Introduction

Phosphaturic mesenchymal tumors (PMT) are uncommon bone or soft tissue neoplasms that cause hypophosphatemia and osteomalacia by secreting a variety of circulating phosphatonins, particularly fibroblast growth factor 23 (FGF23) [1,2,3,4,5,6,7]. By characterizing a large series of cases using RNA sequencing and fluorescence in situ hybridization (FISH), we have previously identified FN1-FGFR1 or FN1-FGF1 fusion genes in nearly half of PMT, thus shedding light on the potentially pivotal roles of the fibroblast growth factor 1 (FGF1)-fibroblast growth factor receptor 1 (FGFR1) axis, as well as the interaction between FN1-encoded fibronectin and other extracellular matrix molecules, in the tumorigenesis of PMT [8, 9]. The tumorigenic drivers remain obscure, however, for the other cases in which FISH detected neither fusion gene, as well as a significant subset of bone tumors where FISH failed, probably as a result of extensive DNA degradation due to acid-based decalcification. It is unknown whether alternative fusion genes, especially those involving the other fibroblast growth factor receptors or fibroblast growth factors, could be responsible for cases negative for both previously identified fusions. Also noteworthy is a patient (case PMT-61 in this series) with a germline rearrangement involving the upstream of the KL gene (coding for α-Klotho) which resulted in its upregulation; the patient developed hypophosphatemic rickets early in life, with 2 PMT identified at the age of 30, neither of which contained either fusion [9, 10]. Whether overexpression of KL plays a tumorigenic role in other PMT, particularly in those fusion-negative tumors as an alternative to FN1-FGFR1/FGF1 fusions, remains to be determined.

Although formalin-fixed and paraffin-embedded (FFPE) tissue specimens represent by far the predominant repository of archived clinical samples, RNA sequencing on FFPE samples has been substantially limited by the highly degraded nature of RNA [11]. Methods to enrich the coding regions in the transcriptome libraries have achieved some success, including the exome capture and anchored multiplex PCR (AMP) techniques [12,13,14]. However, the decalcification of FFPE bone specimens adds yet another level of challenge. The conventional acid-based decalcification process confers extensive disruption to the nucleic acids, greatly hindering molecular assays for such specimens [15, 16]. Therefore, it comes as no surprise that decalcification also significantly decreases the performance of next-generation DNA and RNA sequencing [17,18,19,20]. To our knowledge, RNA sequencing has not been systematically explored in decalcified samples.

Here we employed targeted and transcriptomic RNA sequencing on a set of mostly FFPE PMT samples, in an attempt to (1) identify potential alternative fusions in PMT without FN1-FGFR1/FGF1 fusions, with an emphasis on FN1 and genes coding for the fibroblast growth factors and fibroblast growth factor receptors; (2) validate the previously obtained positive FISH results and characterize the fusion junctions; and (3) test the utility of AMP-based RNA-sequencing technique (AMP-seq) in decalcified samples. Meanwhile, with the gene expression data thereby generated, we set to analyze the differential expression of Klotho among different groups.

Materials and methods

Tumor samples

FFPE samples of 34 previously reported PMT diagnosed between 1997 and 2019 were collected from the archives of National Taiwan University Hospital (Taipei, Taiwan) and Mayo Clinic (Rochester, MN, USA), as well as the consultation files of the authors (ALF and J-CL). The cases included 14 with FN1-FGFR1 fusion revealed by FISH (Group A, with two later reclassified as Group C; see below), 2 with FN1-FGF1 fusion (Group B), 10 with neither fusion (Group C), and 8 with unknown fusion status (Group D, where FISH analysis had failed) [9]. One case (PMT-72) also had snap-frozen tissue. PMT-61 is one of the 2 PMT from a patient with congenital hypophosphatemic rickets associated with a germline translocation resulting in KL overexpression [9, 10]. The clinical and molecular information, as well as the RNA sequencing methods employed, respectively, is summarized in Table 1. This research was approved by the research ethics boards of the respective institutions.

