Increased Expression of Periostin in Vitreous and Fibrovascular Membranes Obtained from Patients with Proliferative Diabetic Retinopathy

Abstract

Purpose.: Preretinal fibrovascular membranes (FVMs) form as a sequela to proliferative diabetic retinopathy (PDR), and their presence can lead to a severe decrease of vision. The purpose of this study was to determine whether periostin, a matricellular protein that plays a role in cell adhesion and migration, is associated with the formation of FVMs.

Methods.: One hundred six vitreous samples and 15 FVMs were obtained during vitrectomy on patients with PDR. Semiquantitative RT-PCR was performed to determine the periostin level of the mRNA. Immunohistochemical analyses were performed to determine the sites of periostin expression in the FVMs. ELISA was used to measure the concentrations of periostin, bFGF, and VEGF in the vitreous.

Results.: The periostin level of the mRNA was high in 10 of 10 FVMs tested but was barely detectable in the control retinas. Sequencing of the periostin PCR products revealed three splice variants of the FVMs. Immunohistochemical analysis showed colocalization of periostin and α-SMA in FVM cells. The concentration of periostin in the vitreous was significantly higher in patients with PDR than in the 31 eyes of patients with a macular hole or an epiretinal membrane (P < 0.001). Among the PDR patients, the mean vitreous level of periostin in eyes with FVMs was significantly higher than in those without FVMs (epicenter only; P < 0.001). The correlation between the vitreous concentrations of periostin and of bFGF and VEGF was not significant.

Conclusions.: These findings indicate that periostin may be involved in the development of FVMs.

Diabetic retinopathy (DR) is one of the leading causes of decreased vision and blindness in industrialized countries. At advanced stages of DR (i.e., proliferative diabetic retinopathy [PDR]), fibrovascular membranes (FVMs) form on the surface of the neuroretina. ^1,2 It has been postulated that FVMs represent a wound-healing process, and blindness can result from FVMs because they can cause intravitreal hemorrhage and tractional retinal detachment. ³  

FVMs are characterized by the migration and proliferation of various types of cells, such as retinal glial cells, macrophages/monocytes/hyalocytes, laminocytes, fibroblasts, and vascular endothelial cells. ⁴ In addition, several molecules, such as basic fibroblast growth factor (bFGF), vascular endothelial growth factor (VEGF), tumor necrosis factor-α (TNF- α), apelin, tumor endothelial marker 7 (TEM7), monocyte chemoattractant protein-1 (MCP-1), erythropoietin, angiopoietin-2, advanced glycation end product, nuclear factor-κB, and activator protein-1, have been detected in FVMs and vitreous fluid collected from patients with PDR. ^{5 –12}  

Despite improvements in vitreal surgical techniques, panretinal photocoagulation, and intravitreal anti-VEGF drugs such as ranibizumab, the prognosis for patients with DR is still poor, especially in those with advanced cases of DR at the proliferative stage. ¹³ It is, therefore, necessary to develop better treatments that are based on the pathogenesis of FVMs. Recently, we performed a comprehensive microarray study on retinas from a mouse model of oxygen-induced retinal neovascularization and determined the existence of several potential factors other than VEGF that could be additional molecular targets for antiangiogenesis therapy. ¹⁴  

In addition, we recently finished sequencing 7000 random sequences of the cDNA library of human FVMs associated with PDR and found that periostin was expressed more strongly in FVMs than in idiopathic epiretinal membranes (ERMs). ¹⁵ In support of this, Takada et al. ¹⁶ used comprehensive proteomic analysis to demonstrate that the expression of periostin was significantly higher in FVMs than in idiopathic ERMs. 

