Mutational landscape and genetic signatures of cell‐free DNA in tumour‐induced osteomalacia

Abstract Tumour‐induced osteomalacia (TIO) is a very rare paraneoplastic syndrome with bone pain, fractures and muscle weakness, which is mostly caused by phosphaturic mesenchymal tumours (PMTs). Cell‐free DNA (cfDNA) has been regarded as a non‐invasive liquid biopsy for many malignant tumours. However, it has not been studied in benign tumours, which prompted us to adopt the targeted next‐generation sequencing approach to compare cfDNAs of 4 TIO patients, four patients with bone metastasis (BM) and 10 healthy controls. The mutational landscapes of cfDNA in TIO and BM groups were similar in the spectrum of allele frequencies and mutation types. Markedly, deleterious missense mutations in FGFR1 and loss‐of‐function mutations in MED12 were found in 3/4 TIO patients but none of BM patients. The gene ontology analysis strongly supported that these mutated genes found in TIOs would play a potential role in PMTs' process. The genetic signatures and corresponding change in expression of FGFR1 and FGF23 were further validated in PMT tissues from a test cohort of another three TIO patients. In summary, we reported the first study of the mutational landscape and genetic signatures of cfDNA in TIO/PMTs.


| INTRODUC TI ON
Tumour-induced osteomalacia (TIO), also known as oncogenic osteomalacia, is a rare paraneoplastic syndrome characterized by hypophosphataemia due to renal phosphate wasting, inappropriately normal or low 1,25-dihydroxy vitamin D, and either elevated or inappropriately normal plasma FGF23 level. 1 Tumour-induced osteomalacia patients are always presented with certain non-specific symptoms, such as bone pain, fractures, muscle weakness and, occasionally, a reduction in height. The tumours inducing this syndrome are mostly benign phosphaturic mesenchymal tumours (PMTs) of variable sizes, which can locate anywhere in the body. 2,3 Previous studies revealed that more than half of PMTs are in the extremities, but almost 40% are not in the limbs, 4 making them difficult to locate. Behaviourally, even though this tumour commonly infiltrates the surrounding connective tissue, it rarely metastasizes. Nowadays, various combinations of functional and anatomical imaging are used for clinical diagnosis, including selective venous sampling for FGF23 measurements, 5 computed tomography (CT) scans, magnetic resonance imaging (MRI), 6 octreotide scintigraphy, 7 fluorodeoxyglucose (FDG) 8 and Gallium-68 DOTATATE positron emission tomography (PET)/CT. 9 However, the mechanisms underlying the tumorigenesis of PMTs and its secretion of those phosphatonins remain obscure. A relevant finding in understanding its tumorigenesis was the identification of fibronectin 1 (FN1) and fibroblast growth factor receptor-1 (FGFR1) translocations which led to an FN1-FGFR1 fusion protein in 60% of studied PMTs by RNA sequencing or FGFR-specific fluorescence in situ hybridization (FISH). 10 Additional studies have confirmed this finding in a larger group of PMTs. 11 In addition to the description of the FN1-FGFR1 translocation, it has been recently reported that liquid biopsy for many malignant tumours. However, it has not been studied in benign tumours, which prompted us to adopt the targeted next-generation sequencing approach to compare cfDNAs of 4 TIO patients, four patients with bone metastasis (BM) and 10 healthy controls. The mutational landscapes of cfDNA in TIO and BM groups were similar in the spectrum of allele frequencies and mutation types.
Markedly, deleterious missense mutations in FGFR1 and loss-of-function mutations in MED12 were found in 3/4 TIO patients but none of BM patients. The gene ontology analysis strongly supported that these mutated genes found in TIOs would play a potential role in PMTs' process. The genetic signatures and corresponding change in expression of FGFR1 and FGF23 were further validated in PMT tissues from a test cohort of another three TIO patients. In summary, we reported the first study of the mutational landscape and genetic signatures of cfDNA in TIO/PMTs.

