocols/90101v10.0.pdf). Biological material and written informed consent from 35 patients with the local diagnosis of IMFT were obtained. No prior treatment to ALK and/or MET inhibitor, including cabozantinib was allowed. Central reference pathology at the University Hospitals Leuven, Belgium confirmed the diagnosis of IMFT in 24 of 35 tumors from a total of 16 primary and eight metastatic lesions, collected before starting crizotinib treatment. Only cases of centrally confirmed diagnosis of IMFT qualified for treatment with crizotinib in the trial and were subject to the current analysis (7). For the purpose of the exploratory research, slides of 4 μm were used for IHC and hematoxylin and eosin (H&E) staining, and 10 μm archival tissue sections were used for H&E-based tissue macrodissection and subsequent DNA and RNA extraction. From 21 of 24 formalin-fixed, paraffin-embedded (FFPE) tumor blocks with sufficient material, a tissue microarray (TMA) was constructed with three cores of 1.5 mm from each block, annotated by the sarcoma reference pathologist involved in this trial. Ethics approval for the published clinical trial was obtained by competent committee(s) and according to national legislation. The study was conducted in accordance with the Declaration of Helsinki, laws and regulations of each participating country/institution, and the International Conference on Harmonization-Good Clinical Practice. The scientific work was approved by the Medical Ethics Committee, University Hospitals Leuven, Leuven, Belgium (S54023).

The CREATE trial included ALK-positive and ALK-negative patients, centrally assessed using both IHC and FISH as described previously in the study protocol. A tumor was deemed ALK-positive if it showed immunopositivity confirmed by IHC with an ALK1 antibody (DAKO) and/or FISH using the Vysis LSI ALK dual color break-apart rearrangement probe (Abbott Molecular), with at least 15% cells staining positive or showing the rearrangement. As part of the current, post hoc exploratory work, further IHC was performed for ALK, ROS1, and NTRK on tissue sections, using the ALK D5F3 antibody (Cell Signaling Technology), ROS1 D4D6 antibody (Cell Signaling Technology), and pan-TRK EPR17341 antibody (Abcam) with 1:50, 1:200 dilutions, and 28 μg/mL, respectively. The Envision Plus detection system (DAKO) was used as the secondary antibody. The anchored multiplex PCR-based targeted next-generation sequencing using the Archer FusionPlex CTL Panel (Archer) was applied to study the presence of gene fusions, including ALK/ROS1 in 24 tumors (15), in which 20 had sufficient material and acceptable RNA quality.

DNA extracted from FFPE tumor samples was used to construct random DNA libraries, which were sequenced on an Illumina HiSeq4000 at low coverage (±0.1×) to evaluate copy-number alterations (CNA). Raw sequencing reads (50 bp) were mapped to the human reference genome (GRCh37/hg19 version) using Burrows-Wheeler Aligner (BWA v0.5.8a, Massachusetts Institute of Technology), sorted with SAMtools (v0.1.19, Massachusetts Institute of Technology), and duplicates were removed with Picard tools. These aligned reads were counted in bins of 50 kb using QDNASeq and segmented by ASCAT (16, 17). The Genomic Identification of Significant Targets in Cancer (GISTIC, Broad Institute) algorithm was used to identify the most frequent and significant chromosomal alterations in tumors. It assigned a G-score to each aberration, which considered the amplitude of the aberration as well as the frequency of its occurrence across samples. FDR (q value) was calculated for the aberrant regions, using the Benjamini and Hochberg method to account for multiple testing (18). A region was considered deleted if the log value was ≤0.1 or gained when the log value was >0.1. A cutoff q value of 0.25 was used to select significant CNAs. Alterations spanning >75% of a chromosomal arm were defined as broad CNAs (arm level), whereas CNAs spanning <75% of a chromosomal arm were considered as focal CNAs (region level).

