Altered brain-derived neurotrophic factor blood levels and gene variability are associated with anorexia and bulimia

Authors

  • J. M. Mercader,

    1. Genes and Disease Program, Center for Genomic Regulation, Barcelona Biomedical Research Park, Barcelona, Catalonia
    Search for more papers by this author
  • M. Ribasés,

    1. Genes and Disease Program, Center for Genomic Regulation, Barcelona Biomedical Research Park, Barcelona, Catalonia
    2. Department of Psychiatry, Hospital Universitari Vall d’Hebron, Universitat Autonoma de Barcelona, Catalonia
    Search for more papers by this author
  • M. Gratacòs,

    1. Genes and Disease Program, Center for Genomic Regulation, Barcelona Biomedical Research Park, Barcelona, Catalonia
    2. National Center of Genotyping (CEGEN), Barcelona Node, Barcelona, Catalonia
    Search for more papers by this author
  • J. R. González,

    1. Genes and Disease Program, Center for Genomic Regulation, Barcelona Biomedical Research Park, Barcelona, Catalonia
    2. National Center of Genotyping (CEGEN), Barcelona Node, Barcelona, Catalonia
    Search for more papers by this author
  • M. Bayés,

    1. Genes and Disease Program, Center for Genomic Regulation, Barcelona Biomedical Research Park, Barcelona, Catalonia
    2. National Center of Genotyping (CEGEN), Barcelona Node, Barcelona, Catalonia
    Search for more papers by this author
  • R. De Cid,

    1. Genes and Disease Program, Center for Genomic Regulation, Barcelona Biomedical Research Park, Barcelona, Catalonia
    2. National Center of Genotyping (CEGEN), Barcelona Node, Barcelona, Catalonia
    Search for more papers by this author
  • A. Badía,

    1. Department of Psychiatry, University Hospital of Bellvitge, L’Hospitalet de Llobregat, Catalonia
    Search for more papers by this author
  • F. Fernández-Aranda,

    1. Department of Psychiatry, University Hospital of Bellvitge, L’Hospitalet de Llobregat, Catalonia
    Search for more papers by this author
  • X. Estivill

    Corresponding author
    1. Genes and Disease Program, Center for Genomic Regulation, Barcelona Biomedical Research Park, Barcelona, Catalonia
    2. National Center of Genotyping (CEGEN), Barcelona Node, Barcelona, Catalonia
    3. Experimental and Health Sciences Department, Pompeu Fabra University, Barcelona, Catalonia, Spain
      *X. Estivill, Genes and Disease Program, Center for Genomic Regulation (CRG), Passeig Maritim, 37-49, 08003 Barcelona, Catalonia, Spain. E-mail: xavier.estivill@crg.es
    Search for more papers by this author

*X. Estivill, Genes and Disease Program, Center for Genomic Regulation (CRG), Passeig Maritim, 37-49, 08003 Barcelona, Catalonia, Spain. E-mail: xavier.estivill@crg.es

Abstract

Murine models and association studies in eating disorder (ED) patients have shown a role for the brain-derived neurotrophic factor (BDNF) in eating behavior. Some studies have shown association of BDNF −270C/T single-nucleotide polymorphism (SNP) with bulimia nervosa (BN), while BDNF Val66Met variant has been shown to be associated with both BN and anorexia nervosa (AN). To further test the role of this neurotrophin in humans, we screened 36 SNPs in the BDNF gene and tested for their association with ED and plasma BDNF levels as a quantitative trait. We performed a family-based association study in 106 ED nuclear families and analyzed BDNF blood levels in 110 ED patients and in 50 sib pairs discordant for ED. The rs7124442T/rs11030102C/rs11030119G haplotype was found associated with high BDNF levels (mean BDNF TCG haplotype carriers = 43.6 ng/ml vs. mean others 23.0 ng/ml, P = 0.016) and BN (Z = 2.64; P recessive = 0.008), and the rs7934165A/270T haplotype was associated with AN (Z =−2.64; P additive = 0.008). The comparison of BDNF levels in 50 ED discordant sib pairs showed elevated plasma BDNF levels for the ED group (mean controls = 41.0 vs. mean ED = 52.7; P = 0.004). Our data strongly suggest that altered BDNF levels modulated by BDNF gene variability are associated with the susceptibility to ED, providing physiological evidence that BDNF plays a role in the development of AN and BN, and strongly arguing for its involvement in eating behavior and body weight regulation.

