To develop a biomarker-based model to predict osteogenic potency of human mesenchymal stem cells (MSCs) from synovial membrane and periosteum.
To develop a biomarker-based model to predict osteogenic potency of human mesenchymal stem cells (MSCs) from synovial membrane and periosteum.
MSC populations were derived from adult synovium and periosteum. Phenotype analysis was performed by fluorescence-activated cell sorting and real-time reverse transcriptase–polymerase chain reaction (RT-PCR). Telomere lengths were determined by Southern blot analysis. In vitro osteogenesis was assessed quantitatively by measurements of alkaline phosphatase activity and calcium deposits. To investigate bone formation in vivo, MSCs were seeded onto osteoinductive scaffolds and implanted subcutaneously in nude mice. Bone was assessed by histology, and the human origin investigated by in situ hybridization for human Alu genomic repeats. Quantitation was achieved by histomorphometry and real-time RT-PCR for human osteocalcin. Analysis at the single-cell level was performed with clonal populations obtained by limiting dilution. Multiple regressions were used to explore the incremental predictive value of the markers.
Periosteal MSCs had significantly greater osteogenic potency than did synovial MSCs inherent to the single cell. Bone was largely of human origin in vivo. Within the same tissue type, there was variability between different donors. To identify predictors of osteogenic potency, we measured the expression levels of osteoblast lineage genes in synovial and periosteal clonal MSCs prior to osteogenic treatment. We identified biomarkers that correlated with osteogenic outcome and developed a mathematical model based on type I collagen and osteoprotegerin expression that predicts the bone-forming potency of MSC preparations, independent of donor-related variables and tissue source.
Our findings indicate that our quality-control mathematical model estimates the bone-forming potency of MSC preparations for bone repair.
Increasing evidence supports the clinical use of stem cells for tissue repair. A major concern is the large structural and clinical variability in outcome. Such variability is at least partly due to donor-related factors and the inconsistency of stem cell preparations. This problem has been highlighted by the recent publication of contrasting results in clinical trials of cardiac cell therapy (1).
The known osteogenic capacity of mesenchymal stem cells (MSCs) (2–5) makes bone repair a natural application of MSC preparations. MSC-based approaches to bone repair would circumvent clinical complications associated with the use of autologous bone grafts, such as donor site morbidity and the long-term discomfort that accompanies bone harvest. Transplantation of allogeneic bone is also used, but is subject to disease transmission and host rejection. Proof-of-concept studies in humans support the clinical utility of MSCs for bone repair (6–8). The incorporation of MSCs into osteoinductive scaffolds accelerates the formation of bone with biomechanical properties equivalent to bone grafts in vivo (9).
Donor-associated factors and MSC preparation protocols influence the bone-forming capacity of MSCs (10). In addition, MSCs isolated from different tissues have distinct differentiation properties (11–15). The resulting variability limits standardization of MSC-based approaches to bone repair and impedes comparison of clinical study outcomes. There is, therefore, a pressing clinical need for assays that allow quantitative estimation of the bone-forming potency of MSC preparations. Such potency assays would allow development of quality controls for efficacy of MSC preparations (16, 17), a prerequisite for routine use in clinical practice.
In this study, we analyzed the variability in osteogenic potency of matched human MSC preparations from synovial membrane (SM) and periosteum and identified biomarkers that, using a mathematical formula modeled on the quantitative outcome of 2 independent assays of bone formation, predict the bone-forming potency of MSC preparations independently of donor-related variables and tissue source.
Samples of human periosteum (wet weight 10–50 mg) were harvested aseptically from the proximal medial tibia of 4 donors of various ages (24, 36, 58, and 83 years). Random biopsies of SM (wet weight 10–50 mg) were obtained aseptically from the knee joints of the same 4 donors. All specimens were obtained within 12 hours after death. Donors had no history of knee joint disorders and did not have active infections or tumors. Periosteal and synovial specimens from each donor were processed in parallel using the same protocol, as previously described (5). Cells were isolated and expanded in monolayer on plastic in growth medium (high-glucose Dulbecco's modified Eagle's medium [DMEM; Life Technologies, Gaithersburg, MD] containing 10% fetal bovine serum [selected lot from Life Technologies] and antibiotics [Life Technologies]) (5).
