Virchows Archiv

, Volume 443, Issue 1, pp 57–66

Differential gene expression in the periprosthetic membrane: lubricin as a new possible pathogenetic factor in prosthesis loosening

Authors

  • Lars Morawietz
    • Institute for Pathology, University Clinic CharitéCharité University Hospital, Humboldt University of Berlin
  • Thorsten Gehrke
    • ENDO-Klinik Hamburg
  • Lars Frommelt
    • ENDO-Klinik Hamburg
  • Petra Gratze
    • Institute for Pathology, University Clinic CharitéCharité University Hospital, Humboldt University of Berlin
  • Andreas Bosio
    • Memorec Stoffel GMBH
  • Johannes Möller
    • Institute for Pathology, University Clinic CharitéCharité University Hospital, Humboldt University of Berlin
  • Bernhard Gerstmayer
    • Memorec Stoffel GMBH
    • Institute for Pathology, University Clinic CharitéCharité University Hospital, Humboldt University of Berlin
Original Article

DOI: 10.1007/s00428-003-0818-y

Cite this article as:
Morawietz, L., Gehrke, T., Frommelt, L. et al. Virchows Arch (2003) 443: 57. doi:10.1007/s00428-003-0818-y

Abstract

About 10% of hip endoprostheses will loosen after 10 years. Prosthesis loosening is caused by two different pathomechanisms: aseptic loosening (AL) and septic loosening (SL). This study evaluated differences in gene expression in AL and SL. Eight hybridizations were performed on PIQOR cDNA arrays. Objects of the study were periprosthetic interface tissue samples from two patients with SL and three patients with AL. Tissue parts directly adjacent to the site of RNA isolation were analyzed immuno/histopathologically in order to overcome the problem of tissue heterogeneity. Thirty-three genes were found constantly differentially expressed, among which were cd11b, cd18, cd68, osteopontin and ferritin heavy-chain upregulated in AL and collagen types 1alpha-1, 3alpha-1, integrin alpha-1, thrombospondin2 and nidogen upregulated in SL. The most striking finding was the strong upregulation (from 20-fold to 323-fold) of megakaryocyte stimulating factor (msf) in all aseptic cases and one of the two septic cases, which was confirmed by real-time reverse transcription-polymerase chain reaction. In this study, msf is linked to prosthesis loosening for the first time. The upregulation in AL suggests an important pathogenetic role: the msf splice product lubricin is responsible for the lubrication of healthy joints, but its excellent lubrication ability may disturb the tight interaction between bone and prosthesis and thereby contribute to prosthesis loosening.

Keywords

Aseptic prosthesis looseningSeptic prosthesis looseningcDNA arraymsfLubricin

Introduction

Because of joint destruction through inflammatory or degenerative diseases, 1.3 million total joint replacements are performed per year world wide. One major problem of total joint replacement is prosthesis loosening. About 10% of hip replacements need revision surgery after 10–15 years [2, 14]. At revision surgery, a periprosthetic membrane that has grown between prosthesis and bone is removed. This membrane is made responsible for prosthesis loosening [3]. It is generated by two different pathomechanisms, aseptic loosening (AL) and septic loosening (SL), which can be discriminated by their histomorphological characteristics and by microbiological detection/absence of bacteria [24, 27].

Constant movement of the prosthesis leads to shedding of polyethylene, metal and cement particles, if used. These wear particles provoke a foreign body reaction, which consists of mainly macrophages and multinucleated giant cells. Also, fibroblasts take part in the formation of periprosthetic interface tissue [12, 20, 39]. T-lymphocytes occur in a low quantity, but antigen specificity and oligoclonal expansion of T-cells were shown together with the transcription of interferon gamma, suggesting a key role of T-lymphocytes in the initiation and perpetuation of periprosthetic inflammation [38]. Osteoclasts and macrophages transformed into osteoclast-like cells on the membrane secrete bone-degrading enzymes [30]. As bone matrix is absorbed, the prosthesis no longer fits tightly and loosens.

In SL, a low quantity of bacteria induces a chronic inflammation characteristic of the presence of macrophages, activated fibroblasts, granulocytes and sprouting capillaries. There are case reports on different bacteria causing a prosthesis infection that suggest that bacteremia after a procedure, e.g., a dental extraction, could lead to bacterial aggregation in the periprosthetic tissue [17, 26]. Other factors contributing to this mechanism are not yet known, so there is no consistent explanation for SL.