Anorexia nervosa (AN) and bulimia nervosa (BN) are complex psychiatric conditions in which genetic and environmental factors are involved (Bulik & Tozzi 2004; Fernandez-Aranda et al. in press; Kaye et al. 2000). A good candidate gene to participate in the pathophysiology of eating disorders (ED) is the gene for the brain-derived neurotrophic factor (BDNF), a gene with a role in satiety, appetite and weight regulation. BDNF encodes a neurotrophic factor with an essential role in neuronal development and synaptic plasticity (Anderson et al. 1995; Thoenen 1995). The administration BDNF in the central nervous system (CNS) induces weight loss and appetite suppression, while BDNF knockout mice develop obesity and hyperphagia (Kernie et al. 2000; Lyons et al. 1999; Pelleymounter et al. 1995; Rios et al. 2001). This obese phenotype has also been observed in mice with a reduced hypothalamic expression of neurotrophin receptor type 2 (NTRK2), the BDNF high-affinity receptor (Xu et al. 2003). Finally, in mice, 48 h of fasting reduces BDNF levels in the arcuate nucleus, while brain BDNF levels are increased by chronic dietary restriction (Duan et al. 2001; Lee et al. 2002; Xu et al. 2003).

Levels of BDNF in plasma or serum have been analyzed for several psychiatric disorders, such as schizophrenia, panic disorder, depression and ED, and in most of these disorders, including ED, altered blood BDNF levels have been found (Karege et al. 2005; Monteleone et al. 2004, 2005; Nakazato et al. 2003; Neumeister et al. 2005; Tan et al. 2005).

Although sources for plasma BDNF are not clearly defined yet, potential sources of plasma BDNF include vascular endothelial and smooth muscle cells (Braun et al. 1999; Donovan et al. 1995; Nakahashi et al. 2000), activated macrophages or lymphocytes (Braun et al. 1999; Gielen et al. 2003; Kerschensteiner et al. 1999) and neurons and glia cells from the CNS (Karege et al. 2002b; Pan et al. 1998; Phillips et al. 1990). Serum BDNF, on the other hand, reflects the amount of BDNF stored in platelets, which is acquired from the circulating plasma and released during the clotting process.

Genetic studies have reported significant and consistent association of BDNF variants with ED (Koizumi et al. 2004; Monteleone et al. 2006; Ribases et al. 2003). The −270C/T BDNF single-nucleotide polymorphism (SNP) is associated to BN, while the Val66Met variant is associated with both AN and BN in different European populations (Ribases et al. 2004, 2005b). In addition, we detected a significant association between NTRK2 and AN, minimum body mass index (BMI) and harm avoidance (Ribases et al. 2005a).

In order to determine whether other variants in tight linkage disequilibrium (LD) with these SNPs configure an extended functional haplotype involved in the susceptibility to ED, we performed a family-based association study (FBAT) for 36 SNPs covering the BDNF gene in a sample of 106 ED nuclear families. Because SNPs in the BDNF gene may influence BDNF protein levels in brain, we also hypothesized that some genetic variants in the BDNF gene could be accompanied by altered blood BDNF levels. We tested 36 BDNF SNPs for association with plasma BDNF levels in a sample consisting of 110 ED patients. We also compared plasma BDNF levels in 50 ED patients with those in their respective siblings. Our data suggest that altered BDNF levels, modulated by BDNF gene variability, might confer susceptibility to ED, providing physiological evidence for a role of BDNF in the development of AN and BN.

Material and methods

Subjects

All patients with ED included in this study were Spanish with a Caucasian origin and were consecutively admitted to the Psychiatric Unit of the Hospital de Bellvitge between 1999 and 2002. Three clinical psychologists, two with master’s degrees and one with a PhD, conducted all assessments and face-to-face interviews. They were all experienced diagnosticians, who had received specific additional training in the administration of the instruments. The clinical sample used for the FBAT consisted of 106 nuclear families (106 trios, 64 of which had an additional unaffected sib) of ED patients (49 AN, 57 BN). The clinical subgroup used to analyze BDNF plasma levels consisted of 110 patients, 49 AN cases [44.5%; 25 restricting AN (ANR) and 24 binge-eating/purging subtype AN (ANP)] and 61 BN patients [55.5%; 58 binge-eating purging BN (BNP) and three nonpurging subtype BN (BNNP)]. All patients were women, fulfilled Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) (APA 1994, 2000) criteria for ED and diagnosed using the Structured Clinical Interview for Mental Disorders, research version 2.0 (SCID-I). More than 3 years of restrictive illness were necessary to classify patients as ANR. The lifetime minimum BMI was 15.3 kg/m2 (SD = 1.37) for AN patients and 19.2 kg/m2 (SD = 2.8) for BN patients. The average age at assessment was 24.5 years (SD = 6.0) for AN and 25.8 years (SD = 5.0) for BN. The average age at onset of weight loss was 15.3 years (SD = 4.3) for AN, and 19.1 years (SD = 4.6) for BN. Diagnosis was blind to genotype, and all patients have been described in previous reports (Gabrovsek et al. 2004; Ribases et al. 2003, 2004, 2005a). The control group for the sib-pair analysis consisted of healthy sisters of 50 ED patients, who were interviewed by the same psychologists to exclude that they had any psychiatric pathology. The mean age of the siblings did not differ from the mean age of the ED patients [mean = 23.8; 95% confidence interval (CI) = 23.2–28.4]. The study had ethics approval from the Bellvitge Hospital research ethics committee, and written informed consent was obtained from all participants.

