Newly Identified Members of FGFR1 Splice Variants Engage in Cross-talk with AXL/AKT Axis in Salivary Adenoid Cystic Carcinoma

Adenoid cystic carcinoma (ACC) of the salivary gland is an infrequent form of head and neck cancer. In the United States, about 1,200 new cases of ACC are diagnosed annually (1). Historically, ACC is viewed as a low-grade malignancy that follows a gradual and indolent course of disease progression.

Complete surgical resection of the tumor remains the standard curative action for patients with ACC (2). Unfortunately, recurrence of ACC is not uncommon, and about 50% of patients develop metastatic disease to lungs, bone, and liver (3, 4). Conventional chemotherapy has limited efficacy, and there is currently no FDA-approved agent for these patients (2, 5, 6). Additional studies to identify and understand the molecular alterations that frequently occur in ACC are vital to advance the development of effective treatment options.

Dysregulation of gene function through aberrant splicing plays a critical role in the development and progression of many cancers (7, 8). The loss of normal gene function due to splicing can also contribute to conferring resistance to cancer therapy (9). Recent advances in RNA sequencing (RNA-seq) technology and analysis have enhanced the ability to profile splice variants expressed in various cell types. For example, reads in RNA-seq detect junctions spanning between exons or exon to intron, and are helpful in the prediction and identification of specific gene variants (10, 11). Complementing these data with splicing-aware bioinformatics analysis can identify gene isoforms produced by alternative splicing (12–16).

We recently developed and applied two junction-based analysis algorithms to compare splice variant expression in tumors relative to normal tissue in head and neck squamous cell carcinomas (17, 18). In this study, we combine these approaches to identify alternative splicing events (ASE) specific to ACC using RNA-seq data from 18 primary ACC tumors and 14 matched normal salivary gland tissues. This analysis revealed the presence of several unique ASEs, including FGFR1 in ACC. Expression and biochemical analysis demonstrate that the novel class of FGFR1 variants mediate FGF/FGFR1-independent function through the AXL/AKT signaling axis and suggest that this process may be involved in mechanisms promoting resistance to FGFR1 inhibition in ACC.

Primary tumors and matched normal salivary gland tissues were obtained from the Johns Hopkins Hospital (Baltimore, MD) and the Salivary Gland Tumor Biorepository, under an approved Institutional Review Board protocol at Johns Hopkins University (Baltimore, MD). These samples has been described and utilized in the previous whole-genome sequencing report (19).

RNA purification and strand-specific sequencing libraries preparation has been reported previously (19). RNA-seq reads were aligned with TopHat to obtain junction expression. Gene expression counts were further obtained with RSEM and upper quantile normalization as described previously (20). Splice variants were prioritized from a combination of outlier statistics (17) and Splice Expression Variation Analysis (SEVA; ref. 18).

For the outlier-based algorithm, we first filtered to junctions that are putative ASEs, have outliers in more than one sample, not located on sex chromosomes, have a biologically significant fold change, and are associated with genes expressed in at least 50% of samples in either class. These filtering criteria reduced the RNA-seq data from 832,317 junctions identified with TopHat to 24,547 candidate junctions in 5,467 genes for statistical analysis. Fisher exact test compared the number of outliers in the ratio of reads in a candidate ASE junction with total gene expression in tumor samples relative to normal samples. Junctions with Benjamini–Hochberg adjusted P values below 0.05 are called statistically significant. The comparison with normal samples enables the outlier statistic to retain only those junctions for candidate ASEs that are tumor specific.

To further discriminate tumor-specific junctions, multivariate statistic from SEVA was applied to model dysregulated junction usage in distinct isoforms expressed in tumor and normal populations. To limit the number of candidate junctions, a more stringent filtering criterion was applied to remove junctions that were expressed in less than in five samples from analysis. We also filtered a junction if the maximum of its expression is less than 10% of the average gene expression in log scale. Finally, we filtered genes with fewer than three junctions. This filtering criterion left 251,676 junctions corresponding to 12,775 genes for analysis. We then applied SEVA to compare the distribution of expression in junctions within a gene that also have overlapping coordinates in the genome using the GSReg R/Bioconductor package (20). Genes with Bonferroni-adjusted P values below 0.01 are called significantly differentially spliced.