Table 1 The clinical and molecular features of the study groups
Full size table

RNA extraction and purification

RNA extraction and purification was performed as previously reported [21]. Briefly, total RNA was extracted with Trizol Reagent (Life Technologies, Carlsbad, CA) for snap-frozen tissue and either RNeasy FFPE kit (Qiagen, Venlo, Netherlands) or Agencourt FormaPure Kit (Beckman Coulter, Brea, CA) for FFPE samples according to the manufacturers’ protocols. RNA was quantified with Qubit® RNA HS Assay Kit (Thermo Scientific, Waltham, MA) after purification. Functional RNA quality was assessed in-line with library generation.

AMP-based RNA sequencing and data analysis

A total of 23 samples of PMT were subjected to AMP-seq, implemented as previously described [21]. For library generation, primers targeting FN1 (as the 5′ partner gene), KL (as the 3′ partner gene), and the genes coding for fibroblastic growth factors (FGFs, as 3′ partner genes) and their receptors (FGFRs, chiefly as the 3′ partner genes) were designed using the web-based ArcherDx Custom Assay Designer tool (ArcherDx, Boulder, CO; see Supplementary Table 1 for the list of targeted genes and exons). Overall, 200 ng of RNA as quantified by Qubit was used as input for library generation with the aforementioned gene-specific primers using Archer FusionPlex Protocol for Illumina, revision LA135.C. RNA quality was assessed in-line with library generation by the PreSeq RNA QC qPCR Assay (ArcherDX), provided with the FusionPlex kit, according to kit protocol. Libraries were then sequenced on an Illumina NextSeq 500 using NextSeq 500 v2 reagents (Illumina, San Diego, CA) for paired-end, 150 base pair reads and dual index reads. Housekeeping genes CHMP2A, GPI, RAB7A, and VCP were assayed in addition to candidate fusion transcripts as a positive control, using the primers from Archer PreSeq RNA QC Assay.

Data were analyzed by Archer Analysis version 6.0. Briefly, consensus sequences were compared with known fusion transcripts in the Quiver database of fusions, and novel transcripts were BLASTed to the reference genome guided by RefSeq. A set of filters were applied to the annotated results to reduce false positives. For gene expression profiling, the unique RNA reads for each gene-specific primer (GSP2) were normalized to the average reads of the four housekeeping genes as mentioned above. The normalized expression data were then retrieved from the Archer Analysis website for heat map generation. The average of expression levels derived from all GSP2 primers targeting any given gene of interest served as the representative expression level of that gene for the comparative purpose; the GSP2 which yielded zero expression level across all tumors were excluded. Mann–Whitney U test was performed to determine the significance of differential expression of the gene(s) of interest between groups, with the cutoff significance level defined as 0.05.

Exome-captured and whole-transcriptomic RNA sequencing and data analysis

Exome-captured and whole-transcriptomic RNA sequencing experiments were performed as previously described [22]. Briefly, RNA was subjected to library construction with the TruSeq RNA Exome kit for exonic region enrichment (Illumina, San Diego, CA, USA) for FFPE samples and TruSeq RNA Sample Prep Kit for poly-A selection for the snap-frozen sample. mRNA sequencing was then performed using a Nextseq or Hiseq system for FFPE or frozen tissue RNA, respectively, for 150-bp paired-end sequencing following the manufacturers’ instructions. The yielded read files (in FASTQ format) were mapped to the hg19 reference by STAR for fusion detection, with the gene expression levels calculated by RSEM [23, 24]. Differential genes were identified by R package using DESeq2 [25].

Validation of genomic fusion junction

Genomic DNA was extracted from the FFPE tissue of selective case(s) as previously described [8]. Primers were designed to target the intronic regions in which the genomic fusion junction was presumed to be located (see Supplementary Table 2 for primer sequences). PCR and bidirectional Sanger sequencing were performed as described [8].

RNA chromogenic in situ hybridization (RNAscope) validation of KL and KLB expression

For semi-quantification of KL expression, non-decalcified FFPE tissue slides recently (<7 days) cut from tissue blocks or preserved at −80 °C immediately after tissue sectioning were subjected to RNA CISH. The RNAscope 2.5 HD Detection kit (ACDbio, Newark, CA) was used following the manufacturer’s instructions as previously described [26]. Briefly, 5-micron-thick tissue sections were deparaffinized and pretreated with heat and protease before hybridization with KL, KLB, and PPIB (control) probes, respectively. RNAscope 2.5 Brown (ACDbio) was used for signal amplification and detection chemistry, followed by counterstaining with hematoxylin. Scoring was performed according to the manufacturer’s recommendation (0: <1 dot/cell; 1+: 1–3 dots/cell; 2+: 4–9 dots/cell; 3+: 10–15 dots/cell or if <10% dots were in clusters; 4+: >15 dots/cell or if >10% dots were in clusters). The result was deemed inadequate for assessment if the control probe yielded a score of 0.

