Mechanistic patterns and clinical implications of oncogenic tyrosine kinase fusions in human cancers

Tyrosine kinase (TK) fusions are frequently found in cancers, either as initiating events or as a mechanism of resistance to targeted therapy. Partner genes and exons in most TK fusions are typical and recurrent, but the underlying mechanisms and clinical implications of these patterns are poorly understood. Here, we investigated structures of > 8,000 kinase fusions and explore their generative mechanisms by applying newly developed experimental framework integrating high-throughput genome-wide gene fusion sequencing and clonal selection called Functionally Active Chromosomal Translocation Sequencing (FACTS). We discovered that typical oncogenic TK fusions recurrently seen in patients are selected from large pools of chromosomal rearrangements spontaneously occurring in cells based on two major determinants: active transcription of the fusion partner genes and protein stability. In contrast, atypical TK fusions that are rarely seen in patients showed reduced protein stability, decreased downstream oncogenic signaling, and were less responsive to inhibition. Consistently, patients with atypical TK fusions were associated with a reduced response to TKI therapies, as well as a shorter progression-free survival (PFS) and overall survival (OS) compared to patients with typical TK fusions. These findings highlight the principles of oncogenic TK fusion formation and their selection in cancers, with clinical implications for guiding targeted therapy.


Main
Tyrosine kinase (TK) gene fusions are common genetic alterations across the cancer types, including both hematologic and solid cancers.They are one of the earliest genomic events that initiate oncogenesis, as demonstrated in functional models 1 , as well as in cancer genome studies [2][3][4] .Furthermore, acquisition of TK fusions, such as ALK or RET fusions, have been also reported during targeted therapy in non-small cell lung cancer (NSCLC) and other tumors, as a mechanism of resistance [5][6][7][8][9] .Identi cation of the functional TK fusions is crucial in the clinic because small-molecular TK inhibitors are highly effective for patients with cancers harboring these TK fusions, often regardless of the tissue of origin [10][11][12] .
TK genes typically fuse with a partner gene that provides an active promoter for the fusion gene's expression and dimerization or oligomerization domains for TK activation through the fused TK domain 13 .Mechanistically, TK fusions are often formed by genomic rearrangements between two DNA double-strand breaks (DSBs) in introns, leading to the transcription of in-frame chimeric gene products.However, the mechanisms by which these recurrent breakpoints are selected among the large pool of potential fusion combinations remain unclear 14 .For example, in patients with NSCLC, EML4 is the most frequent partner gene of ALK fusions and CD74 for ROS1 fusions.However, EML4-ROS1 or CD74-ALK fusions have not been reported, although these fusions can theoretically be functional.Furthermore, multiple introns in ALK can potentially create in-frame ALK fusions fully preserving the kinase domain, but most ALK fusions involve breaks in intron 19, regardless of its partner genes 15,16 .The molecular basis of selecting partner genes, introns, and the clinical implications of different fusion types between the typical and atypical fusions remain unclear.
In this study, we identify mutually exclusive fusion partner selection and speci c exon usage in TK fusions based on the Catalogue of Somatic Mutations in Cancer (COSMIC) datasets.We develop a new experimental framework integrating high-throughput genome-wide gene fusion sequencing followed by clonal selection under the pharmaceutical selective pressure, which we call Functionally Active Chromosomal Translocation Sequencing (FACTS).Through this approach, we identify oncogenic TK fusions that spontaneously occur in the NSCLC cells and confer selective advantages.Furthermore, we determine the critical role of gene transcription and protein stability to explain the recurrent selection of the typical TK fusions.Finally, we highlight their clinical implications that impact the outcome of patients during TKI treatment.

Characterization of kinase fusions across cancer types
We analyzed 8,805 3' kinase gene fusions curated from the COSMIC, focusing on the seven most common kinase fusions involving ALK, RET, ROS1, NTRK1, NTRK3, ABL1, and BRAF genes found in various types of cancers (Supplementary Table 1).In this study, we used 7,751 kinase fusions where the mRNA junction positions were validated (Supplementary Table 1).