MDA-ACC-01 (ACC-01) cell line derived from salivary gland ACC was obtained from Adel El-Naggar (UT MD Anderson Cancer Center, Houston, TX; ref. 21). Cells were cultured in DMEM containing 10% FBS and 1% penicillin-streptomycin. Wherever indicated, cells were stimulated with basic fibroblast growth factor and heparin (STEMCELL). UM-HACC-2A (HACC-2A) cells obtained from Jacques Nor (University of Michigan, Ann Arbor, MI) were cultured on 0.1% gelatin-coated plates in optimal salivary gland media as described previously (22). Both cell lines were short tandem repeat profiled and authenticated at the Bioscience Facility, University of California, Berkeley, CA.

ACC patient-derived xenografts (ACCX) were provided by C.A. Moskaluk (University of Virginia, Charlottesville, VA). ACCX2, ACCX6, ACCX5M1, ACCX9, ACCX11, ACCX20M1/2, ACCX38M, and ACC-01-XP were passaged in NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ mice. The establishment of ACC-01-XP tumors, originating from ACC-01 cell line is detailed in Supplementary Data S1A. ACCX6 and ACC-01-XP second passage tumors were subcutaneously implanted on to the flank of 4–6 week-old female mice. For drug treatment, dovitinib (40 mg/kg), BGB324 (25 mg/kg) or in combination were administered daily by oral gavage for 7–14 days. The dosing for these drugs was based on previously described studies by Dey and colleagues and McDaniel and colleagues (23, 24). Tumors were measured with a caliper —three to four times per week and average relative tumor volume was calculated with the equation, (width² × length)/2. All animals were maintained in animal facilities as approved by the UCSF Animal Care and Use Committee.

Total RNA was extracted from cells and ACCXs using RNeasy Kit (Qiagen). Half a microgram of RNA was used for reverse transcription into cDNA using Superscript First-Strand Synthesis Kit (Invitrogen). For PCR analysis, cDNA and Phusion High-Fidelity DNA Polymerase (Invitrogen) were used with the various FGFR1 splice variant (FGFR1v) primers (Supplementary Data S1B).

Reverse-transcribed cDNA from three representative ACCXs (ACCX6, ACCX5M1, and ACCX20M2) were used for cloning analysis. First, PCR-amplified doublet bands (1.0 and 1.2 kb) using orf-F/orf-R primer sets were gel extracted as single samples, because the two bands ran close to each other. Purified PCR products were then used as templates for second-round PCR with attFGFR1v primers (Supplementary Data S1B) designed for Gateway cloning pDonor vector. Following transformation into One-Shot TOP10 competent E. coli (Invitrogen), individual positive pEntry clones (pDonor with inserted PCR fragment), confirmed by PCR colony screening, were identified as small (1.0 kb) and large (1.2 kb) pEntry clones. Several small and large pEntry clones from each ACCX were then analyzed by DNA sequencing.

For FGFR1v expression constructs, pEntry clones with respective FGFR1v inserts were recombined into the doxycycline-inducible pLVX-N-FLAG-hygro vector using Gateway cloning system (Gateway BP/LR Clonase II, Invitrogen). Identity of the various constructs were further reverified by sequencing. Gateway cloning vectors were obtained from Nevan Krogan (Gladstone Institute, UCSF, San Francisco, CA). All DNA sequencing were performed at Quintara Biosciences. FGFR1v constructs and viral element vectors were transfected into HEK293T cells to produce lentiviral particles. Following viral infection, ACC stable cell lines were generated after selection in hygromycin (200 μg/mL). To express the FGFR1v constructs, stable ACC-01 and HACC-2A cells were cultured with doxycycline (0.5 μg/mL).

The antibodies used in this study are detailed in Supplementary Data S1B. Dovitinib, futibatinib, and BGB324 were purchased from Selleckchem; erlotinib (LC labs); Hygromycin B (Invitrogen); Human phospho-RTK Array Kit (R&D system); Human EGF, doxycycline hyclate; and cycloheximide were from Sigma-Aldrich.

Cell lysates prepared in RIPA extraction buffer (25 mmol/L Tris, pH 7.4, 150 mmol/L NaCl, 1% Triton-X-100, 0.25% sodium deoxycholate, 0.1% SDS, 1 mmol/L sodium pyrophosphate, 2 mmol/L sodium orthovanate, and 10 mmol/L sodium fluoride sodium with freshly added protease inhibitor cocktails) were analyzed by Western blotting as described previously (25). For cell-surface biotin labeling, cells were incubated with 1 mmol/L biotin (EZ-Link Sulfo-NHS-Biotin, Pierce Biotechnology) for 30 minutes on ice. Clarified protein extracts were then immunoprecipitated with either streptavidin agarose-beads (Thermo Fisher Scientific) or anti-FLAG antibody, followed by Western blotting analysis.

