Tumors Associated With Oncogenic Osteomalacia Express Genes Important in Bone and Mineral Metabolism



Oncogenic osteomalacia (OOM) is associated with primitive mesenchymal tumors that secrete phosphaturic factors resulting in low serum concentrations of phosphate and calcitriol, phosphaturia, and defective bone mineralization. To identify overexpressed genes in these tumors, we compared gene expression profiles of tumors resected from patients with OOM and histologically similar control tumors using serial analysis of gene expression (SAGE). Three hundred and sixty-four genes were expressed at least twofold greater in OOM tumors compared with control tumors. A subset of 67 highly expressed genes underwent validation with an extended set of OOM and control tumors using array analysis or reverse-transcription polymerase chain reaction (RT-PCR). Ten of these validated genes were consistently overexpressed in all OOM tumors relative to control tumors. Strikingly, genes with roles in bone matrix formation, mineral ion transport, and bone mineralization were highly expressed in the OOM tumors.


ONCOGENIC OSTEOMALACIA (OOM) is an acquired hypophosphatemic syndrome with biochemical similarity to genetic forms of hypophosphatemic rickets but is associated with the presence of a primitive mesenchymal tumor. OOM is characterized by impaired renal tubular reabsorption of phosphate, low serum concentrations of phosphate and calcitriol {1,25-dihydroxyvitamin D[1,25 (OH) 2D]}, and defective bone mineralization.(1,2) Experimental evidence suggests that the biochemical and skeletal defects in OOM are caused by a humoral factor(s) that is produced by these tumors. Tumor extracts can inhibit phosphate transport in vitro(3–5) and produce phosphaturia and hypophosphatemia in vivo.(6) Furthermore, surgical removal of tumor tissue from affected individuals leads to enhanced renal conservation of phosphate with normalization of serum phosphorus and calcitriol concentrations and subsequent remineralization of bone.(1,2)

The mesenchymal tumors that are associated with OOM are characteristically slow-growing, complex, polymorphous neoplasms. Many of these mesenchymal tumors contain areas of osseous metaplasia, poorly differentiated foci of chondroid tissue, and osteoclast-like giant cells.(7) The presence of these skeletal elements and calcification suggest these tumors express genes that may have important functions in bone formation and mineralization. Furthermore, the profound mineralization defects present within individuals that harbor these small, slow-growing tumors suggest that these mesenchymal tumors produce a factor(s) that may contribute to the abnormal mineralization characteristic of OOM and inherited hypophosphatemic syndromes.

Efforts to identify and isolate the humoral factor(s) produced by OOM tumors have been hampered by the poor growth of these tumors in culture and frequent loss of production of the phosphaturic factor with passage in culture. As an alternative approach for identifying genes important in phosphate homeostasis, bone formation, and mineralization, we used serial analysis of gene expression (SAGE) to compare gene expression profiles from mesenchymal tumors associated with OOM and mesenchymal tumors of similar tissue types that were not associated with hypophosphatemia. Here, we show SAGE profiles of mesenchymal tumors and establish that genes with important roles in bone matrix formation, mineral ion transport, and bone mineralization are highly expressed in OOM tumors.


Criteria for tumor selection

OOM tumors were resected from patients with the following characteristics: (1) acquired hypophosphatemia with no family history of hypophosphatemia or other medical illnesses predisposing to hypophosphatemia, (2) low or inappropriately normal serum levels of 1,25(OH) 2D, (3) reduced tubular reabsorption of phosphate (TRP), (4) biochemical or histological evidence of osteomalacia, (5) normalization of biochemical parameters after complete tumor resection (Table 1).

Table Table 1.. Biochemical Characteristics of Patients With OOM
original image

Four of the five OOM tumors selected for SAGE and/or array analysis had pathological features consistent with hemangiopericytomas. The anatomical locations of the tumors were diverse, including the lower extremity (Hp1 and Hp1B), the sinonasal region (Hp2), the meninges (Hp3), and the upper extremity (Hp4). Hp1B was removed 4 years after Hp1 from the same individual. The fifth OOM tumor (S1), resected from the upper extremity, had pathological features consistent with a leiomyosarcoma.