Periostin is a matricellular protein and is a member of the fasciclin family. It contains an N-terminal secretory signal peptide, followed by a cysteine-rich domain, four internal homologous repeats, and a C-terminal hydrophilic domain. ^17,18 The high degree of structural and sequence homology of periostin with fasciclin 1 and transforming growth factor β–induced suggests that periostin plays a role in cell adhesion and migration. ¹⁸ In addition, periostin is expressed as a complex pattern of transcripts derived by alternative splicing with potentially different activities and biological functions. In humans, three variants of periostin produced by alternative splicing within the C-terminal region have been registered in the GenBank. ¹⁹  

Periostin expression is altered in different diseases, including neoplasias, cardiovascular disease, and wound repair. ²⁰ Periostin is overexpressed in various human cancers such as pancreas, colon, ovary, oral squamous cell carcinoma, and lung, ²¹ and its overexpression is correlated with the aggressiveness of the tumor and with poorer survival. Tumor cell lines engineered to overexpress periostin have accelerated the growth and higher angiogenic and metastatic activity in immunocompromised animals. ^22,23 In the heart, periostin plays an important role in the progression of cardiac valve complex degeneration by inducing angiogenesis and MMP production. ²⁴ Periostin is also a component of bone marrow fibrosis and subepithelial fibrosis of bronchial asthma. ^25,26  

During our investigations of the levels of periostin mRNA expression, we found that the levels of periostin mRNA were significantly higher in the FVMs of patients with PDR than in normal retinas. This led us to hypothesize that periostin plays a distinct role in the development of FVMs. We show that periostin is involved in the development of FVMs and suggest a possible biological role of periostin in the formation of FVMs. 

Subjects and Methods

This study was approved by the ethics committees of the Kyushu University Hospital and Fukuoka University Chikushi Hospital, and the surgical specimens were handled in accordance with the Declaration of Helsinki. All patients gave informed consent before inclusion in the study. Inclusion criteria were tractional retinal detachment with active neovascularization within the FVMs, repeated vitreous hemorrhage with active neovascularization, rubeosis with vitreous hemorrhage precluding additional panretinal photocoagulation, and refractory neovascular glaucoma. Exclusion criteria were age >80 years, renal or hematologic disease, uremia, previous chemotherapy, life support, and chronic abnormalities other than diabetes. ¹¹  

At the beginning of vitrectomy, samples of undiluted vitreous fluid (0.5–1.0 mL) were aspirated under standardized conditions and were immediately transferred to sterile tubes. ¹² The samples were centrifuged for 10 minutes at 3000 rpm (1630g) at 4°C, and the supernatants were divided into aliquots and stored at −70°C until analysis. Vitreous samples were collected from 106 eyes of 84 patients (age, 54.6 ± 8.7 years; 51 men, 33 women ratio) with PDR during the initial pars plana vitrectomy. For control, vitreous samples were collected from 31 eyes of 31 patients (age, 63.1 ± 12.1 years; 15 men, 16 women) who were undergoing vitrectomy for an ERM or a macular hole (MH). The clinical characteristics of these patients are presented in Table 1. 

Table 1.

 

Clinical and Laboratory Data of Patients

Characteristic	Proliferative Diabetic Retinopathy	Nondiabetic Ocular Diseases	P
Age, y	55.7 ± 9.6	63.1 ± 12.1	0.001
Sex, n
Male	51	15	0.002
Female	33	16
Duration of diabetes, y	11.4 ± 7.6	—
Glycosylated hemoglobin, %	8.3 ± 2.7	—
Subgroups, n (%)
Panretinal photocoagulation history	90 (85)	—
Anterior chamber neovascularization	8 (8)	—
Vitreous hemorrhage	85 (80)	—
Fibrovascular membranes	77 (72)	—
Traction retinal detachment	18 (17)	—
Idiopathic macular hole	—	21 (68)
Idiopathic epiretinal membrane	—	10 (32)

 The difference in age and sex ratios between the study group and the control group is examined by Student's t-test and chi-square test, respectively.

FVMs were surgically dissected from the retinal surface with horizontal scissors in 15 eyes of 15 patients with type 2 diabetes with PDR (age, 62.5 ± 14.6 years; duration of diabetes, 15.2 ± 8.6 years) undergoing pars plana vitrectomy. ¹¹ As controls, retinal RNAs (Clontech, Palo Alto, CA) of three healthy persons were used. The removed FVMs were processed by reverse transcription–polymerase chain reaction (RT-PCR) and subcloning. The remaining five FVM specimens (age, 57.2 ± 18.9 years; duration of diabetes, 18.1 ± 12.5 years) were processed for immunohistochemistry. One normal eye was also obtained during orbital surgery and fixed immediately after enucleation in 4% paraformaldehyde for immunohistochemistry. 