Quantification of plasma BDNF levels

Peripheral blood samples were obtained from patients with an active/ongoing ED, and from 50 healthy controls who had never suffered from an ED. Plasma was obtained by centrifugation of fresh peripheral blood samples using ethylenediaminetetraacetic acid as anticoagulant and stored at −20°C until used for the assay. Blood extractions were always carried out between 0900 and 0930 h, and when possible, blood from both sisters was drawn on the same day. This was done in order to exclude any circadian or seasonal effect on BDNF concentration. Neither the patients nor the unaffected sibs were taking any medication at the time of blood extraction. The levels BDNF were measured by the enzyme-linked immunoassay system (BDNF Emax Immunoassay System kit, Promega, Madison, WI, USA), following the manufacturer’s instructions. Levels of BDNF with a variation coefficient higher than 10% were excluded from the analysis. Correlation coefficients for the internal duplicates were 0.994 for the ED patients and 0.996 for the unaffected sibs. The levels of BDNF in unaffected sib-pair analysis were compared by paired t-test.

Selection of BDNF SNPs

A region of 63.8 kb encompassing the BDNF gene was considered for SNP selection within the Celera SNP database (http://www.celera.com) and dbSNP build134 database (http://www.ncbi.nlm.nih.gov/SNP). Only SNPs with a unique mapping location on the NCBI build34 genome assembly were considered for further analysis. The SNPs were submitted to the Applied Biosystems design pipeline to select those that passed the design and genomic rules, ending up with the synthesis of ligation probes for 36 SNPs. Five SNPs were located in the coding exon, and two of them were located in 5′ untranslated exons. The remaining SNPs were located in intron 2 of BDNF. Mean distance between markers was 1.8 kb.

Molecular analysis

All SNPs were genotyped, blind to phenotype and to BDNF blood levels, using the SNPlex™ platform, as described by De la Vega et al. (2005). For the genotyping process, we followed the manufacturer’s protocol for the SNPlex™ Multiplex Genotyping Systems (Applied Biosystems, Foster City, CA, USA). Briefly, 12 μl of DNA at 100 ng/μl were fragmented by boiling at 99°C for 10 min, 2 μl were dispensed into a 384 well microplate and the manufacturer’s specifications were followed through the phosphorylation, oligonucleotide ligation, exonuclease cleanup, polymerase chain reaction and hybridization steps using robotic systems.

Family-based association study

Hardy–Weinberg equilibrium and LD of SNPs were assessed by the Haploview v2.03 software (http://www.broad.mit.edu/mpg/haploview/) (Barrett et al. 2005). To avoid multiple testing and redundant genetic information, we first selected tagSNPs using the tagger algorithm implemented in the Haploview software, which generates marker sets that are captured by tagSNPs. Nine tagSNPs for N different marker sets (Table S1) were selected to capture all the allelic variation within the gene with a minimum R2 of 0.9 (Barrett et al. 2005). The FBAT was performed using the fbat software as described in Horvath et al. (2000). First, the genotype model was used to screen for association with the tagSNPs. When a tagSNP was found to be associated with the disease, the remaining SNPs of the marker set were analyzed. Bonferroni correction was used to account for multiple testing, correcting for the number of tests performed by the fbat software. Haplotype analysis was performed with the hbat algorithm from the fbat software, which computes haplotypes before executing FBAT using biallelic and multiallelic modes. The best fitting model among additive, dominant and recessive models was chosen based on the lowest P value obtained (Horvath et al. 2001). Bonferroni correction for the number of tests performed, including the three models, was performed.

