Zidesamtinib

ROS1 as a ‘druggable’ receptor tyrosine kinase: lessons learned from inhibiting the ALK pathway

ROS1 is one of 58 receptor tyrosine kinases, and one of two orphan receptor tyrosine kinases where its ligand is unknown. ROS1 is evolutionarily related to ALK. ROS1 rearrangement was discovered in glioblastoma in 1987, in non-small-cell lung cancer (NSCLC) in 2007, and in cholangiocarcinoma in 2011. While the clinicopathologic characteristics of ROS1-rearranged glioblastoma and cholangiocarcinoma patients remain to be defined, the clinicopathologic characteristics of ROS1-rearranged NSCLC patients have recently been described. Although ROS1 shares only 49% amino acid sequence homology with ALK in the kinase domains, several ALK inhibitors have demonstrated in vitro inhibitory activity against ROS1. With the recent US approval of crizotinib, a multi-targeted ALK/MET kinase inhibitor, for the treatment of ALK- rearranged NSCLC, attention has turned to ROS1-rearranged tumors, especially NSCLC. The next few years should witness a rapid pace of clinical research in ROS1-rearranged tumors utilizing available ALK inhibitors.

ROS1 as an evolutionarily conserved & unique receptor tyrosine kinase

In 1982, v-ROS1 was identified as a unique oncogenic sequence in the avian sarcoma virus (UR2), a chicken retrovirus, which was different from the other transforming genes in other avian sarcoma viruses [1]. v-ROS1 was fused to the gag gene in UR2 avian sarcoma virus and encoded a chimeric protein with tyrosine kinase activity [2]. Partial sequences of c-ROS1 were reported in 1986, which showed that c-ROS1 likely encoded a transmembrane growth factor-like receptor tyro- sine kinase (RTK) [3–5] and was identical to mf3, which was reported in 1984 to be able to trans- form NIH3T3 cells to form tumors in nude mice [6]. A full-length c-ROS1 cDNA that encoded 2347 amino acids with a molecular weight of 259 kDa was isolated from a glioblastoma cell line in 1990 [7]. The first 1861 amino acids constituted the extracellular domain; the transmem- brane domain is from amino acids 1862 to 1882, and the 464 C-terminal amino acids constitute the cytoplasmic domain. The genomic structure of ROS1 was fully characterized in 2003 and was found to be composed of 44 exons spanning a region of 127 kb on chromosome 6q21 [8].

ROS1 belongs to one of 58 human RTKs and is the sole member of the ROS1 RTK fam- ily, one out of 20 RTK families identified [9]. ROS1 is evolutionarily close to the ALK family, which forms part of the scientific basis of using inhibitors of ALK as inhibitors of ROS1, and we will go into much deeper discussion later in this perspective [10]. The extracellular domain of ROS1 spans more than 1800 amino acids and is postulated to consist of three YWTD domains that fold into three -propeller domains, inter- spersed with potentially nine fibronectin type III domains (FIGURE 1) [10]. The YWTD -propeller domain has a circular folding pattern that brings neighboring protein modules into close proxim- ity [11]. Fibronectin type III domains are scaf- folds that form stable protein domains that are found in many receptors that bind to targets with high affinity [12,13]. However, despite hav- ing one of the largest extracellular domains of all human RTKs, no human ROS1 ligand has been found to date. Among human tissues, the highest level of ROS1 is expressed in lung tissue but its function in humans remains largely unknown [14]. Male ROS1-knockout mice are unable to reproduce. ROS1 controls the regionalization and terminal differentiation of the epididymal epithelium. Disruption of this process in male ROS1- knockout mice interferes with sperm maturation and its ability to fertilize in vivo but not in vitro [15].

ROS1 is evolutionarily conserved from Caenorhabditis elegans to Drosophila to humans. The Drosophila counterpart of ROS1 is sevenless. Sevenless and ROS1 share extensive homology over the cytoplasmic domain and also among stretches of amino acids in the extracellular domain [7,16]. Sevenless is a transmembrane RTK primarily expressed in R7 photoreceptor cell. Bride of sevenless (BOSS) is another seven transmembrane protein and a ligand for sevenless that is expressed on R8 photoreceptor cells. R7 and R8 photoreceptor cells are in close proximity and the binding of BOSS to sevenless activates sevenless signals mainly through the RAS pathway [17,18] and leads to the specialization of R7 to be sensitive to UV light. To date, no human homolog of BOSS has been found and the ligand for ROS1 in humans remains unknown.