K E Y W O R D S
cell-free DNA, fibroblast growth factor receptor-1, mediator complex subunit 12, nextgeneration sequencing, tumour-induced osteomalacia 6% of PMTs had an FN1-fibroblast growth factor 1 (FGF1) translocation. 11 However, these findings still suggest the existence of the tumorigenic drivers behind the fusion-negative cases, and the mutational landscape and genetic signatures of PMTs are yet to be elucidated.
Here, we employed circulating cell-free DNA (cfDNA) analysis as a non-invasive liquid biopsy to study the tumorigenesis of TIO. The cfDNAs mainly come from the fragmented genomic DNA (gDNA) of apoptotic, secretory or necrotic cells. Specifically, the cfDNA released by tumour cells was termed as circulating tumour DNA (ctDNA), 12 which carries tumour-specific genetic mutations, including single-nucleotide variations (SNVs), indels, genome rearrangements and copy number variations (CNVs). 12 With this method, diagnosis and genomic alterations of several cancers have been reported, such as metastatic breast cancer, 13 lung cancer 14 and exocrine pancreatic cancer. 15 In particular, there is a high degree of concordance between the mutation profiles of ctDNA and tumours metastasis. 16 Breakthroughs have increased the sensitivity of ctDNA detection by next-generation sequencing (NGS), contributing to early detections and diagnoses for many kinds of tumours. [16][17][18] Besides, analysis of non-invasive cfDNA has advantages over traditional biopsies in revealing the heterogeneity compared with tissue biopsies, because of its once-for-all capture and detection of mutations present in multiple tumour deposits. 19 Here, we recruited four TIO patients for cfDNA mutation profiling via an NGS approach targeting 422 cancer-relevant genes. The TIO-specific genetic signatures were explored by comparing our sample to cfDNAs of patients with bone metastasis (BM), as the positive control, and that of the healthy (negative) controls (HC). As a proof-of-concept study, we offered more information on the cfDNA landscape of TIO and shed a light on the genetic signatures for non-invasive detection and monitor of TIOs for the first time.

| Patient enrolment and sample preparation
Initially, a case-control design was adopted to identify the genetic signatures of the cfDNAs in TIO and PMT. Then, an additional test cohort of TIO patients was enrolled to validate the previous findings ( Figure 1 The clinicopathological characteristics of all eight patients were retrospectively reviewed. In the validation set, we recruited three more TIO patients and collected their relevant data as well. Written informed consent was obtained from each participant. The study was reviewed and approved by the Ethics Committee of PUMCH. Ten millilitres of peripheral blood was collected from each patient before surgeries and placed into EDTA-coated tubes (BD Biosciences), and samples were also obtained from individuals in the control group. Plasma was extracted within the 2 hours following blood collection, and plasma samples were shipped to the central testing laboratory within 48 hours. Normal tissue adjacent to the tumour was collected at a distance of at least 2 cm away from the tumour margin. 20 Plasma FGF23 was measured using FGF23 ELISA Kit (Kainos).

| Cell-free DNA purification and quantification
The cfDNA from plasma was extracted using the QIAmp Circulating

| Library preparation
Sequencing libraries were prepared using the KAPA Hyper Prep Kit (KAPA Biosystems) with an optimized manufacturer's protocol. In brief, 1 μg of genomic DNA, which was sheared into 350-bp fragments using Covaris M220 instrument (Covaris), or 2-50 ng of cfDNA, underwent end-repairing, A-tailing and ligation with indexed sequencing adapters sequentially, followed by size selection for genomic DNA libraries or purification for cfDNA libraries using Agencourt AMPure XP beads (Beckman Coulter). Finally, libraries were amplified by PCR and purified using Agencourt AMPure XP beads.

| Target gene panel sequencing and data processing
Different libraries with unique indices were pooled together in desirable ratios for up to 2 μg of total library input. Human cot-1 DNA