Cell Cultures

Primary human retinal pigment epithelial (PHRPE) cells were obtained from Lonza Walkersville, Inc. (Walkersville, MD) and were maintained in DMEM/F12 supplemented with 10% FBS, l-glutamine, penicillin, and streptomycin. Cells were cultured in a humidified atmosphere at 37°C and 5% CO₂. 

Antisense Oligonucleotides

The splice variant-specific periostin antisense oligonucleotides used in this study were synthesized and purified as described. ^27,28 Their sequences were as follows: variant II, 5′-TGTgggtctaTAG-3′, and variants I and III, 5′-TGGcttatagACA-3′, where 2′,4′-BNA/LNA modifications were shown in uppercase letters and DNA was shown in lowercase letters. All inner nucleoside linkages are phosphorothioated. Periostin variant II antisense binds specifically to periostin variant II, which lacks exon 17, whereas periostin variants I and III antisense bind specifically to variants I and III, which lack exons 17 and 18. Scrambled oligonucleotide sequences were also used as negative controls. 

Transfection

The day before transfections, PHRPE cells were plated in a 24-well plate at a confluence of 50% to 70%. The antisense oligonucleotides (10 nM) or scrambled oligonucleotide control were mixed with 1 μL reagent (Lipofectamine 2000; Invitrogen, Carlsbad, CA) in 500 μL medium (Opti-MEM; Invitrogen). The mixture was added to the PHRPE cells and incubated for 24 hours. The composite transfection mixture was removed and replaced with FBS-free DMEM/F12 and treated with TGF-β2 (3 ng/mL; Sigma-Aldrich, St. Louis, MO). Cells remained in culture for 24 hours before they were harvested. 

RNA Extraction and Quantitative Reverse Transcription–Polymerase Chain Reaction

All the resected tissues and harvested cells for RT-PCR were snap frozen and stored at −80°C. For the preparation of total RNA, the tissue was homogenized using a bead kit (MagNA Lyser Green Beads; Roche Applied Science, Mannheim, Germany) according to the manufacturer's instructions. Total RNA was extracted from the tissue homogenate with reagent (Trizol; Qiagen, Germantown, MD) and was exposed to DNase (RNase-free DNase set; Qiagen) to eliminate potential genomic DNA contamination. Synthetic oligonucleotide primers based on the cDNA sequences of periostin, VEGF, VEGFR2, and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were prepared as follows: for periostin (expected amplicons between exon 16 and exon 22), 5′-GTGGTAGCACCTTCAAAGAAATCC-3′ (PERIOSTION16F) and 5′-GCAACTTCCTCACGGGTGTGTC-3′ (PERIOSTION22R); for GAPDH, 5′-GAGTCAACGGATTTGGTCGT-3′ and 5′-CTTGATTTTGGAGGGATCTCGC-3′; for VEGF, 5′-TGCCTTGCTGCTCTACCTCC-3′, and 5′-TCACCGCCTCGGCTTGTCAC-3′; and for VEGFR2, 5′-GATGTGGTTCTGAGTCCGTCT-3′ and 5′-CATGGCTCTGCTTCTCCTTTG-3′. 

PCR was carried out semiquantitatively, as described. ^6,29 First-strand cDNA was synthesized using Moloney murine leukemia virus reverse transcriptase (Invitrogen, San Diego, CA). RNA (1 μg) and random hexamers (100 ng) were denatured at 65°C for 10 minutes and added to the reverse transcription mixture as instructed by the manufacturer. After incubation at 37°C for 1 hour, cDNA from the reverse transcription mixture was subjected to PCR in a 10-μL volume containing 5 pmol of the primer pair and 0.5 U Platina Taq (Applied Biosystems, Foster City, CA) using a DNA thermal cycler (Applied Biosystems). The DNA was denatured for 9 minutes at 95°C followed by 35 PCR cycles. Each cycle included a 30-second denaturation at 94°C, 30-second primer annealing at 55°C, and 45-second polymerization at 72°C. An 8-μL aliquot of each RT-PCR reaction mixture was analyzed by electrophoresis on a 2% agarose gel and stained with ethidium bromide. The density of the ethidium bromide luminescence was measured by a CCD image sensor (Densitograph AE-6920M; Atto, Japan). Real-time quantitative RT-PCR (qRT-PCR) was performed on the harvested cells as described. ¹⁴  