To determine cytotoxicity and cell proliferation, we used the Cell Counting Kit-8 (Dojindo Molecular Technologies). Briefly, cells (1–4 × 10³ cells/well) seeded on 96-well microplates ± doxycycline were cultured overnight and then treated with various inhibitors. At different time durations, cells were treated with the WST-8 reagent and absorbance at 450 nm was measured per the manufacturer's manual.

Custom Stealth RNAi siRNA targeting three different sites on the intronic region of FGFR1v and negative control siRNA were obtained from Invitrogen (Supplementary Data S1B). The ON-TARGETplus SMARTpool siRNA for AXL, as well as the nontargeting control siRNA were purchased from Dharmacon. All siRNAs were transfected using Lipofectamine RNAiMAX (Invitrogen).

Data from National Organization for Rare Disorders, Adenoid Cystic Carcinoma 2020 were accessed through https://rarediseases.org/rare-diseases/adenoid-cystic-carcinoma/.

Data reported are representatives from replicate independent experiments wherever indicated with values shown as mean ± SD. For IC₅₀ and statistical calculations, GraphPad Prism v7 was used. Unpaired t tests between groups with P < 0.05 were considered statistically significant.

To identify ASEs in ACC, we subjected RNA-seq output from 18 primary ACC tumor and 14 matched normal salivary gland (NSG) samples through outlier statistics (17) and SEVA (18). Initially, we obtained 283 and 546 putative junctions by applying outlier and SEVA methods, respectively. After normalization of junctions based on gene expression using RSEM calculations and combined filtering criteria that passed the adjusted P value threshold, we identified 138 ASEs uniquely expressed in ACC. These 138 ASEs consisted of insertion/deletions representing 58 different genes (Fig. 1A; Supplementary Data S2).

Pathway enrichment analysis of these 58 genes based on KEGG and WikiPathways database in Enrichr (26) revealed important pathways in the top three modules including, PI3K-AKT signaling pathway (Supplementary Fig. S1). The genes featured in the PI3K-AKT pathway included CDK6, IGFR1, COL9A1, ITGA9, YWHAZ, and FGFR1. In this article, we focused on the FGFR1v, as the dysregulation of this receptor family is implicated in a variety of cancer types (27) and is considered an important therapeutic target including in ACC (28, 29). FGFR1 junction expression located at Chr8:38277253-38277775 was positive in 61% (n = 11) of patient with ACC but not in NSG samples (Fig. 1B, left). The intensity of FGFR1 junction expression, which represents an intronic insertion, varied across the ACC samples. The overall expression of wild-type FGFR1 gene was, however, comparable in both ACCs and NSGs (Fig. 1B, right).

On the basis of the RNA-seq junction read, the putative FGFR1v contains an intron segment between exon 8 and exon 9 of FGFR1 (Fig. 2A). Accordingly, we used a forward primer from exon 8 (e8F) and a reverse primer (iR2) from the intron region (Fig. 2A) to validate FGFR1v expression. Results show the amplification of a predicted 1.3 kb band in all the ACCXs examined, albeit at variable expression levels (Fig. 2B). However, no PCR band was evident from the NSG samples. As in ACCXs, FGFR1v was also expressed in the ACC-01 cells (Fig. 2C). To demonstrate that the PCR product was a sequence-specific reaction by iR2 primer, we used another iR1 reverse primer (Fig. 2A). Results show that the e8F/iR1 primer set amplified the predicted 850 bp band (Supplementary Fig. S2A), thus confirming the presence of FGFR1v resulting from intron retention.

To assess the functional relevance of FGFR1v, we generated three siRNAs targeting different regions in the intronic locus of the splice variant (Fig. 2A). Initially, siFGFRv-transfected ACC-01 cells were assessed by RT-PCR analysis to verify the knockdown efficiency. Result shows that siFGFRv-i498 produced greater FGFR1v knockdown effect compared with siFGFRv-i446 and siFGFRv-i543 (Fig. 2D). Parallel immunoblotting analysis show that these siRNAs did not affect the overall expression of FGFR1^wt protein (Supplementary Fig. S2B). Notably, our result shows that FGFR1v silencing decreased cell growth (Fig. 2E; Supplementary Fig. S2C). The siFGFRv-i498 exhibited the highest growth inhibitory effect as compared with the other siRNAs, correlating with the higher FGFR1v knockdown by siFGFRv-i498 (Fig. 2D). These results indicate that the novel FGFR1v may be involved in cell growth regulation in ACC.