The mesenchymal tumors selected as controls were obtained from the Johns Hopkins Hospital Tumor Bank and had been resected from patients who had no history of hypophosphatemia. The two control hemangiopericytomas (C1 and C2) that were selected for SAGE profiling had similar histological appearance and were derived from similar anatomical locations as OOM tumors Hp1 and Hp2. Four additional control mesenchymal tumors were selected for array analysis. These included three additional hemangiopericytomas isolated from the chest (C3), brain (C5), and lower extremity (C6), and one mesenchymal tumor isolated from the pancreas (C4). OOM and control tumors used in this study were resected as part of the clinical care of each patient and surplus tumor tissue was obtained in compliance with the guidelines of the Joint Commission on Clinical Investigation of The Johns Hopkins University School of Medicine and the Mayo Clinic.

Tissue RNA extraction

Tissues were obtained immediately after surgical resection, frozen in liquid nitrogen, and stored at −80°C. Total cytoplasmic RNA was extracted in Trizol (all control tumors, Hp1, Hp2, Hp4, and S1; Life Technologies, Gaithersburg, MD, USA) or Stat-100 (Hp3; Tel-Test Laboratories, Inc., Houston, TX, USA) according to the recommendations of the manufacturer. PolyA+ RNA was isolated with the Oligotex messenger RNA (mRNA) Midi Kit (Qiagen Inc., Valencia, CA, USA).

SAGE analysis

SAGE analysis was performed as previously described.(8,9) Briefly, double-stranded complementary DNA (cDNA) was synthesized from each tumor RNA sample using biotinylated oligo dT primer. The cDNA was cleaved by the restriction enzyme NlaIII and the 3′ portion of the cleaved DNA was selected by binding to streptavidin beads. Then, isolated cDNA fragments were ligated via the terminal NlaIII site to linkers containing the recognition sequence for the type II enzyme BsmFI, and, subsequently, the linked cDNA was released from the streptavidin beads by digestion with BsmFI. The released fragments (SAGE tags) were ligated, amplified, and digested with NlaIII to release the SAGE ditags. The ditags were then concatenated, cloned into pZERO (Invitrogen Life Technologies, Carlsbad, CA, USA), and sequenced. The sequence and abundance of each of the transcript tags were determined using SAGE software.(10) SAGE libraries were independently sequenced to the following tag depths: C1 = 44,559; C2 = 43,386; Hp1 = 48,210; Hp2 = 43,916; Hp3 = 23,066; and S1 = 22,493. All values are reported after normalization to 40,000 tags. Sequence tags are represented by the 3′ most NlaIII site followed by a unique 10 base pair (bp) sequence. For convenience, the NlaIII recognition sequence (CATG) is not reported as part of the tag sequence. To assign a rank order of the highest to lowest of the overexpressed OOM sequences, transcripts were sorted-based on the ratio between the minimum tag count observed within the OOM tumors and the maximum tag count observed in either control sample (MIN/MAX sort).

λ-cDNA library construction

PolyA+ RNA from Hp1 and Hp2 were combined in equal amounts. A total of 5 μg of pooled RNA was used to generate an OOM λ-phage cDNA library using the ZAP-cDNA synthesis kit (Stratagene, La Jolla, CA, USA).

cDNA isolation

Partial cDNAs encompassing the 3′ end of selected candidates were isolated using three independent methods. When possible, plasmids containing expressed sequence tags (ESTs) or full-length cDNAs representing genes identified by SAGE were purchased from commercial sources (IMAGE Consortium clones from Research Genetics, Carlsbad, CA, USA) and used as templates for amplification via polymerase chain reaction (PCR) to create partial cDNAs. Sequences not found in public databases were procured by two methods. The preferred method was to isolate fragments using an anchored PCR technique directly from OOM-derived cDNA that were inserted into vectors containing a 3′ M13 priming site. Amplicons were generated using forward primers complementary to the SAGE tag and M13F1 as the reverse primer, TA cloned into vectors (Invitrogen Life Technologies), and confirmed by sequencing. In several cases, cDNAs were isolated directly from OOM λ-phage libraries using the 15-mer corresponding to the SAGE tag of interest as a probe for oligomer hybridizations.