Cell cloning was performed by limiting dilution as described elsewhere (4). First-passage synovial and periosteal cells were suspended in growth medium and plated at a density of 0.5 cells/well in 96-well, flat-bottomed culture plates. Cell populations arising from single cells were subcultured upon reaching confluence with serial 1:4 dilution passages.
Genomic DNA was extracted from passage 7 periosteal and synovial MSC monolayers using a standard protocol, and telomere lengths were determined semiquantitatively by Southern blotting using TeloTAGGG telomere length assay, according to the recommendations of the manufacturer (Roche, Indianapolis, IN). Mean telomere length was determined as described previously (18).
Culture-expanded (passage 4 to passage 6) MSCs were used for flow cytometry at 105 cells/test. Test antibodies were as follows: phycoerythrin (PE)–conjugated low-affinity nerve growth factor receptor (LNGFR)/p75, CD106 (vascular cell adhesion molecule 1), CD146 (Muc18), CD166 (activated leukocyte cell adhesion molecule), CD73/SH3 (all from PharMingen, San Diego, CA), CD105/SH2 (Serotec, Oxford, UK), fluorescein isothiocyanate (FITC)–conjugated CD45 (Dako, Carpinteria, CA), and CD13 (Serotec). D7-FIB–PE was labeled in-house from purified D7-FIB (Serotec). Hybridoma cells B4-78 against bone and liver isoforms of alkaline phosphatase were obtained from the Developmental Studies Hybridoma Bank (Iowa City, IA). Hybridoma supernatant was produced in-house, and antibody labeling was detected using secondary goat anti-mouse FITC (Serotec). Isotype-specific negative control antibodies were purchased from Serotec. Dead cells were gated out based on propidium iodide exclusion (Sigma, St. Louis, MO). All flow cytometry data were analyzed with WinMDI, version 8 software (Scripps Research Institute, La Jolla, CA).
The in vitro osteogenesis assay was performed as described previously (5). Alkaline phosphatase activity was determined colorimetrically as described previously (4), using a commercially available kit (Thermo Electron Corporation, Waltham, MA). Protein content was determined by the Bradford method (Bio-Rad, Richmond, CA), using bovine serum albumin (Sigma) as standard. Alkaline phosphatase activity was expressed as arbitrary units per microgram of protein content. To determine calcium content, cell layers were rinsed twice with phosphate buffered saline and scraped off the dish in 0.5N HCl. The cell layers were extracted by shaking for 4 hours at 4°C, were centrifuged at 1,000g for 5 minutes, and the supernatant was used for calcium determination using a commercially available kit, according to the recommendations of the manufacturer (Thermo Electron Corporation). Total calcium was calculated from standard solutions and expressed as micrograms per microgram of protein content (determined in parallel wells).
Animal experiments were performed in compliance with the UK Home Office and the Ethics Committee for Animal Research at the Katholieke Universiteit. MSCs were seeded onto the osteoinductive Collagraft matrix (NeuColl, Campbell, CA) and implanted subcutaneously into nude mice as described previously (4). Briefly, 3-mm3 Collagraft cubes were wetted in DMEM, blotted onto a gauze compress, and immersed into 50 μl of MSC suspension (105 cells/μl in growth medium) to allow attachment of the cells to the Collagraft matrix. After 1 hour of incubation at 37°C, cells that did not attach to the Collagraft matrix but remained in suspension in the medium were counted and subtracted from the total number of seeded cells (5 million) to calculate the seeding efficiency as the percentage of cells that remained in the Collagraft matrix after 1 hour of incubation. The seeding efficiency was independent of cell type and routinely varied between 40% and 70%. The cell–Collagraft constructs were implanted subcutaneously into nude mice. At different time points postimplantation, mice were killed, and the constructs dissected. The explants were then cut in two, and one half was used for total RNA extraction and the other fixed overnight in 4% formaldehyde. Fixed samples were decalcified overnight in Decal (Serva, Heidelberg, Germany), paraffin-embedded, and sectioned (into 7-μm–thick sections) at 3 different levels.
Sections were stained with hematoxylin and eosin (H&E) and then assessed qualitatively for the presence of bone tissue and quantitatively by bone histomorphometry. For each section, the total area and the area covered with bone tissue were measured using Scion Image (Scion, Frederick, MD). The amount of bone tissue was then calculated as a percentage of the total implant area.