Libraries prepared for low-coverage whole-genome sequencing (LC-WGS) were enriched for exomic sequences using the SeqCapV3 exome enrichment kit (Roche) following the manufacturer's instructions and were sequenced on HiSeq4000 using a flow cell resulting in 2 × 150 bp paired-end reads, which were mapped and sorted as described above. Duplicate reads were removed using Picard tools (Broad Institute), followed by base recalibration and local realignment using the Genome Analysis Tool Kit (GATK, Broad Institute). Substitutions were called with GATK's haplotype caller and small insertions and deletions were called with Dindel. Low-quality mutations <×10 coverage and a quality score <30 for substitutions and <50 for indels were excluded. Because no germline samples were available, a stringent filtering strategy based on publicly available databases was applied to limit the common SNPs. Mutations occurring in large databases (ESP, 1 kg, ExAC) with an allelic frequency >0.001 were removed. Mutations occurring in smaller, appropriate databases (bitsTrio, inhouseDB, cg69, GoNL, Illumina) were removed if they occurred in more than one individual. The cancer gene consensus-associated genes (CGC) were then selected using Catalogue of Somatic Mutations in Cancer databases (COSMIC v89, Wellcome Trust Sanger Institute) and analyzed further (19). PolyPhen-2 (Harvard Medical School) was applied to predict the possible impact of amino acid substitutions on the structure and function of human proteins (20). The deconstructSigs was used to characterize the mutational signatures that are compatible with the COSMIC single-base substitutions (SBS) catalogue (21).

The radiological response to crizotinib was evaluated using RECIST v1.1 as previously reported (7). The primary endpoint of the clinical trial was the proportion of patients who achieved an objective response [i.e., a complete response (CR) or partial response (PR) according to RECIST]. For the purpose of the current exploratory analysis, patients achieving CR or PR were considered as responders (n = 7), whereas patients with stable disease (SD) or progressive disease (PD) were defined as non-responders (n = 12). To assess the correlation between molecular findings and clinical outcome of patients, Fisher's exact test was used to test whether response groups were significantly enriched for certain alterations. Survival analysis with Kaplan–Meier estimates and comparisons with the log-rank test was used to assess the correlation between molecular findings and clinical outcome of patients. Progression-free survival (PFS) was calculated from the starting date of crizotinib treatment until the date of disease progression (last updated on April 25, 2019). Statistical analysis was performed using Prism 7 (GraphPad) and a P value of <0.05 was considered significant.

Archival tumor material was available from 24 patients with centrally confirmed diagnosis of IMFT. This included 16 ALK-positive and 8 ALK-negative cases, with ALK status determined according to the CREATE protocol. The male-to-female ratio was 1.6 and the median age at diagnosis was 48 years (range, 15–79). Five out of 20 treated cases were still receiving crizotinib treatment at the clinical data cutoff on April 25, 2019. Among 20 patients who received crizotinib, 19 were evaluable by RECIST v1.1: 7 achieved an objective response (CR or PR), 11 had SD as best response, and PD was observed in only one case. The 5 patients who were still receiving active treatment at least achieved SD. Key clinicopathological variables are summarized in Table 1.

In this exploratory series, the majority (n = 20 out of 24 patients) presented concordant results between ALK1 IHC and FISH used for ALK evaluation, as described previously in the study protocol. Both methods were discordant in four ALK-positive tumors: two were negative on IHC but had positive FISH results (with 15% and 21% of cells with split ALK signal, respectively), whereas two had ALK1 immuno-positive FISH results that were below the predefined cutoff (0% and 11% of tumor cells with split signal, respectively). To confirm the presence of an ALK rearrangement, an Archer Fusion CTL panel was performed in a subset of 20 evaluable cases. Fusion transcripts were identified in 13 out of 20 specimens of which RNA quality was acceptable. All but one fusion involved ALK with a total of 11 different fusion partners (see Table 1 for details). In one case, we unexpectedly detected an ETV6–NTRK3 fusion. The ETV6 translocation and NTRK positivity were confirmed by FISH and IHC with a pan-TRK antibody, respectively (Supplementary Fig. S1). In three out of the four discordant cases, various fusions were detected, all involving the ALK gene with one of three different partners (Table 1). The only ALK-positive case that lacked ALK expression had 21% cells with split signal on FISH and was found to have an underlying EML4–ALK fusion that is known not to be well recognized by the ALK1 antibody, as shown in ALK-driven non–small cell lung cancer (NSCLC; ref. 22). Interestingly, none of the analyzed cases had a ROS1 rearrangement in FISH or showed ROS1 expression in IHC. On the basis of the experience in ALK-driven NSCLC, we also tested the D5F3 antibody for ALK, which showed 100% sensitivity in NSCLC, whereas ALK1 only had 66% sensitivity (23). In our study, we observed 13 positive and 10 negative tumor samples with ALK–D5F3 antibody, of which four were discordant with ALK1 immunostaining. An overview of ALK/ROS1/NTRK molecular and histopathological evaluation is summarized in Table 1.