Effect of the SNPs on plasma BDNF levels

Plasma BDNF levels were logarithm transformed in order to normalize and homogenize variance among all the groups. Box plots representing the BDNF levels for different genotypes of every SNP were represented to check for possible association between SNPs and plasma BDNF levels. This relationship was then analyzed using a general linear model. For each SNP genotype, the mean and its 95% CI were computed. In all cases, the reference class was set as the homozygosity for the major allele among controls. Because there is no generally accepted answer to the question of which single SNP test to use (Balding 2006), we fitted four different genetic models. Thus, analyses were done under a codominant model (three genotypes considered independently), a dominant model (heterozygotes grouped with the homozygotes for the rare allele), a recessive model (heterozygotes grouped with the homozygotes for the common allele) and an additive model (genotypes coded as numeric values: 1, homozygous for the common allele; 2, heterozygous and 3, homozygous for the rare allele). P values were derived from likelihood ratio tests. The best model was chosen using the Akaike information criteria. Bonferroni correction for 23 nonmonomorphic SNPs was used to correct for multiple comparisons. An additional factor of correction of 2.5 was applied to account for the use of four different genetic models. This factor of 2.5 was obtained from a simulation study as the effective number of tests performed when four genetic models are used because the four tests are not independent (data not shown). Using this criterion, the corrected level of significance was set equal to 0.0009 (23 SNPs × 2.5 effective tests = 57.5 comparisons). All analyses were carried out using the SNPassoc R library (González et al. in press). Haplotypes were reconstructed using the phase software, version 2.0 (Stephens & Scheet 2005; Stephens et al. 2001), and a global test of hypothesis for every haplotype was done using a linear model.

Prealbumin concentration assessment

Serum prealbumin concentrations for 105 ED patients and 46 unaffected sisters were analyzed by immunoturbidimetry (ITC Diagnostics, Barcelona, Spain) in an ILab 1600 automated chemistry analyzer (Instrumentation Laboratories, Milan, Italy). The levels of patients and controls were compared by paired t-test. Pearson correlation was used to test the link between prealbumin and BDNF levels.

Results

Haplotype structure of the BDNF gene

As tagSNPs may be population specific, it has been suggested that they should be newly assessed in the local population where the association study is going to be performed (Carlson et al. 2004; Thompson et al. 2003; Weale et al. 2003). Thus, a total of 36 SNPs covering the BDNF region were selected and genotyped in 106 nuclear families of ED patients (49 AN, 57 BN). Thirteen of the SNPs were discarded because they were either monomorphic or had minor allele frequencies lower than 1%. The remaining 23 variants were included in the analysis to determine the LD structure of the region, after verification of Hardy–Weinberg equilibrium. Using the four-gamete rule, the region encompassing BDNF is composed of three LD blocks. The larger block spans 44 kb, it contains the BDNF coding region and encompasses 15 SNPs, including the functional Val66Met variant (rs6265). The other two blocks are smaller and include the 3′ untranslated region (UTR) and the BDNF 5’ flanking region, respectively (Figure S1). From the 23 polymorphisms in the region, we selected nine tagSNPs, which captured the allelic variation within the gene with a minimum R2 of 0.9 (Table S1).

Family-based association study

We performed a FBAT using the fbat software, considering the genotypes of nine tagSNPs in a sample of 106 nuclear families (106 trios, 64 of who had an additional unaffected sib). As no significant association was found for the tagSNPs when ED was considered comparing the expected vs. observed transmission of each possible genotype, we further stratified the sample according to the ED subtype.

Anorexia nervosa

In the anorexia subgroup (49 nuclear families), an undertransmission of the AA genotype (Z =−2.52; P = 0.01) of the rs7934165 SNP, tag for marker set 2, and an overtransmission of the CT genotype (Z = 2.04; P = 0.04) of the −270 C/T SNP, tag for marker set 4, was found associated with the disease, although after Bonferroni correction, statistical significance was no longer supported (Table 1). When stratifying for the AN subtypes, the association remained significant for rs7934165 (Z =−2.33; P = 0.01) and −270 C/T (Z = 2.36; P = 0.02) in the ANR subtype (31 nuclear families) (Table S4). In a subsequent step, we performed haplotype analyses. The two-locus haplotype containing both tagSNPs showed undertransmission of haplotype AC under an additive model (Z =−2.64; P = 0.008) (Table 2). This AC haplotype includes seven SNPs (marker sets 2 and 4) (Table S1). This result was significant after correcting for multiple testing.

Table 1.  FBAT of BDNF SNPs in AN
MarkerMarker setGenotypeFrequencyFamiliesSE (S)Var (S)ZP
  • SNPs with less than 10 informative families were excluded from the analysis. tagSNPs are squared. S, observed transmission of genotype to affected offspring; E, expected transmission under Mendelian inheritance; Var, variance S − E; P, two-tailed P value.

  • *

    Significant P value < 0.05.

  • *****

    Less than the minimum of 10 informative families to perform the test.