The discovery of ROS1 rearrangement in solid malignancies ROS1 rearrangement in glioblastoma multiforme

The first rearrangement in ROS1 in solid tumor was discovered in 1987 [19], when Southern hybridization of genomic DNA revealed that ROS1 was altered in the U118MG glioblastoma cell line [19]. The full-length fusion partner to ROS1 in glio- blastoma was sequenced in 2001 and was termed ‘fused in glioblastoma’ (FIG) [20]. There is no FIG counterpart in yeast or Drosophila. FIG encodes a protein of 454 amino acids with a molecular weight of 49 kDa. FIG contains three potential pro- tein kinase C and seven casein kinase II phosphorylation sites. FIG also contains two coiled-coil domains (amino acids 78–131 and 144–197) and a PDZ domain (amino acids 276–363) in the C-terminal half of FIG. There is a putative leucine zipper (amino acids 167–188) within the second coiled-coil domain. FIG is localized to the Golgi apparatus and binds to syntaxin 6, a Golgi-associated protein, through the second coiled-coil domain, but is independent of the leucine residues [20]. FIG is located distally to ROS1 on chromosome 6 and both genes are transcribed in the same orientation [8]. A small, 240-kb, inter- stitial homozygous deletion on chromosome 6q21 leads to the fusion of the ubiquitously expressed FIG to the tyrosine kinase domain of ROS1 in the human glioblastoma cell line U118MG [8]. This genomic rearrangement results in an in-frame fusion of exon 7 of FIG and exon 36 of ROS1. This FIG–ROS1 transcript in GBM is encoded by seven FIG exons and nine ROS1 exons and results in a constitutively active tyrosine kinase protein. A total of 90% of FIG sequences are retained in the FIG–ROS1 fusion protein, including the two coiled-coil motifs and the PDZ domain (FIGURE 2). Although FIG–ROS1 retains the two coiled- coil domains (and the putative leucine zipper within the second coiled-coil domain), FIG–ROS1 occurs as a monomer in vivo [21]. The second coiled-coil domain of FIG mediates the targeting of FIG–ROS1 to the Golgi apparatus. Indeed, the constitutive acti- vation of FIG–ROS1 requires localization to the Golgi apparatus. The fused protein product FIG–ROS1 is a potent oncogene, and its transforming potential resides in its ability to interact with and become localized in the Golgi apparatus at the cytoplasm. Deletion of FIG sequences crucial for Golgi localization from the fusion protein eliminates the transformation capacity of FIG– ROS1 [19]. Direct injection of a conditional FIG–ROS1 transgene into the basal ganglion of adult mice resulted in the formation of malignant astrocytoma. When FIG–ROS1 was injected into mice with homozygous deletion of the AFR tumor suppressor, there was increased aggressiveness, decreased latency and increased tumor penetrance of the malignant astrocytoma formed com- pared with mice harboring wild-type ARF tumor suppressor [22]. Taken together, FIG–ROS1 is sufficient to induce high-grade astrocytoma in mice, but the process is accelerated by deletion of other important tumor suppressor genes.

ROS1 rearrangement in non-small-cell lung cancer

ROS1 rearrangement in non-small-cell lung cancer (NSCLC) was originally discovered in 2007, in one of the two seminal papers describing the discovery of ALK rearrangement in NSCLC. Rikova et al. performed a large-scale survey of tyrosine kinase activation in 41 NSCLC cell lines and approximately 150 Chinese NSCLC patients’ tumor samples [23]. In addition to identifying ALK rearrangement in NSCLC, the authors also identified ROS1 rearrangement in one NSCLC cell line (HCC78) and one NSCLC patient tumor sample. Reverse transcription-PCR (RT-PCR) and DNA sequencing identified two separate ROS1 fusion products (FIGURE 3). One of the two ROS1 fusion products in NSCLC was discovered in the HCC78 cell line. Exon 4 of SCL34A2 was fused to either exon 32 or 34 of ROS1. This resulted in a hybrid of the N-terminal region of SLC34A2 (a transmembrane solute car- rier protein), ending just after the first transmembrane region, fusing to the transmembrane region of ROS1, and producing a truncated fusion with potentially two transmembrane domains in the HCC78 cell line (FIGURE 3). SLC34A2 belongs to the family of type II sodium phosphate cotransporters. SLC34A2 is located on chromosome 4p15 and is hypothesized to span the membrane eight times. SLC34A2 is expressed in small intestine, testis, liver and secreting mammary glands [24]. Furthermore, SLC34A2 is expressed in the atypical membrane of murine type II alveolar epithelium and may be involved in the reuptake of phosphate necessary for the synthesis of surfactant [25]. Indeed, mutations in SLC34A2 are associated with pulmonary alveolar microlithi- asis [26]. The location of SLC34A2 may provide the rationale for the pathogenesis of NSCLC in patients with SLC34A2–ROS1 rearrangement. Thus, alterations in SLC34A2 seem to specifically lead to lung pathology, despite its rather ubiquitous expression in human tissues.