LC-WGS and GISTIC were performed on 24 tumor samples to detect chromosomal regions recurrently affected by CNAs. Across all samples, the frequency of the broad (chromosomal arm level) and focal (region level) events ranged from 21% to 58% and 21% to 71%, respectively. The most frequent CNA was the loss of chromosome 22q, which was present in 15 out of 24 (58%) IMFTs. Other broad alterations included losses in 6p (25%), 6q (21%), 13q (29%), 16p (33%), 16q (29%), 18q (21%), 19p (25%), and 19q (29%). We also detected recurrent focal alterations at 27 loci as well as involved CGCs. Ten regions were gained and 17 deleted, as summarized in Supplementary Table S1. In 54% of cases 2p21 (EPAS1, SIX2, and EML4) was gained, whereas recurrent losses were observed in 22q12.3 (71%; TIPM3, ISX), 7q36.3 (54%; MNX1), 1p36.32 (50%; RPL22, SKI, TNFRSF14, CAMTA1, PRDM16), 8p23.3 (50%; ARHGEF10), 10q26.3 (54%; MGMT, DUX4), 12q24.33 (46%; POLE), and 11q13.4 (33%; CCND1). Interestingly, losses of 19p and 19q co-occurred in six cases, suggesting chromosome 19 monosomy. Losses of 22q12.3 and 22q13.22 were also observed to co-occur in 15 cases. The copy-number gains and losses, which include both broad and focal alterations, are summarized in Fig. 1A. Moreover, we also summarized the percentage of genome that was altered by CNAs in each tumor in Table 1, showing an average of 8.1% (range, 0%–39.6%) in 16 ALK-positive tumors and 1.3% (0%–3.7%) in eight ALK-negative tumors. To determine the association between CNA profile and clinical outcome, we correlated the detected CNAs with PFS and the response to crizotinib. From all broad and focal CNAs, only chromosome 19 monosomy was found to be associated with shorter PFS in patients with IMFT receiving crizotinib (P = 0.008; Fig. 1B).

To determine the correlation between CNA profiles underlying IMFT and the response to crizotinib treatment, GISTIC analysis was performed in two subgroups of IMFT, that is, responders (CR+PR) versus non-responders (SD+PD; according to RECIST 1.1). Responders had a lower number of CNAs than non-responders (12 vs. 30 regions; Fig. 2A). Using Kaplan–Meier estimates, we found that losses of 12q24.33 and 19q12 were associated with decreased PFS in non-responders (Fig. 2B), whereas other changes did not correlate with PFS. We also performed a similar subgroup analysis in ALK-positive and ALK-negative IMFTs. A higher number of significant regions affected by CNAs was detected in ALK-positive IMFT compared with ALK-negative cases (25 vs. 2 regions; Fig. 2C). Correlations between CNA profiles and clinical features (PFS and response to crizotinib) were also made in ALK-positive and ALK-negative subgroups, and we found loss of 19q to be associated with decreased PFS on crizotinib treatment in the ALK-positive patients (Fig. 2D).