  • For marker sets, with significant association for the tagSNPs, the rest of the SNPs were analyzed.

rs71034111CC0.082***** 
CT0.342215.0011.005.501.710.09
TT0.58206.0010.005.00−1.790.07
rs79341652AA0.24171.005.753.56−2.520.01*
AG0.473017.0015.007.500.730.46
GG0.292412.009.255.311.190.23
rs10767665 AA0.24151.005.253.19−2.380.02*
 AG0.482615.0013.006.500.780.43
 GG0.282010.007.754.441.070.29
rs2030324 CC0.24171.005.753.56−2.520.01*
 CT0.473017.0015.007.500.730.47
 TT0.292412.009.255.311.190.23
rs7103873 CC0.24161.005.253.31−2.330.02*
 CG0.473017.0015.007.500.730.48
 GG0.292512.009.755.560.950.34
rs110301023CC0.60219.009.505.00−0.220.82
CG0.372411.0012.006.00−0.410.68
GG0.037***** 
−270CT4CC0.60236.0010.505.50−1.920.06
CT0.362417.0012.006.002.040.04*
TT0.045***** 
rs2049045 CC0.043***** 
 CG0.352214.0011.005.501.280.21
 GG0.61217.0010.005.13−1.330.19
rs6265 AA0.045***** 
 AG0.362415.0012.006.001.230.22
 GG0.60238.0010.505.50−1.070.29
rs110301195AA0.087***** 
AG0.392512.0012.506.25−0.200.84
GG0.53219.009.755.06−0.330.74
rs71244426CC0.089***** 
CT0.402713.0013.506.75−0.190.85
TT0.522310.0010.255.44−0.110.92
rs108352107AA0.20142.004.752.94−1610.11
AC0.452815.0014.007.000.380.71
CC0.352311.009.255.190.770.44
Table 2. BDNF haplotypes for AN, estimated from the two tagSNPs significantly overtransmitted or undertransmitted following additive model
Haplotype270CTrs7934165FrequencyFamilySE (S)Var (S)ZP
  • S, observed transmission of haplotypes to affected offspring; E, expected transmission under Mendelian inheritance; Var, variance S − E; P, two-tailed P value.

  • **

    Significant P value after Bonferroni correction; P < 0.00833.

  • *****

    Less than the minimum of 10 informative families to perform the test.

  • Number of informative families.

H1CA0.4622.917.5524.146.22−2.640.0082**
H2CG0.3324.025.4522.866.411.020.31
H3TG0.2122.418.5514.145.451.890.06
H4TA0.011.7***** 

Bulimia nervosa

In the case of BN patients (57 nuclear families), an overtransmission of genotypes CC (Z = 2.33; P = 0.02) of rs11030102, tag for marker set 3, and TT (Z = 3.07; P = 0.002) of rs7124442, tag for marker set 6, and an undertransmission of the genotypes AG (Z =−2.60; P = 0.009) of rs11030119, tag for marker set 5, was observed. Only the overtransmission of the TT genotype for rs7124442 remained statistically significant after Bonferroni correction, considering 16 tests (Table 3). When stratifying for the distinct BN subtypes, the association was significant for rs11030102 (Z = 2.62; P = 0.008), rs11030119 (Z =−2.86; P = 0.004) and rs7124442 in BNP (Z = 3.362; P = 0.0008) (55 nuclear families) (Table S4). Consistently, the most common haplotype for these three tagSNPs showed an overtransmission of the CGT haplotype, assuming a recessive model (Z = 2.64; P dominant = 0.008) (Table 4). The haplotype includes seven SNPs (marker sets 3, 5 and 6) (Table S1). This result was significant after correcting for multiple testing.

Table 3.  FBAT of BDNF SNPs in BN
MarkerMarker setGenotypeFrequencyFamiliesSE (S)Var (S)ZP
  • SNPs with less than 10 informative families were excluded from the analysis. tagSNPs are squared. S, observed transmission of genotype to affected offspring; E, expected transmission under Mendelian inheritance; Var, variance S − E; P, two-tailed P value.

  • *

    Significant P value <0.05.

  • **

    Significant P value after Bonferroni correction (P < 0.002).

  • For marker sets with significant association for the tagSNPs, the rest of the SNPs were analyzed.

rs71034111CC0.087***** 
CT0.342514.0012.506.250.600.55
TT0.58229.0010.505.38−0.650.52
rs79341652AA0.24188.007.754.190.120.90
AG0.473317.0016.508.250.170.86
GG0.29208.008.754.69−0.350.73
rs110301023CC0.602115.009.755.062.330.02*
CG0.37278.0013.506.75−2.120.03*
GG0.039***** 
rs10835211 AA0.039***** 
 AG0.36278.0013.506.75−2.120.03*
 GG0.612215.0010.005.252.180.03*
rs11030107 AA0.631914.008.754.562.460.01*
 AG0.34235.0011.505.75−2.710.007*
 GG0.037***** 
−270C/T4CC0.60208.008.754.69−0.350.73
CT0.362111.0010.505.250.220.83
TT0.046***** 
rs110301195AA0.089***** 
AG0.39256.0012.506.25−2.600.009
GG0.532014.009.004.752.290.02*
hCV7469136 CC0.088 *****  
 CT0.38246.0012.006.00−2.450.01*
 TT0.542014.009.004.752.290.02*
rs71244426CC0.087***** 
CT0.40275.0013.506.75−3.270.001**
TT0.522318.0010.755.563.070.002*
rs7127507 CC0.096***** 
 CT0.40246.0012.006.00−2.450.01*
 TT0.512016.009.504.872.940.003*
rs108352107AA0.20147.005.753.190.70.48
AC0.452915.0014.507.250.190.85
CC0.35207.008.754.69−0.810.42
 