The second ROS1 fusion product was discovered from the tumor of a NSCLC patient. In this NSCLC patient, exon 6 of CD74 was fused to exon 34 of ROS1, resulting in the N-terminal half of CD74 (a type II transmembrane protein) fusing to the C-terminal half of ROS1, creating a fusion protein with potentially two transmembrane domains (FIGURE 3). Of note, this CD74–ROS1 NSCLC patient was a female never-smoker with adenocarcinoma [23]. The function of CD74 includes serving as a receptor for the macrophage migration inhibitory factor, a lymphokine involved in the regulation of macrophage function in host defense; and functioning as a MHC class II chaperone protein, which plays a critical role in peptide presentation to CD4-positive lymphocytes [27]. The pathogenesis of CD74–ROS1 rearrangement in NSCLC remains unknown. Finally, Rikova et al. did not find any muta- tions in the ROS1 kinase domain, indicating that the pathogenesis of NSCLC is due to ROS1 rearrangement rather than mutation in ROS1 itself [23]. In vitro and in vivo transforming activity of SCL34A2–ROS1 has been demonstrated by Gu et al. [28], while in vitro or in vivo transforming activity of CD74–ROS1 has yet to be reported.

The same group of investigators then further characterized the incidence of ROS1-rearranged NSCLC by screening 656 different formalin-fixed paraffin-embedded NSCLC tumors from China by immunohistochemistry using a sensitive ROS1 antibody, D4D6, and by break-apart FISH assay. At least seven (and potent- ially up to nine) tumors were positive for ROS1 rearrangement by break-apart FISH, indicating an incidence of approximately 1.1–1.4%. In comparison, 24 out of 656 (3.8%) tumor samples were positive for ALK rearrangement by break-apart FISH. There was no overlap between ROS1- and ALK-positive NSCLC. The clinicopathologic characteristics of the seven ROS1-rearranged NSCLC patients were not reported [29]. This report indicated that the incidence of ROS1-rearranged NSCLC is approximately a third as common as ALK-rearranged NSCLC [29].