We rst analyzed the prevalence of the seven 3' kinase fusions across 16 tissue types.Except for ABL1 fusions, all kinase fusions were identi ed in multiple tissue types at variable frequencies between the tissue types (Fig. 1a and Supplementary Table 2).We observed several common patterns of kinase fusions, in terms of partner genes and intron usage.ALK fusions were predominantly observed in lung cancer and lymphoma, while RET, ABL1, and BRAF fusions have been identi ed most frequently in thyroid cancer, leukemias, and pediatric low-grade gliomas, respectively.EML4 was the most frequent partner of ALK fusions in lung cancers, while NPM was the case in lymphomas (Fig. 1b and Supplementary Table 1).The other TK fusions, including RET, ROS1, BRAF, and ABL1 fusions, showed several frequent partner genes (Fig. 1b and Supplementary Table 3).Partner genes were largely speci c to kinase genes, with several exceptions (Fig. 1c).For example, seven partner genes (9.6%), including KIF5B, TPM3, ERC1, and ETV6, were shared across the 73 kinase fusions (Fig. 1c, d and Supplementary Table 1).In contrast, most frequent partner genes, i.e., EML4, CCDC6, CD74, BCR and, KIAA1549 were exclusively associated with ALK, RET, ROS1, ABL1 and BRAF fusions, respectively (Fig. 1c, d and Supplementary Table 3).In summary, this analysis shows tissue type-and kinase gene-speci c partnering in fusion oncogene formation.
Functionally Active Chromosomal Translocation Sequencing (FACTS) identi es oncogenic ALK fusions genome-wide Genome-wide techniques identifying chromosomal translocations, such as HTGTS and TC-seq, have been widely used to study the mechanisms of translocation in normal and tumor cells by cloning chromosomal junctions generated within a few days of inducing a programmed DNA DSB at a speci c site [17][18][19] .However, not all these translocations result in functional fusion genes, and their impact on oncogenesis cannot be determined by these technologies alone.To overcome this limitation, we developed FACTS to speci cally map functional oncogenic fusions genome-widely in the setting of pharmaceutical selective pressure.Inspired by the recent reports indicating fusion oncogene formation as a mechanism of resistance to EGFR inhibitors, we applied FACTS into the in vitro model of EGFR inhibitor resistance using PC-9 cells, harboring EGFR-activating mutation (EGFR E746-A750del) 20 .

PC-9 cells have been extensively used to characterize various mechanisms of resistance to EGFR
inhibitors.Under the EGFR inhibitor treatment, PC-9 cells typically undergo cell-cycle arrest, and a small number of cells undergo persistence.Emergence of fully resistant clones require additional genetic alterations, including secondary mutations in EGFR that prevent TKI binding 21 or another oncogenic driver events that bypasses EGFR inhibition, such as MET ampli cations or the acquisition of fusion oncogenes 22 .Therefore, we reasoned that PC-9 cells under selective pressure from a selective EGFR inhibitor osimertinib 23 would be an ideal setting to test the functional outcome of various kinase fusions induced by genome-wide translocations.
ALK fusions are the most frequent TK fusions found in 3-7% of patients with NSCLC 15 and drive resistance to targeted therapy in patients with EGFR-mutant or KRAS G12C -mutant cancer [7][8][9] .Consistently, PC-9 cells expressing two sgRNAs targeting the relevant introns in EML4 (intron 6 or intron 13) and ALK (intron 19) to force the formation of typical EML4-ALK fusions generated resistant clones expressing the EML4-ALK fusions at a frequency of 1.16%-1.34%upon osimertinib selection (Extended Data Fig. 1a-d), which was consistent with the frequency of translocations induced by two DNA DSBs in previous studies 24, 25 .Furthermore, EML4-ALK fusion expressed from the endogenous EML4 promoter rapidly induced osimertinib resistance in PC-9 cells by producing active and phosphorylated EML4-ALK fusion proteins (Extended Data Fig. 1e).Next, we applied FACTS to study ALK fusion formation from a single programmed DSB with any partners in the genome under the osimertinib treatment.We induced a programmed DSB in intron 19 of ALK, the hotspot of genomic rearrangements causing EML4-ALK fusions in NSCLC 15 (Fig. 2a).Multiple osimertinib-resistant clones developed during selection, with an estimated frequency of ~ 1 clone/million cells (Fig. 2b).Analysis of single clones showed that each expressed ALK proteins with a wide range of sizes from approximately 60 kDa to 400 kDa, likely indicating formation of fusion oncoprotein in different sizes (Extended Data Fig. 1f).In contrast, when we introduced a DSB in intron 6 of EML4 gene, no ALK expression was observed in osimertinib-resistant clones (Extended Data Fig. 1g), indicating that the DSB at ALK is critical in fusion formation.The ALK fusions formed after the DSBs at ALK intron 19 showed a strong phosphorylation of the kinase domain, which resulted in sustained activation of the MAPK pathway despite the presence of osimertinib (Fig. 2c), explaining the mechanism of resistance.Both ALK and ERK1/2 phosphorylation were completely blocked by adding ALK-speci c inhibitor lorlatinib (Fig. 2c).Consistently, the growth of osimertinib-resistant clones was inhibited by lorlatinib (Fig. 2d).These data showed that a DSB in ALK led to the formation of in-frame fusions, of which expression conferred resistance to osimertinib.