FGFR1v sequence harbors a termination codon at nucleotide 147 of intronic segment, thereby producing a putative truncated FGFR1 (Fig. 2A). Thus, the full-length FGFR1v was PCR amplified using orf-F/orf-R primers (Fig. 2A). Interestingly, the PCR generated two major bands (∼1.2 and 1.0 kb) in ACC-01 and ACCXs examined but not in the NSGs (Fig. 2F). FGFR1v expression was also evaluated in a panel of cell lines and patient-derived xenografts of head and neck squamous cell carcinomas (HNSCC; ref. 30). However, no PCR products was detected (Fig. 2G and H), suggesting that FGFR1v expression may be unique to ACC.

Next, the two PCR bands from ACCXs were extracted, subcloned, and analyzed by DNA sequencing. Interestingly, sequence analysis revealed that the approximately 1.2 kb PCR band consisted of 1,230 and 1,224 bp isoforms, while the 1.0 kb band represented a 957 bp isoform. Alignment of each sequence with the human FGFR1 transcript databases at NCBI established that the 1,230 bp isoform overlapped with extracellular domain (ECD) of FGFR1 (NM_023110.3), while the 1,224 bp product corresponded to the ECD of FGFR1 (NM_015850.4) that lacks six in-frame base pairs at exon 4, and the shorter 957 bp isoform corresponded to the ECD of FGFR1 (NM_023106.2), which lacks alternate in-frame exon 3 and six nucleotides on exon 4. All three variants contained the intron segment, which was identical to the intron section present in a soluble FGFR1 (31). On the basis of the number of encoding amino acids, the individual FGFR1v hereafter are also referred to as FGFR1-v409, FGFR1-v407, and FGFR1-v318 (Fig. 2I; Supplementary Data S3). Together, these findings not only confirmed unique FGFR1v expression, but also revealed three previously unreported, cytoplasmic domain-lacking FGFR1 isoforms in ACC.

To further study FGFR1v function, we generated an inducible system to individually overexpress the three FGFR1v in ACC-01 (Fig. 3A). Western blotting analysis using different antibodies demonstrated and verified the efficient FGFR1v expression (Fig. 3B). While anti-FLAG detected all three constructs, FGFR1-Ab1 antibody, which reacts with an epitope spanning exon 2 and 3 of FGFR1 (Fig. 3A), did not detect FGFR1-v318 as expected. Predictably, FGFR1-Ab1 antibody detected both FGFR1-v407 and FGFR1-v409 constructs, shown as approximately 70 kDa protein. We were unable to assess the status of endogenous FGFR1-v318 type with available FGFR1 antibodies. ACC-01 cells also expressed at least three larger FGFR1 isoforms, as detected by FGFR1-Ab2 antibody that specifically recognizes cytoplasmic domain of the receptor (Fig. 3A) and designated here as wild-type FGFR1 (FGFR1^wt) to distinguish them from FGFR1v. In the parental and FGFR-v318 expressing ACC-01 cells, FGFR1-Ab1 antibody also detected a band size similar to the two constructs, which likely represented the endogenous FGFR1v (FGFR1-v407 or FGFR1-v409 or both). However, FGFR1v level was substantially lower than that of FGFR1^wt (Fig. 3C).

To assess whether the three FGFR1v are soluble and secreted, we subjected culture media collected from doxycycline-induced FGFR1v-expressing cells to immunoblot analysis. However, we did not detect the presence of FGFR1v. Therefore, we performed cell-surface biotin labeling assay to determine whether the FGFR1v were membrane bound proteins. Results show that all three FGFR1v were readily accessible to the biotin labeling (Fig. 3D and E), thus indicating that FGFR1v are likely expressed as cell surface proteins. Furthermore, we evaluated FGFR1v protein expression in a panel of ACCXs. Consistent with the PCR analysis, the result shows that all ACCXs examined expressed the 70 kDa FGFR1v in addition to the three larger isoforms representing FGFR1^wt (Fig. 3F). However, the ratio of FGFR1v expression was at much lower level than the FGFR1^wt (Fig. 3G).

To determine whether the three novel FGFR1v lacking the cytoplasmic kinase domain participate in cell signaling, we analyzed FGFR1v-expressing cells on the phosphorylation status of key FGFR1-associated downstream targets, like FRS2 and AKT (32). Results show that overexpression of FGFR1-v318, FGFR1-v407, or FGFR1-v409 did not influence the pFRS-2 activity (Fig. 4A). Interestingly, FGFR1-v407 or FGFR1-v409 expression elevated pAKT levels with modest increase in pERK1/2 activities (Fig. 4A, lanes 3 and 4). However, in the FGFR1-v318–expressing cells, the pAKT and pERK1/2 levels remained similar to that of the control ACC-01 cells (compare lane 1 vs. 2).