Explants were homogenized in TRIzol reagent (Life Technologies) using a Polytron homogenizer. Total RNA was isolated using TRIzol, according to the recommendations of the manufacturer. After DNase treatment, complementary DNA (cDNA) samples were obtained by reverse transcription of 2 μg of total RNA (ThermoScript; Life Technologies) using oligo(dT)20 as primer. Real-time quantitative PCR was carried out with SYBR Green using the Opticon real-time PCR cycler (MJ Research, San Francisco, CA). Gene expression of human cells within mouse tissue was evaluated using primers specific for human cDNA samples as described previously (4, 21, 22). Mouse bone was used to ensure specificity of the primers for human cDNA samples. Primer sequences are listed in Table 1.
|Gene||Primer sequence||Amplicon (bp)|
|Type I collagen||Forward 5′-CAGCCGCTTCACCTACAGC-3′||85|
Each assessment was performed on 2 or 3 cultures or constructs per cell source. Parametric data were analyzed using Student's t-test and are presented as the mean ± SD. Analysis of variance with Bonferroni adjustment for multiple comparisons was used to evaluate alkaline phosphatase activity over time in SM MSCs versus periosteal MSCs. Multiple regressions were used to explore the incremental predictive value of the markers, taking into account any significant interactions. Polynomial and logarithmic transformations of the markers were explored. Agreement of the measurements of calcium with the model was assessed using mixed linear regressions, to account for the effect of the donor, as well as the Bland and Altman method and intraclass correlation coefficients (ICCs), and 95% confidence intervals (95% CIs) were calculated.
We quantified the osteogenic potency of matched human MSC preparations from SM and periosteum. In vitro, passage 7 SM MSCs and periosteal MSCs were treated with osteogenic medium for 20 days and assessed for alkaline phosphatase activity at different time points. In both MSC populations, there was a significant increase in alkaline phosphatase activity, which peaked on day 8 and then returned to almost basal levels by day 20 (Figure 1A). A first important source of variability that we identified was the tissue of origin of MSCs. Indeed, at all time points except days 0 and 20, periosteal MSCs displayed significantly higher levels of alkaline phosphatase activity than did SM MSCs (Figure 1A). By day 20, calcium deposits were significantly higher in periosteal MSCs than in SM MSCs (Figure 1B). Under similar conditions, human dermal fibroblasts failed to undergo osteogenesis (data not shown). These data indicate greater osteogenic potential of periosteal MSCs than SM MSCs in vitro.
To investigate whether the difference in osteogenic potential was dependent on in vitro conditions, we examined the capacity of SM MSCs and periosteal MSCs to form bone in vivo. We used an established assay consisting of seeding cells onto osteoinductive Collagraft scaffolds (composed of hydroxyapatite, tricalcium phosphate, and type I collagen, routinely used in orthopedic clinical practice) and implanting the constructs under the skin of immunodeficient nude mice (4, 20). The seeding efficiency of SM MSCs and periosteal MSCs into Collagraft scaffolds was comparable (∼40–70%), with no correlation between MSC seeding efficiency and bone formation in vivo. Histologically, no bone was evident at 4 weeks. At 8 weeks, areas of woven bone were observed after H&E staining of both SM MSC–seeded (Figures 1C and D) and periosteal MSC–seeded scaffolds (Figures 1E and F).
In contrast, no bone was retrieved from empty scaffolds or scaffolds seeded with expanded human dermal fibroblasts (data not shown), indicating that human MSCs are necessary for bone formation in vivo under these conditions. In situ hybridization for human Alu genomic repeats demonstrated that human nuclei were present within bone areas in both SM MSC and periosteal MSC constructs (Figure 1G).
To confirm differentiation of human MSCs into an osteoblast phenotype at the molecular level, we monitored the expression of human osteocalcin (OC) by quantitative real-time RT-PCR using human-specific primers. The levels of human OC messenger RNA (mRNA), normalized to human β-actin, were significantly higher in the periosteal MSC constructs at both 4 and 8 weeks (Figure 1H), indicating that periosteal MSCs had greater osteogenic potential in vivo than did SM MSCs.
To quantify bone formation, we measured the bone area in serial sections of MSC constructs by histomorphometry, following H&E staining. At 8 weeks, the percent area occupied by bone was significantly higher in periosteal MSC constructs than in SM MSC constructs (Figure 1I). The levels of OC mRNA detected with human-specific primers were significantly higher in periosteal MSC than in SM MSC implants (Figure 1H) and correlated significantly with the percent bone area (gradient 121.4 [95% CI 112, 131]; R2 = 0.997, P = 0.000) (Figure 1J).