Next, we identi ed unknown 5' partner genes of these ALK fusions by using a 3' end-directed fusion assay 26 .Several in-frame ALK fusions were identi ed (Fig. 2e and Supplementary Table 4), and a subset of them was further validated in single clones using RT-PCR (Extended Data Fig. 1h).Three of these fusion partners, EML4, STRN, and ATIC, are on chromosome 2, where ALK is located, whereas others were spread in the genome (Fig. 2e and Extended Data Fig. 1i).Remarkably, several of these spontaneous ALK fusions were identical to those described in NSCLC 27 or in other tumor types 4 (Supplementary Table 4).For example, EML4-ALK fusions joined e2, e6, e13, or e18 of the EML4 gene to e20 of the ALK gene (Fig. 2e and Supplementary Table 4), exactly as seen in patients with NSCLC and other tumors 28, 29 .In addition, some fusion partners showed exclusive exon usage, such as e3 in STRN and e31 in CLTC, while others showed variable usage (Fig. 2e and Supplementary Table 4).These exon usages were identical to the corresponding ALK fusions found in NSCLC and thyroid cancer (Supplementary Table 4).We also identi ed a list of new ALK fusions that have not been described in human tumors (Supplementary Table 4), which may indicate rare functional fusion events yet to be discovered.Several of them (e.g., QKI, TRAF2, TRAF3, and TP53BP1) were the genes previously reported in fusions with other kinases [30][31][32] and contain dimerization or oligomerization domains, further supporting their functionality (Extended Data Fig. 1j and Supplementary Table 4).Taken together, FACTS demonstrated functional fusion oncogene formation through genome-wide translocations after a single DSB at ALK intron 19 and reproduced ALK fusion landscape in human cancers.
To test whether oncogenic ALK fusions can also be generated in non-cancerous cells, we applied FACTS to the bronchial epithelial BEAS-2B cells.These cells can grow in vivo once they are transformed by oncogenic drivers 33 .As a positive control, we injected mice with BEAS-2B cells where two DSBs were induced in EML4 intron 6 or 13 and ALK intron 19.As expected, all mice in this group developed tumors (Extended Data Fig. 1k-m).When we injected mice with BEAS-2B where a single DSB was introduced in ALK intron 19, we observed tumor formation at a lower rate and with a slower growth kinetics (Extended Data Fig. 1l, m).FACTS and RT-PCR validation revealed that tumors expressed various ALK fusions identical to those in PC-9 cells and patient samples (Extended Data Fig. 1n and Supplementary Table 4).Protein expression of these fusions was most likely determined by the fusion partner, with some fusions being expressed at higher levels than others (Extended Data Fig. 1o, p).Thus, by applying FACTS to immortalized normal-like bronchial epithelial cells, we demonstrated that a single DSB in ALK intron 19 produced functional ALK fusion oncogenes that led to a malignant transformation in vivo.

FACTS identi es oncogenic RET, ROS1, and NTRK1 fusions
We next applied FACTS to other kinase fusions.RET, ROS1, and NTRK family gene fusions are found in approximately 4% of patients with NSCLC 34 .We designed FACTS by introducing one DSB in their intron that is most frequently involved in chromosomal translocations, (i.e., intron 11 for RET, intron 33 for ROS1, and intron 11 for NTRK1; Extended Data Fig. 2a-d).Resistant clones developed after 4 weeks of osimertinib selection at a frequency comparable to what was observed from clones with ALK fusion (Fig. 2b and Extended Data Fig. 2e).FACTS identi ed several in-frame chimeric proteins with RET, ROS1, and NTRK1 and their joined partners across the genome (Extended Data Fig. 2f-n and Supplementary Table 4).We validated some of these acquired fusions by RT-PCR and Sanger sequencing and con rmed that they were identical to the RET fusions described in patients with NSCLC (Extended Data Fig. 2o, p).Other fusions were novel and not yet described in patients (Supplementary Table 4).We con rmed that acquired RET fusions conferred resistance to osimertinib, as demonstrated by the reversal of resistance phenotype by selpercatinib (Extended Data Fig. 2q).The fusion partner genes identi ed here also contained dimerization or oligomerization domains (Extended Data Fig. 2i-k and Supplementary Table 4).Intriguingly, FACTS identi ed exon fusion variants involving different ROS1 exons (e34, e35, and e36) or NTRK1 exons (e12 and e13) but only e12 of RET (Extended Data Fig. 2f-h and Supplementary Table 4), which is consistent with what was reported from patients 27,[35][36][37] .

