Author + information
- Received August 25, 2006
- Revision received December 6, 2006
- Accepted January 9, 2007
- Published online May 15, 2007.
- Joseph P. Mathew, MD⁎,⁎ (, )
- Mihai V. Podgoreanu, MD⁎,
- Hilary P. Grocott, MD⁎,
- William D. White, MPH⁎,
- Richard W. Morris, PhD⁎,
- Mark Stafford-Smith, MD⁎,
- G. Burkhard Mackensen, MD⁎,
- Christine S. Rinder, MD‖,
- James A. Blumenthal, PhD†,
- Debra A. Schwinn, MD⁎,‡,§,
- Mark F. Newman, MD⁎,
- PEGASUS Investigative Team
- ↵⁎Reprint requests and correspondence:
Dr. Joseph P. Mathew, Box 3094, Duke University Medical Center, Durham, North Carolina 27710.
Objectives We hypothesized that candidate gene polymorphisms in biologic pathways regulating inflammation, cell matrix adhesion/interaction, coagulation-thrombosis, lipid metabolism, and vascular reactivity are associated with postoperative cognitive deficit (POCD).
Background Cognitive decline is a common complication of coronary artery bypass graft (CABG) surgery and is associated with a reduced quality of life.
Methods In a prospective cohort study of 513 patients (86% European American) undergoing CABG surgery with cardiopulmonary bypass, a panel of 37 single-nucleotide polymorphisms (SNPs) was genotyped by mass spectrometry. Association between these SNPs and cognitive deficit at 6 weeks after surgery was tested using multiple logistic regression accounting for age, level of education, baseline cognition, and population structure. Permutation analysis was used to account for multiple testing.
Results We found that minor alleles of the CRP1059G/C SNP (odds ratio [OR] 0.37, 95% confidence interval [CI] 0.16 to 0.78; p = 0.013) and the SELP1087G/A SNP (OR 0.51, 95% CI 0.30 to 0.85; p = 0.011) were associated with a reduction in cognitive deficit in European Americans (n = 443). The absolute risk reduction in the observed incidence of POCD was 20.6% for carriers of the CRP1059C allele and 15.2% for carriers of the SELP1087A allele. Perioperative serum C-reactive protein (CRP) and degree of platelet activation were also significantly lower in patients with a copy of the minor alleles, providing biologic support for the observed allelic association.
Conclusions The results suggest a contribution of P-selectin and CRP genes in modulating susceptibility to cognitive decline after cardiac surgery, with potential implications for identifying populations at risk who might benefit from targeted perioperative antiinflammatory strategies.
Postoperative cognitive dysfunction (POCD), including impairments in attention, memory, language, and processing time, serves as a marker of long-term cognitive decline and is associated with a reduced quality of life despite patients’ expectations that recovery of physical status will generally improve their lives (1). Although many adverse events related to cardiac surgery have been minimized, little progress has been made in reducing POCD, occurring in as many as 53% of patients at hospital discharge and 36% at 6 weeks (2). The etiology of perioperative neurologic injury is multifactorial and includes cerebral embolism and hypoperfusion, but preliminary reports of heritability of cognitive decline (3,4) suggest that genetic factors may also modulate response to this type of neurologic injury. We therefore hypothesized that polymorphisms in candidate genes regulating biologic pathways for inflammation, cell matrix adhesion/interaction, coagulation-thrombosis, lipid metabolism, and vascular reactivity are associated with incidence of POCD after CABG surgery.
Patients enrolled in the study were part of the PEGASUS (Perioperative Genetics and Safety Outcomes Study), an ongoing Institutional Review Board-approved prospective longitudinal study at Duke University Medical Center. The present substudy targeted a cohort of patients undergoing isolated coronary artery bypass graft (CABG) surgery using cardiopulmonary bypass (CPB) between April 1994 and May 2002 in whom detailed genotyping and prospective cognitive testing were performed. Patients were excluded if they had a history of symptomatic cerebrovascular disease, psychiatric illness, renal failure, active liver disease, bleeding disorders, or less than a seventh-grade education.