F I G U R E 1
The workflow of the searching biomarker in cfDNA of TIO patients. Blood samples were obtained from four patients with tumour-induced osteomalacia, four patients with tumour and bone metastasis and ten healthy controls. cfDNA was extracted from blood samples and underwent panel NGS to compare SNV, indel and CNV among three groups so that biomarkers could be found. NGS, nextgeneration sequencing; TIO, tumour-induced osteomalacia; BM, bone metastasis; HC, healthy control; SNV, single-nucleotide variation; CNV, copy number variation; LoF, loss-of-function  29 is used for transcriptome mapping followed by isoform and gene-level quantification performed by RSEM (version 1.3.0). 30 Differential expression analysis was conducted by R packages DESeq2 (version 1.16.1) 31 and edgeR (version 3.18.1). 32 Differentially expressed genes were selected by fold change >2 and P < .05. Corresponding volcano plots and heatmaps were generated by in-house R scripts.

| Fusion detection
Three tools were applied for fusion detection of RNA-seq data.

| Gene ontology analysis
To gain biologically functional insights of the gene clusters identified in TIO-only, BM-only and TIO&BM-both group, we conducted gene set enrichment analysis using R (version 3.4.3; https ://www.r-proje ct.org/). For annotation, the biological process terms of the Gene Ontology (GO) project 38,39 were involved as the biological knowledge, and the clusterProfiler package 40 was used for over-representation test. All biological process terms with P value adjusted <.01 were retained for further analysis. For reducing the functional redundancy of enriched terms, we calculated the similarities among GO terms and remove those highly similar terms by choosing the most representative term with the GOSemSim package. 41 For data visualization, only the top ten most enriched terms were presented.

| Statistical analysis
The somatic variants were identified by comparing the variants of cfDNA and tumours to the normal tissue or leucocytes in peripheral blood. With the starting cut-off point of 0.5% mutant allele frequency (MAF), comparisons among TIO, BM and HC were able to attain by statistical analysis. Association analysis was performed for allelic and genotypic association utilizing the SPSS software v15.0 (SPSS). The difference of MAFs and mutation types among different groups was compared using the chi-square test. The characteristics were presented as means ± standard deviations. Odds ratio (OR) with 95% confidence interval (CI) was used to assess the difference of mutation types in the cfDNA in three groups. The two-sided P < .05 was considered as statistically significant.

| Clinical characteristics of the patients with tumour-induced osteomalacia and bone metastasis
To investigate the TIO-specific genetic signatures, we concurrently utilized both positive and negative control groups. Four bone metastasis patients were recruited as positive controls considering that their clinical performance was similar to TIO and cfDNA was reported to be good liquid biopsy for detecting advanced cancer. 12,13 Another 10 unrelated healthy individuals whose gender and age were matched with the patients in the TIO and BM groups were considered as negative controls to exclude non-specific mutations and systematic errors ( Figure 1). As the discovery cohort, two female and two male TIO patients with a mean age of 40 ± 10 years and one female and three male BM patients with a mean age of 49 ± 13 years were enrolled ( Table 1) (Table 1).
Also, the TmP/GFR (ratio of the renal tubular maximum reabsorption rate of phosphate to glomerular filtration rate) also decreased, indicating the reduction of renal reabsorption of phosphorus. We also obtained the elevated FGF23 levels of these seven patients before operation, ranging from 62.83 to 374.67 pg/mL comparing to the normal level of 10-50 pg/mL. 42 The pathological results indicated that tumours of TIO patients were all PMTs ( Figure S1) and the locations of all the lesions were also been confirmed, consistently with our previous study. 3 In this study, all the PMTs were around or inside the bone, but in different positions, including the medial right shoulder joint, left lateral femoral condyle, L5 left pedicle and left distal ulna and radius. In addition, the corresponding primary tumours of the BM group are shown in Table 1.