Although both diseases have a different etiology, they display a similar histopathology, with macrophages as the predominant cell type. The mechanism of how the periprosthetic membranes of AL and SL are generated and which pathways are responsible for bone destruction is not understood.

cDNA array technology was introduced a few years ago; it has already given new insights into the pathogenesis of various diseases by means of whole tissue analysis [36] and has been applied to AL [11]. Using this approach, we expected to detect differentially expressed genes specific for AL and SL that have not yet been considered relevant to prosthesis loosening. A comparative hybridization with RNA from AL and SL was chosen because, in general, they have a similar morphology with macrophages as the predominant cell type, but, in particular, because genes contributing to their pathomechanisms should, in part, be different, as both mechanisms have different etiologies and their morphologies differ in detail. In septic cases, prosthesis loosening does not occur without infection. As it generates only few wear particles and a slight foreign body reaction, SL should thus have a low expression of genes linked to the pathomechanisms of AL. Therefore, regarding the initiation of AL, tissue from SL may be regarded as control tissue. However, tissue from AL shows no signs of infection and may be considered a control for SL. One should also consider that genes involved in both mechanisms might be upregulated in both AL and SL and, therefore, may not be found using this approach at all. This problem derives from the fact that there is no way to obtain periprosthetic tissue from prostheses that have not loosened. In closely fitting prostheses, there is hardly any periprosthetic membrane [13], and, since there is no need for surgery, this tissue could only be acquired at necropsy, which greatly reduces the RNA quality.

Materials and methods

Patients and tissue samples

The subjects of the study were five patients with total knee arthroplasty due to osteoarthritis. Tissue samples were derived from periprosthetic tissue at revision surgery of three patients with AL and two patients with SL. In all cases, the examined tissue came from the femoral part of the prostheses. The time between primary arthroplasty and revision surgery was 4 years on average (minimum 9 months, maximum 7 years). One patient was female, four patients were male, and the mean age at revision surgery was 70.2 years (minimum 66 years, maximum 80 years) (Table 1). At revision surgery, all cases showed radiographic osteolysis around the implants.
Table 1.

Patient data. AL aseptic loosening, SL septic loosening

Patient no.

Age

Sex

Time between primary and revision arthroplasty

Loosening type

Cemented

85

66 years

Male

4 years

AL

Yes

87

70 years

Female

2 years

AL

No

88

80 years

Male

6 years

SL

Yes

91

66 years

Male

7 years

AL

No

92

69 years

Male

9 months

SL

Yes

Some tissue samples were large enough to be divided into more than one segment. The sample from patient 92 was divided into nine segments, which still provided enough material for array hybridization. One part of each segment was directly used for bacterial culturing techniques (aerobe and anaerobe cultures), another part was fixed in phosphate-buffered formalin and embedded in paraffin for immunohistochemistry and routine microscopy. The main part of each segment was snap frozen for RNA isolation and hybridization.

Standard microbiological culturing was applied for the detection of bacteria to confirm the diagnosis of SL. Cultures were positive in two cases that also showed an infiltrate with neutrophil granulocytes in standard hemalaun-eosin staining, so that SL was diagnosed.

Histopathological diagnosis included the description of the inflammatory infiltrate in the periprosthetic membrane, the amount of fibroblast or fibrocytes, respectively, and the vascularization. Furthermore, detection of wear particles was performed. Polarized light microscopy was used in defining birefringent polyethylene particles, while metal particles were identified as very small black particles (about 1–2 μm) showing only a circumferential birefringence and were easily distinguishable from polyethylene particles because of their opaque appearance without polarization. Cement particles were detected without polarization as bizarre shaped particles of various sizes. The same microscope was used (Leitz DMRBE).

RNA isolation

RNA was isolated from snap-frozen full periprosthetic tissue by applying the guanidinium thiocyanate method [6] using TRI Reagent (Sigma-Aldrich, Munich, Germany) according to the supplier's manual: 1 ml TRI Reagentwas added per 50–100 mg of mechanically pulverized tissue. The preparation was homogenized and incubated at room temperature. Per 1 ml TRI Reagent, 0.2 ml CHCl3 was added, incubated at room temperature and centrifuged for 15 min at 4°C. Isopropyl alcohol (0.5 ml) was adjoined to the aqueous phase to precipitate the RNA. After repeated incubation and centrifugation, the RNA pellet was washed with 75% EtOH. The RNA was solubilized in 100 μl H2O per 200 mg original tissue.

RNA was isolated from 200–300 mg tissue from each sample. The amount of RNA ranged from 121 μg to 351 μg, and the concentration from 0.40 μg/μl to 1.25 μg/μl. The quality of the isolation process was controlled with electropherograms and calculation of the 28S/18S RNA ratio, which was, in all samples, at least 1.5.