Results

Evaluation of RNA quality for AMP-based sequencing

PreSeq Cq-values were determined in all but five cases (all belonging to Group D; Table 2). The Cq-values ranged from 26.11 to 34.56 (median: 30.64) and significantly correlated with sample ages (r = 0.808 and P = 3.3 × 10−18; Fig. 1a). All the three Group D cases studied had Cq-values greater than 31 and erred on the upper side of the regression line.

Table 2 Results of AMP-based RNA sequencing (ArcherDx)
Full size table
Fig. 1
figure 1

Parameters of AMP-seq reflecting the RNA quality. The Cq-values correlate with sample ages (a); the Group D (decalcified) cases err on the upper side of the regression line. RNA read numbers in relation to sample group (b) and sample age (c). Samples with lower read numbers are enriched by Group D cases

Full size image

In addition, the actual numbers of RNA reads derived from sequencing ranged from 4K to 1171K (median: 807K; Table 2 and Fig. 1b). With the highest number and the lowest number excluded as outliers, the median read number minus twice the standard deviation was arbitrarily defined as the quality threshold (220K). Three cases had RNA read numbers lower than this threshold, including the two oldest specimens (sampled 19 and 17 years ago, respectively) and two Group D bone tumors. These three cases were considered having significantly worse RNA quality than the other cases, and hence a higher probability of falsely negative results.

Further, Group D as a whole had significantly lower RNA reads compared with the other groups (median: 505K vs. 923K, Student’s t test P = 0.005). In addition, the ages of the samples negatively correlated with RNA read numbers (r = −0.684 and P = 1.34 × 10−10; Fig. 1c), and this held true when cases of Group D and the other groups were considered separately (for Group D, r = −0.637 and P = 4.93 × 10−8; for the others, r = −0.581 and P = 1.63 × 10−9).

Fusion genes detected by AMP-based sequencing

Of the eight Group A cases, the results of AMP-seq were consistent with those of FISH in 6, showing in-frame FN1-FGFR1 fusion transcripts (Table 2). The exact fusion junctions of PMT-1 and PMT-2 were identical to those having been demonstrated by whole-transcriptome sequencing previously [8]. The breakpoints of FN1 rested on the 3′ ends of exon 16, 22, 23, and 24 in 1, 1, 3, and 1 cases, respectively. For the FGFR1 gene, breakpoints at the 5′ end of exon 3 were detected in four cases, all of which had concomitant transcripts with an exon 4 fusion junction that probably represented alternative splicing as previously suggested [8]. PMT-41 had exon 4 as the sole FGFR1 breakpoint, implying a genomic breakpoint in intron 3. It is noteworthy that PMT-64 harbored the fusion transcripts with novel breakpoints in FGFR1 exon 9, either at the 5′ end or, as a minor variant, 67 bp into the exon (immediately 3′ to an “AG” sequence, supposedly a de novo splice acceptor).

Both Group B cases were confirmed to have in-frame FN1-FGF1 fusions, with exons 23 and 24 of FN1, respectively, fused to exon 2 of FGF1 at the nucleotide immediately 3′ to an “AG” sequence. The fusion junction of PMT-50 was identical to that previously demonstrated by RNA sequencing [9].

Two Group D bone tumors contained detectable fusions. PMT-26 harbored an in-frame FN1-FGF1 fusion with the fusion junction identical to that identified in PMT-16 (exon 26 to exon 2)[9] and therefore was reclassified as Group B (“D > B”). PMT-27 had a novel THRB-FGF10 fusion, with the breakpoints located in the 3′ UTR of THRB (coding for thyroid hormone receptor beta) and at the 104th nucleotide into the coding sequence of FGF10, respectively; the significance of this fusion is uncertain, as the FGF10 part is not predicted to be translated into peptide in the form of chimeric protein. RT-PCR to validate the fusions in both cases failed, probably because of previous tissue decalcification (data not shown). None of the Group C cases had a detectable fusion. The results of AMP-based sequencing are summarized in Table 2 and Supplementary Table 3.