| Quality of the sequencing data
In the DNA sample set of eight patients and ten healthy controls,

| The mutational landscape of cfDNA in the tumour-induced osteomalacia and bone metastasis groups
To comprehensively analyse genetic differences among the three groups, we compared the distributions of the allele frequencies (AF) in three groups (Figure 2A). Interestingly, there was no obvious difference between the TIO and BM groups (P = .946).
Nevertheless, the differences between the case groups (TIO and BM) and the control group (HC) were both extremely significant (P < 10 −100 ).
In addition to frequency distribution analysis, the mutation types were also analysed. Similarly, TIO and BM were closely alike (P = .495) and HC varied widely (P = 3.05 × 10 −58 and P = 7.55 × 10 −50 , respectively) ( Figure 2B). More specifically, there were significantly more LoF

| Exploring the genetic signatures of tumourinduced osteomalacia in cfDNA
To identify the potentially deleterious mutations, we focused on the functional somatic variants, including rare missense and LoF mutations. The somatic variants were identified by comparing the variants of cfDNA to the gDNA from the peripheral blood. With the strict exclusion criteria, we next identified and compared statistically significant mutations among TIO, BM and HC groups ( Figure 3, Tables S3 and S4). Generally, more mutation genes of rare missense were detected than that of LoF. Meanwhile, only seven shared genes of LoF between TIO and BM were identified.
There were 47 unique rare missense mutations for TIO and 35 for BM ( Figure 3A), and 40 LoF mutations for TIO and 28 for BM, respectively ( Figure 3B).
We then analysed the detailed distributions of the above pathogenic mutations. We selected candidate genes in which mutation distribution was different between TIO and BM by over 50% (no <2 in 4 cases). With such screening criteria, we further narrowed down the mutations to 45 mutation genes, including 39 rare missense mutations and seven LoF mutations ( Figure 4,

| Gene ontology enrichment analysis revealing biological processes responding to tumour-induced osteomalacia and bone metastasis
To verify the specific functions the genes related to the TIO and BM patients may involve in the disease process, we conducted an enrichment analysis for GO related to biological processes to gain insights into the molecular mechanisms. As shown in Figure 5A, among the top ten processes enriched by 21 genes of the TIO group, the most significant biological process was phosphorylation, so did another two  TIO01   TIO02   TIO03   TIO04   BM01   BM02   BM03   BM04   FGFR1  FANCE  MED12  BRIP1  SRC  SDHC  RECQL4  ALK  MSH2  QKI  RB1  BRD4  CBL  RARA  MCL1  SDHD  CYP2B6  SDHA  CHEK2  NBN  PAK3  KDR  CDC73  TEK  NOTCH2  IRF2  SETD2  BRCA1  TNFRSF14  PIK3CA  STAG2  PALB2  MDM2  XPC  FLCN  CCND1  AKT2  GSTT1  PTPN11  AMER1  GATA1  AXL  ESR1  GATA2  PRKCI   HC01   HC02   HC03   HC04   HC05   HC06 HC07 HC08 HC09

HC10
Loss-of-function Rare missense the regulation of response to the stimulus, programmed cell death and apoptotic process with high enrichment scores. These biological processes were essential for metastatic cancer to survive in migration. 43 Besides, common processes showed that cell communication and cell proliferation were important for both disease development ( Figure S2).