Gene transcription, rather than chromatin accessibility, dictates the selection of partner genes in TK fusions
Next, we investigated how ALK fusions are selected among many potential rearrangements, and which mechanistic factors dictate the choice of ALK fusion partners in the genome.Among all reported partner genes of ALK fusion from the patients with NSCLC, those identi ed by FACTS in our PC-9 model showed a signi cantly higher level of transcription (Extended Data Fig. 3a, b).In contrast, we found no difference in terms of chromatin accessibility measured by ATAC-seq or histone activation marks between the FACTS-identi ed and -unidenti ed genes (Extended Data Fig. 3c, d).Recurrent translocation partners consistently showed active chromatin marks (Extended Data Fig. 3e-j).These ndings were consistent in partner genes of RET, ROS1, and NTRK1 fusions [35][36][37] (Extended Data Fig. 3k-p).Taken together, the partner genes selected in FACTS were associated with higher level of transcription, compared to the other partner genes not identi ed by FACTS but reported in patients.
Next, we further explored whether gene transcription was su cient to induce the formation of oncogenic fusions.PC-9 cells express very low to undetectable levels of HLA-DR molecules and the invariant chain CD74 that is essential for the assembly and subcellular tra cking of the MHC class II complex 38 (Extended Data Fig. 4a-d).We hypothesized that this undetectable expression could explain why CD74 or HLA-DR fusions with kinases 39 were not identi ed by FACTS in PC-9 cells.Because expression of both CD74 and HLA-DR can be induced by the Class II transactivator (CIITA) 40 (Extended Data Fig. 4e), we asked whether induction of HLA-DR or CD74 expression by CIITA was su cient to generate fusions of HLA-DR or CD74 with ROS1.FACTS was applied to PC-9 cells expressing CIITA that showed signi cantly increased HLA-DR and CD74 mRNA and protein levels (Extended Data Fig. 4f-m).By introducing DSBs in intron 33 of ROS1 (Extended Data Fig. 5a), we identi ed genome-wide oncogenic fusions including HLA-DRB1-ROS1 fusions in which the breakpoint in the HLA-DRB1 gene was identical to that observed in patients with HLA-DRB1-MET fusion 39 (Extended Data Fig. 5b, c and Supplementary Table 5).We estimated the frequency of HLA-DRB1-ROS1 fusions at 6.7% using single clone analysis (Extended Data Fig. 5d).In contrast to HLA-DRB1-ROS1 fusions, CD74-ROS1 fusions were not detected, which suggests that induction of transcription for the CD74 gene was not su cient to trigger CD74-ROS1 translocations.However, when we simultaneously introduced DSBs in both CD74 and ROS1 genes in either control PC-9 or CIITA-expressing PC-9 cells, resistant clones rapidly emerged only in CIITA-expressing PC-9 cells (Extended Data Fig. 5e, f).While DNA junctions were detected in both cells, CD74-ROS1 fusion transcripts were detected only in CIITA-expressing PC-9 cells (Extended Data Fig. 5g-i).These results suggest that increased gene expression of the partner gene is su cient to induce the formation of TK fusions in loci such as HLA-DRB1, which is located on the same chromosome with ROS1, and point out that the detection of DNA junctions is insu cient to determine oncogenicity of resulting TK fusions without evidence of e cient transcription of the TK fusion.