Labeling reaction and sample cleanup

Labeling was performed according to the PIQOR protocol (Memorec Stoffel GMBH, Cologne, Germany), which is in short: 8 μl 5x first strand buffer, 2 μl 10x hexanucleotide mix (Roche Diagnostics), 2 μl "low C dNTP" (10 mM dATP, 10 mM dGTP, 10 mM dTTP, 4 mM dCTP), 2 μl FluoroLink Cy3-dCTP (or Cy5-dCTP, respectively), 4 μl 0.1 M DTT, 1 μl RNasin (40 U), 2 μl of each control RNA 1 prior mRNA isolation and control RNA 2 prior reverse transcription and 15 μl of the isolated probe aRNA (2 μg) were combined in an RNase-free microcentrifuge tube. After vortexing, the solution was incubated at 65°C for 5 min and then cooled down to 42°C. SSII enzyme (1 μl; 200 U) was added, mixed and incubated at 42°C for 30 min. Another 2 μl of SSII enzyme was added, mixed and incubated at 42°C for 30 min. RNase H (0.5 µl) was added, and the solution was incubated at 37°C for 20 min.

For sample cleanup, we used QIAquick (Qiagen) according to the supplier's manual: labeled samples were combined with 400 μl buffer PB. The sample was applied to a QIAquick column inside a 2-ml collection tube and centrifuged for 30–60 s. After flow-through was discarded, the sample was washed with 0.75 ml buffer PE and centrifuged for 30–60 s, then the flow-through was discarded and the column again centrifuged for 1 min. DNA was eluted in a new 1.5-ml microfuge tube with 30 μl H2O pH 8. After 1 min, it was centrifuged for 1 min. The eluate was diluted to a final volume of 55 μl and added to 55 μl of 2× hybridization solution. Finally, the sample was stored at room temperature in the dark until the array pre-hybridization procedure was finished.

Array selection

A variant of the commercially available PIQOR rheumatology array (Memorec Stoffel GMBH) containing 820 immunological relevant cDNAs and 6 cDNAs from housekeeping genes was chosen for hybridization. Each glass slide chip contains four piezoelectric spotted sets of 832 cDNAs, each consisting of those 826 being examined, four positive and two negative controls. Positive controls are four DNA fragments from E. coli. In vitro transcribed E. coli RNA is added to the target to control the hybridization process. One negative control is salmon sperm DNA, one is buffer.

The examined cDNA fragments were selected from a database research and computational based addition of cDNAs with relevant sequence motifs. The fragments were 200–400 bp long and fulfilled the following criteria: absence of repetitive elements, less than 85% sequence homology to other cDNAs, and the chosen fragment covered all known splice and polyadenylation variants.

Array hybridization and evaluation

Pre-hybridization solution was heated at 98°C for 2 min, then quick spun and cooled to 42°C. Of this solution, 110 μl was applied according to manufacturers guidelines (Memorec Stoffel GMBH) using a GeneTAC hybridization station (Perkin Elmer, Langen, Germany).

After prehybridization, 110 μl of the labeled probes were pipetted onto the fixed and prehybridized array within the Gene TAC hybridization station and the hybridization process was performed overnight. After washing, the array was dried by centrifugation at 500 g for 3 min, then stored in a dust-free cassette.

Eight comparative hybridizations were performed. One array compared two regions of SL from patient 92 (intraindividual control); another array compared two SL patients, 88 and 92 (intraspecific control). These arrays should control the variability of the gene expression in one disease. Four arrays compared different areas of the periprosthetic membrane from patient 91 (AL) with different areas of tissue from patient 92 (SL); one array compared patient 85 (AL) and one array patient 87 (AL) with patient 92 (SL).

Fluorescence signal intensity was measured with a laser scanner (ScanArray 3000, General Scanning Inc., Watertown, Mass., USA).

Data analysis

Calculation of intensities was performed using ImaGene 4.1 (BioDiscovery, Marina del Rey, Ca., USA) software. Genes were regarded as detected, when the signal intensity in either the Cy3 or the Cy5 probe was at least twice as high as the intensity of the negative controls (buffer and salmon sperm). After the Cy5/Cy3 quotients for the detected genes were calculated, these quotients were normalized over the median quotient of all genes on that chip. Then, for each gene, the mean quotient of all four hybridizations on one chip was calculated.

Two-dimensional hierarchical Cluster analysis was carried out using Gene Cluster and Tree View software [7].

Immunohistochemistry

Immunostaining was performed on microsections (1–2 µm) of paraffin-embedded periprosthetic tissue. Sections were deparaffinized with xylene and cooked for 5 min in 0.01 M citrate buffer for antigen retrieval.

For the detection of T-lymphocytes, we used rabbit anti-human CD3 antibodies, B-lymphocytes were stained with mouse anti-human CD20, dendritic cells were detected with mouse anti-human CD23, macrophages/giant cells were stained with mouse anti-human CD68, blood vessel endothelials were stained with mouse anti-human CD34 and proliferating cells were detected with mouse anti-human Ki67. Staining was performed according to the standard supplier's protocol. Immunohistochemical localization was carried out with the labeled Streptavidin Biotin method using the LSAB+ System (Dako, Denmark). Fuchsin+Substrate-Chromogen-System (Dako) was used as chromogene, giving a red staining. Counterstaining was performed using hemalaun.