Measurement of cognitive function phenotype
Cognitive function was assessed the day before surgery and at 6 weeks after surgery by investigators experienced in neuropsychologic testing and who were blinded to the genetic data. In accordance with the Consensus Statement on Assessment of Neurobehavioral Outcomes After Cardiac Surgery (5), we used a cognitive test battery comprising the following 5 instruments: Short Story module of the Randt Memory Test (6), Modified Visual Reproduction Test from the Wechsler Memory Scale (7), Digit Span subtest of the Wechsler Adult Intelligence Scale-Revised (WAIS-R) exam (8), Digit Symbol subtest of the WAIS-R (8), and the Trail Making Test (Part B) (9).
To account for correlation among cognitive test scores, we performed factor analysis on the 10 cognitive test scores from baseline, as previously described (2). We used SAS Proc Factor (SAS/Genetics version 8.02, SAS Institute, Cary, North Carolina) with a principal components method without priors. Factors were rotated by an orthogonal varimax transformation yielding uncorrelated rotated factors and independent scores representing 4 cognitive domains: 1) verbal memory and language comprehension (short-term and delayed); 2) attention, psychomotor processing speed, and concentration; 3) abstraction and visuospatial orientation; and 4) figural memory. The factor analysis was performed on baseline (preoperative) scores from all eligible patients in our ongoing prospective post-CABG cognitive database (n = 867), of which the patients in the present study were a subset. Factor weights were then used to score the patients in the present study, both at baseline and at 6 weeks. Postoperative cognitive deficit was defined as a decline from baseline score of at least 1 standard deviation for 1 or more of the 4 domain scores at 6 weeks after surgery. Baseline cognitive score was defined as the mean of the 4 preoperative factor scores. A complete description of the factor analysis, including a scree plot and a table of factor loadings, is available in the online appendix.
Candidate gene and polymorphism selection
A candidate gene set, representing pathways putatively involved in cognitive dysfunction, was selected a priori based on comprehensive analysis of existing expression studies, linkage data, neuropharmacology, population-based association studies, and expert opinion. Principal among these are inflammation, cell matrix adhesion/interaction, thrombosis, lipid metabolism, and vascular reactivity pathways. Preference was given to genes previously implicated in memory, learning, cognition, or mental retardation in both humans and experimental animal models of cognition (10). A panel of 37 specific polymorphisms in these candidate genes was identified from public databases, with emphasis on variants with demonstrated or high likelihood of functionally significant effects (Table 1).
Isolation of genomic DNA and genotype analysis
After isolation of genomic DNA from whole blood, genotyping assays were conducted by matrix assisted laser desorption/ionization time-of-flight mass spectrometry on a Sequenom MassArray system (Sequenom, San Diego, California) at Agencourt Bioscience Corporation (Beverly, Massachusetts). Primers used and polymorphism details are available at: http://anesthesia.duhs.duke.edu/pegasus/cognition/1/cognition-webTable1.htm. Genotyping accuracy of the Sequenom MassArray system was estimated at 99.6% (11). Using direct sequencing on an ABI3700 capillary sequencer (Applied Biosystems, Foster City, California), genotyping reproducibility in this study was validated to be >99% by scoring a panel of 6 polymorphisms in 100 randomly selected patients. The ACEinsertion/deletion polymorphism was genotyped by polymerase chain reaction amplification followed by size-fractionation through electrophoresis (12).
Categoric and continuous demographic characteristics were compared between POCD groups with Pearson chi-squared and Wilcoxon rank sum tests, respectively. To test for association between the 37 candidate gene polymorphisms and incidence of POCD, we used a previously described 2-stage analysis approach: marker selection by set association followed by model building (13). This method has been shown to be much more powerful than individual marker analysis when multiple genes are likely involved in a disease phenotype (14).
Allele and genotype frequencies were calculated for each polymorphism, and Hardy-Weinberg equilibrium was evaluated using an exact test. All association analyses were based on 2 genotypic classes, distinguished by the presence or absence of at least 1 copy of the minor allele (homozygote minor and heterozygote versus homozygote major). In the first stage, the set association approach (15) was used to identify a set of markers jointly associated with POCD as follows: Pearson’s chi-square statistics were computed for each of the 37 polymorphisms and ordered from largest to smallest (χ1> χ2> … χ37); sums (si; i= 1, 2, … , 10) were formed from the 10 largest single-locus statistics; the empirical significance level piassociated with each siwas evaluated based on 3,000 random permutations of the data set; the set of markers associated with the smallest piwas selected for further analysis. A series of logistic regression models was then developed to test for association between each pair of markers in the selected set (both as main effects and in interaction) and POCD. Model p values were Bonferroni corrected for multiple comparisons.