Table 4. BDNF haplotypes for BN, estimated from the three tagSNPs significantly overtransmitted or undertransmitted under a recessive model
HaplotypeRs7124442rs11030102rs11030119FrequencyFamilySE (S)Var (S)ZP
  • S, observed transmission of haplotypes to affected offspring; E, expected transmission under Mendelian inheritance; Var, variance S − E; P, two-tailed P value.

  • **

    Significant P value after Bonferroni correction; P < 0.0125.

  • *****

    Less than the minimum of 10 informative families to perform the test.

  • Number of informative families.

H1TCG0.7322.018.011.835.472.640.008**
H2CGA0.187.0***** 
H3CCA0.040.0***** 
H4CCG0.020.0***** 
H5CGG0.010.0***** 
H6TCA0.0080.0***** 
H7TGA0.0070.0***** 

Relationship between SNPs and BDNF plasma levels

The genotype distribution of all BDNF SNPs was compared to plasma protein levels determined in a total of 110 ED patients. Thirteen SNPs were nominally associated with BDNF plasma levels: rs7124442 (P additive = 0.0006), rs1030102 (P recessive = 0.001), rs2049045 (P additive = 0.04), rs11030107 (P recessive = 0.002), rs10835210 (P recessive = 0.01), rs7103873 (P dominant = 0.004), rs7127507 (P recessive = 4.7 E−5), −270 C/T (P additive = 0.004), rs2030324 (P dominant = 0.002), rs1030119 (P recessive = 4.7 E−5), rs7934165 (P dominant = 0.004), rs10767665 (P dominant = 0.003) and hCV7469136 (P recessive = 0.0001), (Figure 1, Table S2), although only four remained significant after Bonferroni correction (P < 0.0009) (see Effect of the SNPs on Plasma BDNF Levels). These results are consistent, as the same trend was observed when stratifying for AN or BN status (Figures S2 and S3), and when considering unaffected sibs alone, association was found for rs712442 (P additive = 0.04), rs10835210 (P dominant = 0.002), rs7103873 (P dominant = 0.004), rs2030324 (P dominant = 0.004), rs7934165 (P dominant = 0.004) and rs10767665 (P dominant = 0.002). Although none of these results remained significant after Bonferroni correction (P < 0.002; 32 effective tests), the same trend found in cases was observed (Table S3, Figure S4).

Figure 1.

Box plot representing BDNF plasma levels for BDNF SNPs analyzed under codominant, dominant and recessive models of ED patients. The SNPs are ordered by genomic location in the BDNF gene. The SNPs shown in blue are those that show mean statistically different BDNF plasma levels depending on the genotype. *Statistically significant after Bonferroni correction (P < 0.002). P values are based on logarithm of the concentration and are adjusted by anorexia or bulimia status. A, common allele; a, rare allele.

Relationship between ED-associated haplotypes and BDNF plasma levels

We compared the effects of the BDNF haplotypes associated with AN and BN on BDNF plasma levels. No differences in BDNF levels were detected when the AN patients carrying the haplotype associated with AN were compared to the other AN patients. However, BN patients carrying the overtransmitted haplotype in BN had significantly higher BDNF levels than noncarriers, explaining 10.7% of the variability (mean TCG carriers = 43.6 ng/ml vs. mean others 23.0 ng/ml, P = 0.016) (Table 5). P values were considered statistically significant when P < 0.025, accounting for subgrouping AN and BN.

Table 5.  Means, 95% CI, and general linear model of BDNF plasma levels according to the haplotypes associated with AN and BN
GenotypesGeneral linear model
Mean (95% CI)R2BetaSE of betaP
  • AN patient carriers of rs7934165A and −270C haplotype undertransmitted in AN vs. noncarriers.

  • BN patient carriers of rs7124442T, rs11030102C and rs11030119G haplotype found to be overtransmitted in BN patients vs. noncarriers.

  • P values based on BDNF plasma concentration.

  • **

    Significant P value after Bonferroni correction (P < 0.025).