Confirmation of genomic fusion junction and clinicopathologic review of PMT-64

In contrast to all other FGFR1 breakpoints we have so far identified in PMT, which are located 5′ to exon 3, 4, or 5 and thus include at least two receptor domains of FGFR1 in the chimeric protein, PMT-64 had a breakpoint in exon 9, which included the transmembrane domain and a minimal extracellular part (47 bp) of FGFR1. To exclude the possibility of this unusual fusion junction representing alternative splicing rather than reflecting the true genomic junction, genomic DNA PCR and Sanger sequencing was performed and confirmed the genomic breakpoint in FN1 intron 23 and FGFR1 intron 8 (Fig. 2a). The predicted chimeric protein domains are shown in Fig. 2b.

Fig. 2
figure 2

Molecular and pathologic features of PMT-64. a The sequence flanking the genomic fusion junction. b The presumed fusion protein contains no ligand-binding domain (LBD) of FGFR1. c The histology is typical of PMT

Full size image

The exceptional fusion junction of this case prompted us to look for its clinical and pathological difference from the other cases. The tumor manifested itself with classical TIO/hypophosphatemia and was resected from the left maxillary sinus of a 49-year-old man. It was one of the only four sinonasal cases we have reported, with the other three cases harboring neither fusion [9]. Histologically, the tumor exhibited typical features of PMT, consisting of spindled cells with bland nuclei and eosinophilic cytoplasm, in a richly vascularized and collagenous stroma interspersed with prominent foci of so-called ‘grungy’ calcifications (Fig. 2c). Like many other PMTs, it was immunoreactive for FGFR1 and FGF1 and positive for FGF23 mRNA by chromogenic in situ hybridization [9]. The clinicopathological features of this case were essentially identical to those of PMT showing more conventional fusion junctions.

Reappraisal of the FISH results of the inconsistent or borderline cases

The discrepancy between previous FISH and the current AMP-seq results in PMT-10 and PMT-18 prompted us to review the FISH data of both cases as well as two other cases with borderline FISH results. PMT-18 was a 17-year-old sample and, despite a convincingly positive FN1-FGFR1 FISH result, in retrospect had an RNA quality among the worst of this cohort, to which a falsely negative sequencing result might be attributed. On the other hand, PMT-10, as well as PMT-8 and PMT-9, was originally classified as Group A based on borderline positive FISH results (10–15% of cells harboring a fusion signal, i.e., a 5′ FN1 signal and a 3′ FGFR1 signal less than half the average width of a signal apart, after meticulously counting >200 cells) [8]. Nonetheless, based on our later experience, cases positive for FN1-FGFR1 fusion usually have cells more readily found to have a fusion signal where the two constituent signals just abut or even overlap with each other, instead of being slightly separate as was observed in the three borderline cases in retrospect. Therefore, PMT-8, PMT-9 (not included in this study), and PMT-10 were reclassified as Group C, where the occasional juxtaposition of FN1 and FGFR1 signals observed by FISH probably represented fortuitous proximity (i.e., falsely positive results).

Exome-captured and whole-transcriptomic RNA sequencing confirmed the fusion-negative status of selective Group C cases

To cross-validate their fusion status, PMT-8 and PMT-10, as well as three additional Group C cases, were subjected to RNA exome sequencing (N = 4) or transcriptomic sequencing (N = 1). Consistent with the results of FISH analysis/reappraisal and AMP-seq (in the case of PMT-10), no FN1-FGFR1/FGF1 or other convincing fusion could be identified in any of the five cases (data not shown). Collectively, these RNA sequencing data suggested fundamentally different mechanisms might be responsible for PMT without FN1-FGFR1/FGF1 fusion.

Group C showed significantly higher KL or KLB expression

The heat map of gene expression profiles derived from AMP-seq is illustrated as Supplementary Fig. 1. Five cases with the lowest total RNA reads and/or zero expression of all four control genes were excluded from the analysis because of the inferred poor sample quality. In keeping with our previous observations, FN1, FGFR1, FGF1, and FGF23 were the most constantly expressed genes across all tumor groups.