Oncogenic TK fusions originate after selection of pools of rearrangements spontaneously occurring in fusion partner and TK genes
In some tumors such as lymphoma, recurrent translocations are the result of the activity of the activationinduced cytidine deaminase (AID) enzyme that targets speci c regions of the genome 41 .Therefore, we asked whether the selection of speci c partners or exons by TK fusions is mechanistically determined by the formation of DSBs in speci c positions of genes or rather by the selection of random genomic DSBs.To this end, we generated libraries of DNA junctions by HTGTS which allows for an unbiased detection of genome-wide chromosomal rearrangements before selection 17 (Extended Data Fig. 6a, b).HTGTS yielded 111,811 genomic translocation breakpoints before selection, which were distributed throughout the genome with enhanced clustering in the 2 Mbp regions surrounding the ALK DSB (Fig. 3a-c and Extended Data Fig. 6c).We identi ed 154 hotspots with signi cantly enriched breakpoint clustering (Fig. 3d and Supplementary Table 6).Only 2.6% (4/154; EML4, SQSTM1, TRAF2, and CLTC) of these hotspots occurred in genes that are known partners of ALK fusions (Fig. 3d and Supplementary Table 6).In sharp contrast, HTGTS performed with resistant clones after osimertinib selection yielded 5,005 DNA breakpoints with 81% hotspots (13/16) occurring in genes leading to the transcribed ALK fusions identi ed by FACTS (Fig. 2e and Fig. 3a and Supplementary Tables 4 and 6).Several strong genomic translocation hotspots observed before selections completely disappeared after selection (Fig. 3e-g and Extended Data Fig. 6d), most likely because the resulting rearrangements did not generate a functional ALK fusions.EML4 was the gene most frequently translocated with ALK after selection (Fig. 3h).Breakpoints before selection did not show a preferential strand bias, which is consistent with previous works of genome-wide cloning of unselected translocations 17,42 .In contrast, the breakpoints after selection showed a strong bias for an orientation of the gene leading to a functional fusion with ALK (Fig. 3i), with DNA breakpoints markedly enriched for junctions occurring in gene introns (Fig. 3j).Within individual partner genes, we observed a selective enrichment of breakpoints occurring in introns leading to in-frame functional fusions with ALK (Fig. 3k and Extended Data Fig. 6e-h).Overall, these data indicate that the formation of ALK fusion is the result of a functional selection of transcribed translocations based on the location and orientation, not just a re ection of DNA break frequency.
Next, we focused on TK genes and generated HTGTS libraries in BEAS-2B and PC-9 cells by inducing a DSB in EML4 as bait to capture breaks spontaneously occurring in TK genes (Extended Data Fig. 7a).We looked at the distribution of breakpoints in ALK, RET, ROS1, NTRK1, as well as other kinase genes known to generate oncogenic fusions in lung cancer, such as EGFR, ERBB4, MET, FGFR3, and EPHA2 43 .Breakpoints identi ed in these kinases were evenly spread throughout the gene body including introns and exons without clear clusters (Extended Data Fig. 7b-j).In both BEAS-2B and PC-9 cells, more breakpoints were observed in ALK than in other kinases (Extended Data Fig. 7k, l and Supplementary Table 7), most likely because EML4 and ALK are proximally located on the same chromosome 2 44,45 .More breakpoints in EGFR were detected in PC-9 cells than in BEAS-2B cells (Extended Data Fig. 7f, k, l), most likely due to the presence of > 4 copies of the EGFR gene in PC-9 22 , and frequent breakpoints were observed also in EPHA2 gene, which is highly transcribed in these cells (Extended Data Fig. 7j, l).All combined, these data suggest that the preferential usage of speci c partners or exons during oncogenic TK fusion formation is the result of a selection process among multiple combinations of junctions created by DSBs spontaneously generated in the genome, rather than due to the presence of pre-existing clusters of breakpoints like in the case of AID-initiated translocations.

Protein stability determines the selection of TK fusion partners
Next, we investigated the process of TK fusion selection.Consistent with the COSMIC analysis in patients, TK fusion partners obtained by FACTS were mutually exclusive in most cases 30,[40][41][42] , with only a few partners shared by multiple TK fusions (Fig. 4a and Supplementary Table 4).Interestingly, some fusion partner genes, such as TPM3 and ETV6, used the same exons when they generated oncogenic fusions with different TKs (Fig. 4b).Thus, we explored functional basis of fusion-partner speci city to each kinase.We engineered all combinations of EML4 and CD74 fusions with ALK, RET, ROS1, and NTRK1 (Extended Data Fig. 8a, b).While all of the fusion junctions were detected equally at the genomic DNA levels (Extended Data Fig. 8c,d), some of the kinase fusion combinations did not yield resistant clones under osimertinib selection (Fig. 4c,d).While thousands of EML4-ALK, EML4-RET, or EML4-NTRK1 clones rapidly emerged, no clones with EML4-ROS1 fusions were observed (Fig. 4c).Likewise, while thousands of CD74-ROS1 clones emerged, no clones with CD74-ALK, CD74-RET, or CD74-NTRK1 fusions emerged in PC9 cells expressing CIITA (Fig. 4d).Next, we isolated single cell-derived clones harboring different fusions for further characterization (Extended Data Fig. 8e).Clones with CD74-ROS1 fusion displayed abundant protein that was phosphorylated as expected, but clones with EML4-ROS1 fusion showed very low abundance of the EML4-ROS1 protein that was also poorly phosphorylated (Fig. 4e).