AN
 rs7934165A-270C carriers (= 31)49.0 (39.8–58.2)0.033.37.80.6
 Non-rs7934165A-270C (= 18)45.7 (33.6–57.8) 
BN
 rs7124442T-rs11030102C-rs11030119G carriers (= 55)43.6 (37.5–49.8)0.1120.610.00.02**
 Non-rs7124442T-rs11030102C-rs11030119G carriers (= 6)23.0 (4.4–41.6) 

Analysis of BDNF plasma levels in ED discordant sib pairs

We assessed BDNF protein levels in 50 ED patients and in their respective available unaffected sibs/sisters as a control group. We found significantly higher BDNF levels in ED patients relative to their sibs (mean 52.7 vs. 41.0 ng/ml, respectively; P = 0.004). In a subsequent step, we divided the sib-pairs group according to the clinical subtype of the affected cases, and the differences remained significant only in the AN sample (mean 57.8 vs 42.6 ng/ml; P = 0.02) (Table 6).

Table 6.  Means and 95% CI, and mean paired differences and 95% CI of BDNF plasma levels in patients with ED and the different clinical subtypes
Phenotype (n)Mean patient (95% CI)Mean sibs (95% CI)Mean paired differences (95% CI)Paired t-test P value
ED (50)52.7 (45.8–59.6)41.0 (35.1–46.8)11.8 (4.0–19.5)0.004
AN (21)57.8 (46.0–69.7)42.6 (31.8–53.3)15.3 (2.4–28.2)0.023
BN (29)49.1 (40.4–57.7)39.8 (32.8–46.9)9.2 (−1.0–19.4)0.076

Correlation of BDNF levels with BMI and prealbumin levels

Correlation between BDNF levels and prealbumin levels, a nutritional status marker, was assessed to discard unpaired nutritional status as a cause of the higher BDNF levels found in ED patients. Mean prealbumin levels were lower in ED patients than in their unaffected sisters (mean ED = 0.20 g/l, mean unaffected sibs = 0.26, P = 5.3 E−5), although they were not out of the reference values (0.10–0.4 g/l). However, there was no correlation between BDNF levels and prealbumin (Pearson correlation = 0.016, P = 0.86). No correlation was found between BDNF blood levels and BMI (data not shown).

Discussion

Our study provides new insights into the influences of genetic variation of the BDNF gene on the susceptibility to AN and BN, and on the regulation of BDNF blood levels. Our FBAT considering tagSNPs did not show a common haplotype associated with ED as a single group. However, a stratified analysis for each ED subtype showed significant associations for both AN and BN. As ANP subtype of AN might be more similar to that of bulimia than to other AN subtypes (ANR), and because it has been shown that subtypes often interchange during the disease course (Milos et al. 2005), we also performed a stratified analysis for the different clinical subtypes. When dividing by AN and BN subphenotypes (ANP, ANR, BNP, BNNP), the statistical significance for the associated SNPs remained significant for the ANR group and the BNP group, respectively (Table S4). However, we cannot exclude the association of these SNPs in the other subgroups (ANP and BNNP), as there were not enough informative families to perform the FBAT test.

These results replicate the associations of BDNF gene polymorphisms with ED (Gratacòs et al. in press; Ribases et al. 2004, 2005b). Interestingly, other variants rather than the Val66Met and the −270C/T appear to be more strongly associated with AN and BN in our present study. This would indicate that the Val66Met and −270C/T variants would not be the real causative ones, but they would be capturing information from other variants, including the real causal locus, due to strong LD. However, because many of the associated SNPs pertain to the same LD block, it is difficult to define a critical susceptible variant conferring risk to either AN or BN.

Interestingly, the associated haplotype composition was different for AN and BN. This result raises the possibility that different susceptibility variant(s) could be involved in the development of each phenotype. This is not surprising if we take into account the fact that, although both are considered ED subtypes, they show different physiological and psychopathological trait characteristics (Bulik et al. 1995; Fassino et al. 2004). Although the variability in BDNF seems to be altering eating behavior, the biological consequences might be slightly different, evolving to either AN or BN, depending on the specific genetic variant. Indeed, different misregulations of the expression of this gene could lead to binge-eating episodes, present in BN, while other deregulations could lead to long restricting periods, leading to AN.

We have also looked at BDNF blood levels in 110 ED patients and 50 discordant sibs. Several SNPs are associated with BDNF blood levels in both ED patients (irrespective of the AN or BN status) and unaffected sibs (controls). The low power, due to the small number of controls (n = 50), might explain the lack of statistical significance for some SNPs that are associated with BDNF blood levels in ED patients but not in controls (Table S3, Figure S4). Apart from this significant global association using single markers, we also investigated the BDNF levels in carriers of the AN- or BN-associated haplotypes. There were no differences in BDNF levels in the AN patients depending on the AN-associated haplotype. However, in the case of BN, the same haplotype that confers susceptibility to the phenotype is also associated with higher BDNF levels. This finding suggests that, in the case of BN, genetic variation in the BDNF gene is linked to BDNF blood levels, probably contributing to the development of BN. In the case of AN, the haplotype might predispose to the disease in a different way, rather than altering BDNF blood levels.