A second source of variability within each tissue was related to the individual donor. The coefficients of variation of OC expression levels and percent bone area in vivo were 55% and 90% for SM MSCs and 36% and 41% for periosteal MSCs, respectively. This interindividual variability was not due to unreliable measurements, since within each sample, the values of OC expression and percent bone area, obtained using 2 independent systems, displayed a high degree of correlation (Figure 1J).
Calcium deposition (normalized to protein content) after 20 days of osteogenic treatment in vitro correlated significantly with both percent bone area (gradient 0.015 [95% CI 0.01, 0.02]; R2 = 0.941, P = 0.0000) (Figure 1K) and expression levels of human OC (normalized to human β-actin; gradient 0.0001 [95% CI 0.0001, 0.0002]; R2 = 0.944, P = 0.001) (Figure 1L) at 8 weeks in vivo. This indicates that, under our experimental conditions and at the specified time points, calcium deposition (normalized to protein content) in the in vitro osteogenesis assay was sufficient to measure osteogenic outcome quantitatively.
Growth potential and cell phenotype could account for the differential bone-forming capacities of MSC populations (23). During culture expansion, the growth curves of SM MSCs and periosteal MSCs from the 4 donors tested overlapped and remained linear over at least 30 population doublings (Figure 2A). As expected, and consistent with the findings of our previous studies (4, 5), there was a progressive age-associated decline in the growth rate of MSCs, which was comparable for both SM MSCs and periosteal MSCs (Figures 2B and C). We then determined semiquantitatively the length of the telomeres of SM MSCs and periosteal MSCs at passage 7, when the osteogenesis assays were carried out. Despite undetectable telomerase activity (4, 5), the telomeres of the human MSC populations tested were long, comparable with those of an immortalized cell line, irrespective of donor age, and with no significant difference between the 2 tissue sources (Figure 2D). The mean ± SD telomere length was 11.50 ± 0.53 for SM MSCs and 11.45 ± 0.47 for periosteal MSCs (P = 0.86).
Next, we compared the phenotype of the expanded human MSCs using fluorescence-activated cell sorting. Both synovial and periosteal cell populations displayed the conventional MSC phenotype (3, 24, 25), with no significant difference in the expression levels of the cell surface markers tested (Figure 2E). Histograms are shown in Supplementary Figure 1, available on the Arthritis & Rheumatism Web site at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/. As expected, CD45, a marker of hematopoietic lineage cells, was not detected in any of the MSC samples tested. Expression of CD73, CD166, CD105, CD13, and D7-FIB was detected uniformly in all MSC samples. CD146 and CD106 displayed variable expression between donors, with no correlation with donor age. LNGFR was undetectable in expanded MSCs (Figure 2E), consistent with findings reported for MSCs from human bone marrow (24). To exclude osteoblast contamination or early osteogenic commitment of MSC populations, we analyzed the cell surface expression levels of the osteoblast lineage marker alkaline phosphatase (26, 27). Both synovial and periosteal MSCs were negative for alkaline phosphatase (Figure 2E), consistent with the notion that they were undifferentiated MSCs with no detectable commitment to the osteogenic lineage.
MSC populations were expanded as polyclonal cultures. Thus, the different osteogenic capacities of SM MSCs and periosteal MSCs could either result from variability in cell subset compositions of these possibly mixed cell populations or be inherent to the single MSC. To address this, we compared the osteogenic potency of single-cell–derived clonal populations obtained by limiting dilution from both synovium and periosteum. Of a total of 91 clones analyzed, 31 stopped growing prior to 10 population doublings. The remaining 60 clones displayed heterogeneous proliferation potential in vitro (Figure 3A).
To eliminate the bias of different growth rates and to compare like with like, we selected 8 clones (4 SM and 4 periosteal clones; 1 clone per tissue per donor) that displayed similar and efficient growth curves (Figure 3B) and similar telomere lengths (Figure 3C). Mean ± SD telomere length was 7.08 ± 0.63 for synovial clones and 7.53 ± 0.18 for periosteal clones (P = 0.23). As assessed by flow cytometry, expanded clonal populations displayed the conventional MSC phenotype (representative histograms are shown in Supplementary Figure 2, available on the Arthritis & Rheumatism Web site at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/). All selected clones were multipotent, capable of chondrogenic, osteogenic, and adipogenic differentiation in vitro (data not shown). After 22 population doublings, and while still in the linear phase of their growth curves, the clonal cell populations were assessed quantitatively for their capacity to undergo osteogenesis in vitro. Periosteal clones displayed a significantly greater osteogenic potential than did SM clones, as determined by alkaline phosphatase activity (Figure 3D) and calcium deposition (Figure 3E), indicating that the superior osteogenic potency of periosteal MSCs is intrinsic to the single multipotent MSC.