Treatment with proteasome inhibitor MG132 stabilized the EML4-ROS1 fusion protein and its phosphorylation substantially increased (Fig. 4e).Crizotinib, primarily a MET inhibitor with an activity on ROS1, inhibited the growth of clones harboring CD74-ROS1 fusions but not EML4-ROS1 fusions, suggesting that only stable and abundant kinase fusions could create oncogenic dependency (Fig. 4f).
Protein stability determines the speci c exon usage of oncogenic TK fusions An additional nding of COSMIC analysis was the preferential usage of speci c exons in TK fusions (Fig. 1e, f).To better understand molecular basis of preferential exon usage in kinase fusions, we engineered EML4-ALK variants by CRISPR/Cas9 that fuse the same EML4 exon 6 to different ALK exons (e18, e19, or e20) (Extended Data Fig. 8f).All these fusions are predicted to be in frame, which could potentially lead to functional ALK fusions.However, these different fusion variants have been detected in different frequencies in patients, the EML4-ALK E6;A20 fusion being far more frequent than the E6;A18 or E6;A19 fusions (less than 1% among ALK fusions) 46,47 (Fig. 1e), which was consistently observed in FACTS (Fig. 5a).To understand the cause of these differences, we generated clonal lines for each fusion variant.The mRNA transcription levels were comparable among the variants, likely due to their regulation by the same promoter (Fig. 5b and Extended Data Fig. 8g).However, the protein abundance and the level of phosphorylation were markedly different (Fig. 5c and Extended Data Fig. 8h).The E6;A20 fusion protein was highly expressed and phosphorylated, whereas the E6;A18 or E6;A19 fusions were much less abundant with barely detectable phosphorylation (Fig. 5c and Extended Data Fig. 8h).Consequently, the E6;A20 variant showed a greater potency in rescuing MAPK pathway activation compared to the E6;A18 or E6;A19 fusions in osimertinib-treated PC-9 cells (Fig. 5c).Treatment with MG132 stabilized the E6;A18 or E6;A19 fusions and led to their phosphorylation (Fig. 5d).Next, we investigated whether the different functional features of these EML4-ALK fusion variants were due to differences in subcellular localization, given recent evidence showing that the oncogenic activity of EML4-ALK is dependent on its subcellular localization and formation of protein granules in the cell cytoplasm 48 .The three EML4-ALK fusion variants showed comparable intracellular localization in confocal microscopy analysis (Extended Data Fig. 8i, j), with weaker signals with the E6:A18 or E6:A19 fusions, likely due to their low protein abundance.Functional assay showed that lorlatinib inhibited the growth of cells harboring E6;A20 fusions but not of the E6;A18 and E6;A19 fusions, suggesting that only E6;A20 fusions are stable enough to confer oncogenic dependence (Fig. 5e).These ndings imply that the usage of speci c exons in TK fusions is likely dictated by protein stability rather than transcription or subcellular localization of the resulting fusions, and that only an abundant expression of TK fusion proteins creates a dependency that might determine the e cacy of TKI treatment.