How the identified SNPs in the BDNF gene can correlate with altered functional activity of the protein remains to be clarified. A possibility would be to act through an altered transcription or translation. Although we have genotyped all available SNPs in the genomic region, we cannot be sure that the genetic risk allele is being directly tested, and we assume that a genetic effect at this SNP locus is capturing the causal locus itself. To identify the functional variant, all existing variants in the same marker set should be analyzed in a larger sample and, once the candidate region is more precisely mapped, a mutational screening analysis should be performed. Among all variants we genotyped, the only one that has been functionally tested is Val66Met (rs6265), which has been shown to affect intracellular trafficking and activity-dependent secretion of BDNF protein (Egan et al. 2003). However, a microsatellite polymorphism located approximately 1.0 kb upstream the translation-initiation site has been recently reported to affect transcriptional activity (Okada et al. 2006).

Finally, we found that BDNF plasma levels are about 25% higher in ED patients than in their unaffected sibs. These results are consistent with previous findings showing that CNS intraventricular administration of BDNF induces starvation and body weight loss, while BDNF heterozygous knockout mice develop obesity and hyperphagia (Kernie et al. 2000; Lyons et al. 1999; Rios et al. 2001). Accordingly, increased BDNF plasma levels could reflect high protein levels in the CNS that would alter eating behavior in ED patients. Alternatively, taking into account that BDNF expression is induced by long periods of food restriction, the altered plasma levels could also reflect the eating patterns of ED patients, instead of indicating a direct causative role in the development of the disease. However, we did not find a correlation between BDNF and BMI or prealbumin levels, a marker of nutritional status (Beck & Rosenthal 2002; Fuhrman et al. 2004). This argues against impaired nutritional status or altered eating patterns as the main cause of the observed BDNF differences. The analysis of BDNF concentration after weight recovery, in the case of AN, should provide additional clues about the BDNF contribution to the pathogenesis of ED.

Despite the robustness of our findings regarding plasma BDNF protein levels, which are supported by murine and association data (Kernie et al. 2000; Lyons et al. 1999; Pelleymounter et al. 1995; Ribases et al. 2003, 2004, 2005b), the results also challenge previous reports in which BDNF serum concentration was found decreased in AN patients (Monteleone et al. 2004, 2005; Nakazato et al. 2003). However, these studies measured serum BDNF levels, while we determined plasma BDNF levels. Although serum BDNF levels can be considered a long-term marker of varying plasma BDNF, it has recently been reported that the correlation between BDNF levels in plasma and in platelets, the major source of BDNF in serum, is not very high (Lommatzsch et al. 2005). This indicates that the results obtained from both specimens are not fully comparable.

In addition, it is well known that short-term fasting reduces BDNF levels in the hypothalamus, while long dietary-restriction periods increase the expression of this neurotrophin in the CNS (Duan et al. 2001; Xu et al. 2003). These results suggest that the observations reported by Monteleone et al. (2004), where blood collection was performed after overnight fasting, could reflect the negative effects of short fasting periods on BDNF levels and have masked the basal BDNF levels in both ED patients and controls. In our study, blood samples were withdrawn during the morning from patients or controls who did not fast to avoid the short-term fasting response of BDNF. As well as having a substantially larger sample than those studied in other reports (Karege et al. 2002a, Koizumi et al. 2004; Toyooka et al. 2002), the control sample we used here consisted of unaffected sibs, which avoids the potential stratification effects and allows the comparison of each ED patient with her own sib. Indeed, blood samples from each sib pair were collected the same day, and all specimens were treated equally before the analytical procedures, which reduces possible environmental and assessment effects on the observed results.

In summary, our findings confirm the role of BDNF gene variability in ED, through, at least in part, the modulation of plasma BDNF levels. It will be important to elucidate which variants are causative and to disclose whether BDNF is a biological marker of the disease.

Acknowledgments

We thank the patients for participation in the study, and Laura Giménez, Raquel Solano, Carolina Casanovas and Araceli Núñez, for their help on collecting the data. We also thank Heidi Howard for her helpful comments after reading the manuscript. Financial support was received from the Psychiatry Genetics Network (G03/184) (Carlos III Research Institute), the Ministry of Education and Science (SAF2005-01005), the ‘Fondo de Investigaciones Sanitarias de la Seguridad Social, FIS (PI40632 and PI052307) and FIS (PI040619; CIBER-CB06/03/0034), the Department of Health (Generalitat de Catalunya), and the Department of Universities, Research and Information Society (2005SGR00008; 2005SGR322) (Generalitat de Catalunya). Josep M. Mercader was supported by the CRG under project SAF2002-00799 (Spanish Ministry of Science and Education), and by fellowship of the Danone Institute. Marta Ribasés was recipient of a BEFI fellowship from the FIS (Spanish Ministry of Health).

Ancillary