To account for covariable effects and population substructure, self-reported ethnicity, age, baseline cognitive score, and years of education were subsequently included in multiple logistic regression modeling. Interactions between genotype and these covariates were also tested. Patients representing ethnic groups other than European American (EA), African American (AA), and Native American (NA) were excluded from analyses involving race. Furthermore, the structured association method was used to control for the possibility of cryptic heterogeneity within the EA patients (16,17); multilocus genotypes from a panel of 52 unlinked bi-allelic null markers, evenly distributed across the genome, were analyzed using Structure version 2.1 software (Division of Biological Sciences, University of Chicago, Chicago, Illinois) to identify clusters of genetically similar individuals. Posterior probabilities of subpopulation membership (based on 2 putative subpopulations) were subsequently used in the logistic regression models that included age, baseline cognitive score, and years of education as covariates.
Haplotypes were inferred from unphased genotypic data using Phase version 2.0 software (Department of Statistics, University of Washington, Seattle, Washington) (18); subsequent haplotype analyses were carried out using Haplo.stats software (Division of Biostatistics, Mayo Clinic, Rochester, Minnesota) (19), restricted to haplotypes with an inferred frequency of >1%. All statistical analyses were performed using SAS and SAS/Genetics version 8.02 (SAS Institute).
Mechanistic subset analyses
Based on positive genetic associations, platelet activation was measured in a subset of 93 patients for whom serial arterial blood samples had been collected before induction of anesthesia, before aortic cross-clamp release, 10 min after cross-clamp release, at end of CPB, and at end of surgery. After incubation with saturating concentrations of monoclonal antibodies, blood samples were analyzed on a FACScan flow cytometer (Becton-Dickinson, Mountain View, California). The percentage of platelets expressing P-selectin (CD62P) was determined as a marker of platelet activation. Similarly, C-reactive protein (CRP) levels were serially measured in a subset of 239 patients from whom serum had been collected before induction of anesthesia and at 4.5, 24, and 48 h after cross-clamp removal. Immunoassays were forward immunometric assays performed by Biosite Diagnostics (San Diego, California) in 384-well microtiter plates using a Tecan Genesis RSP 200/8 Workstation (Tecan, Research Triangle Park, North Carolina).
The association between percentage of P-selectin–positive platelets and SELPgenotype and between serum CRP levels and CRPgenotype was tested using repeated measures analysis of variance (ANOVA) based on log transformation and using an unstructured covariance model. Because baseline concentrations differed between SELPgenotypes, concentration levels at subsequent times were expressed as ratio to baseline. Genotype differences at individual event times were tested using Wilcoxon rank sum tests when analysis of variance was significant at p < 0.05.
Demographic characteristics of the 677 enrolled patients are presented in Table 2.Of these, 164 (24%) did not complete 6-week testing, leaving 513 patients for the final analyses; nonreturners had a lower educational level and total CPB time, were older, and were more likely to suffer from diabetes and pulmonary and vascular disease. Allele frequencies for the single-nucleotide polymorphisms (SNPs) identified below were not different between those who did and did not return for 6-week testing. The incidence of POCD in the study sample was 35.7% (183 out of 513), and patients with a deficit were similar to those without a deficit with the exception of baseline cognition (Table 2).
Initial demographic comparisons revealed differences in cognitive deficit among racial groups (EA 34.3%, AA 39.1%, NA 50%) after accounting for age, baseline cognition, and years of education in multivariable logistic regression. Therefore, all subsequent analyses were limited to the subset of EA patients (n = 443). Minor allele frequencies for the 37 polymorphisms examined among the EA patients without cognitive deficit are presented in Table 1. Four polymorphisms had deviations from Hardy-Weinberg equilibrium and were excluded from subsequent analyses.
The marker selection process identified 5 SNPs whose sum of chi-square statistics had a minimum p value of 0.121. Genotype frequencies for these 5 SNPs in patients with and without POCD are presented in Table 3.All 10 possible pairs of 5 SNPs were subjected to logistic regression analysis for association with POCD. The smallest Bonferroni-corrected model p value (p = 0.002) among models with main effects only occurred for SNPs CRP1059G/C and SELP1087G/A. This pair of SNPs also had the smallest Bonferroni-corrected model p value (p = 0.005) among models including both main and interaction effects. The interaction effect, however, was not significant (p = 0.425).