ROS1-rearranged NSCLC was discovered at the same time as ALK-rearranged NSCLC. Because there was a clinical trial with crizotinib, an ALK/MET multi-targeted kinase inhibitor, at the time of the discovery of ALK rearrangement in NSCLC and which started to screen for and enroll patients with ALK- rearrangement NSCLC [30], the clinicopathologic characteristics of ALK-rearranged NSCLC patients were quickly defined even though ALK-rearranged NSCLC constituted only approximately 5% of all NSCLC [31,32]. Without the benefit of a concurrent ROS1 inhibitor trial, the clinicopathologic characteristics of ROS1-rearranged NSCLC patients were only recently reported [33]. Bergethon et al. screened 1073 NSCLC tumors and identified 18 (1.7%) patients with ROS1 rearrangement by break-apart FISH assay [33]. The median age of the 18 ROS1-rearranged NSCLC patients was 49.8 years (range: 32–79 years), 78% were never- smokers and all of them presented with adenocarcinoma. Out of the 18 patients, fusion partners to ROS1 were detectable in six patients (five CD74 and one SLC34A2) by RT-PCR. Eight patients had no readily identifiable ROS1 fusion partner and four patients did not have sufficient tumor tissue for RT-PCR. Unlike the slight abundance of signet-ring morphology among adenocarcinoma with ALK rearrangement [34,35], there was no particular subtype of adenocarcinoma that could be ascribed to ROS1-rearranged NSCLC at this time. More importantly, there is significant overlap in the clinicopathologic characteristics between ROS1- and ALK-rearranged NSCLC patients (young, nonsmokers and adenocarcinoma). The median age of diagno- sis between 31 ALK-rearranged NSCLC patients identified in the same screening process (median: 51.6 years) and 18 ROS1- rearranged NSCLC patients (median: 49.8 years) were similar. Of note, ROS1-rearranged NSCLC patients constituted 5.9% while ALK-rearranged NSCLC patients constituted 5.4% of the never- smokers in this screening study. In addition, Bergethon et al. dem- onstrated that crizotinib inhibited the growth of the HCC78 cell line, which harbors SLC34A2–ROS1 translocation, and inhibited the phosphorylation of ROS1 in HEK293 cells transfected with CD74–ROS1 cDNA [33]. Finally, a ROS1-rearranged NSCLC patient had a promising response to crizotinib in a clinical trial (NCT00858195) that is investigating the efficacy of crizotinib in ROS1-rearranged NSCLC patients. Recently, Li et al. demon- strated that ROS1-rearranged NSCLC constitutes approximately 1% of East Asian never-smokers with NSCLC and presented with the youngest median age of diagnosis among East Asian NSCLC patients with various driver mutations [36]. Furthermore, they also identified a second CD74–ROS1 fusion product that resulted from fusion of exon 6 of CD74 to exon 32 of ROS1, thus indicating exon 32 or exon 34 of ROS1 are common chro- mosomal breakpoints in ROS1 (FIGURE 3) [36]. In summary, while ROS1-rearranged NSCLC is uncommon, this study provided the roadmap to enrich for NSCLC patients to be screened for ROS1 rearrangement. Given that ROS1 and ALK are evolutionarily related and ROS1-rearranged NSCLC patients are very similar to ALK-rearranged NSCLC, it will be important to distinguish whether there is any difference in the triggering event that led to some patients developing ROS1-rearranged NSCLC as opposed to ALK-rearranged NSCLC.

ROS1 rearrangement in cholangiocarcinoma

The same group of investigators who discovered ROS1 rearrange- ment in NSCLC also recently reported the ROS1 rearrangement in cholangiocarcinoma using the same survey of phosphotyrosine kinase activation in cholangiocarcinoma [28]. In two out of 23 (8.7%) cholangiocarcinoma samples, ROS1 was found to be truncated while no mutations were detected in the ROS1 kinase domain. 5´ rapid amplification of cDNAs in both patient samples revealed two different forms of FIG–ROS1 transcripts. The longer FIG–ROS1(L) is the same FIG–ROS1 fusion transcript found in glioblastoma multiforme (GBM), where exon 7 of FIG was fused to exon 35 of ROS1 and contains the N-terminal 209 amino acids of FIG and C-terminal four amino acids of ROS1. The shorter FIG–ROS1(S) transcript is generated by the fusion of exon 3 of FIG to exon 36 of ROS1, which contains the N-terminal 299 amino acids of FIG and C-terminal 421 amino acids of ROS1 . The major difference between FIG–ROS1(L) and FIG–ROS1(S) is that the PDZ domain is deleted in the FIG–ROS1(S) (FIGURE 4). More importantly, the two FIG–ROS1 fusion proteins have dif- ferent subcellular localizations. FIG–ROS1(L) protein is local- ized to the Golgi apparatus as expected but FIG–ROS1(S) is scattered throughout the cytoplasm. In addition, FIG–ROS1(S) possesses four-times the kinase activity of FIG–ROS1(L) and consequently resulted in much higher anchorage-independent growth of NIH3T3 cells transfected with the FIG–ROS1(S) than FIG–ROS1(L). Whether the difference in the subcellular location plays a role in the difference in the kinase activities and transforming potential of the two FIG–ROS1 fusion proteins remains to be determined. Interestingly, in the same experiment SLC34A2–ROS1, which is rearranged in NSCLC, is localized to the para-nuclei compartment and has the level of transforming activity similar to FIG–ROS1(L). As some cholangiocarcinoma can present as an intrahepatic mass resembling hepatocellular car- cinoma (HCC), the authors did not detect FIG–ROS1 transcripts in a comprehensive survey of 60 HCC tumor samples [28].