Of note, Group C tumors, including PMT-61 which was a case of germline rearrangement-induced systemic KL upregulation [10], expressed significantly higher levels of KL by AMP-seq in comparison with Groups A and B combined (P = 0.0078; Fig. 3a, b). During the conduction of this study, patient of PMT-58 developed a metastatic PMT to the liver (PMT-58-2), where FISH revealed no FN1-FGFR1/FGF1 fusion (data not shown), and therefore PMT-58 was reclassified as Group C (“D > C”). By AMP-seq, PMT-58 also expressed a high level of KL (Fig. 3a).

Fig. 3
figure 3

Differential expression of KL and KLB in various groups of PMT and non-PMT mesenchymal tumors by cross-platform RNA sequencing. a, b The KL expression levels derived from the three primers by AMP-seq are significantly higher in Group C than Groups A and B combined. The expression level of zero was reassigned as 0.001 for loge transformation. PMT-26 was reclassified as Group B (“D > B”) based on the AMP-seq results; PMT-10 was reclassified as Group C (“A > C”) after revisiting the previous FISH data; PMT-58 was reclassified as Group C (“D > C”) because a metastatic hepatic tumor (PMT-58-2) was later confirmed to be fusion-negative by FISH. c Three additional Group C PMT, PMT-8, PMT-73, and PMT-74, expressed the highest levels of KL or KLB (in normalized count) among the 30 mesenchymal tumors subjected to the RNA exome sequencing (“Others” indicates non-PMT tumors). d An additional Group C PMT (PMT-72) exhibited the highest level of KL (in FPKM) among the 10 mesenchymal tumors with snap-frozen tissue subjected to whole-transcriptomic sequencing (“Others” indicates non-PMT tumors). All the other tumors revealed very low KL and KLB expression, including the three fusion-containing PMT. The numbers on top of each bar represent the rank of gene expression /total number of expressed genes of respective tumor

Full size image

By RNA exome and whole-transcriptomic sequencing, respectively, two additional Group C cases were also found to express high levels of KL (Fig. 3c, d). Two other Group C cases, despite showing low KL expression, presented the highest levels of KLB instead among the 30 mesenchymal tumors analyzed with the same platform (i.e., RNA exome; Fig. 3c) [22]. Taken together, four of the five fusion-negative PMT (except PMT-10) expressed the highest levels of either KL or KLB among the 40 PMT or non-PMT mesenchymal tumors, while both genes were barely expressed in the fusion-positive PMT [8, 9], in keeping with the AMP-seq results. Parenthetically, our data also showed that PMT tended to express higher levels of selective genes related to osteoblast/osteocyte or osteoclast, similar to the findings of a previous report (Fig. 3c, d, and Supplementary Fig. 2) [27].

To further validate these findings, RNAscope was performed to semi-quantify KL and KLB RNA expression. Adequate samples were available in 11 Group A, 1 Group B, and 10 Group C cases (Table 1). Three samples failed the RNAscope analysis with zero expression of the control gene (and KL/KLB). Consistent with the AMP-seq results, the Group C cases expressed significantly higher levels of KL in comparison with Groups A and B combined (P = 0.0208) (Fig. 4a, b). Moreover, the KL expression levels derived from RNAscope and AMP-seq (ArcherDx) were highly correlated [RNAscope vs. log(ArcherDx): regression r2 = 0.822, P = 2.98 × 10−4; Spearman r = 0.8847, P = 0.0011], which could serve as a mutual validation of the credibility of both platforms in the characterization of gene expression (Fig. 4c). Furthermore, two of the four KL-negative Group C cases (PMT-73 and PMT-74) expressed the highest levels of KLB, consistent with their RNA exome sequencing results, while the other cases showed zero or merely detectable KLB expression (Fig. 4a). The Group C cases demonstrated significantly higher combined KL/KLB scores than Groups A and B (P = 0.0081; Fig. 4d). Collectively, by RNAscope, 8 of the 10 Group C cases showed overexpression (defined as a score of 2~4) of either gene in a mutually exclusive manner, in stark contrast with Groups A and B (Fig. 4e).