Logistic regression in EAs identified age, baseline cognition, and years of education in addition to SNPs CRP1059G/C and SELP1087G/A to be significantly associated with POCD (model p < 0.0001) (Table 4).No interaction was found between the SNPs and covariates. Presence of the minor allele at both CRPand SELPpolymorphisms had a protective effect; the incidence of cognitive deficit was 16.7% in carriers of minor alleles at both of these loci compared with 42.9% in patients homozygous for the major allele (Fig. 1).Moreover, the absolute risk reduction in the observed incidence of POCD was 20.6% for carriers of the CRP1059C allele and 15.2% for carriers of the SELP1087A allele. The CRPand SELPSNPs were not associated with baseline cognition. Detailed analyses of cognitive responses separately by factor and genotype are available in the online appendixand at: http://anesthesia.duhs.duke.edu/pegasus/cognition/1/cognition-webAppendix.htm.
Pairwise estimates of linkage disequilibrium in EAs between 2 CRP, among 5 SELPSNPs, and between the CRPand SELPSNPs were low, suggesting no strong correlation between the POCD-associated SNPs and other genotyped SNPs. For the CRPgene, containing 2 exons and 1 intron spanning 7 kb, SNPs 1059G/C and 2147C/T exhibited r2= 0.121 (p < 0.001). For the larger SELPgene, containing 17 exons and 16 introns and spanning >50 kb, r2values between the POCD-associated SNP 1087G/A, and SNPs 1969A/G, 1902A/G, 2013G/T, and 2361A/C were 0.019, 0.011, 0.019, and 0.008, respectively, with the first and third r2values significant at p = 0.01. Sixteen unique SELPhaplotypes were inferred from the 5 common SELPSNPs, of which 10 exhibited frequencies of >1%. Tests for association between SELPhaplotypes and POCD were significant for 2 haplotypes, each bearing the A allele of SELP1087G/A. These haplotypes had frequencies below 5% and add little to the observed association of SELP1087G/A to POCD (data not shown).
In separate logistic regression models that included self-reported race in the entire sample, or subpopulation membership probabilities in EAs only, the SNP effects remained significant. However, none of the 2-way interactions between CRPand SELPSNPs with race or subpopulation membership was significant, suggesting that race and population substructure had no effect on the SNP associations described. In non-EAs, the effects of the CRPand SELPSNPs trended toward being protective but were not statistically significant.
For serum CRP levels, a significant interaction between time and CRP1059G/C genotype was seen (p = 0.011). Serum CRP levels were significantly lower at the 24-hour sampling time in patients homozygous (CC) or heterozygous (CG) for the minor allele compared with patients homozygous (GG) for the major allele (p < 0.001) (Fig. 2).Similarly, a significant genotype and time difference was seen when the effect of the SELP1087G/A genotype on the percentage of P-selectin–positive platelets was examined. Platelet activation was significantly lower upon release of the cross-clamp (p = 0.043) (Fig. 3)in patients homozygous (A/A) or heterozygous (G/A) for the minor allele compared with patients homozygous (G/G) for the major allele.
Cognitive dysfunction remains a frequent complication after cardiac surgery, occurring in approximately 36% of patients at 6 weeks after surgery. In the present study of 513 CABG patients tested with a detailed cognitive test battery before and after surgery, we report a potential genetic basis for this cognitive decline. In addition to the previously described risk factors of age, level of education and baseline cognition, we found that the risk of POCD was significantly lower in patients carrying at least 1 copy of the CRP1059C or SELP1087A alleles, with an additive effect between loci. Furthermore, we provide preliminary evidence that perioperative serum levels of CRP and platelet activation were reduced in patients with these polymorphisms, providing biologic underpinning to the observed allelic association.