Custom array generation and analysis

Sixty-seven candidates were selected for array analysis from the 364 SAGE-identified genes in which expression was at least twofold greater in OOM compared with control tumors. Because array analysis was initiated before receipt of tumors Hp3 and S1, the 67 target sequences represented the most highly induced genes based on a MIN/MAX sort of the Hp1 and Hp2 versus C1 and C2 SAGE data (see description of MIN/MAX sort mentioned previously). DNA target sequences representing each of these 67 genes were amplified by PCR using T3/T7 or M13 universal primers and purified using the QIAquick PCR purification kit or gel extraction kit (Qiagen, Inc.). Arrays consisted of cDNA sequences spotted in quadruplicate onto nylon membranes (Biotrans; ICN, Costa Mesa, CA, USA) at a concentration of 2 ng of DNA/spot. To synthesize radiolabeled cDNA probes, polyA+ RNA was isolated from tumor tissue as described previously and control tumor polyA+ RNA was treated with DNase I (Ambion, Inc., Austin, TX, USA). Aliquots of polyA+ RNA (70 ng) was converted simultaneously to cDNA and radiolabeled with [32P]deoxycytosine triphosphate (dCTP) using a modified reverse transcriptase reaction (GeneFilters protocol, 12/98; Research Genetics, Inc., Huntsville, AL, USA). Arrays were prehybridized in 6× SSC, 5× Denhardt's solution, 0.1% sodium dodecyl sulfate (SDS), 50% deionized formamide (Fluka, Milwaukee, WI, USA), and 0.1 mg/ml of heat-denatured salmon sperm DNA for 2 h at 42°C and then hybridized overnight at 42°C with radiolabeled probes in prehybridization solution containing 10 μg of human COT-1 DNA (Life Technologies) and 8 μg of poly dA (Research Genetics, Inc.). The arrays were washed twice in 2× SSC and 0.1% SDS at room temperature, followed by three 30-minute washes in 0.1× SSC and 0.1% SDS at 65°C. Hybridization intensities were quantified on a STORM PhosphorImager using ImageQuant software (Molecular Dynamics, Sunnyvale, CA, USA). Expression values are reported as the average of four replicate samples per gene relative to the average value of replicates for four housekeeping genes (elongation factor 1 [ EF-1], tubulin [ TUBULIN], α-actin [ ACTA], and cyclophilin [CYP]).

Reverse-trancsription PCR

Reverse-transcription (RT) PCR was performed with 1 μg of total RNA as previously described.(11) PCR was performed using 2 μl of first-strand cDNA in a 25-μl volume containing Taq polymerase buffer (1.5 mmol/liter of magnesium chloride), 1 mmol/liter of each primer, 0.1 mmol/liter of each deoxyribonucleotide triphosphate, and 2.5 U of Taq polymerase (Perkin-Elmer, Foster City, CA, USA). The following sets of sense and antisense primers were used for PCR: 5′-CCAATGACTTCAGTTTCTGTT-3′ and 5′-TAAGCTTGTCAAATTATT CTCAG-3′ for FRP-4, 5′-GTATGCCACAAGGGAAAGG-3′ and 5′-GTAATATCGGAAATTATCTT-3-′ for MEPE, 5′-AATGAGTGGATGGATGCAGGAA-3′ and 5′-AAGAAAGGCTTCTGGAGCTC-3′ for PHEX, 5′-GCTCTGGGTCTGTGCCTTGT-3′ and 5′-TCGGGAGCTCCTGTGAACAG-3′ for FGF-23, 5′-GGTGAGTGCCAAGACAGAG-3′ and 5′-TTCTAGCATTGCACGCAAAC-3′ for DMP-1, and 5′-GAGAAGGCAACCAAAGTGCAG-3′ and 5′-GAAGCACTGGATCACTTGCG GCG-3′ for GNAS1. The PCR cycling profile consisted of an initial 15 minutes of denaturation at 94°C, followed by 29 cycles of annealing (50°C for FRP-4 and GNAS1; 47°C for MEPE; 55°C for PHEX; 45 s, 60°C for FGF-23; and 55°C for DMP-1), extension (72°C, 60 s) and denaturation (94°C, 20 s), with a final 5-minute extension. PCR reactions without added reverse transcriptase were performed in parallel to exclude DNA contamination.