Fig. 4
figure 4

RNA CISH (RNAscope) to detect KL and KLB expression. a PMT-58-2 and PMT-73 showed a score of 4 for KL and KLB, respectively, while PMT-2 was negative for both genes (insets: PPIB control). Also, note the normal expression of KLB in the adjacent liver tissue in PMT-58-2 (asterisk). b Group C as a whole expressed significantly higher levels of KL than Groups A and B combined. c The KL expression levels derived from RNAscope and AMP-seq (ArcherDx) highly correlated with each other. d With two of the KL-negative Group C cases (PMT-73 and PMT-74) highly expressing KLB instead, Group C also revealed significantly higher sums of KL and KLB scores compared with Groups A and B. e The overexpression (when defined as a score of 2~4) of either gene appeared mutually exclusive (cases with scores of zero in both genes were not shown)

Full size image

In summary, these findings suggested that aberrant expression of α-Klotho or β-Klotho might possibly serve as an alternative to the FN1-FGFR1/FGF1 fusion genes as a tumorigenesis mechanism in fusion-negative PMT.

Discussion

In prior studies, we have identified FN1-FGFR1 or FN1-FGF1 fusion gene in nearly half of PMT cases [8, 9]. As the activated FGFR1 pathway is known for its oncogenic roles in other tumor types and is a canonical driver of FGF23 production in osteocytes, our findings suggest a central role for the FGFR1 pathway in both the tumorigenesis and FGF23 hypersecretion phenotype of PMT, and hence a rationale for FGFR inhibitors to serve as therapeutic agents [28,29,30,31,32,33]. However, the tumorigenesis mechanisms remain elusive in more than half of PMT which harbor neither fusion gene. It is reasonable to speculate that alternative fusions might be in play, particularly those involving paralogous genes in the FGFR or FGF family as well as the FN1 gene. In this study, we employed three different RNA-seq techniques to study a large cohort of PMT with a special focus on these paralogous genes and the fusion-negative (Group C) cases. Somewhat to our surprise, none of these tumors were found with an alternative fusion except for the THRB-FGF10 fusion identified in a previously decalcified (Group D) case, whose significance is uncertain as the THRB breakpoint falls in the 3′ UTR. Therefore, the current data suggest the tumorigenic mechanisms behind the PMT without either known fusion gene might be categorically different and probably do not involve genetic fusions, or at least not those detectable at the RNA level.

Rather, as the most important finding of this study, these fusion-negative PMT exhibited significantly higher KL or KLB expression despite no associated fusion detected. KL encodes α-Klotho, an obligatory co-receptor for FGF23-FGFR1 binding, and its expression is mostly limited to renal tubules, choroid plexus, and parathyroid gland [34]. As we previously hypothesized in view of the unique case PMT-61, aberrant expression of α-Klotho in FGF23-secreting cells, such as osteocytes and their precursors, might possibly facilitate the establishment of an FGF23-FGFR1 autocrine loop, which might in turn drive the upregulation of FGF23 and tumorigenesis through the activated FGFR1 pathway, thereby underpinning the formation of PMT [9, 29,30,31,32,33]. Echoing our findings, a subset of PMT with α-Klotho expression was recently reported, although the respective authors failed to characterize the fusion status of their cases, except for 1 case where RNA sequencing also revealed KL overexpression and no fusion gene [27, 35]. On the other hand, KLB, coding for β-Klotho, is a paralog of KL. To our knowledge, there has been no report of β-Klotho expression in PMT. β-Klotho is known as a co-receptor for FGF19/21-FGFR binding [36]. However, the expression of FGF19 and FGF21 was low, if any, in all PMT studied including those overexpressing KLB. Therefore, our findings might suggest an unrecognized role for β-Klotho in the binding and expression of FGF23 or FGF1, another frequently overexpressed fibroblast growth factor in PMT. Importantly, the implication of KL/KLB overexpression in a majority of fusion-negative PMT suggests the FGFR pathway probably also plays a pivotal role in this group of tumors, hence the potential utility of FGFR inhibitors in treating PMT regardless of their fusion status. The mechanisms behind the upregulation of KL/KLB and how they lead to PMT formation remain to be investigated. It is noteworthy that the germline chromosomal rearrangement in the patient with PMT-61 involved a locus 49 kb upstream of KL, which cannot be detected at the RNA level [9, 10]. Going forward, it will be interesting to see whether similar mechanisms could have a role in the overexpression of KL/KLB in other fusion-negative PMT.