Additionally Giemsa staining was performed for the detection of mast cells, elastica-van-Giesson staining was used for the differentiation of fibrin and collagen and Berlin blue staining for the detection of metal debris and hemosiderin.

Reverse transcription and real-time polymerase chain reaction

Of the same total RNAs used for the microarray hybridizations, 2 μg was reverse transcribed and 40 ng of the reverse-transcription reaction product was used as template for further analysis. Transcript levels were measured by real-time quantitative reverse-transcription polymerase chain reaction (RT-PCR) using ABI 7000 SDS (PE Applied Biosystems, USA). The sequences of megakaryocyte stimulating factor (msf) forward and reverse primers are identical to those used for the generation of the cDNA probe on the microarray. The msf cDNA probe fragment spans the region from position 3084 to 3482 of the mRNA, which corresponds to parts of exon 6 and 7 from msf. These exons are known to be a constituent of the msf splice product lubricin [15]. The following primer pairs were used: msf forward 5′-AAAAGCCAACCAAAGCACC–3′ and reverse 5′-CAGTCCATCTACTGGCTTACC-3′, Primer sequences for glyceraldehyde 3-phosphate dehydrogenase (GAPDH): Forward 5′-GAGCTGAACGGGAAGCTCAC–3′ and reverse 5′-CACCACCCTGTTGCTGTAGC-3′. Real-time RT-PCR assays were performed in triplicate. Threshold cycle, which correlates inversely with the target mRNA levels, was measured as the cycle number at which the SYBRgreen emission increases above a threshold level. Microarray analysis indicated that transcript levels of GAPDH did not vary dramatically (less than twofold), therefore, GADPH was used to normalize real-time RT-PCR data. The following cycle conditions were used: 95°C for 10 min followed by 50 cycles of 95°C for 15 s and 60°C for 1 min. For each amplified product, melting curves were determined according to the supplier's guidelines (Applied Biosystems prism model 7000 SDS software) ensuring specific amplification of either msf or GAPDH in the tissues analyzed. mRNA transcript levels were expressed as fold differences between the indicated aseptic or septic tissues. For the determination of transcript levels, the "relative quantification method", using a standard curve, was used as described in the manual (Abi Prism 7700 Sequence Detection System – User Bulletin, Perkin Elmer) [19].

Results

The periprosthetic tissue samples chosen for hybridization were clearly diagnosed as AL or SL. Samples from three patients with negative microbiological culturing exhibited the typical histopathology of AL (patients 85, 87 and 91), and samples from two patients with positive culturing showed the aspect of SL (patients 88 and 92).

Histopathology

Aseptic cases

The typical dense accumulation of CD68+ macrophages and multinucleated giant cells recognizing polyethylene particles, metal particles and cement particles could be observed (Fig. 1a, c) in all aseptic cases. Tissue from patient 87 contained fewer wear particles and macrophages than tissue from patients 85 and 91. Polyethylene particles were identified as highly birefringent fragments when viewed under polarized light. They were mainly found in the cytoplasm of multi-nucleated giant cells as well as in large sized macrophages. Abundant metal particles appeared as small black particles (about 1–2 μm) in hematoxylin and eosin staining and as dense blue areas in Berlin blue staining. They were found predominantly in large-sized macrophages. Few large-sized cement particles were found in the cemented cases. Aseptic cases also exhibited a low chronic inflammatory infiltrate with some CD3+ T-cells and CD20+ B-cells. This infiltrate was strongest in sample 85, in which some mast cells could also be found (Giemsa staining). Angioneogenesis, made visible through CD34+ endothelial cells, was low in AL. In all samples some fibrin and also hemosiderin deposits could be found. Tissue sample 87 was not as typical for aseptic loosening as the samples from patient 85 and 91, as it contained a lesser amount of CD68+ macrophages/giant cells but stronger angioneogenesis. However, no infiltrate of granulocytes could be found and no bacteria detected either by microscopy or by microbiological culturing in any of the tissues. Single Ki67+ proliferating cells occurred in all tissue samples.
Fig. 1. a

Histopathology of aseptic loosening (AL) (patient 91) with macrophages and multinucleated giant cells (arrows) containing different kinds of wear particles (arrowheads polyethylene particles; HE staining, ×200). b Histopathology of septic loosening (SL) (patient 92) with granulation tissue, macrophages (arrows) and infiltrate with granulocytes (arrowheads; HE staining, ×200). c Dense accumulation of CD68+ macrophages and multinucleated giant cells (arrows) in AL (CD68 immunohistochemistry, ×200). d Loose infiltrate with CD68+ macrophages (arrows) in SL (CD68 immunostaining, ×200)

Septic cases

The septic cases exhibited the characteristic pattern of chronic granulation with focal fibrosis, edema and abundant sprouting capillaries (CD34+). The moderate inflammatory infiltrate consisted mainly of neutrophil granulocytes, whereby septic loosening could be diagnosed. Microbiology found an infection with Streptokokkus oralis in sample 88 and an infection with Enterococcus faecalis in the samples from patient 92. The infiltrate of CD68+ macrophages was low and there were far fewer wear particles in the cytoplasm of these cells than in the aseptic tissues (Fig. 1b, d). Fibrin deposits of various sizes appeared throughout the tissue. Very few Ki67+ proliferating cells were found in all tissue samples.