SAGE analysis of OOM tumors

We analyzed the gene expression of a series of mesenchymal tumors resected from patients with OOM, an acquired hypophosphatemic syndrome. In each case, hypophosphatemia was accompanied by phosphaturia with a markedly depressed TRP and an inappropriately normal or low serum calcitriol level (Table 1). Complete surgical resection of the tumors led to normalization of serum phosphorous and calcitriol concentrations (Table 1). In one case (Hp3), residual tumor was present after surgery and biochemical improvement was incomplete. Conditioned media from primary tumor cultures markedly inhibited phosphate transport in an opossum kidney cell phosphate transport assay (data not shown). The clinical and biochemical characteristics of all patients are consistent with tumor secretion of one or more phosphaturic factors.

SAGE was performed with histologically matched OOM (Hp1, Hp2, and Hp3) and control (C1 and C2) hemangiopericytomas. SAGE libraries were constructed and ∼20-40,000 tags (representing individual transcripts) were sequenced from each sample. Three hundred and sixty-four genes were identified that exhibited an at least twofold increase in expression in OOM tumors relative to control tumors. Although most differentially expressed genes were overexpressed, 285 genes were identified that were repressed twofold in OOM tumors compared with control tumors. Overall gene expression was similar between OOM and control tumors as evidenced by the relatively consistent tag numbers observed for housekeeping genes EF-1, ACTA, TUBULIN, and CYP (Table 2B).

Table Table 2.. SAGE Tags for Differentially Expressed and Housekeeping Genes
original image

Fifty-five of the most highly expressed genes are listed in Table 2. These candidates were identified by a conservative rank order sort that was based on the ratio between the minimum tag count observed within the Hp1, Hp2, and Hp3 libraries and the maximum tag count observed in either control sample. Of these 55 genes, approximately one-third coded for previously identified proteins of unknown function (GenBank) and 7 genes were novel in that sequences encoding a complete open-reading frame were not present in GenBank. The remaining overexpressed genes encoded a diverse spectrum of proteins with known functions that are located intracellularly, at the plasma membrane, or are secreted (Table 2).

Validation and prioritization of candidate genes

Because of the statistical nature of sampling, the ability to detect all expressed genes within a given SAGE library is dependent on the total depth of sequencing. Theoretically, it is possible that a small number of genes overexpressed at relatively low levels in OOM tumors were not identified by our strategy, and therefore, are not listed in Table 2. Conversely, because our sort strategy is dependent on the assumption that for each control tumor the tag number of 0 accurately reflects the absence of expression in these tumors, it also is possible that genes listed in Table 2 could include a small number of false positives. In addition, a specific 10-bp SAGE tag may map to more than one gene in a small number of instances. For these reasons, an independent method was used to verify that the assigned gene representing a given SAGE tag was indeed overexpressed in OOM tumors relative to controls.

Using additional tumor and control cDNAs, array expression profiling was performed for selected candidate genes to provide validation of the SAGE data and to extend the analysis to additional control and OOM data. The arrays were comprised of partial cDNAs encoding 67 overexpressed candidate genes and four housekeeping genes (EF-1, TUBULIN, ACTIN, and CYP). Arrays were probed with [32P]-labeled cDNAs prepared by RT of tumor RNA generated from several tumors previously analyzed by SAGE (C1, Hp1, and Hp2), four additional control hemangiopericytomas (C3, C4, C5, and C6), and two additional OOM tumor samples (Hp1B and Hp4). Relative expression of each candidate was defined as the ratio of the average expression of the gene in OOM tumors to the average expression in control tumors. Overexpression of a gene was defined as an array ratio greater than or equal to 2.0. Array analysis confirmed the SAGE profiles of several genes including MEPEand FRP-4, because they remained highly expressed in OOM tumors relative to controls in an expanded set of tumors and had array ratios >2.0 (Fig. 2 and Table 3). In contrast, many of the genes had a relative expression ratio of <2.0 because they were not expressed in additional OOM tumors and/or were substantially overexpressed in several additional control tumors. For example, PHEX, milk fat globule-EGF 8 protein (MFG-EGF8), and fibronectin (FN) exhibited equivalent or greater expression in some control tumor samples compared with OOM tumor samples (Fig. 1). Based on the array analysis, two-thirds of the candidate genes were assigned a lower priority.