Furthermore, we analyzed the FGFR1v signaling capacity at varying time points following growth factor (GF) stimulation. GF readily induced pFRS-2, but no difference in pFRS-2 activity was detected because of FGFR1v overexpression (Fig. 4B). While GF-induced pAKT induction appeared to be transient in parental and FGFR1-v318 cells, pAKT showed a propensity for prolonged activity in FGFR1-v407– or FGFR1-v409–expressing cells. GF also induced a sustained activity of pERK1/2 in the FGFR1-v407 or FGFR1-v409 cells. This differential signaling responses were discernable in spite of comparable levels of FGFR1^wt either with or without the three FGFR1v expression.

Because FGFR1-v407 or FGFR1-v409 appeared to be associated in pAKT activity, we sought to assess the stability of FGFR1-v409 relative to the endogenous FGFR1^wt. We examined protein extracts from parental and FGFR1-v409–expressing cells treated with cycloheximide, to prevent synthesis of de novo receptor protein. In both the parental and FGFR1-v409–expressing cells, FGFR1^wt levels decreased with comparable half-life of approximately 2 hours (Fig. 4C, top right). However, FGFR1-v409 levels remained stable during the same time period with half-life extending beyond 2 hours. In the presence of GF, FGFR1^wt degradation was more rapid in both the parental and FGFR1-v409–expressing cells with half-life approximately 1 hour (Fig. 4D). In contrast, GF showed no effect on FGFR1-v409 protein levels. Further analysis indicated that endogenous FGFR1v expression was unaltered with or without GF (Supplementary Fig. S3A). We also confirmed a similar pattern of increased stability for FGFR1-v407 in FGFR1-v407–expressing cells (Supplementary Fig. S3B). In addition, cell-surface FGFR1-v409 biotin-labeled assay substantiated that GF decreased FGFR1^wt levels, in accordance with ligand binding–mediated receptor internalization and degradation (Fig. 4E). Conversely, GF showed no effect on the cell surface–exposed FGFR1-v409 protein level. These results show that FGFR1v did not respond to GF, and that FGFR1v displayed greater stability than FGFR1^wt, whether in the absence or presence of GF stimulation. Concomitantly, we evaluated the impact of FGFR1v overexpression on cell growth. However, FGFR1v overexpression did not support an increase in cell proliferation whether in presence or absence of GF.

On the basis of the above results, we hypothesized that FGFR1v function is FGF/FGFR1 independent, and that FGFR1-v407 or FGFR1-v409 expression may desensitize cells exposed to inhibitors targeting FGFR1^wt. Dovitinib is a small-molecule multikinase inhibitor that blocks FGFR activity and has been widely evaluated in clinical trials of multiple cancers, including ACC (28, 33, 34). Indeed, unlike in parental ACC-01 cells (Supplementary Fig. S3C), dovitinib decreased pFRS2 in a dose-dependent manner but was unable to block the FGFR1-v409– or FGFR1-v407–induced pAKT activation (Fig. 4F). In addition, dovitinib induced a dose-response cytotoxic effect on control and FGFR1v–expressing cells (Fig. 4G). However, FGFR1-v407– or FGFR1-v409–expressing cells exhibited a 2-fold higher IC₅₀ compared with that of control and FGFR1-v318 cells. Conversely, siFGFR_v-i498–mediated FGFR1v knockdown in control ACC-01 cells enhanced sensitivity to dovitinib (Fig. 4H). Similar results were obtained using futibatinib, a highly selective FGFR inhibitor where FGFR1-v409–expressing cells displayed reduced sensitivity to futibatinib, while FGFR1v knockdown in control cells enhanced sensitivity to futibatinib (Supplementary Fig. S3D–S3F). Together, our results show that FGFR1-v409 or FGFR1-v407 is involved in signaling and cell growth regulation. However, its function appears to operate through a mechanism that is primarily not dependent on the FGF/FGFR1 complex.