Hybridization

From 826 genes examined on each array, 569 different genes were detected in at least one of the eight arrays. The number of detected genes per array ranged between 290 and 510 with a mean number of 411 genes. After normalization, a mean number of 120 genes (minimum 54, maximum 172) were at least twofold differentially expressed, meaning that the normalized, mean signal intensity of the four Cy3 targets was increased twofold in the Cy5 targets or vice versa. The differential expression was risen fivefold in a mean number of 26 genes per array (minimum 5, maximum 53) (Table 2).
Table 2.

Number of detected and differentially expressed genes on eight arrays. AL aseptic loosening, SL septic loosening

Cy5-labeled target

91 (AL)

92 (SL)

92 (SL)

85 (AL)

91 (AL)

91 (AL)

91 (AL)

87 (AL)

Mean

Cy3-labeled target

92 (SL)

92 (SL)

88 (SL)

92 (SL)

92 (SL)

92 (SL)

92 (SL)

92 (SL)

 

Number of cDNAs detected

441

510

426

354

426

290

424

420

411

>Two-fold differential

131

54

66

172

101

123

151

165

120

>Five-fold differential

21

6

5

53

24

30

22

43

26

Comment

 

Intrain-dividual control

Intra-septic control

      

The two control arrays showed by far the lowest number of twofold (54 and 66, respectively) and fivefold (6 and 5, respectively) differentially expressed genes. This is well in line with the fact that these arrays did not compare AL with SL but two different sites in one patient with SL (intraindividual control) or two patients both with SL (intraspecific control). The differential expression of these genes might either be due to the heterogeneity of the periprosthetic tissue or to artifacts.

Cluster analysis

In the two-dimensional computerized cluster analysis of all hybridization experiments, array 2 (intraindividual control) and array 8 were grouped together whereas the other arrays formed a main cluster with array 3 (intraspecific control) as a separate subcluster (Fig. 2). This is in good correlation with the histomorphological findings of the tissue used in these hybridizations. As array 2 and array 3 are a comparison among SL, it makes sense that they appear in a separate cluster or subcluster, respectively. It is somewhat surprising that clustering grouped together array 2 and array 8. However, the aseptic tissue sample from patient 87 used in array 8 is not typical for AL because it contained only few multinucleated foreign body cells but a strong angioneogenesis. This correlation between histomorphology and cluster analysis demonstrates that the array data go well in line with histological findings and that major artifacts concerning RNA isolation and hybridization, for example, can be excluded.
Fig. 2.

Two-dimensional hierarchical clustering of genes detected in at least 60% of experiments (406 genes). Rows clustering of genes, columns clustering of arrays with array 2 (intraindividual control) and array 8 (aseptic patient 87 with untypical histomorphology) in one cluster and array 3 (intraspecific control) in a subcluster

Differentially expressed genes

On at least one array, 388 of 826 genes were found at least twofold differentially expressed. In four or more arrays, 103 of these were found, among which were 33 genes that were differentially expressed in at least four of the six arrays comparing AL with SL, but not in the two control arrays. These 34 genes are of greatest interest, as a differential expression in the intraindividual or intraspecific control would rather be due to the heterogeneity of the tissue. A list of these genes is given in Table 3.
Table 3.

Genes at least two-fold differentially expressed on at least four arrays comparing AL with SL and less than two-fold differentially expressed on both control arrays. AL aseptic loosening, SL, septic loosening

Overexpressed in AL

On no. of arrays

Differential expression

BCL2-related protein A1

5

2.4–8.0

CD 11b

5

3.1–4.0

CD 18

4

2.5–5.7

CD 36

5

2.2–3.7

CD 37

4

2.4–5.1

CD 52

5

7.3–13.5

CD 53

4

2.3–4.6

CD 68

5

2.3–3.5

CD 74

4

2.1–5.0

CD 83

5

2.4–7.8

CD 86

4

2.1–4.8

Complement 3a receptor 1

4

2.0–3.6

Ferritin heavy-chain

4

2.1–4.3

Immunoglobulin gamma Fc receptor IIc

4

3.5–4.5

Lysosome-associated membrane glycoprotein 2

5

5.0–31.5

Proto-oncogene tyrosine-protein kinase

5

4.4–8.6

Small inducible cytokine B 16

4

2.3–3.7

Overexpressed in SL

  