Subcloning of Periostin PCR Products

RT-PCR was performed to amplify the region spanning exon 16 to exon 22 using PERIOSTIN16F-PERIOSTIN22R primer pairs to detect exon-specific splice variants for periostin. The amplicons were subcloned into a cloning kit (TOPO-2; Invitrogen) according to the manufacturer's protocol. ^6,30 This was followed by nucleotide sequencing using T7 primer with a cycle sequencing kit (Taq Dyedeoxy Terminator; Applied Biosystems). Sequencing reactions were resolved on an automated sequencer (ABI 3130; Applied Biosystems). ^{31 –38}  

Immunohistochemistry

Immunohistochemistry was performed essentially as described. ^{39 –41} Surgically resected FVMs were fixed in 4% paraformaldehyde in phosphate-buffered saline (PBS) and embedded in paraffin. Thick sections (3 μm) were cut and, after removal of the paraffin, were rehydrated, blocked, and incubated for 1 hour at room temperature with a monoclonal antibody against mouse monoclonal α-smooth muscle actin (SMA; 1:100 dilution; DAKO, Glostrup, Denmark) or a rat monoclonal antibody against periostin (SS5D). ²⁵ The bound antibody was made visible by a conventional avidin-biotin-peroxidase protocol with 3-amino-9-ethylcarbazole as the substrate. For negative controls, mouse nonimmune IgG was used as the primary antibody. 

Dual-Color Immunofluorescence Immunohistochemistry

Dual-color immunofluorescent staining was performed on paraffin sections ⁴² by staining with anti-periostin and mouse monoclonal anti-α-SMA antibody (1:100 dilution; DAKO). Periostin was made visible with a PE-conjugated anti-rat IgG (1:40,000 dilution; Santa Cruz Biotechnology, Santa Cruz, CA), and α-SMA by a biotin anti-mouse IgG (1:40,000 dilution; Santa Cruz Biotechnology) followed by rhodamine-streptavidin (Vector Laboratories, Burlingame, CA). Sections were examined with a fluorescence microscope (Leica, Wetzlar, Germany). 

ELISA of Periostin

Two rat anti-human periostin monoclonal antibodies (clones SS18A and SS17B) were used for sandwich ELISA. The sandwich ELISA was prepared as follows: SS18A (2 μg/mL) was incubated overnight at 25°C in 96-well plates (Thermo Fisher Scientific, Pittsburgh, PA), which were then blocked with blocking buffer (0.5% casein in Tris-buffered saline (TBS). After three phosphate-buffered saline (PBS-T; 0.05% Tween 20 in PBS) washes, diluted vitreous samples (1/100) or recombinant periostin standards were incubated overnight at 25°C. After five washes, biotin-labeled SS17B (50 ng/mL) was added, followed by incubation for 1.5 hours at 25°C. After five washes, diluted peroxidase-labeled streptavidin (1/15,000; Stereospecific Detection Technologies, Baesweiler, Germany) was added to the plates, which were then incubated for 1 hour at 25°C. The plates were washed five times and then incubated for 10 minutes with reaction solution (0.8 mM 3,3′,5,5′-tetramethylbenzidine, 2.5 mM H₂O₂). The reaction was terminated by the addition of the stop solution (0.7 N HCl). Absorbances were measured at 450 nm (primary wavelength) and 550 nm (secondary wavelength) using a microplate reader (Bio-Rad Laboratories, Hercules, CA). 