Signaling pathways transduced by ROS1

Like all RTKs, ROS1 engages multiple signaling pathways to exert its transforming activity (FIGURE 5). Before ROS1 rearrange- ment was discovered in solid tumors, most of the studies on ROS1 were performed using a genetically engineered chimeric ROS1 protein rather than the full-length protein because its large extracellular domain is hard to express in vitro. Mutagenesis of the tyrosine amino acid residues in the cytoplasmic domain of ROS1 indicated that PLC, MAP kinase, Shc proteins, IRS-1 (a signaling molecule to PI3K), cytoskeleton and cell–cell interac- tion (1 integrin, tensin, -, - and -catenin, N-cadherin and p190Rho/GAP) proteins can all interact with ROS1 [37]. However, only phosphorylation of the cytoskeleton proteins by ROS1 mediated morphological transforming activity. It is unknown whether ROS1 directly phosphorylates these cytoskeleton and cell–cell interaction proteins [38]. Subsequently, phosphorylation of STAT3, which leads to increased DNA binding activity of STATs, was shown to be mediated by a chimeric ROS1 protein. More importantly, dominant-negative STAT3 mutants were able to block the STAT3 phosphorylation and DNA-binding activity of wild-type STAT3 and the transforming activity of the chimeric ROS1 protein [39].

More refined experiments were performed to distinguish which signaling pathway is important for ROS1 transforming activity. In an experimental system to investigate the transformation activ- ity of ROS1 using a chimeric DNA construct of EGFR–ROS1 (containing the extracellular portion of EGFR and the intracyto- plasmic portion of ROS1) transfected into NIH3T3 cells or avian sarcoma virus UR2 transformed chicken embryo fibroblasts, the most important signaling pathway for anchorage-independent growth is the PI3K/Akt pathway [40]. However, anchorage-inde- pendent growth was largely intact when MEK was inhibited by PD98059, a MEK inhibitor, indicating that the RAS/MAPK/ MEK pathway was not involved in the ROS1 transformation in NIH3T3 cells [40]. Rapamycin, an mTOR inhibitor, inhibited the anchorage-independent growth, but not to the extent of the inhibition by LY294002, indicating that the mTOR pathway may only be partially engaged by the PI3K [40]. Indeed, it has subsequently been demonstrated that phosphorylation of cyclin A-associated Cdk2 kinase was markedly decreased by LY294002. Addition of Cdk2 inhibitors, roscovitine and olomoucine, resulted in dose-dependent loss of anchorage growth in avian sarcoma virus UR2 transfected chicken embryo fibroblasts [41]. Thus it seems likely that PI3K directly phosphorylates Cdk2, which is responsible for the bulk of ROS1 transforming activity.

Two tyrosine residues that are autophosphorylated in the C-terminal region of FIG–ROS1 (Y805 and Y865, which cor- respond to Y2274 and Y2334 in the full-length ROS1) are key to ROS1’s capacity to transform cells. Mutating both tyrosine residues to phenylalanine eliminated the transforming activity of FIG–ROS1 in vitro [21]. Protein tyrosine phosphatase SHP-2 binds to FIG–ROS1 in a phosphotyrosine-dependent manner via its SH2 domain and allows for its oncogenic potential. Inhibition of SHP-2 by calpeptin in a FIG–ROS1-dependent astrocytoma cell line led to marked reduction of phospho-Akt and phospho-S6 indicat- ing the PI3K/Akt/mTOR pathway was engaged by FIG–ROS1 [21].