Table Table 3.. Validated Differentially Expressed Genes
original image
Figure FIG. 1..

Array analysis of selected genes confirmed and narrowed candidate genes. Arrays consisted of 67 cDNA target sequences identified in the initial SAGE analysis and were probed with reverse-transcribed polyA+RNA that was isolated from OOM and control tumor tissue. Expression values are reported as the average of four replicate samples per gene relative to the average value of replicates for four housekeeping genes. The relative gene expression values for MEPE, FRP-4, SP-11, PHEX, MFG-EGF-8, and FNare shown in graphic form. Five control tumors (C1, C3, C4, C5, and C6) and four OOM tumors (Hp1a, Hp1b, Hp2, and Hp4) were used to probe the arrays. Expression patterns for MEPE, FRP-4, and SP-11were consistently overexpressed in the OOM tumors and agreed with the SAGE data (compare with Table 2). PHEXwas expressed in all the OOM tumors; however, a comparable level of expression was also observed in C4, C5, C6. MFG-EGF-8appeared to be differentially expressed by the OOM tumors in the SAGE analysis, however, the expression in two control samples appears to exceed that of all the OOM tumors and is not specific for OOM. Similarly, in the initial SAGE analysis, FNappeared to be differentially expressed in OOM tumors. The addition of more control tumors showed that elevated FNexpression was not specific to OOM tumors.

Figure FIG. 2..

Verification of SAGE by RT-PCR of selected candidate genes. cDNA prepared from total RNA from OOM tumors (Hp1, Hp2, and S1) and control tumors (C1 and C2) was amplified with primers for FRP-4, MEPE, PHEX, DMP-1, FGF-23, and the Gsα-gene (GNAS1). FRP-4, MEPE, DMP-1, and FGF-23are highly and specifically expressed in Hp1 and Hp2 and to a lesser degree in S1. In contrast, no amplification is detected in control tumors. PHEXis expressed by all OOM tumors and faint signal in the control tumors lanes is detected (compare with Table 4). The GNAS1gene is expressed ubiquitously and shows that an equal amount of cDNA was used in each reaction.

Table Table 4.. Highly Expressed Genes With Functions Related to Mineral Ion Homeostasis and Mineralization
original image

RT-PCR of selected candidate genes was performed as an additional means of validating SAGE. Similar to our SAGE profiles, FRP-4, MEPE, DMP-1, and FGF-23were expressed highly in Hp1 and Hp2, lower in S1, and not at all in control tumors (Fig. 1). PHEXmRNA was present in Hp1, Hp2, and S1 and barely detectable levels were expressed in control tumors. As expected, GNAS1is expressed at similar levels in control and OOM tumors (Fig. 2).

Using the results of array analysis and RT-PCR, the candidate genes were prioritized further to identify a subset of genes that were overexpressed in all OOM tumors. Table 3 represents a list of genes originally identified by SAGE that met the following criteria: (1) the assigned gene was shown to correctly reflect the specific SAGE tag and (2) the gene was substantially overexpressed in all OOM tumors compared with control tumors based on an array expression ratio of ≥2.0 and/or because RT-PCR revealed undetectable levels in control tumors. Highly expressed genes that did not always correlate with the OOM phenotype and had array expression ratios of <2.0 were not included in this prioritized list. Because only 67 of the original 364 candidate genes identified by SAGE were validated by array analysis or RT-PCR, it is likely that Table 3 represents only a subset of genes in which expression is always elevated in OOM tumors. However, it is unlikely that highly differentially expressed genes have not been detected using this strategy.