Statistical Analysis

Statistical analysis was performed using a commercial statistical software package (JMP, version 7.0; SAS Institute, Cary, NC). The distribution of the data was examined first by Shapiro-Wilk tests. The significance of the differences in the periostin levels among the different groups was analyzed with the Mann-Whitney U test. To determine whether a significant correlation existed between periostin and bFGF and VEGF, the measurements of the periostin and the bFGF and VEGF concentrations were transformed into a decadic logarithm scale, and Pearson and Spearman correlation tests were used. Two-tailed P < 0.05 was considered statistically significant. 

Results

Upregulation of Periostin in FVMs

In an earlier study, we detected the expression of the mRNA of periostin in FVMs by expressed sequence tag (EST) analysis. ¹⁵ We thus performed semiquantitative RT-PCR on the RNAs extracted from FVMs. The mRNA of periostin was detected in all 10 of the FVMs obtained from the PDR patients who were tested but was barely detected in the three normal retinas (Fig. 1, control). In addition, RT-PCR with the 16F and 22R primer pairs yielded multiple bands suggesting the presence of splice variants of periostin in the FVMs. The mRNAs of VEGF and VEGFR2 were also detected in 10 of 10 FVM specimens, but, in contrast to periostin, these angiogenic molecules were also upregulated in some of the control retinas (VEGF, 2 of 3; VEGFR2, 3 of 3). 

Figure 1.

 

RT-PCR analyses of periostin (POSTN), VEGF, VEGFR2, and GAPDH in FVMs derived from patients with proliferative diabetic retinopathy (PDR; lanes 4–13) and from control retinas (lanes 1–3). After 35 cycles, 8 μL each sample was electrophoresed through a 2% Tris-acetate-EDTA agarose gel, and the fractionated products were stained with ethidium bromide. Note the high expression of the mRNA of periostin in the FVMs derived from patients with PDR compared with control retinas.

Presence of Several Splice Variants of Periostin in FVMs

The presence of multiple bands in the RT-PCR using the 16F and 22R primers of periostin (Fig. 1) led us to search for the splice variants of the periostin gene. Three splice variants of human periostin in addition to the wild-type (WT) are registered in GenBank. ¹⁹ These variants lack exons 17, 18, and 21 in variant I (accession no. AY918092), exons 17 and 21 in variant II (accession no. AY14046), and exons 17 and 18 in variant III (accession no. D13665). ¹⁹  

Sequencing the subcloned PCR products of periostin amplified by primer pairs 16F-22R from 10 PDR specimens revealed that all three spliced variants of human periostin, in addition to WT, were present in all 10 FVMs tested (Fig. 2). 

Figure 2.

 

Nucleotide sequence showing several splice variants of periostin expressed in FVMs. Vertical line: exon boundaries. (A) Subcloned sequence of the PCR amplicon spanning exon 6 to exon 22 of the periostin cDNA from FVM contains exon 16 just before exon 19. (B) Same subcloned sequence of the amplicon also contains exon 22 after exon 20, demonstrating the existence of periostin variant I (accession no. AY918092) in the FVMs. (C) Another subcloned sequence of the amplicon spanning exon 6 to exon 22 of the periostin cDNA from FVM contains exon 16 just before exon 18. (D) Same subcloned sequence of PCR amplicon contains exon 20 just before exon 22, demonstrating the existence of periostin variant II (accession no. AY14046) in the FVMs. (E) Another subcloned sequence of the PCR amplicon spanning exon 6 to exon 22 of the periostin cDNA from FVM contains exon 16 just before exon 19. (F) The same subcloned sequence of PCR amplicon also shows transcripts consisting of exon 20 and exon 21, demonstrating the existence of periostin variant III (accession no. D13665) in the FVMs. (G) Another subcloned sequence of the amplicon spanning exon 6 to exon 22 of the periostin cDNA from FVM contains exon 17 just before exon 18, demonstrating the existence of wild-type periostin in the FVMs. (A, C, E, G) Sequencing from antisense strand is shown.