Whole-exome sequencing (WES) was successfully performed in 22 IMFTs with an average coverage depth of ×93, whereas two IMFT samples had insufficient sequencing reads (coverage less than ×10). We detected a total of 8,615 mutations with an average of 392 (range, 212–766) mutations per sample. Single-nucleotide variants (SNV) represented the majority of mutations (5,378 nonsynonymous substitutions, 2,841 synonymous substitutions, 210 nonsense, and 57 splicing mutations) and 132 insertions and deletions (73 frameshift and 60 non-frameshift) were detected. The mutational load profile was calculated in each case. We found an average of 6.8 mutations (range, 4.8–12.8) per coding Mb per sample (Fig. 3A). The spectrum of single base-pair substitutions for IMFT was characterized by a predominance of C>T transitions, of which the dominant mutational signatures were SBS5 (82.9%) and SBS1 (17.1%) that were seen across vast majority of tumors (Fig. 3B; Supplementary Table S2). To identify potential cancer-related genes, we focused on those that were found to be mutated in CGCs. Excluding synonymous SNVs, we identified a total of 178 unique mutations, affecting 143 genes with an average of 8 (range, 3–35) per case. Mutations in 30 CGCs were detected in more than one tumor (see Supplementary Table S3), and alterations identified in 13 of 30 genes (FAT1, BIRC6, MUC4, BCR, CBLB, CTNND2, FAM46C, FCRL4, MYH11, PIK3CA, PCM1, TP53, and SETD2) were considered as damaging by PolyPhen-2, including indels, nonsense, or missense mutations. The proportion of mutations and the number of damaging mutations is presented in Fig. 3C. The most frequent alterations were mutations in FAT1 (four out of 22 samples), which encodes a cadherin-like protein functioning as a negative regulator of Wnt signaling (24). Furthermore, we observed mutations in TP53, PIK3CA, CBLB, BIRC6, SMAD4, FAMC46C, CALR, and MYH11 exclusively in the subgroup of non-responders. Mutations in TP53, SMAD4, and FAM46C were exclusively detected in ALK-negative IMFT. Correlations between recurrent mutations and clinical features revealed that PIK3CA alterations, identified in two tumors, were associated with decreased PFS in IMFT (P = 0.0475) albeit to a limited amount (Fig. 3D). No associations between mutational load and PFS, response to crizotinib, disease status (primary tumors vs. metastatic lesions) or ALK status were found in the entire study population (Fig. 3A). However, one tumor that did not respond to crizotinib treatment revealed the highest number of mutations (overall 766 mutations and 35 mutated CGCs). These mutations were found in genes that encode proteins involved in DNA repair mechanism (e.g., BRCA2, RB1, TP53) as well as in others (e.g., AKT1, CDKN1B, PI3KCA). Moreover, this tumor had an increased proportion of SBS2 and SBS13 that are commonly seen in some cancers presenting local hypermutation (Supplementary Table S2).

We next explored the ALK-related alterations in addition to ALK translocation because these events can affect the sensitivity to the ALK inhibitor. In three ALK-positive cases we observed ALK locus amplification. In one of them it was an isolated 2p23.2 high-level copy-number gain (Fig. 4A), whereas in two others it was associated with chromosome 2 chromothripsis (Fig. 4B and C), a phenomenon with clustered chromosomal rearrangements occurring in a single event in restricted genomic regions in one or a few chromosomes (25). Interestingly, chromothripsis-caused ALK amplification was associated with a breakpoint in the ALK locus, which might suggest the amplification of the ALK fusion in both cases. Among the three ALK-positive cases harboring ALK amplification, 2 patients had SD with PFS of 8.5 (SeqID132) and 23.9 months (SeqID197), respectively, whereas clinical data were missing for the remaining one (SeqID001). In addition, we identified a case of an ALK mutation in the analyzed cohort. The ALK mutation (p.N571K) was identified in an ALK-negative patient who achieved SD as best response to crizotinib. Unlike previously reported crizotinib-resistant mutations in tyrosine kinase domain, this mutation is located in the meprin/A5-protein/PTPmu (MAM2) domain, a highly conserved region associated with cell–cell interaction. This substitution was previously described as a variant of uncertain significance in neuroblastoma (26), but we detected it in IMFT for the first time (Fig. 4D).

Using the molecular findings, we sought to identify potential targets for this orphan disease based on the genetic alterations underlying IMFT. An overview of genes affected either by fusions, CNAs, and/or mutations in our series of 24 IMFT cases is summarized in Fig. 5A. Results suggest that the receptor tyrosine kinase (RTK) signaling (22/24), DNA damage repair mechanisms (19/24), Wnt signaling (16/24), and cell-cycle and cell death pathway (13/24) are dysregulated in IMFT (Fig. 5B). Using the drug–gene interaction database (DGIdb; ref. 27), we identified 17 potentially actionable recurrent gene aberrations in IMFT (ATRX, BRD4, CALR, CBLB, CLTCL1, FAM46C, FAT1, FCRL4, FOXO1, NUTM2B, PCM1, PDE4DIP, PIK3CA, PIK3CB, SETD2, SMAD4, and TP53).