Furthermore, rapamycin was able to inhibit a FIG–ROS1- transformed malignant astrocytoma cell line in a dose-dependent fashion [21]. Again, in FIG–ROS1-dependent astrocytoma cells, the RAS/MAPK/MEK pathway does not seem to be involved. This is consistent with the fact that the RAS pathway is not significantly involved in ROS1’s transforming activity, although the mTOR pathway was engaged by PI3K in this case. ROS1 has also been found to colocalize with SHP-1, a protein tyrosine phosphatase (which also has an SH2 domain), in epididymal epithelium of mice [42]. SHP-1 can bind to ROS1 via phosphotyrosine 2267 residue in a yeast two-hybrid system [42]. Finally, recent experiments involv- ing BaF3 cells transfected with both the short and long versions of FIG–ROS1, TAE684, a selective ALK inhibitor and a potential ROS1 inhibitor, resulted in a dose-dependent growth of BaF3 cells and led to a decrease in phosphorylation of STAT3, Akt, ERK and SHP-2. This result indicates that the JAK/STAT, PI3K/Akt and RAS/MAPK pathways were all inhibited when ROS1 was directly inhibited [28]. The authors could not dissect which signaling path- way is important in FIG–ROS1 signaling in cholangiocarcinoma; any ALK rearrangement [23]. This observa- tion provided the first evidence that an ALK inhibitor can also act as a ROS1 inhibitor. Subsequently, the transforming activity of the two FIG–ROS1 fusion transcripts found in cholangiocarcinoma was again success- fully inhibited by TAE684 in vitro [28]. Crizotinib, a ALK/MET inhibitor, has also been shown to inhibit growth of the ROS1- dependent HCC78 cell line and phosphor- ylation of ROS1 when FIG–ROS1 was trans- fected into HEK293 cells [33]. Furthermore, using the HotSpot kinase Ki-determination studies, a second-generation ALK inhibitor, AP26113, revealed an IC50 of 0.69 nM against ALK and IC50 of 1.9 nM against ROS1 [47]. Finally, another selective ALK inhibitor, WZ-5–126, has also been demonstrated to harbor ROS1 inhibitory activity with an IC50 of 3.4 nM for ALK and IC50 of 8.2 nM for ROS1 [45]. Thus, there is a growing body of evidence that at least four ALK inhibitors can also act as ROS1 inhibitors.one of the limitations could be that TAE684 may inhibit other signaling pathways in addition to ROS1.

Figure 5. Signaling pathways transduced by ROS1.

ALK inhibitors as ROS1 inhibitors In vitro inhibition of ROS1 by ALK inhibitors

Despite sharing approximately 49% amino acid sequence homol- ogy in the kinase domain between ALK and ROS1 (FIGURE 6), sev- eral ALK inhibitors have been shown to inhibit ROS1. In com- parison, the amino acid sequence homology between ALK and MET kinase domains is only 36% but crizotinib, a multi-targeted kinase inhibitor has proven to be a potent inhibitor of both ALK and MET RTKs [43,44].

The first indication that ALK inhibitors could also inhibit ROS1 came from the observation of McDermott et al. [45]. Using an automatic platform to examine the molecular basis of drug sens- itivity, McDermott et al. examined the sensitivity of 602 estab- lished cancer cell lines [45] derived from a variety of tumor types to TAE684, a potent and selective ALK inhibitor [46]. Ten cell lines derived from either NSCLC, neuroblastoma or anaplastic large cell lymphoma exhibited marked sensitivity to TAE684. One of the ten cell lines was HCC78, a NSCLC cell line, which subsequently was shown to harbor SLC34A2–ROS1.

To investigate the amino acid interactions between an ALK inhibitor and the ALK or ROS1 kinase domains, an in silico method was used. Using the Yet Another Scientific Artificial Reality Application (YASARA) molecular modeling program, we mod- eled the drug–amino acid interactions between TAE684 and the ROS1 kinase domain using a resolved x-ray structure of TAE684 complexed in the ALK kinase domain as a reference. In both the ROS1 and ALK kinase domains, we observe a high number of conserved residues in the sites that interact with TAE684. Our analysis indicated there are potentially 22 amino acid residues in the ALK kinase domain interacting within 5 Å with TAE684, while 26 amino acid residues from the ROS1 kinase domain are potentially interacting with TAE684. We performed a similar amino acid interaction simulation with crizotinib in both the ALK (FIGURE 7A & 7B) and the ROS1 (FIGURE 7C & 7D) kinase domains. We found there are potentially 20 amino acid residues in the ALK kinase domain that can interact with crizotinib, including the gatekeeper amino acid Leu1196 (FIGURE 7B) [48], while 18 amino acid residues within the ROS1 kinase domain can potentially interact with crizotinib (FIGURE 7D). Crizotinib fits well in the kinase pocket of both domains, which further supports the observation that ALK inhibitors can act as ROS1 inhibitors. However, it is important to note that so far ALK inhibitors have lower IC50 (better inhibitory activity against) for ALK than ROS1 in vitro; they are still bet- ter ALK inhibitors. It remains to be seen if ALK inhibitors, especially crizotinib, can achieve significant clinical efficacy in patients with ROS1-rearranged tumors. ROS1 inhibitors are currently being devel- oped but the IC50 is in the range of 200 nMs which is unlikely to be specific enough to be clinically useful [49]. It will be interest- ing to see if a selective ROS1 inhibitor can reciprocally function as an ALK inhibitor.