Next, we postulated that FGFR1v function via alternative non-FGFR1 receptor tyrosine kinases (RTK). To test this possibility, we used a phospho-RTK array to identify the RTK potentially active in FGFR1-v409–expressing cells. This revealed that EGFR and AXL were the two mostly activated receptors in the FGFR1-v409–expressing cells compared with ACC-01 cells (Fig. 5A). Notably, in the ACC-01 cells, pAXL activity was relatively higher compared with other RTKs (Fig. 5A, top). A follow-up analysis indicated that inhibition of EGFR activity in the absence or presence of EGF by erlotinib did not alter the pAKT status in FGFR1-v409 cells (Supplementary Fig. S4A and S4B), suggesting that EGFR is not the primary cause of AKT activation in the FGFR1-v407/409–expressing cells.

AXL is a member of RTK family that is involved in activating multitude of downstream signaling pathways, including the PI3K-AKT (35). To determine whether FGFR1v transduces AKT signaling through AXL, we first confirmed the presence of pAXL activity in FGFR1-v407/v409–expressing cells (Fig. 5B). Next, we analyzed pAKT activity in FGFR1-v409–overexpressing cells following siRNA-mediated AXL silencing. Results confirmed an efficient AXL knockdown in both the control ACC-01 and FGFR1-v409 cells (Fig. 5C). Interestingly, AXL knockdown significantly reduced the FGFR1-v409–induced pAKT levels (Fig. 5C, compare lane 3 vs. 4). AXL depletion also downregulated pAKT levels in the ACC-01 cells (Fig. 5C, compare lane 1 vs. 2). In addition, we used BGB324 (bemcentinib), a highly selective AXL inhibitor (36, 37) to block AKT activity in FGFR1-v409 cells. Consistent with the AXL knockdown data, BGB324 treatment abolished pAKT activity in the FGFR1-v409–expressing cells (Fig. 5D). Similar effect of BGB324 on pAKT activity was obtained in FGFR1-v407 cells (Supplementary Fig. S4C). These results demonstrate that FGFR1v-mediated AKT activation is associated with AXL.

Next, we assessed the cytotoxic effect of BGB324 using control and FGFR1-v409–expressing cells. While FGFR1-v409–expressing cells showed decreased sensitivity against dovitinib (Fig. 4G), both the control and FGFR1-v409–expressing cells showed similar dose-dependent reduction in cell viability to BGB324 with comparable IC₅₀ (Fig. 5E).

We further explored to test whether FGFR1v silencing would enhance sensitivity of ACC-01 to AXL inhibition. ACC-01 cells transfected with siFGFR_v-i498 were either analyzed by Western blot analysis to verify FGFR1v knockdown or treated with varying concentration of BGB324. Results show that siFGFR1_v-i498 was able to specifically knockdown FGFR1v expression but not FGFR1^wt (Fig. 5F). Importantly, FGFR1v knockdown was accompanied by reduced pAXL/pAKT activity. Furthermore, FGFR1v knockdown cells exhibited enhanced sensitivity to BGB324 with IC₅₀ of approximately 1 μmol/L, compared with siRNA-control transfected cells with IC₅₀ of approximately 2 μmol/L (Fig. 5G).

To corroborate the findings further, we evaluated HACC-2A, a newly validated ACC cell line (22). We confirmed that HACC-2A cells expressed FGFR1v, AXL, and FGFR1^wt similar to ACC-01 cells (Supplementary Fig. S5A and S5B). Furthermore, siFGFR1v-i498–mediated FGFR1v knockdown resulted to decreased cell growth (Fig. 6A). Importantly, exogenous expression of FGFR1-v409 elevated pAXL/pAKT activity in HACC-2A cells (Fig. 6B, lane 1 vs. 2), and that dovitinib was unable to completely abolish the FGFR1-v409–mediated pAKT activity (Fig. 6B, lane 2 vs. 4). In addition, BGB324 and dovitinib treatment also increased cytotoxic effect in a dose-dependent manner with IC₅₀ of approximately 1.7 μmol/L and approximately 3.8 μmol/L, respectively (Supplementary Fig. S5C). Notably, knockdown of endogenous FGFR1v diminishes pAXL and pAKT levels (Supplementary Fig. S5D) and rendered HACC-2A cells more responsive to FGFR and AXL inhibitors (Fig. 6C and D; Supplementary Fig. S5E).