Coronary artery bypass graft surgery using CPB is associated with ischemia-reperfusion injury, inducing a complex inflammatory response that impacts not only the heart but also the brain, lungs, kidneys, and gut (20). C-Reactive protein is an acute-phase reactant produced primarily in the liver in response to tissue injury or inflammation, implicated not only as a marker, but also a potential participant in the pathogenesis of inflammatory-mediated processes (21). Mean CRP concentrations have been reported to rise as much as 83-fold from the preoperative period to 72 h after CABG surgery, and variation in the extent of increase appears to be influenced by CRPgenotype (22). Recently, the relationship of patterns of SNP variation at the CRPlocus to plasma CRP levels has been extensively characterized in both EA and AA populations (23,24). The synonymous 1059G/C in exon 2 (L184L) has been associated with lower CRP levels in several studies (25–27). Zee and Ridker (26) reported significantly lower CRP levels in G/C heterozygotes than in G/G homozygotes (1.05 vs. 1.38 mg/l); Suk et al. (25) found a 29% lower baseline CRP level in carriers of the minor allele at 1059. Furthermore, 1059G/C is the only SNP that tags a haplotype associated with the lowest plasma CRP levels; in a large population sample 1059G/C decreased plasma CRP an average of 1.5 mg/l/copy (23). Therefore, it is likely that this polymorphism directly influences CRP levels, although the mechanism for a direct effect remains unclear.
The immune system and the central nervous system form a bidirectional communication network. Host defense against infection and recovery from tissue injury includes not only immune activation, but also an integrated neuroendocrine response coordinated by the central nervous system, and several proinflammatory cytokines have been identified as the signaling molecules for immune-to-brain communication (28). In a double-blind crossover study of 20 healthy male volunteers given an intravenous injection of endotoxin, significant positive correlations were found between cytokine secretion and endotoxin-induced decreased verbal and nonverbal memory functions (29). These findings are consistent with reports that memory impairment is a common adverse effect of cytokine therapy and viral infection (30,31). Inflammatory mechanisms and immune activation have been hypothesized to play a role not only in neurodegenerative conditions such as Alzheimer’s disease (32) and vascular dementia (33) but also in age-associated cognitive decline (34–36). Yaffe et al. (36) followed 3,031 well-functioning subjects for 2 years and reported that a high baseline level of serum markers of inflammation, most notably CRP, was associated with poor cognitive performance and greater risk of cognitive decline over the follow-up period. Similar associations between elevated baseline CRP levels and greater cognitive decline have been reported in patients followed for 5 to 6 years (34,35). In the present study, the lower incidence of POCD in patients with the CRP1059G/C polymorphism may be related to a lesser degree of perioperative inflammation as evidenced by the lower levels of serum CRP.
P-selectin (CD62P) is a membrane glycoprotein that is rapidly mobilized to the surface of activated platelets and endothelial cells where it mediates leukocyte-platelet and leukocyte-vascular endothelial cell adhesion, respectively. Moreover, P-selectin expression on activated platelets appears important for the formation of large stable platelet aggregates and amplification of leukocyte recruitment process (37) and may prime monocytes for tissue factor and cytokine up-regulation (38). In the setting of CPB, activation of platelets, as measured by CD62P expression, peaks 2 to 4 h after CPB and returns to baseline 18 hours after CPB (39). Through increased platelet CD62P expression, CPB results in formation of monocyte-platelet and, to a lesser extent, neutrophil-platelet conjugates (40).
The gene coding for P-selectin (SELP) has been reported to be highly polymorphic (41). The SELP1087G/A SNP results in a nonsynonymous amino acid change (S290N) in the consensus repeat (sushi) domain; this extracellular domain has been shown to be important for P-selectin binding to its ligand on leukocytes (42,43). In a study of 582 subjects with myocardial infarction (MI) and 630 age-matched control subjects, Tregouet et al. (44) reported that the risk for MI associated with the SELP1087G/A SNP was reduced but differed according to the haplotype background. In the present study of cardiac surgical patients, we found the 1087G/A variant to be associated with POCD in both genotype and haplotype analyses and provide preliminary evidence for association with lower levels of platelet activation during CPB. Although several studies revealed associations between SELPpolymorphisms and soluble P-selectin levels (45,46), which appear highly heritable (47), to our knowledge this is the first report of an association between SELPgenotype and P-selectin expression on circulating platelets. However, based on this preliminary data, we cannot rule out that 1087G/A results in altered stability of membrane-bound P-selectin or differential binding of antibodies used for flow cytometry, and further functional characterization of this SNP is required.