To investigate the molecular basis of this differential osteogenic potency and to identify potential predictors of osteogenic outcome, we examined the expression levels of osteoblast lineage genes in both SM MSC and periosteal MSC clonal populations prior to osteogenic treatment. The expression levels of type I collagen, osteoprotegerin (OPG), osteonectin, and osteopontin, normalized to GAPDH, were significantly higher in the periosteal clones than in the synovial clones, whereas there was no significant difference in the expression levels of runt-related transcription factor 2, osterix, alkaline phosphatase, bone sialoprotein, or OC (Figure 3F). The basal expression levels of osteoblast lineage genes in clonal MSCs were log scales lower than in periosteal MSCs after full osteogenic differentiation in the Collagraft scaffold in vivo (Figure 3F), indicating that prior to osteogenic treatment, neither synovial nor periosteal clonal MSCs displayed an osteoblast phenotype.
Capitalizing on these results, we endeavored to identify predictors of osteogenic potency of MSCs that would be independent of the tissue source and of the donor. To this end, we evaluated the correlations between the basal gene expression levels in the 8 selected (4 synovial and 4 periosteal) clonal MSCs prior to osteogenic treatment and the values of calcium deposition at 20 days of osteogenic treatment in vitro (Table 2). In a multiple linear regression model (R2 = 98%; adjusted 98%) of the individual markers studied, type I collagen proved to be the most important predictor, followed by OPG. Although the bivariate correlation between calcium and OPG was moderate (R2 = 0.36), a curve fitting exercise indicated a strong quadratic trend (R2 = 0.91), thus confirming OPG as a valid predictor. Calcium increased on average by 0.43 (95% CI 0.3, 0.53; P = 0.000) for each unit increase in type I collagen and by 0.012 (95% CI 0.001, 0.02; P = 0.04) for each unit increase in OPG. The derived model was as follows:
There was no interaction with the type of clonal MSC population, indicating that this model can predict osteogenic outcome regardless of the tissue of origin of MSCs.
|Type I collagen||0.973||0.000|
The model showed excellent agreement with the actual calcium scale (Figure 4A). The Bland and Altman plot (Figure 4B) proved that the bias was independent of the level of calcium assessed (the ICC was 98%). The estimate of the average discrepancy in relation to the actual calcium scale was 0.01, with an SD of 0.09, yielding 95% limits of agreement of −0.07 (95% CI −0.08, −0.06) to 0.09 (95% CI 0.08, 0.10).
Several studies have demonstrated the therapeutic utility of MSCs in humans (6–8, 28–31). However, donor-related variability and inconsistency of MSC preparations limit standardization and impede comparisons of clinical study outcomes. In this proof-of-concept study, we proposed a strategy for tackling this problem. We developed a simple mathematical model based on the expression of only 2 biomarkers, type I collagen and OPG, which predicts the osteogenic potency of adult human MSC preparations from synovium and periosteum, independently of donor-related variables and tissue source.
The mathematical model was obtained using single-cell–derived clonal populations. As with clonal populations, periosteal polyclonal MSCs displayed a significantly greater osteogenic potency than did synovial polyclonal MSCs. However, within each tissue source of MSCs, there was no statistically significant difference in the osteogenic outcome between polyclonal and clonal cell populations. In spite of this, when we applied the mathematical model to polyclonal synovial or periosteal MSC populations, it yielded an ICC of 25%, thus showing low predictive power. In addition, it was not possible to generate by regression analysis any model predictive of the osteogenic outcome of polyclonal cell populations, at least with the molecular markers tested. A possible explanation is that in polyclonal mixed cell populations, noncompetent, nonclonogenic cells would influence the overall gene expression prior to osteogenic treatment, but are subsequently lost during selective osteogenic conditions. In support of this hypothesis, in the osteogenic cultures a large proportion of cells underwent apoptosis within the first few days of treatment, resulting in the selection of a subpopulation that underwent osteogenesis (data not shown).