Because FGFR1v knockdown increased sensitivity against AXL inhibition in the two cell lines evaluated, we tested whether blockade of AXL/AKT axis desensitizes ACC cells against FGFR1^wt inhibition. Combination of BGB324 and dovitinib displayed higher visible cell killing effect compared with either BGB324 or dovitinib alone (Fig. 6E). Biochemical analysis showed that dovitinib reduced pFRS2 levels but reduced pAKT is only observed when cotreated with BGB324 (Fig. 6F). Notably, combined treatment caused induction of apoptotic indicators such as active caspase-3, PARP1 degradation, and loss in survivin levels, with reduced cell proliferative marker such as proliferating cell nuclear antigen (PCNA). In addition, cells exposed to the combination treatment showed an enhanced cytotoxic effect compared with either single-agent treatments (Fig. 6G). Together, these results demonstrate that AXL inhibition enhances sensitivity to FGFR1 inhibition causing an increased cytotoxic effect in ACC cells.

Finally, we used ACCX6 and ACC-01-XP tumor model to evaluate the effects of combined inhibition of FGFR1 and AXL in vivo. Initially, we verified AXL expression in ACCX6 (Supplementary Fig. S5F). Subcutaneous tumor-bearing mice were treated with vehicle, dovitinib, BGB324, or combination of both. Although treatment with dovitinib or BGB324 alone did not inhibit tumor growth significantly, tumors size was significantly smaller in the combined treatment group in both the ACCX6 and ACC-01-XP tumor model (Fig. 7A and C). To examine further the impact of the drugs, tumors harvested 24 hours after final treatment were subjected to Western blotting analysis. Results show that the combined inhibitor treatment caused significant reduction in antiapoptotic factor like survivin, and cell proliferative markers such as cyclin D1 and PCNA (Fig. 7B and D). We noted that dovitinib also impacted the apoptotic and proliferative factors expression in ACCX6 tumors, that appears to correlate with tumor growth inhibition, while the effect was not statistically significant. Together, these results indicate that combined FGFR1 and AXL inhibitors exerts increased effect in inhibiting ACC tumor growth.

We report here the findings of novel FGFR1vs in ACC, which are cytoplasmic domain-lacking forms of truncated FGFR1. We found that members of the FGFR1v can mediate FGF/FGFR1-independent function through the AXL/AKT signaling axis, which suggests that these FGFR1 variants may be involved in resistance mechanism to FGFR1 inhibitors in ACC.

Novel RNA-seq junction expression analysis using outlier statistic and SEVA methods (17, 18), allowed us to identify ASEs in 58 genes that were overrepresented in ACC. A pathway enrichment analysis indicated that the ASEs included a number of genes that are involved in PI3K-AKT signaling. The aberrant expression of these genes including FGFR1, thus may contribute to a PI3K-AKT signaling dysfunction in ACC.

FGFR1 is involved in many normal biological processes, but in ACC and many other cancer types, FGFR1 expression is elevated and is a potential therapeutic target (27, 29, 38). We found that 61% of the patients with ACC represented tumors with unique FGFR1 ASEs. Although we were unable to access these original patient tissues for further evaluation, a series of PCR-based analysis allowed us to validate and demonstrate the transcriptional expression of the FGFR1v in ACC patient-derived xenografts and cell lines.

FGFR1 is notable among RTKs with multiple isoforms resulting from exon switching, exon deletion, and intronic sequence inclusion (39). Our study revealed that the novel isoforms, characterized by intron inclusion, produced three previously unreported intracellular domain-lacking FGFR1v. Importantly, we found that FGFR1v silencing impacted cell growth activity. The doxycycline-inducible model employed in this study clearly represented higher FGFR1v levels than the physiologic FGFR1v expression level found in the ACC cell lines. However, complementary approach involving knockdown of endogenous FGFR1v enabled us to gain insight into the possible functional relevance of FGFR1v in ACC. Our findings suggested that FGFR1v are cell surface–linked protein, and that exogenous FGFR1v expression augmented AKT signaling. Aberrant gene splicing with intron insertion or retention may undergo nonsense-mediated decay and formation of truncated variants, leading to loss of function (40), or generate functional new peptides and isoforms (41–43). On the basis of our findings, FGFR1v may represent a functional isoform despite lacking the kinase domain.