When interpreting positive findings from any genetic association study, several epidemiologic limitations should be considered, including inadequate sample size, poorly matched control groups, subgroup analyses, multiple testing, and population stratification (48). With regard to these concerns, strengths of our study include a relatively large sample of cardiac surgery patients (n = 513) who have undergone intensive cognitive evaluation and a prospective cohort design that reduces the selection bias inherent in case-control designs. Although serum CRP and platelet CD62P data were available only for a subset of patients, we were able to provide preliminary evidence of a genotypic effect on both CRP level and platelet activation. However, it is possible that the CRP1059G/C and SELP1087G/A SNPs are in linkage disequilibrium with other regulatory (causal) variants not within regions previously scanned for polymorphisms and therefore not included in this study. This is particularly true for the CRPSNP which is silent at the amino acid level. Further, our structured association analysis revealed no evidence of population stratification in these data. Although we did find differences in cognitive deficit among racial groups, we could not detect a genetic effect that differed between races. It is possible that a larger sample of patients with varying ethnicity may have demonstrated such an effect. The definition of “significant” cognitive change, although commonly used, also remains controversial. Our primary aim was to investigate the role of genetic variability in overall postoperative cognitive function rather than specific brain functions and independent constructs. Post hoc analyses using individual factor scores as well as the raw test scores found no significant associations between the CRPand SELPpolymorphisms and individual cognitive domains, which may only mean that genetic variations in perioperative inflammatory responses have a broad impact on cortical and subcortical function. Evaluation of other SNPs specifically involved in memory and learning may, however, yield a domain-specific effect.
In summary, using a prospective cohort study design, we found 2 candidate gene polymorphisms with additive effects associated with a reduction in the incidence of cognitive decline after cardiac surgery. Functionally, these CRPand SELPpolymorphisms were associated with reductions in serum CRP and platelet activation, respectively, raising the possibility that therapies aimed at reducing the perioperative inflammatory state may be beneficial. Moreover, using cardiac surgery as a model of neurologic injury, these results provide insight into the biologic factors modulating cognitive performance in humans and further evidence for a genetic basis of cognitive deterioration, which should translate into more precise identification of patients at risk.
For a list of the members of the PEGASUS Investigative Team, and a complete description of the factor analysis, including a scree plot and a table of factor loadings, please see the online version of this article.
Supported in part by grants AG09663 (to Dr. Newman), HL54316 (to Dr. Newman), AG17556 (to Dr. Schwinn), HL075273 (to Dr. Schwinn), and M01-RR-30 (Duke Clinical Research Centers Program) from the National Institutes of Health and grants 0256342U (to Dr. Mathew), 9951185U (to Dr. Mathew), 9970128N (to Dr. Newman), and 0120492U (to Dr. Podgoreanu) from the American Heart Association. Measurement of C-reactive protein levels was supported by Biosite Diagnostics. The first two authors contributed equally to this work. Members of the PEGASUS Investigative Team are acknowledged in the Appendix.
- Abbreviations and Acronyms
- African American
- coronary artery bypass graft
- cardiopulmonary bypass
- C-reactive protein
- European American
- Native American
- postoperative cognitive deficit
- P-selectin gene
- single-nucleotide polymorphism
- Received August 25, 2006.
- Revision received December 6, 2006.
- Accepted January 9, 2007.
- American College of Cardiology Foundation
- Newman M.F.,
- Grocott H.P.,
- Mathew J.P.,
- et al.
- Randt C.,
- Brown E.
- Wechsler D.
- Wechsler D.
- Reitan R.M.
- Gabriel S.B.,
- Schaffner S.F.,
- Nguyen H.,
- et al.
- Rigat B.,
- Hubert C.,
- Corvol P.,
- Soubrier F.
- Brull D.J.,
- Serrano N.,
- Zito F.,
- et al.
- Szalai A.J.,
- Wu J.,
- Lange E.M.,
- et al.
- Russell A.I.,
- Cunninghame Graham D.S.,
- Shepherd C.,
- et al.
- Dantzer R.,
- Wollman E.E.,
- Yirmiya R.
- Kansas G.S.
- Rinder C.S.,
- Bonan J.L.,
- Rinder H.M.,
- Mathew J.,
- Hines R.,
- Smith B.R.
- Herrmann S.M.,
- Ricard S.,
- Nicaud V.,
- et al.
- Patel K.D.,
- Nollert M.U.,
- McEver R.P.
- Ruchaud-Sparagano M.H.,
- Malaud E.,
- Gayet O.,
- Chignier E.,
- Buckland R.,
- McGregor J.L.
- Tregouet D.A.,
- Barbaux S.,
- Escolano S.,
- et al.
- Barbaux S.C.,
- Blankenberg S.,
- Rupprecht H.J.,
- et al.