Collagen type 1 a2

4

2.6–14.3

Collagen type 3 a1

5

2.1–7.1

Collagen type 4 a1

4

2.3–4.8

Collagen type 4 a2

4

2.1–7.7

Collagen type 7 a1

4

4.0–7.1

Collagen type 8 a1

4

2.4–5.3

Collagen type 11 a1

6

3.9–50.0

Collagen type 12 a1

4

4.0–16.7

Collagen type 16 a1

4

2.7–12.5

Cadherin 5

4

2.0–2.6

Integrin alpha-1

4

2.0–5.0

Microfibril-associated glycoprotein

5

2.5–6.7

Nidogen

4

2.2–4.4

Platelet-derived growth factor receptor B

4

2.1–4.2

Regulator of G-protein signaling 3

4

3.3–9.1

Thrombospondin 2

4

2.0–4.6

The strongest overexpression in AL showed msf. The Cy5/Cy3 ratios were 27.7, 87.8, 90.1, 142.9, 188.5 and 323.0. On the intraspecific control array, a ratio of 20.0 was found, indicating some variability of the msf expression. Yet the strong overexpression in aseptic tissue is striking. Therefore, the expression of msr was further investigated using real-time RT-PCR.

Real-time RT-PCR analysis of msf expression

In order to verify the strong induction (AL versus SL comparison) of msf using a different methodology, real-time RT-PCR analysis from six tissue samples from both AL and SL was performed (patient 85, AL; patient 87, AL; samples b, I and g from patient 92, SL; patient 88, SL), which corresponded to the samples used in three microarray experiments. Although the absolute values of msf ratios obtained by the two methodologies varied somewhat, overall a good correlation of whether msf gene transcripts were overexpressed or repressed could be observed (Fig. 3).
Fig. 3.

Comparison of megakaryocyte stimulating factor ratios determined by microarray and real-time reverse-transcription polymerase chain reaction analysis. RNA from two aseptic loosening patients (85 and 87) and two septic loosening patients (92 and 88) were included for the comparative analysis

Discussion

On at least one array, 388 of 826 genes were at least twofold differentially expressed. As this is a high number of genes, it is possible that some expressions are only due to chance [23]. To find those genes that were consistently differentially expressed, we applied two criteria: as two arrays (intraindividual and intraspecific control) compare septic with aseptic tissues, the genes differentially expressed on those arrays seemed to resemble the heterogeneity of the process. So (with the exception of msf, which showed the strongest differential expression of all genes and therefore could not be ignored) we did not further investigate genes that appeared on any control array, even if they were also differentially expressed on some of the arrays comparing AL and SL. The second criterion was to investigate only genes that were differentially expressed on at least four of the six arrays comparing AL and SL. These genes are given in Table 3.

Lubricin

The strongest overexpression in aseptic cases of all genes showed msf. The differential expression showed ratios between 27.7 and 323.0. However, a ratio of 20.0 was found on the intraspecific control array. So we must assume that one SL patient expressed hardly any msf, whereas the other SL patient had a stronger expression. Examination of msf RNA levels in a large number of patients will show whether msf is suitable to differentiate between AL and SL. Only then might it be regarded as a marker gene. However, the strong overexpression in AL tissue is striking, suggesting an important pathogenetic function of msf protein for at least AL and perhaps for some SL patients. msf was first described in thrombocytopenic bone marrow transplant patients (Turner et al. 1991) and was found to have stimulating ability on megakaryocytes. It is composed of 12 exons encoding for 1404 amino acids [22]. However, the msf gene codes for two other products with possible relevance in joint pathology. Homologous to msf precursor protein is a particular cartilage superficial zone protein, which is a proteoglycan synthesized by chondrocytes and synovial lining cells [10]. The other product of the msf gene is lubricin, which is secreted by synovial fibroblasts. Lubricin consists of at least msf exons 6 through 9 [15] and is a semi-rigid rod-shaped structure made of about 800 amino acids. It has a repetitive structure and extensive O-glycosylation which is most likely responsible for its lubricating ability [16, 34]. It is the one factor responsible for lubrication of joints [33]. In vitro tests showed that its lubricating ability is equal to that of unfractioned synovial fluid [16].

To further validate the differential msf expression in periprosthetic tissue with a microarray-independent approach, we performed real-time RT-PCR analysis of selected tissue samples. The same primers for RT-PCR analysis comprising parts of exon 6 and 7 were chosen as for the msf cDNA probes present on the array. The amplified RT-PCR product encodes for the lubricin protein. Overall a good match of msf ratios derived by microarray and real-time RT-PCR analysis could be observed. The differences in the absolute ratios might, at least in part, be explained by the different normalization procedures used. The real-time RT-PCR approach confirmed that mRNA coding for lubricin is present in both aseptic and septic periprosthetic tissues. However, the tissue samples derived from patient 92 seemed to lack significant msf expression. Here, all comparative approaches either with aseptic or septic samples indicated low expression of msf transcripts (Fig. 3). Still, all other tissues analyzed independently, whether they stemmed from aseptic or septic origin, expressed high amounts of msf transcripts.