Our transcriptome-wide analyses of PMT also revealed upregulation of osteocyte- or osteoclast-related genes as indicated in a previous study, suggesting the osteoblastic differentiation and a possible role of osteoclast in the tumorigenesis of PMT (Fig. 3c, d and Supplementary Fig. 2) [27]. Of potential interest is SPP1 (coding for osteopontin), which is involved in osteoclast attachment to bone mineral and a variety of cellular processes including tumorigenesis [37]. While the fusion-negative PMT expressed significantly higher levels of SPP1 than non-PMT mesenchymal tumors, all the 3 fusion-positive PMT analyzed showed zero expression of this gene (Supplementary Fig. 2). Whether SPP1 might play differential roles in PMT of different fusion status might warrant further study to determine.

The current study characterized the capability of the AMP-seq to detect fusion transcripts from FFPE samples with highly degraded nucleic acids. In the present series, fusion detection was successful in samples as old as 11 years. Further, we were able to identify FN1-FGF1 fusion in PMT-26, the decalcification of which had probably caused the failure of previous FISH analyses. To our knowledge, this was the first demonstration of fusion detection by AMP-seq in decalcified samples. Interestingly, all four cases with the FN1-FGF1 fusion were bone tumors [9], suggesting a possible context/microenvironment-dependence in the tumorigenic mechanism of this fusion. Meanwhile, the current study also clearly demonstrated the limitations of AMP-seq, whose scope of surveillance is confined to a very focused panel of target genes and thus heavily hypothesis-based, and whose performance was obviously compromised by the old sample age and previous tissue decalcification. As shown, the RNA quality demonstrated a significantly reverse relation to these factors. Similar to our findings, recently Lam et al. failed to identify genetic fusion in any decalcified tissue using the same AMP-seq technique and could attribute the suboptimal sequencing quality to the decalcification process per se in two cases [20].

Knowledge of the exact fusion junctions derived from RNA sequencing, and hence the inferred protein domains included in the chimeric proteins, is valuable for dissecting the functional aspects of the fusions, which cannot be provided by FISH. In contrast to previously characterized FN1-FGFR1 fusions, all of which included the ligand-binding domain of FGFR1, implying that the ligand binding might add to the tumorigenic pathway as an autocrine positive feedback loop, the current data have revealed one exception (PMT-64) that does not contain this ligand-binding domain, suggesting that the pathogenesis of PMT may not always be related to ligand binding (Fig. 2b). This finding appears to echo the preliminary observations in a recent phase 2 clinical trial, where a neutralizing anti-FGF23 antibody did not seem to confer a significant tumoricidal anti-PMT effect as shown by an FGFR1 inhibitor [38, 39]. What might be essential instead, in terms of the contribution of FGFR1 (other than its kinase domain) to the chimeric protein, is the transmembrane domain of FGFR1, which could anchor the fusion protein onto a membrane to avoid being secreted, a hypothesis that is supported by the unusual FGFR1 breakpoints seen in PMT, as opposed to the classical exon 10 breakpoint seen in most other FGFR1-rearranged neoplasms [9].

There are limitations of this study. The current panel of gene-specific primers for AMP-seq did not cover FN1 outside exons 15–30, because FN1 breakpoints in PMT had been located between exons 22 and 28 when this panel was designed. However, an exon 42 breakpoint was later identified in a calcifying aponeurotic fibroma with FN1-EGF fusion and a gastrointestinal leiomyoma with FN1-ALK fusion [40, 41], underscoring the insufficient coverage of the current panel and the possibility that some fusion-positive cases might have been overlooked. Furthermore, the FGF16 gene was inadvertently left out of this panel. Moreover, although the Group C cases were shown to express significantly higher KL/KLB at the RNA level by AMP-seq and/or RNA CISH, how the overexpression of the KL/KLB mRNA could be translated into the protein level requires further validation.

In conclusion, the current data demonstrated the absence of alternative fusion gene in PMT without FN1-FGFR1/FGF1 fusion, which might probably harbor significantly distinct aberrations. In particular, the overexpression of Klotho, which was noted in the majority of fusion-negative tumors in stark contrast with those containing either known fusion, could possibly replace those fusion genes as a novel tumorigenic driver by activating the FGFR pathway through enabling an FGF23-FGFR autocrine loop. If proved, this hypothesis would legitimize the utility of FGFR inhibitors to treat those Klotho-overexpressing fusion-negative PMT where necessary. Further studies are required to confirm this phenotype in a larger group of fusion-negative PMT and to characterize the genomic or epigenetic underpinnings.