Inhibition of α-SMA Production by Antisense Oligonucleotides against Splice Variant-Specific Periostin

To examine the periostin variant-specific regulation of gene expression, we constructed antisense oligonucleotides against variants I and III of periostin or variant II of periostin. To determine whether these variant-specific periostins were specifically involved in TGF-β–induced α-SMA expression, the level of α-SMA, a marker for epithelial mesenchymal transformation, was measured in the cellular fraction after PHRPE cells were treated with antisense oligonucleotide (10 nM) directed against variants I and III of periostin or variant II of periostin for 24 hours, followed by exposure to TGF-β2 (3 ng/mL). The cells were incubated for an additional 24 hours, and the cellular fractions were collected and subjected to real-time PCR. Treatment with TGF-β2 increased the level of α-SMA by fivefold (Fig. 3). Neither the scramble nor the antisense oligonucleotide directed against variants I and III of periostin had any effect. In contrast, administration of the antisense oligonucleotide directed against variant II of periostin inhibited TGF-β2–induced the production of α-SMA by approximately 50% (P < 0.01). 

Figure 3.

 

Inhibition of α-smooth muscle actin production by periostin splice variant-specific antisense oligonucleotides. Primary human retinal epithelial cells were plated onto a 24-well plastic plate in DMEM/F12 medium containing 10% FBS. After the cells were confluent, 10 nM scramble or antisense periostin (variant I and III) or periostin (variant II) oligonucleotide was added to the medium and incubated for 24 hours. Cells were treated then with TGF-β2 (3 ng/mL). After an additional 24-hour incubation, the cellular fraction was collected, and α-smooth muscle actin expression was determined by real-time quantitative PCR. *P < 0.01, statistically significant difference compared with the value obtained with the scramble oligonucleotide.

Localization of Periostin Protein in FVMs

To determine the location of the protein of periostin in the FVMs, we stained the FVM sections with an anti-periostin monoclonal antibody and an antibody to α-SMA (Fig. 4). The periostin antibody specifically labeled the α-SMA cells in the FVMs (Figs. 4A, 4B) but barely stained any cells in the normal retina (Fig. 4D). Nonimmune IgG did not label any cellular structures (Fig. 4C). Periostin protein showed a similar expression staining pattern in all the FVM specimens from the five PDR subjects examined. 

Figure 4.

 

Periostin expression in FVMs from a 42-year-old patient with a 7-year history of diabetes (A–C) and in the normal retina from a 50-year-old control subject (D). (A, D) Immunohistochemical staining with SS5D antibody to detect sites of periostin expression. (A) Positive staining is seen in pericytes and fibroblast-like cells. (B) Anti–α-smooth muscle actin antibody was used as a control for smooth muscle cell staining. (C) Signals are not seen with negative control of FVM stained with mouse nonimmune IgG. (D) Signals are not seen in the normal retinal section. Scale bar, 100 μm.

We then double stained the FVMs with antibodies to periostin and to α-SMA. Results showed colocalization of α-SMA and periostin in the cells of FVMs (Fig. 5). 

Figure 5.

 

Double staining for periostin and α-smooth muscle actin in the FVM from a 56-year-old patient with a 15-year history of diabetes. (A) Pericytes and fibroblast-like cells are visible after specific staining with α-smooth muscle actin in the FVM. (B) Specific staining for periostin in the same section shows an identical staining pattern. (C) Double staining for periostin and α-smooth muscle actin in the same sample shows cells positive for both antibodies. The yellow staining is caused by the overlapping of the red and the green colors, indicating a colocalization of periostin with the marker of α-smooth muscle actin. Scale bar, 50 μm.

Vitreous Levels of Periostin

We next examined the amount of periostin in the 106 vitreous samples of patients with PDR collected during vitrectomy, and in the 31 vitreous samples obtained from patients during MH or ERM surgery. The concentration of periostin in the vitreous was significantly higher in the patients with PDR (9.09 ng/mL; range, 0.00–118.86 ng/mL) than in the eyes with MH (0.08 ng/mL; range, 0.0–0.74 ng/mL; P < 0.001) or ERM (0.02 ng/mL; range, 0.00–0.1 ng/mL; P < 0.001; Fig. 6). 

Figure 6.

 

Periostin levels in vitreous samples from eyes with nondiabetic (epiretinal membrane and macular hole) ocular diseases and eyes with proliferative diabetic retinopathy. ERM, epiretinal membrane. *P < 0.001.