The excellent lubricating ability of the encoded glycoprotein lubricin is essential for healthy joints. However, its synthesis in periprosthetic membranes does not seem desirable. Bone and prosthesis should interact so that the shaft is tightly held by the surrounding tissue. Lubrication of this tissue would reduce the adherence between bone and prosthesis and thereby cause prosthesis loosening. Future studies focusing on the analysis of lubricin protein levels and the site of lubricin secretion inside the periprosthetic membrane will reveal whether our above-mentioned working hypothesis holds true.

Alternatively, as msf can be isolated from urine [35], clinical examinations could focus on the urine msf level in prosthesis carriers to possibly have a tool for early and non-invasive diagnosis of prosthesis loosening.

Genes resembling the cellular composition of periprosthetic tissue

For this study, whole tissue samples were used and no cell separation was performed. Although arguments have been raised against this approach [9], it has been applied with great success in other fields [36]. Still, there are limits to this method. A high RNA level of certain genes is not necessarily due to an abundant transcription, but can also be explained by the abundant presence of a certain cell type that usually expresses this gene. Fortunately, this problem can be overcome by performing a detailed immuno/histopathological examination of tissue samples. The comparison of those genes with histomorphological and immunohistochemical findings will serve as plausibility control of cDNA array analysis. By applying histopathology and using immunostaining, we determined the different amounts of lymphocytes (CD3, CD20), macrophages (CD68), granulocytes, fibrocytes and endothelial cells (CD34) in aseptic and septic tissue so that we were able to explain the differential expression of a number of genes. Histopathology assured that those tissue sections used for the arrays were characteristic for AL or SL, respectively.

Overexpression in aseptic cases

In general, aseptic tissue contains more T- and B-lymphocytes, macrophages and giant cells, and this is reflected by an overexpression of a multitude of leukocyte surface markers. CD68 is overexpressed between 2.3-fold and 3.5-fold in all arrays except for one array, in which patient sample 87 did not have the typical aseptic appearance but showed fewer macrophages and wear particles and more angioneogenesis.

Other genes whose overexpression in AL can be explained by a higher number of lymphocytes, macrophages or giant cells are encoding CD 11b, CD18, CD36, CD37, CD52, CD53, CD74, CD83, CD107b, Bcl2-related protein A1, C-X-C chemokine receptor type 4, small inducible cytokine B16 (also known as CXCL16, which is a strong transmembrane chemoattractant) [21], low affinity immunoglobulin gamma Fc region receptor II-C and complement factor 3a receptor 1 (Table 3).

Not much is known about the tissue distribution of complement factor 3a receptor 1, which is present on macrophages, platelets, granulocytes and mast cells [25]. As the density of macrophages in AL is higher than that of granulocytes in SL, the complement factor 3a receptor 1 overexpression in AL is plausible; however, the involvement of the complement system in AL is surprising.

CD 11b and CD 18 together form the complement factor 3 receptor, which occurs not only on macrophages and granulocytes but is also a marker of natural killer (NK) cells. Little is known about their role in prosthesis loosening. Case et al. (2000) found a significant increase in the relative proportion of CD16-positive NK cells in the peripheral blood at revision arthroplasty compared with primary arthroplasty in a study of 83 patients [4], whereas Savarino et al. (1999) found a significant decrease of NK cells in the blood of patients with AL relative to patients without joint replacement [31]. Further investigation seems required for a better understanding of the role of NK cells in prosthesis loosening.

Osteopontin functions include macrophage and lymphocyte recruitment as part of a nonspecific immunological response [28]. This fits the overexpression of osteopontin in AL. It is also one of the most abundant noncollagenous proteins in bone, and it plays an important role in bone formation and turnover [32]. Bone turnover occurs in both AL and SL. The overexpression of osteopontin in AL might indicate a pathogenetic contribution to AL, whereas SL might be osteopontin independent.

Very surprisingly, some expected genes with known contribution to chronic inflammation were not constantly differentially expressed, and some were not even detected. Among these were all interleukin subtypes 1 to 18, tumor necrosis factor (TNF) subtypes alpha and beta and interferon gamma. Interferon gamma was only detected on one array, TNFα and TNFβ were not detected on any array. Interleukin subtypes were detected, but not constantly differentially expressed. Regarding these data, one must say that interleukins, interferons and TNF have no key roles as mediators in prosthesis loosening.