Ongoing trials of crizotinib in ROS1-rearranged tumors

Given the in vitro data from TAE684 and internal data from Pfizer, the original Phase I crizotinib trial (NCT00585195) has been amended since May 2010 to include patients with solid tumors that harbor ROS1 rear- rangement. Detection of ROS1 rearrange- ment will be determined primarily using the ROS1 break-apart FISH assay [33], similar to the ALK break-apart FISH assay. So far the known fusion partners for ROS1 in NSCLC are on different chromosomes so the break- apart FISH should be easily applicable. For FIG–ROS1 in cholangiocarcinoma and GBM, which is generated from a small intra-chromosomal deletion event, the break-apart FISH assay will only detect the 3´ ROS1 signal since the 5´ ROS1 signal will be lost [30] and should be augmented by RT-PCR if additional tissue is available.

Differences between ROS1-rearranged NSCLC & ALK-rearranged NSCLC and implications for treatment strategies & potential resistance mechanisms

The EML4–ALK fusion protein in ALK-rearranged NSCLC is located in the cytoplasm, while SLC34A2–ROS1 and CD74–ROS1 fusion proteins in ROS1-rearranged NSCLC are predicted to be transmembrane. The difference in the subcellular locations between FIG–ROS1(S) and FIG–ROS1(L) in ROS1- rearranged cholangiocarcinoma may potentially account partially for the difference in the potency of the kinase activities of the two FIG–ROS1 fusion proteins. It is now well known that hsp90 inhibitors have clinical activity in ALK-rearranged NSCLC [50,51]. The scientific basis for the observed clinical activity is that the heat-shock proteins act as chaperone proteins for the aberrantly expressed EML4–ALK protein in the cytoplasm by shepherd- ing EML4–ALK to its substrates and protecting it from rapid degradation. Inhibiting hsp90 with IPI-504, a hsp90 inhibitor, in cell-based assays revealed that EML4–ALK was rapidly degraded within 3 h, while other transmembrane RTKs such as EGFR and HER2 were degraded much slower over a period of 24 h [52]. Hsp90 inhibitors are being developed to overcome resistance to crizotinib. Hsp90 inhibitors may have a more significant clini- cal role in ROS1-rearranged tumors such as cholangiocarcinoma and GBM, where the FIG–ROS1 fusion proteins are postulated to be cytoplasmic where they are presumably to be chaperoned by the heat-shock proteins than in ROS1-dependent NSCLC, where the SLC34A2–ROS1 and CD74–ROS1 fusion proteins are postulated to be transmembrane and less likely to be dependent on heat-shock proteins for function.

Expert commentary

ROS1-rearranged NSCLC was discovered simultaneously with ALK-rearranged NSCLC in 2007 [23]. It took an unprecedented rapid 4 years from the publication of EML4–ALK rearrangement in NSCLC in August 2007 to the conditional approval of crizotinib for the treatment of ALK-rearranged NSCLC in August 2011 in the USA. The rapid development of crizotinib as an ALK inhibitor led to the rapid characterization of ALK-rearranged NSCLC patients and the recognition that ALK-rearranged NSCLC is a unique and distinct molecular subset of NSCLC. On the other hand, without the benefit of the concurrent clinical development of a ROS1 inhibi- tor, the characterization of ROS1-rearranged NSCLC patients was delayed until recently and even then was mainly spurred by the availability of a clinical trial investigating crizotinib as a ROS1 inhibitor. The surprise finding is the very similar clinicopathologic characteristics between ROS1-rearranged and ALK-rearranged
activity of crizotinib in ROS1-rearranged tumors (NCT00858195). Given the clinicopathologic characteristics of ROS1-rearranged NSCLC patients show striking similarities to ALK-rearranged NSCLC patients, following the roadmap that led to the successful development of crizotinib in ALK-rearranged NSCLC, the potential efficacy of crizotinib in ROS1-rearranged NSCLC patients will be answered soon.

NSCLC patients (young and similar in median age of diagnosis, never-smoker with adenocarcinoma) and that ROS1 and ALK are evolutionarily related. Despite sharing only approximately 49% amino acid sequence homology in the kinase domains between ALK and ROS1, there is now growing evidence that at least several ALK inhibitors are also potent ROS1 inhibitors in vitro. The origi- nal Phase I clinical trial of crizotinib is now investigating the clinical.