TKI therapy is less effective in patients with atypical ALK fusions
Since atypical ALK fusions showed reduced functionality and oncogenic signaling in PC-9 cell models (Fig. 5c, e), we investigated whether these ndings were re ected in patients by studying clinical responses to ALK TKIs in patients carrying either typical or atypical ALK fusions.We analyzed 108 patients with metastatic NSCLC who tested positive for ALK fusions by next-generation sequencing (NGS) and received ALK TKI treatment and divided them into two groups based on the ALK gene fusion breakpoints: typical (breakpoints in ALK intron 19) and atypical (breakpoints in other ALK introns/exons or atypical fusion partner).There were 97 typical ALK fusions with ALK breakpoints in intron 19, and 11 atypical fusions cases with breakpoints in introns 16, 17, 18, and 20 or inside exon 20 (Extended Data Fig. 9a and Supplementary Table 8).Patients with atypical ALK fusions had clinical characteristics comparable to patients with typical ALK fusions in terms of age, gender, smoking history, ECOG performance status, and ALK inhibitor treatment (Extended Data Fig. 9b).The typical ALK fusion group had 88.7% (88/97) of EML4-ALK fusions or other known oncogenic ALK fusions, such as HIP1-ALK 49 , whereas the group of atypical ALK fusions was composed of 54.5% (6/11) of EML4-ALK fusions with non-intron 19 breakpoints (Extended Data Fig. 9c) or ALK fusions with atypical partners.Strikingly, the atypical group showed signi cantly lower objective response rate (ORR) to ALK TKI compared to the group of patients with typical ALK fusions (54.5% versus 88.7%, p = 0.01) (Extended Data Fig. 9d), resulting in a signi cantly shorter progression-free survival (PFS; 5 months versus 20.5 months, HR: 0.18 [95% CI: 0.08-0.38],p < 0.001) and overall survival (OS; 20.5 months versus 83.0 months, HR: 0.20 [95%CI: 0.09-0.45],p < 0.001) (Fig. 6a,b).We also con rmed that atypical ALK fusions retained a signi cant association with shorter PFS and OS after adjusting for potential confounders in multivariable Cox regression models (Extended Data Fig. 9e).We further examined co-occurring mutations in cases with typical or atypical ALK fusions.The most frequently mutated gene was TP53 in typical and atypical ALK fusions, with a signi cantly higher frequency in atypical fusions (62.5% versus 25.9%, p = 0.046), which may have also contributed to the worse outcomes to ALK TKIs observed in this subset of patients (Fig. 6c).In addition, atypical ALK fusions were associated with a higher rate of mutations in alternative oncogenic driver genes, including BLM, FLT4, RAF1, RB1, and TCF3 51 , compared to typical ALK fusions (Fig. 6c and Extended Data Fig. 9f).Overall, these results demonstrate that atypical TK fusions are weaker oncogenic driver, are associated with increased co-mutation of other oncogenes, and respond poorly to ALK inhibition, providing a biomarker predictor for response to ALK TKI in patients.

Discussion
In this work we provide mechanistic explanation and clinical relevance of the recurrent patterns of oncogenic TK fusion in cancers.TK fusions show mutually exclusive fusion partners, with just a few partner genes shared by multiple TK genes.In addition, TK and partner genes employ a preferential usage of speci c exons.
To explain these patterns, we developed FACTS as a novel approach to identify genome-wide functional chromosomal translocations that drive solid tumor growth and resistance to TKI inhibition.Current techniques to map genome-wide chromosomal translocations are mostly focused on early, unbiased mechanistic events of DNA translocation formation largely in B lymphocytes 17,18 , without interrogating the oncogenicity of the cloned translocations.In contrast, by FACTS we found that oncogenic ALK, RET, ROS1, and NTRK1 fusions form spontaneously in normal or tumoral lung epithelial cells when a DSB is introduced in the TK gene.These spontaneous translocations occur not only in PC-9 lung cancer cells selected in vitro by the pressure of osimertinib, but also in non-tumoral BEAS-2B cells that are transformed in vivo in mice, indicating that FACTS can be applied to tumoral or non-tumoral cells subjected to different selection modalities.By extension, it is conceivable that FACTS could be applied virtually to almost any TK gene or any normal or tumoral cells for which a method of selective pressure is available.Furthermore, it is likely that FACTS application could be expanded to non-TK fusions, such as translocations recurrently found in sarcomas, hematologic malignancies, or other tumors.By applying FACTS to lung epithelial cells, we discovered key factors leading to the formation of chromosomal translocations in solid tumors.We found that mRNA expression level of the partner gene was essential to the point that reactivation of transcription, such as in the case of HLA-DRB1, was su cient to induce the spontaneous formation of functional fusions (Extended Data Figs. 4 and 5).Gene transcription was required not only to express the resulting fusion but also to increase the probability of DSBs occurring within a gene, in keeping with the knowledge that transcription levels in a gene correlate with DSB frequency 52,53 .This mechanism has implications for the understanding of the cell of origin in lung cancers driven by chromosomal translocations.For example, lung cancers with HLA-DR or CD74 fusions are most likely to originate in cells that express these genes robustly, such as alveolar type II cells 54,55 (Extended Data Fig. 10a-c).