As reported with single-cell–derived colonies of bone marrow–derived stromal cells (32–34), synovial and periosteal clonal populations became heterogeneous during culture expansion, as determined by flow cytometry for the cell surface markers tested (Supplementary Figure 2). Nonetheless, the growth curves and the osteogenic outcomes of individual clonal populations were distinct, and it was possible to predict their osteogenic potential using molecular markers. The phenotypic heterogeneity within each clonal population may be due to the different functional status of the individual cells of the single-cell progeny (e.g., different phases of the cell cycle, symmetric or asymmetric cell division, or different stages of differentiation). However, in spite of the phenotypic heterogeneity during culture expansion, clones would maintain distinct, possibly inherent biologic features. The coexistence, in different proportions, of clones with diverse osteogenic potential within the same polyclonal population would result in a variability too complex to be predicted by our marker analysis.
We are aware that the use of clonal populations in the clinic would be impractical and costly, and therefore, we do not advocate the use of our model for quality control of MSC populations for clinical use with the current technologies. However, our biomarker-based model represents a biologic platform upon which to devise strategies for the prospective purification of MSC subpopulations with predictable and consistent bone-forming potency independently of the donor and tissue source. Such purification, which could be either from fresh synovial/periosteal tissue digests or from expanded polyclonal cell populations, would not only increase the consistency of MSC preparations for clinical use, but also facilitate the estimation of their potency by using biomarkers, thus allowing standardization of therapeutic protocols using MSCs.
Periosteal MSCs formed more bone in vivo than did SM MSCs, as determined by histomorphometry. The bone tissue retrieved appeared to be largely of human origin. Although we do not exclude contribution of the mouse host to bone formation in vivo, the difference in bone amount is unlikely to be solely due to periosteal MSCs having a greater capacity for recruiting mouse host osteoprogenitor cells, since the levels of OC mRNA detected with human-specific primers were significantly higher in periosteal MSC implants than in SM MSC implants and correlated significantly with the percent bone area. Notably, no bone was retrieved from empty scaffolds or scaffolds seeded with expanded human dermal fibroblasts, indicating that human MSCs are necessary for bone formation in vivo under these experimental conditions.
The development of a biomarker-based model that predicts the bone-forming potency of human MSC preparations is of considerable clinical relevance. It may facilitate the selection of individuals that qualify for an MSC-based bone repair approach. It may also help identify the best source and preparation protocol of human MSCs or adjust the dose in bone-forming MSC units from patient to patient. It remains to be investigated whether the same formula can be applied to MSC preparations from other tissue sources, such as bone marrow, or to any other stem cell preparations, whether embryonic, fetal, or adult. We predict that, in addition to the properties intrinsic to the cell preparation, other factors such as inflammation, biomechanics, and patient comorbidity will influence orthotopic bone formation and clinical outcome. Preclinical studies and clinical trials of bone repair using MSC-based protocols will thus be necessary for prospective validation of our potency assays and related model.
The generation of consistent MSC-based therapeutic protocols is a prerequisite for translating these technologies into routine clinical application. There is evidence that MSCs from mismatched donors could be poorly immunogenic in recipients in some in vivo systems (35, 36). Also, the induction of tolerance to allogeneic stem cells is a field of intense investigation (37). Once the immunologic barriers are overcome, potency assays and related quality control measures will provide a solid basis for establishing batches of certified stem cell products with specific clinical indications for allogeneic transplantation. This approach will increase consistency and decrease costs of MSC therapies, thus accelerating translation to routine clinical use.
Dr. De Bari had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study design. De Bari, Dell'Accio, Fisk, Luyten, Pitzalis.
Acquisition of data. De Bari, Dell'Accio, Karystinou, Guillot, Jones, McGonagle, Khan, Archer.
Analysis and interpretation of data. De Bari, Dell'Accio, Guillot, Fisk, Jones, Khan, Archer, Mitsiadis, Donaldson.
Manuscript preparation. De Bari, Dell'Accio, Guillot, Fisk, Jones, McGonagle, Khan, Mitsiadis, Donaldson, Luyten, Pitzalis.
Statistical analysis. De Bari, Dell'Accio, Jones, Khan, Donaldson.
We thank the staff of the mortuary at the University Hospitals, Katholieke Universiteit Leuven.