Because the levels of periostin mRNAs were upregulated in FVMs while those in the control retinas were low (Fig. 1), we hypothesized that periostin was particularly associated with the formation of FVMs. To test this hypothesis, we subdivided the PDR patients into those with FVMs and those without FVMs (epicenter only). For the 106 eyes with PDR, the mean vitreous level of periostin was 12.73 ng/mL (range, 0.00–118.86 ng/mL) in the 77 eyes with FVMs and 1.17 ng/mL (range, 0.00–6.84 ng/mL) in the 29 eyes without FVMs (epicenter only). This difference was statistically significant (P < 0.001; Fig. 7A). 

Figure 7.

 

Intravitreous levels of periostin, bFGF, and VEGF according to the presence or absence of FVMs. (A) Intravitreous level of periostin in eyes with and without FVM. *P < 0.001. (B) Intravitreous level of bFGF in eyes with and without FVM. NS, not significant. (C) Intravitreous level of VEGF according to the presence or absence of FVM. NS, not significant.

Because bFGF is a factor that is probably involved in the generation of FVMs and because VEGF is a potent angiogenic factor closely associated with the pathogenesis of PDR, we also determined the concentration of bFGF and VEGF in the same 106 vitreous samples from the PDR eyes. The bFGF and VEGF concentrations in the 77 eyes with FVMs (3.57 pg/mL [range, 0.90–7.91 pg/mL] and 1109.56 pg/mL [range, 0.00–8240.72 pg/mL], respectively) were higher than those in 29 patients without FVMs (epicenter only) (3.43 pg/mL [range, 1.15–7.05 pg/mL] and 652.00 pg/mL [range, 0.00–2120.06 pg/mL]), although the difference was not statistically significant (P = 0.169; Figs. 7B, 7C). 

When we determined the correlation between periostin and bFGF and VEGF, there was no statistically significant correlation between the vitreous concentration of both periostin and bFGF and VEGF in the 106 eyes with PDR (r = 0.100 [P = 0.400] and r = 0.167 [P = 0.095], respectively; Spearman correlation coefficient; Fig. 8). 

Figure 8.

 

Scatter plot showing the association between the vitreous concentrations of periostin and bFGF (A) and VEGF levels (B) in patients with PDR. The correlations between the two parameters are not significant (P = 0.400 and P = 0.095, respectively).

Discussion

Our results showed that the concentration of periostin in the vitreous of patients with PDR was significantly higher than that in the vitreous of patients without PDR (Fig. 6). The concentration of periostin in the vitreous of patients with PDR was significantly correlated with the presence of FVMs but that of VEGF or bFGF was not correlated (Fig. 7). This is similar to our RT-PCR results that VEGF mRNA expression was also elevated in some of the retinas (Fig. 1). The differences in the correlations between periostin and VEGF are probably because VEGF is upregulated in the retina at an earlier stage in response to ischemia before the development of FVMs. ^14,43  

In contrast, immunohistochemical analysis of periostin in the FVMs showed staining in the α-SMA–positive cells, and these cells were probably vascular pericytes and myofibroblasts in the FVMs. In addition, periostin expression was specifically enhanced in the FVMs associated with PDR (Fig. 1). Taken together, these results indicate that periostin may play specific roles in the development or maintenance of FVMs, presumably in an autocrine fashion. 

Periostin is a secreted extracellular matrix (ECM) protein that is found in areas of normal fibrogenesis or pathologic fibrosis and that can directly interact with other ECM proteins such as fibronectin, tenascin-C, collagens I and V, and heparin. Our earlier EST analysis of FVMs demonstrated an expression of periostin, collagen, acidic cysteine-rich (SPARC), fibronectin (FN), and other cellular adhesion components. ¹⁵ These findings suggested that the cells that make up the FVM actively produce a variety of adhesion molecules that interact with periostin, and these molecules are involved in the migration and proliferation of the cells of the FVM. 