In this study, we used archival tumor samples to reveal the molecular alteration profile of IMFT and further analyzed recurrent alterations that may have an impact on the response to the ALK/ROS1 inhibitor crizotinib. Appropriate identification of the ALK/ROS1 status in patients with IMFT is crucial for accurate diagnosis and therapeutic decision-making. Using the Archer panel, we detected ALK fusions in 12 out of 20 tested samples, involving 11 different fusion partners. Because of this molecular heterogeneity it was not possible to draw any conclusions about the sensitivity to crizotinib regarding specific fusion proteins. However, we were able to show a good concordance between conventional IHC staining with the ALK1 antibody and the presence of ALK rearrangement with FISH, confirming that ALK1 is a good surrogate marker for ALK-rearranged IMFT, as shown previously (8). The only case that was immunonegative with the ALK1 antibody but positive with FISH (15% rearranged cells) was found to have an EML4–ALK fusion with Archer. This rearrangement is known to produce a fusion protein that is less frequently detected with the ALK1 antibody (28). Interestingly, ALK–D5F3 immunostaining was also negative in this case, even though this antibody is considered to be more specific for the EML4–ALK fusion, as shown in studies in NSCLC (28). Although it was described previously that ALK–D5F3 outperformed to detect this hybrid protein in NSCLC, it might also cause false-positive staining in some EML4–ALK-negative tumors (29), as shown in two cases in our study. As a result, we recommend the ALK1 antibody for IHC in the context of IMFT, but diagnosis supplemented by other methodologies (FISH and/or Archer) should be considered for a better disease management approach.

Notably, all but one responding patient were ALK-positive with either IHC and/or FISH, as described previously in the original article (7). The remaining case with a partial response to crizotinib presented the ETV6–NTRK3 fusion with the Archer panel, which was further confirmed by FISH (split ETV6 signal) and IHC with a pan-TRK monoclonal antibody. IMFT tumors with this fusion were previously described in a small subset of ALK-negative cases (6). Published in vitro studies showed that crizotinib inhibited the proliferation of ETV6–NTRK3-dependent tumor cells even more potently than its primary target ALK, indicating its antitumor potency in vivo (30). There are also anecdotal cases of tumors with NTRK fusions (e.g., infantile fibrosarcomas) that were successfully treated with crizotinib (31). These data suggest that crizotinib can also be considered for ALK-negative patients, if diagnostic tests reveal other actionable targets that match the kinase inhibitory profile of crizotinib. Furthermore, specific TRK inhibitors, such as larotrectinib or entrectinib could be considered as next-line therapy for patients with IMFT with documented NTRK rearrangements, as these compounds are associated with high response rates (>75%), regardless of tumor histology (32). We did not identify any ROS1-positive IMFT using IHC, FISH, or the Archer panel in our series. ROS1-driven IMFTs are very rare and occur in children and young adults, whereas our study population mainly comprised adult patients with a median age of 48 years (range, 15–69).

Next, we performed LC-WGS on 24 tumors focusing on the identification of recurrent CNAs in the study cohort. The most common CNAs we detected were losses of 22q (58%) and 22q12.3 (71%) as broad and focal events, respectively, suggesting an association between IMFT and these alterations. As previously reported, loss of 22q causing a deletion of putative tumor suppressor genes was frequently seen in patients with epithelioid sarcoma, adenocarcinoma, and colorectal carcinoma (33–35). It is possible that tumor-suppressor genes affected by these CNAs influence the tumorigenesis or progression of IMFT. Furthermore, a frequent loss of 22q12.3 and the deletion of the tissue inhibitor of metalloproteinases-3 (TIMP3) caused by 22q loss of heterozygosity is implicated in disease progression of glioblastoma (36). Another cluster analysis in metastatic STS also revealed that TIMP3 overexpression was associated with good prognosis and decreased metastatic potential (37). We therefore believe that TIMP3 could be an attractive candidate for further investigation in IMFT. We also observed chromosome 19 monosomy that was associated with poor PFS in patients receiving crizotinib. This suggests that the loss of chromosome 19 may serve as a potential biomarker to predict the outcome of patients with IMFT on ALK inhibitor treatment. In the subgroup analysis, we observed a higher number of CNAs in non-responders compared with responders. Thus, a complex CNA profile may cause decreased sensitivity to crizotinib. Likewise, a more complex CNA profile and a higher proportion of the altered genome were observed in ALK-positive IMFT (n = 16) compared with ALK-negative IMFT (n = 8). The distinct CNA profiles between ALK-positive and ALK-negative IMFT suggest that they may represent different branches of this disease. We also observed associations between decreased PFS and CNAs in subgroups, but due to the small sample size we were not able to determine their specific prognostic or predictive value.