Five-year view

We are hopeful that crizotinib will be able to demonstrate its efficacy in ROS1-rearranged NSCLC within the next 2 years, thus establish- ing crizotinib as a bona fide ROS1 inhibitor. Given the conditional approval of crizotinib for ALK-rearranged NSCLC in the USA in 2011 and likely approval in other regions such as Japan, Korea and the EU, there will be a broad and concerted effort to investigate crizotinib as a ROS1 inhibitor in NSCLC, cholangiocarcinoma and GBM. As part of this larger effort, the clinicopathologic character- istics of ROS1-rearranged cholangiocarcinoma and GBM patients will eventually be defined and may unravel the pathogenesis of these tumor types.

Furthermore, if crizotinib is indeed a ROS1 inhibitor then the current second-generation ALK inhibitors will likely also be investigated as ROS1 inhibitors. It is likely that gatekeeper muta- tions in ROS1 will develop after prolonged inhibition by crizotinib and that second-generation ROS1 inhibitors (or ALK inhibitors) will be needed to overcome this resistance.

One of the major potential differences between ALK-rearranged and ROS1-rearranged NSCLC is that EML4–ALK is localized to the cytoplasm while SCL34A2–ROS1 and CD74–ROS1 are predicted to be transmembrane. Subcellular localization may modulate the kinase activity of the fusion kinase and its trans- forming ability and result in differences in the clinicopathologic presentations and natural history of the tumor. This has been partially shown by the difference in kinase and transforming activities between FIG–ROS1(S) and FIG–FOS(L), which may be partly attributed to the difference in subcellular localization. We should know more about this potential mechanism in the next 5 years, as heat-shock protein inhibitors may be very useful in overcoming resistance in ROS1 fusion proteins located in the cytoplasm but less effective in overcoming resistance in fusion ROS1 kinase proteins that are membrane anchored. Within the next 5 years we should witness tremendous clinical success in the identification and treatment of ROS1-rearranged tumors, a group of tumors driven by a currently relatively obscure and under- studied RTK, heralded by the development of ALK inhibitors.

Key issues

• ROS1 is one of 58 human receptor tyrosine kinases (RTKs) and the sole family member of the ROS1 family of RTKs, one of 20 RTK families.
• ROS1 rearrangement has been discovered in glioblastoma multiforme, non-small-cell lung cancer (NSCLC) and cholangiocarcinoma. However, fusion partners to ROS1 are different in NSCLC (SLC34A2–ROS1, SDC4–ROS1, TPM3–ROS1, EZR–ROS, LRIG–ROS1 and CD74–ROS1) from in cholangiocarcinoma and glioblastoma (fused in glioblastoma–ROS1).
• SLC34A2–ROS1 and CD74–ROS1 are putative transmembrane proteins, while fused in glioblastoma–ROS1 is located either in the cytoplasm or the Golgi apparatus. Differences in the subcellular location may reflect potential differences in the natural history of ROS1-rearranged tumors.
• Clinicopathologic characteristics of ROS1-rearranged NSCLC patients have been described recently (younger, nonsmoker, adenocarcinoma, with wild-type EGFR and ALK) and are very similar to ALK-rearranged NSCLC patients, which may reflect the evolutionary relatedness of these two RTKs.
• Clinicopathologic characteristics of ROS1-rearranged cholangiocarcinoma and glioblastoma patients remain undefined.
• Although there is only 49% amino acid sequence homology in the kinase domains between ALK and ROS1, several ALK inhibitors have shown in vitro inhibitory activity against ROS1.
• In silico methods have been used to provide a compelling insight into the drug–amino acid interactions between ALK inhibitors and the kinase pockets in both ALK and ROS1 kinase domains.
• A clinical trial investigating the efficacy of an ALK inhibitor, crizotinib, is ongoing against ROS1-rearranged tumors (NCT00858195).
• It will be interesting to see in the future if an inhibitor specifically developed against ROS1 will have reciprocal inhibitory activity against ALK.

Notes added postwriting

Takeuchi et al. have recently identified four more novel ROS1 fusions in NSCLC including TPM3 –, SDC4–, ERZ– and LRIG3–ROS1 [53]. The incidence of ROS1 rearranged NSCLC is about 1.2% among all adenocarcinoma screened. The character- istics of these 13 ROS1-rearranged NSCLC patients were similar to those 18 described by Bergethon et al. [33].Zidesamtinib The known fusion partners to ROS1 in NSCLC are listed in TABLE 1.