Alternative splicing events occur within the C-terminal region of periostin, which is a key region that regulates cell invasiveness and metastasis. ¹⁹ We confirmed that three spliced variants and the WT of human periostin were present in FVMs (Fig. 2) and that the periostin splice variant specifically regulated α-SMA gene expression (Fig. 3). The C-terminal region is devoid of known domains and contains few known sequence motifs. However, Hoersch et al. ⁴⁴ studied periostin and identified 13-amino acid repeat units within the C-terminal region whose secondary structure was predicted to be consecutive β strands. They suggested that these β strands may mediate binding interactions with other proteins such as FN or collagen through an extended β-zipper. Because cell-specific isoform profiles and isoform-specific biological properties of periostin have been demonstrated, ⁴⁵ the existence of the different transcripts of periostin in FVMs may be used to vary the binding properties of periostin to other ECM proteins. This can then lead to the deregulation of crucial cellular processes such as adhesion, proliferation, differentiation, and invasion. How the alternative splicing of periostin is related to the formation of FVMs awaits further studies. Such efforts may yield the basis for the development of isoform-specific molecular targeting therapeutic strategies. 

Immunohistochemical analyses with the anti-periostin antibody demonstrated that periostin was expressed in the vascular pericytes that were α-SMA positive (Figs. 4, 5). Periostin is reported to induce angiogenesis by an upregulation of the VEGF receptor, Flk-1/KDR, by endothelial cells through an integrin αVβ3-focal adhesion kinase-mediated signaling pathway. ⁴⁶ This suggests that periostin may play a role in promoting or maintaining vasculature in FVMs in a paracrine fashion. However, some earlier studies showed that periostin is produced by endothelial cells enriched from cardiac grafts, suggesting that periostin may exhibit tissue-specific pattern of expression. ⁴⁷ Additionally, periostin was also expressed in myofibroblast-like cells in the stroma of FVMs. It has been reported that a stable expression of a periostin transgene in 293T cells causes the cells to undergo fibroblast-like transformation, and the cells expressing ectopic periostin increased cell migration, invasion, and adhesion. ⁴⁸ These findings indicate that periostin-expressing myofibroblast-like cells may play a role in the invasive properties of FVMs. Together with the upregulation of the mRNA in the FVMs (Fig. 1), the increased periostin concentrations in the vitreous of the eyes with PDR (Fig. 6) might be caused by the local production of periostin, primarily from FVMs. 

Recently, several trials of anti-VEGF therapy have been performed on patients with intraocular neovascular diseases. However, undesirable side effects such as brain and retinal vein occlusion have been reported. ^49,50 This was partly attributed to a steady level of VEGF mRNA and protein expression in the normal retina, suggesting a role of VEGF in keeping normal homeostasis of the retina (Fig. 1). ^51,52 The manipulation of the VEGF pathway to inhibit pathologic neovascularization could result in unexpected disturbances of the normal homeostasis in the retina and thus should be approached carefully. ⁵³ Therefore, finding additional targets that can be treated less invasively is certainly still a goal. Because the vitreous concentrations of periostin were not significantly correlated with those of VEGF in the patients with PDR (Fig. 8), it may be inferred that periostin and VEGF do not act in a directly synchronized manner in the development of FVMs. Moreover, in contrast to VEGF, periostin is assumed to be nonfunctional in normal retinas, in keeping with the very low levels of periostin in the normal control retinas (Figs. 1, 4). These results raise the possibility that periostin might be a potential therapeutic target to regulate “disease-specific” pathways in the development of FVMs while minimizing the unfavorable side effects to the normal retina. Therefore, modulating the expression of periostin by antibodies or antisense oligonucleotides directed against the molecule could be a potential therapeutic strategy for inhibiting the development of FVMs associated with PDR. 

Footnotes

 Supported in part by grants from the Ministry of Education, Science, Sports and Culture, Japan (TI, SY), and the Japan Diabetes Foundation (SY).

Footnotes

 Disclosure: S. Yoshida, None; K. Ishikawa, None; R. Asato, None; M. Arima, None; Y. Sassa, None; A. Yoshida, None; H. Yoshikawa, None; K. Narukawa, None; S. Obika, None; J. Ono, None; S. Ohta, None; K. Izuhara, None; T. Kono, None; T. Ishibashi, None

The authors thank Mari Imamura and Aiko Kuni (Kyushu University) for their excellent technical assistance.