In addition to the CNA analysis, we also investigated the IMFT mutational profile using WES. To the best of our knowledge, this has not yet been done in this tumor subtype. Tumor mutational burden is an emerging predictive biomarker with relevance for tumor treatment with immune checkpoint inhibitors. In other settings, a higher mutational burden is associated with poor response to certain treatments, as in the case of EGFR inhibition in NSCLC (38). The average of 7 mutations per Mb detected in our cohort suggests an intermediate mutational burden in IMFT as compared with other solid tumors (39). We did not find a significant association between mutational burden and response to crizotinib. Furthermore, no association between loss of 12q24.33 and mutational burden was found, suggesting that the POLE deletion does not influence the mutational profile in the analyzed cohort. The mutational signature of IMFT was characterized by predominant C>T transitions, as seen in most solid tumors (40). In addition, cytosine deamination in FFPE material is known to induce an artefact of C>T transitions with a signature highly similar to SBS30 (41). However, the clock-like signatures of SBS5 and SBS1 observed in our IMFT cohort suggest the etiology linked with ageing and allow us to disregard the artefactual contribution from FFPE material in this study. Interestingly, the highest number of mutations (766 mutations overall, including 35 mutated CGCs) were found in the IMFT of the unique patient in our series who did not benefit from crizotinib (PD). In this case, we found alterations in TP53, RB1, PIK3CA, BRCA, POLE, and MGMT as well as an increased proportion of hypermutation-associated signatures (SBS2 and SBS13) by sequencing, suggesting DNA repair deficiency in that unique case.

Next, we focused on recurrent mutations identified in our cohort and analyzed them using PolyPhen-2. The most frequently mutated gene detected was FAT1, a gene-suppressing cancer cell growth via binding β-catenin and repressing its translocation to the nucleus (24). Dysregulation of β-catenin is known to play an important role in the Wnt signaling pathway, promoting tumorigenesis not only in various types of carcinoma but also in mesenchymal tumors (e.g., in endometrial stromal sarcoma or solitary fibrous tumor; ref. 42). We believe that mutations in FAT1 and aberrant Wnt signaling could contribute to the tumorigenesis in IMFT. A blockade of Wnt signaling can be suggested as an experimental treatment approach for patients with IMFT, with agents such as PORCN inhibitors or FZD antagonists (43).

TP53 mutations (TP53 p.R81X, TP53 p.C3X) and PIK3CA mutations (PI3KCA p.E453K, PI3KCA p.E542K) were found exclusively in cases with no objective response (SD+PD) to crizotinib. Mutations in both genes were previously reported in individual IMFT cases (13, 14). In NSCLC, TP53 mutations were found to be associated with poor response to crizotinib (44, 45), and PIK3CA mutations were reported in cases after acquiring resistance to the agent (46, 47). Interestingly, we also identified an association between PIK3CA mutation and decreased PFS, suggesting the potential predictive value of PIK3CA in IMFT treated with crizotinib. However, it should be pointed out that we only detected PIK3CA mutations in two out of 22 evaluable cases. Nevertheless, a PIK3CA inhibitor (alpelisib) developed for breast cancer could be considered as an experimental therapeutic option for PIK3CA-mutated IMFT (48, 49).

Because the amplification in the ALK gene or in ALK fusions as well as a number of mutations in the ALK tyrosine kinase domain have been identified as mechanisms of resistance to crizotinib in ALK-rearranged NSCLC (50–52), we also determined the presence of ALK-related alterations in addition to translocation in our group of IMFTs. Four alterations were identified, including an ALK gene amplification in one ALK-positive IMFT (SeqID132), chromothripsis-associated ALK fusion amplification in two ALK-positive IMFTs (SeqID001, SeqID197), as well as an ALK mutation (p.N571K) in an ALK-negative IMFT (SeqID118). In two cases of ALK-positive IMFT harboring ALK gene or fusion amplifications, no objective response but SD was observed. However, the diverging PFS (23.9 vs. 8.5 months) of these two cases makes it difficult to determine the relevance of ALK amplification. In the ALK-negative case, the ALK mutation we detected was located in the MAM2 domain, which was previously described in neuroblastoma (53). However, compared with the ALK mutations in the tyrosine kinase domain (e.g., p.F1174L) responsible for crizotinib resistance (54, 55), this specific mutation had not yet been fully described previously in terms of its functionality. In addition to ALK-related alterations, bypass signaling pathways or alternatively activating alterations have been shown to play a role in the resistance to crizotinib. We detected an amplified region of 8q24.21 (with the MYC locus) in an ALK-positive IMFT from a patient with SD. MYC amplification was recently identified as a potential mechanism of primary resistance to crizotinib in ALK-rearranged NSCLC (56). This may indicate a potential impact of this aberration on crizotinib treatment in this patient with IMFT, but additional investigation is required.