OOM tumors express genes implicated in bone and mineral metabolism

SAGE profiles of OOM tumors reveal that many of the most differentially expressed genes encode proteins with known functions related to bone matrix formation (DMP-1(12,13) and fibronectin(14)) and mineralization (MEPE/osteoregulin,(15)) as well as mineral ion transport (Pit-1/Glvr-1(16) and Ank-A)(17) (Table 4). In addition, we observed differential expression of several genes involved in tumorigenesis and/or inflammation (CD44,(18)osteopontin ( OPN),(19)TIAM-1,(20) and FN.(14) The results of our SAGE profiles are in close agreement with those of Shimada et al., who found through differential cDNA screening of an OOM tumor that DMP-1, MEPE/osteoregulin, FN, and CD44were among the most overexpressed genes in many OOM tumors.(21) MEPE/osteoregulin is a recently identified secreted protein that exhibits structural features of an extracellular matrix protein and is highly expressed in bone marrow and differentiated osteoblasts.(22,23) Evidence from MEPE-deficient mice that exhibit increased bone density because of enhanced bone mineralization(15) paired with evidence for overexpression of MEPEin OOM tumors associated with impaired bone mineralization signifies that MEPE is an important negative regulator of bone mineralization. Several additional candidate genes have been implicated in mineral homeostasis and regulation of bone mineralization including the type III sodium phosphate transporter Pit-1/Glvr-1 that controls osteoblast-directed mineralization( 24) and transforming growth factor (TGF) β1 that up-regulates Pit-1/Glvr-1 transcription.(24) Furthermore, aberrant activity and/or expression of candidate genes Ank-Aand SP-11, are associated with abnormal calcification of articular surfaces and osteoarthritis.(17,25)

Two genes FGF-23and PHEXhave been implicated in the pathogenesis of inherited hypophosphatemic disorders X-linked hypophosphatemic rickets (XLH) and autosomal dominant hypophosphatemic rickets (ADHR), which are clinically and biochemically similar to OOM.(26,27)PHEX, the defective gene in XLH,( 28) is expressed at relatively high levels in some hemangiopericytomas associated with OOM tumors.( 29) SAGE and array analysis showed that significant PHEXexpression was detected in two of four OOM tumors (Table 4) and in three of five control tumors (Fig. 2), indicating that elevated PHEXexpression is not specific to OOM.

Missense mutations in FGF-23, which are presumed to activate the protein, have been identified in patients with ADHR.(30) Furthermore, FGF-23mRNA was detected in several tumors associated with OOM and immunoblot analysis of whole cell extracts from one tumor showed the presence of FGF-23 protein.(31,32) FGF-23 inhibits phosphate transport in vitro in opossum kidney (OK) cell assays( 33) and promotes phosphaturia in vivo in mice.(21) However, FGF-23 may not be the only phosphaturic factor elaborated by these tumors. Recently, we have shown that FRP-4, another gene overexpressed in OOM tumors, encodes a protein that inhibits phosphate uptake in vitro.( 33) This suggests that at least two proteins can function independently or in concert to promote the phosphaturia associated with OOM.

In conclusion, we have used expression profiling to identify genes that are highly and differentially expressed in tumors associated with OOM. Several of these genes are important in bone matrix formation, mineral ion transport, and control of mineralization. Furthermore, we have identified two candidate genes FGF-23 and FRP-4 in which encoded proteins inhibit phosphate uptake in vitro. These observations underscore the possibility that the phosphaturic activity observed in OOM may require the concerted action of multiple proteins acting in a common phosphate regulatory pathway.


We are grateful to our clinical collaborators, Elizabeth Streeten, Michael Sharon, Steve Marx, and Liliana Uribe who evaluated patients with OOM and provided tumor samples. We thank Lisa Hire, Michael Phipps, Sheri Procious, Weiqun Zhang, Mindy Zhang, and the Genzyme sequencing team for expert technical assistance and Mariana Nacht, Sharon Morgenbesser, and Kathy Klinger for helpful discussions. This work was supported by grant DK02652 from the National Institutes of Health (NIH; to S.J.) and a grant from Genzyme (to R.K.).