AXL is a member of the Tyro3, AXL, MERTK (TAM) RTK family involved in regulating critical downstream signaling pathways, including PI3K/AKT (35). AXL signaling pathways are associated with cancer cell survival, proliferation, invasion, and drug resistance (35, 44). Our study demonstrated that FGFR1v-associated AKT activation and function is mediated by AXL in ACC cells. BGB324 (bemcentinib) is an orally administered AXL inhibitor, tested in the treatment of cancers such as, AML (45), triple-negative breast cancer (46), metastatic melanoma (47), non–small cell lung cancer (48), and myelodysplastic syndrome (49). Abrogating AXL activity by BGB324 abolished FGFR1 variant–mediated AKT activation accompanied by dose-dependent cytotoxic effect. In addition, downregulating endogenous FGFR1v increased the cytotoxic effect by BGB324. The precise biochemical mechanisms linking AXL/AKT with FGFR1v expression, however, remains to be determined. AXL activation occurs via ligand-dependent homodimerization and ligand-independent mechanisms through transcellular homophilic AXL adhesion, aggregation, and interaction with other TAM members or RTKs such as EGFR (48). In leukemic B cells, AXL has been reported to bind with FGFR3 (50). It may be possible that AXL/AKT signal competence is involved through complex formation with FGFR1v, leading to induction of FGFR1v-associated function in ACC.

We did not observe growth promoting advantages with exogenous FGFR1v expression contrary to the reduced cell growth activity by FGFR1v silencing. Cells used in the study expressed endogenous FGFR1v, and it is possible that a redundant growth activity is exhibited in exogenously expressed FGFR1v cells. However, FGFR1v-dependent function played a role in dovitinib resistance in ACC. The resistance mechanism can be partially attributed to FGFR1v cross-talk with the AXL/AKT axis, as not only did FGFR1v expression enhanced AXL activation, but AXL inhibition by BGB324 also led to increased dovitinib sensitivity in ACC cell line and in vivo tumor models. Moreover, the revelation that AXL promotes FGFR1v-mediated function supports the dual targeting of FGFR1/AXL to inhibit tumor growth in ACC (Fig. 7E).

Dovitinib was evaluated in phase II clinical studies for recurrent/metastatic ACC (33, 34), with only modest clinical activity. AXL inhibitors are in clinical development in several cancers, and it may be worthwhile to consider combining a FGFR inhibitor and an AXL inhibitor in clinical trial as a novel therapeutic strategy for ACC.

In conclusion, this is the first study to use RNA-seq data to search for novel splice variants in ACC. Several FGFR1 variants were identified and appear to have effects on associated signaling pathways, namely, AXL. Further studies of other splice variants in ACC and other cancers are likely to lead to the identification of additional variants that can have mechanistic relevance to cancer development and progression.

B. Leonard reports grants from NIH during the conduct of the study and is currently employed by Genentech. This work was done during postdoc and minimal work was done by B. Leonard while also employed by Genentech. A.V. Favorov reports grants from NIH and RAS during the conduct of the study and grants from RFBR outside the submitted work. E.J. Fertig reports grants from NCI, Adenoid Cystic Carcinoma Research Foundation, and National Institute of Dental and Carniofacial Research during the conduct of the study and personal fees from Viosera Therapeutics and Champions Oncology outside the submitted work. H. Kang reports other funding from Mitoimmune, PIN therapeutics, Prelude Therapeutics, GSK, Genentech, and Bayer; grants and other funding from Kura Oncology; grants from Eli-Lilly, Elevar Therapeutics, Exelixis, Novartis, and Advaxis outside the submitted work. P.K. Ha reports grants from NIH/NIDCR during the conduct of the study; personal fees from Rakuten Medical and Loxo Oncology; other funding from Stryker, Johnson & Johnson, and Medtronic outside the submitted work. No disclosures were reported by the other authors.

J.O. Humtsoe: Conceptualization, resources, formal analysis, supervision, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. H.-S. Kim: Data curation, investigation, methodology, writing-review and editing. B. Leonard: Formal analysis, supervision, investigation, writing-review and editing. S. Ling: Formal analysis, validation, investigation, methodology, writing-review and editing. B. Keam: Investigation, methodology, writing-review and editing. L. Marchionni: Data curation, software, formal analysis, writing-review and editing. B. Afsari: Data curation, software, formal analysis, writing-review and editing. M. Considine: Data curation, software, formal analysis, investigation. A.V. Favorov: Data curation, software, formal analysis, writing-review and editing. E.J. Fertig: Data curation, formal analysis, investigation, writing-review and editing. H. Kang: Validation, methodology, writing-review and editing. P.K. Ha: Conceptualization, resources, supervision, funding acquisition, validation, methodology, project administration, writing-review and editing.

This study was supported by the Adenoid Cystic Carcinoma Research Foundation and NIH/NIDCR R01 grant DE023227. A.V. Favorov was supported by the Russian Academic Project, 0112-2019-0001. E.J. Fertig, B. Afsari, and A.V. Favorov further acknowledge funding from the NIH P30CA006973 grant. The authors also thank Leilani Jones for providing technical support.

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.