Because IMFT is so rare, the small size of our cohort and limited amount of available biological material is an inevitable challenge for this study. Nevertheless, EORTC 90101 “CREATE” probably resulted in the largest collection of archival tissue samples from this ultra-rare malignancy. This unique collection enabled us to study the disease biology and to identify other potential therapeutic targets in IMFT, in addition to ALK and ROS1. This underlines the importance of mandatory collections of biological material in trials with rare malignancies. To our knowledge, this is the first study characterizing gene alterations and associated molecular pathways in IMFT using high-throughput WGS/WES. In the absence of germline samples, we had to use stringent computational strategy, focusing on cancer-related genes, which might miss rare or uninterpreted alterations. Even with a rigorous filtering, it is still likely to have a certain percentage of germline polymorphisms included in the analysis, which may impact some results (e.g., tumor mutational loads and mutational signatures). However, this factor does not substantially influence the main observations described previously in this study. With the accumulation of genome-wide datasets over time, further exploration and the experimental validation of our findings should be considered.

In summary, our study comprehensively describes molecular alterations in IMFT, suggesting a certain extent of genomic and mutational heterogeneity in this disease. The comparison of different approaches with assess ALK status and other fusions shows the importance of using an appropriate antibody and other diagnostic tools, especially in ALK-negative cases. We have established correlations between molecular alterations and clinical outcome and identified recurrent alterations with predictive value for patients with IMFT on crizotinib treatment. On the basis of these analyses, we provide a deeper insight into the molecular profile of this ultra-rare malignancy, which may lead to novel strategies of disease management for patients with IMFT.

C.-J. Lee reports grants from Stichting tegen Kanker and Pfizer, as well as non-financial support from EORTC and BioRep during the conduct of the study. P. Schöffski reports grants from Stichting tegen Kanker and Pfizer, as well as non-financial support from EORTC and BioRep during the conduct of the study. P. Schöffski also reports personal fees from Deciphera, Boehringer Ingelheim, Exelixis, Medscape, Guided Clarity, and Ysios Capital; personal fees and other support from Blueprint Medicines, Ellipses Pharma, and Transgene; and other support from Adaptimmune, Intellisphere, and Advanced Medical outside the submitted work. J. Sufliarsky reports other support from EORTC during the conduct of the study. J. Sufliarsky also reports other support from Merck, Novartis, Amgen, and Pierre Fabre; non-financial support and other support from Roche; and non-financial support from Sweedish Orphan outside the submitted work. J.-Y. Blay reports personal fees from Roche and Bayer during the conduct of the study. A. Wozniak reports grants from Stichting tegen Kanker and Pfizer, as well as non-financial support from EORTC and BioRep during the conduct of the study. No disclosures were reported by the other authors.

C.-J. Lee: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. P. Schöffski: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing, proposed and developed the clinical trial and facilitated the translational work. E. Modave: Software, formal analysis, methodology, writing–review and editing. T. van Wezel: Software, formal analysis, methodology, writing–review and editing. B. Boeckx: Software, formal analysis, methodology, writing–review and editing. J. Sufliarsky: Resources, writing–review and editing. H. Gelderblom: Resources, writing–review and editing. J.-Y. Blay: Resources, writing–review and editing. M. Debiec-Rychter: Methodology, writing–review and editing. R. Sciot: Methodology, writing–review and editing. J.V.M.G. Bovée: Resources, writing–review and editing. D. Lambrechts: Resources, writing–review and editing. A. Wozniak: Conceptualization, resources, supervision, funding acquisition, validation, project administration, writing–review and editing.

This work was financially supported by Stichting tegen Kanker, Brussels (Belgium) and Pfizer, New York (US). The authors would like to acknowledge all physicians for their recruitment and treatment of patients with IMFT in this study, EORTC for providing material and clinical information, and BioRep, Milano (Italy) for the management of biological material. The computational resources and services used in this work were provided by the VSC (Flemish Supercomputer Center), funded by the Research Foundation - Flanders (FWO) and the Flemish Government—Department EWI. C.-J. Lee gratefully acknowledges the financial support from the KU Leuven—Taiwan scholarship program, a collaboration of KU Leuven, Leuven (Belgium) and the Taiwanese Ministry of Education, Taipei (Taiwan).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.