| Validating the genetic signatures of cfDNA in phosphaturic mesenchymal tumours
To validate the reliability of the genetic signatures which were previously found in the cfDNA of patients with TIO, a prospective cohort of three more TIO patients (TIO05, TIO06 and TIO07) was enrolled.
The TmP/GFR rate and the blood phosphorus level of these patients were also low before the surgery. The corresponding primary tumours are shown in Table 1.  (Table S7). Two in-frame deletion variants within MED12 (c.6165_6167delGCA and c.6165_6167delGCA) were shared in cfDNA and PMT. However, a frameshift variant within MED12 (c.5285delA) was only found in cfDNA with an AF of 0.43%, and two MED12 frameshift variants which were previously found in the TIO patients in the discovery cohort (c.1080delT in TIO01 and TIO04, and c.4901delA in TIO02) were identified only in the PMT tissue of TIO05 with the AFs less than 1%. Additionally, a missense variant within FGFR1 (c.239G>A) with an AF of 0.25% was only found in cfDNA but not in PMT.
To identify the origin of the deleterious mutations of FGFR1 and MED12 in cfDNA, the DNA from the PMT tissues and normal tissues of the three TIO patients in the validation cohort was sequenced using the NGS panel targeting 422 cancer-relevant genes ( Table   S1). As a result, intriguingly, the FGFR1 missense variant (c.239G>A) which was found in cfDNA of TIO05 was also found in somatic variants in the PMT tissues in TIO06 and TIO07 with relatively AFs of 0.16% and 29%, respectively. However, the LoF variant of MED12 was only found in the PMT tissue of TIO05. The germline variants were also curated according to the ACMG guideline, 26 while no deleterious mutation was identified for these three patients (Table S8).
To illustrate the correspondence changes in the transcriptome of PMT to the genetic signatures, the gene expression analysis of a pair of the PMT tissue and normal tissue of patient TIO07 was conducted by total RNA sequencing. Comparing to the normal tissue, the expression of FGF23, FGF1 and FGFR1 was all significantly up-regulated with the Log2FoldChange of 5.97, 3.22 and 1.78, respectively, and the P value of 5.80 × 10 −38 , 2.18 × 10 −21 and 2.19 × 10 −6 , respectively ( Figure S3). In addition, several gene fusions including the FN1-FGFR1 fusion were identified (Table S9). Similarly, to better functional enrichment analysis suggested that these differentially expressed genes mainly participated in bone-associated biological processes, such as ossification, skeletal system development, which was consistent with the top ten processes enriched from mutated genes of the cfDNA of the TIO patients in the discovery cohort ( Figure S4).

| D ISCUSS I ON
Our study aimed to elucidate the mutational landscape and the ge- have also shown that an isolated benign tumour cell could not be distinguished from an isolated malignant tumour cell. 48 Genetically, cfDNA also failed to discriminate between malignant and benign breast lesions. 49 To avoid the false positivity, large cohorts were recommended. 49 A recent study reported that endometriotic lesions, commonly considered to be benign inflammatory lesions, exhibited cancer-like features such as local invasion and resistance to apoptosis, which harboured some specific somatic cancer driver mutations. 50 Therefore, our findings were not only consistent with those previous studies but also extended the spectrum of similarities between malignant and benign tumours to shared mutations and mutational landscape. For example, we identified 79.46% (147/185) genes with rare missense mutations in the BM group and 71.71% (147/205) in the TIO group were shared ( Figure 3A,  63 The protein encoded by AXL is a member of the Tyro3-Axl-Mer receptor tyrosine kinase subfamily. The increased expression of AXL in cancer acts as a mechanism of acquired drug resistance, frequently accompanied by epithelial-to-mesenchymal transition. 64 ESR1 encodes oestrogen receptor 1, which can initiate or enhance gene transcription in response to oestrogen stimulation. Oestrogen has a multifunctional role in affecting the growth, differentiation and functions of different tissues, which is reported to be activated in hormone-resistant bone-metastatic breast cancer. 65 Oestrogen also has an important role in regulating bone growth and bone remodelling in adults.
Mutations on oestrogen are believed to be associated with severe osteoporosis. 66 The major limitation of our study is its relatively small sample size, as TIO is a rare disease. Validations in larger cohorts are still needed. Additionally, due to the fact that PMTs responsible for the osteomalacia were small in size, there was not enough tumour mass leftover for DNA or RNA sequencing after routine pathological diagnoses. Also, the target gene strategy prevented us from finding all the mutations, CNVs and gene fusions within the genome.
Finally, the exact functions of FGFR1 and MED12 mutations need to be confirmed by further functional studies in cell lines and animal models, which would confer much greater knowledge base for this rare disease.
In conclusion, we reported the first study of the mutational landscape and genetic signatures of cfDNA in TIO and PMT. Mutations such as FGFR1 might have important clinical implications. As a pioneer study, this cfDNA analysis represents a unique opportunity in the studies of rare tumours genetics and gives some reference for the future studies.

ACK N OWLED G EM ENTS
We thank all the individuals, families and physicians involved in the study for their participation. We thank the nurses from the

CO N FLI C T O F I NTE R E S T
The authors have no conflict of interests to declare. The raw data that support the findings of this study are openly available in the supplementary materials.