By comparing patterns of rearrangements before selection by HTGTS to those after selection by FACTS, we further gained insights in the process of oncogenic TK fusion formation.Before selection, HTGTS detected no clusters of DSB breakpoints in TK genes (Extended Data Fig. 7b-j) and only few clusters in partner genes (Fig. 3k and Extended Data Fig. 6e-g), suggesting that most TK fusions originate by a selection of translocations that arise from spontaneous DSBs dispersed throughout the genome without pre-determined hotspots.These DSBs are likely generated by various mechanisms, including the formation of R-loops, G4 quadruplex, stalled replication forks, correlate with active transcription 56 and might be facilitated by enzymatic activity of APOBEC enzymes 57 .Thus, in solid tumors recurrent translocations might be selected by different mechanisms than in hematologic malignancies, such as Bcell lymphoma, in which recurrent translocation are largely dictated by the off-target activity of the activation-induced cytidine deaminase (AID) and the recombination activating gene (RAG)1/2 enzymes 52,58,59 .Furthermore, recurrent TK translocations do not appear to occur at a high frequency when compared to other translocations, but oncogenic translocations are heavily selected due to their potential to drive cancer cell survival and proliferation.
During the selection process, only genomic breakpoints located in introns with the correct orientation that result in fusion proteins with functional activity were enriched, whereas breakpoints leading to out-offrame proteins disappeared.Proximity of the genes also played a role because we consistently identi ed fusions with genes located on the same chromosome that constitute a topological domain with a higher probability of contact 44,45,60 .While the generation of in-frame and highly transcribed fusions with a partner that have dimerization domains are expected mechanistic factors, they are insu cient to explain the speci c exon usage or partner usage of each TK fusion as it is observed in patients.For example, EML4 is the most frequent fusion partner for ALK, but it is very rarely translocated with RET, ROS1 or NTRK1; likewise, CCDC6-RET and CD74-ROS1 fusions are frequently found in NSCLC patients, but CCDC6 fusion with ALK or ROS1 and CD74 fusion with ALK have never been described 27,[35][36][37] (Extended Data Fig. 10d-g).We discovered that protein stability is a key determinant factor that could be not predicted simply based on the characteristic of the fusion partner.By investigating EML4-ALK fusions with different ALK exons, we demonstrated that a speci c exon usage was critical to provide stability to the fusion protein (Fig. 5).We found that inhibition of the proteasome by MG-132 in the unstable EML4-ALK E6;A18 and E6;A19 variants increased protein abundance (Fig. 5d).Recent study suggested that degrons, short motifs that affect protein degradation rate (i.e.D box, KEN box, SPOP motif, and PEST sequences), could regulate the expression of fusion proteins by the degron loss mechanism 61 .PEST sequences 62 are present in the unstable EML4-ALK E6;A18 and E6;A19 variants but not in the stable oncogenic EML4-ALK E6;A20 variant, implying that degron gain in the EML4-ALK E6;A18 and E6;A19 variants might contribute to their rapid degradation through the ubiquitin-proteasome pathway.Similarly, we observed that oncogenic fusions were stable only when each TK portion was paired with a speci c partner but with other partners (Fig. 4).While the CD74-ROS1 fusion was stable and phosphorylated, the EML4-ROS1 fusion, which is not seen in patients, was unstable and poorly phosphorylated (Fig. 4e).Thus, the selection process of the speci c fusion partner for each TK depends not only on the availability of an inframe dimerization domain but also on the stability of the resulting fusion.Furthermore, the pattern of recurrent oncogenic TK fusions may be dictated by the protein stability of the resulting fusion proteins rather than the enriched DSBs in speci c locations in the cancer genome.
Accurate detection of oncogenic TK fusions is critical to effective treatment for cancer patients with TK fusions.Several methods, such as uorescence in situ hybridization (FISH), immunohistochemistry (IHC), quantitative real-time PCR (qRT-PCR), targeted DNA sequencing, and targeted RNA sequencing, are routinely employed to diagnose TK fusions.Although these techniques are su cient for detecting TK fusions, they do not provide functional evidence for their oncogenic activity.Indeed, recent studies showed that some patients do not respond to ALK TKIs despite IHC and NGS con rmation of ALK fusions 63,64 .We demonstrate here that the stability and functionality of the TK fusion proteins is a key factor for the TKI response (Figs. 4 and 5), indicating that the detection of TK fusion junctions by DNAand RNA-based sequencing does not fully predict the response to TKI therapy.We showed that patients with atypical EML4-ALK rearrangements encoding for unstable fusion proteins responded poorly to ALK TKIs and had shorter PFS and OS compared to patients with typical EML4-ALK (Fig. 6a, b and Extended Data Fig. 9d).Although these data are still limited by the small sample size, they suggest that a strong oncogenic dependency by tumor cells develops only when the TK fusion is stably expressed, and that this dependency predicts TKI response in patients.Thus, studies on novel ALK inhibitors should stratify results by TK junction as patients with atypical ALK junctions may not respond as well.Figure 6

Figure 4 The
Figure 4