According to this, it is surprising that ferritin heavy-chain (FTH) is overexpressed in AL. It has been shown that TNFα and interleukin 1 stimulate FTH synthesis [37], suggesting that FTH is involved in inflammatory processes. Its overexpression in AL might be explained by the fact that FTH was found to be expressed by T-helper 1 lymphocytes after stimulation with mycobacterial antigen [18]. Thus, it is possible that FTH is involved in the macrophage–T-cell interaction and the formation of multinucleated cells.

Proto-oncogene tyrosine-protein kinase, which is part of intracellular signaling pathways, is upregulated in AL, but its function is hard to interpret. Gentzsch et al. (2002) demonstrated an overexpression of nine cytokines (e.g., calgranulin A and B, interleukin 10, thymosin beta 10) in AL. None of these cytokines was found overexpressed in our present study. This can be explained by the use of a different reference tissue, which was skeletal muscle in the study of Gentzsch et al., by the use of different microarrays and by different evaluation algorithms [11].

Overexpression in septic cases

Except for the granulocytes, septic cases exhibit morphologically a lesser inflammatory infiltrate of CD68+ macrophages, but exhibit dense angioneogenesis and fibroblasts, which was proven by the immune and routine histopathological analysis. The overexpression of many different collagen transcripts resembles the fibroblast activity. This is especially true for array 4 (patient 85, AL vs patient 92, SL), in which genes encoding all collagens from 1 to 18 except for 2, 9 and 13 are overexpressed in SL. Collagen subtype genes overexpressed in at least four experiments are (alpha-1 or alpha-2 chain, respectively) collagen 1a2, 3a1, 4a1, 4a2, 7a1, 8a1, 11a1, 12a1 and 16a1. It is unclear why fibroblasts inside the periprosthetic membrane apparently produce different kinds of collagens that are normally not ubiquitously expressed. Interestingly, RNA for collagen type 2 was not detectable in the tissues analyzed. The absence of collagen type 2 was noticed before by Wooley et al. (1999) in an analysis of proteins bound to polyethylene components. They examined periprosthetic tissue of 42 explants and detected collagen type 1 and immunoglobulin in most specimens, aggrecan proteoglycans in some specimens, but collagen type 2 in no specimens [40].

The overexpression of integrin alpha-1 in SL can be explained by the higher density of mesenchymal cells (namely fibroblast). Integrins mediate cell adhesion to extracellular matrix components. A recent study of mice demonstrated the importance of collagen-integrin interaction in fracture healing [8], which is closely related to the conditions in the periprosthetic compartment where tissue damage and tissue regeneration occur permanently. Also microfibril-associated glycoprotein, which is a widespread product of mesenchymal/connective tissue cells [5], and thrombospondin 2, which has been shown to be produced by fibroblasts [1], are overexpressed in SL.

Beta platelet-derived growth factor receptor (also known as CD140b) is also overexpressed in SL. It is found in endothelial cells and can therefore be explained by the strong angioneogenesis in septic tissue [29]. The same is true for Nidogen, which is a basement membrane protein. Matrix metalloproteinase (MMP) 14, known equally as membrane-type 1 MMP, which has recently been shown to be involved in angioneogenesis [41], was also overexpressed in SL.

MMPs, enzymes responsible for extracellular matrix digestion, would be expected to be overexpressed in AL, as it is known that their secretion leads to bone loss and hence to prosthesis loosening [42]. However, no MMP subtype is consistently differentially expressed, although most appear also on the control arrays and most are, in fact, overexpressed in SL like MMP 13 (overexpressed on six arrays from 2.4-fold to 16.7-fold). This suggests a more important role for MMPs in SL than in AL.

Two genes of intracellular signaling pathway proteins are differentially expressed. However, their function regarding prosthesis loosening is hard to interpret. These genes are an encoding regulator of G-protein signaling and secreted frizzled-related sequence protein 4.

The complete list of differentially expressed genes can be downloaded from our homepage: http://www.charite.de/ch/patho/.

This study was undertaken to reveal genes relevant for the pathogenesis of prosthesis loosening. It is viewed as a preliminary search for potentially interesting genes. msf showed a strong expression in AL and one SL patient, suggesting an important pathogenetic role of lubricin. To prove definite pathomechanisms, further investigation of lubricin and other candidate genes by means of immunohistochemistry and other established methods is promising.

Acknowledgements

We thank Prof. Reinhold Schäfer and Mrs. Claudia Campisi for their lectorship. The technical assistance of Mrs. Gabriele Fernahl and Mrs. Janine Karle is gratefully acknowledged. The present work was partially supported by the SFB421 and by the Gemeinnütziger Verein ENDO-Klinik e.V., Hamburg (project W1/01, T.G., L.F., M.D., V.K.).

Copyright information

© Springer-Verlag 2003