Variants in Doublecortin- and Calmodulin Kinase Like 1, a Gene Up-Regulated by BDNF, Are Associated with Memory and General Cognitive Abilities

Background Human memory and general cognitive abilities are complex functions of high heritability and wide variability in the population. The brain-derived neurotrophic factor (BDNF) plays an important role in mammalian memory formation. Methodology / Principal Finding Based on the identification of genes markedly up-regulated during BDNF-induced synaptic consolidation in the hippocampus, we selected genetic variants that were tested in three independent samples, from Norway and Scotland, of adult individuals examined for cognitive abilities. In all samples, we show that markers in the doublecortin- and calmodulin kinase like 1 (DCLK1) gene, are significantly associated with general cognition (IQ scores) and verbal memory function, resisting multiple testing. DCLK1 is a complex gene with multiple transcripts which vary in expression and function. We show that the short variants are all up-regulated after BDNF treatment in the rat hippocampus, and that they are expressed in the adult human brain (mostly in cortices and hippocampus). We demonstrate that several of the associated variants are located in potential alternative promoter- and cis-regulatory elements of the gene and that they affect BDNF-mediated expression of short DCLK1 transcripts in a reporter system. Conclusion These data present DCLK1 as a functionally pertinent gene involved in human memory and cognitive functions.


Introduction
Empirical evidence for genetic factors underlying cognitive variation is strong, but identification of specific genetic variants has proved challenging. The study of differences in human mental abilities, measured with psychometric tools, has defined a hierarchical structure of human cognition. General cognitive ability (g) stands at the pinnacle, accounting for about 50% of the total variance in test performances. Correlated with g, there are additional, separable group factors representing distinct cognitive domains, with memory as a prominent example [1,2]. Behavioural studies suggest that the impact of genetic factors on specific cognitive domains, e.g. memory, derives from both inherited contributions to general cognitive ability and from genetic variation that more selectively influences memory. This heritability is thought to be polygenic but, to date, few molecular genetic studies have provided replicated genetic associations with cognitive function [3].
Variations in the gene for the brain-derived neurotrophic factor (BDNF) have been examined for associations between cognitive traits and psychiatric disorders. Several studies have identified association of a single nucleotide polymorphism (SNP -rs6265)which corresponds to an amino acid change (Val66Met) -and verbal memory, general cognitive ability, age related change in reasoning skills and hippocampal functions, but others have failed to replicate these findings [4,5,6,7,8,9]. The same Val66Met genetic variant of BDNF has been largely studied as a risk factor for bipolar affective disorder, schizophrenia and other psychiatric disorders (for review see [10,11]). Since several of the cognitive traits associated with BDNF variants have been reported potentially deficient in major psychosis (e.g. verbal memory and general cognition in bipolar disorder and/or schizophrenia [12,13]), these common associations might reflect genetic associations to clinical endophenotypes of these disorders. Indeed, neurocognitive traits have been proposed as endophenotypes for psychosis that, because they are less clinically heterogeneous and have high heritability, would be more powerful in identifying genetic factors of susceptibility [14,15,16].
Functional convergent genomics approaches, which use a set of candidate genes identified in gene expression-based relevant models [17,18,19], have been successful to identify new genes for schizophrenia, antipsychotic induced weight gain and bipolar disorders [20,21,22]. In order to broaden the study of BDNF in cognitive traits, we chose to develop a functional convergent genomics approach to characterise genes up-regulated by BDNF, and their implication in cognition. BDNF plays a critical role as a trigger of memory formation and transcription-dependent enhancement of synaptic strength [23], via neuronal activity induced gene expression [24]. Previously, we used a rat model where BDNF is infused in the dentate gyrus region of the hippocampal formation, and identified a panel of 14 genes that were upregulated following BDNF treatment that induces long-term potentiation -LTP, in a similar way as the immediate early gene Arc [25]. Among these genes, the five most strongly differentially expressed genes were further examined by in situ hybridization and also found to be up-regulated in another LTP inducing model (afferent high frequency stimulation of the synaptic strength in the dentate gyrus). In the present study, we selected haplotype-tagging SNPs to screen those genes (ARL4L, NEURITIN, DCLK1, KLF10 and NPTX2) together with BDNF and ARC for influence on human cognitive functioning in samples of healthy individuals who volunteered for testing of their memory and general intellectual function (IQ).

Samples descriptions
All Norwegian participants read an information sheet and signed a statement of informed consent approved by the regional ethical committee for medical research (Project ID: S-03116). Permission to obtain and store blood samples for genotyping together with cognitive and MRI data in a biobank was given by the Department of Health, and permission to establish a registry with relevant information for a time period of 10 years was given by the Department of Health. For the Scottish samples, ethics permission for the study protocol was obtained from the Multi-Centre Research Ethics Committee for Scotland (MREC/01/0/ 56) and from Lothian Research Ethics Committee (LREC/2003/ 2/29). The research was carried out in compliance to the Helsinki Declaration. All subjects gave written, informed consent.
The Norwegian Cognitive NeuroGenetics (NCNG) sample. consists of 271 participants aged 46-75 years (mean age of 62.6 years, s.d. 7.9) that were recruited through advertisements in local newspapers in the Oslo and Bergen urban areas (see previous report in Espeseth et al., 2006 [26]). Candidates were first interviewed by phone according to a check list about health and previous illness or injuries. Exclusion criteria were previously diagnosed neurological or psychiatric illness, any other chronic illness that might influence test performance, or sensory or motor impairments. Participants with a history of alcohol or substance abuse, or current addictive disorders, were also excluded. Participants had to be native speakers of Norwegian and have completed obligatory basic education (7 years for this age group) without diagnosed reading or learning disorders. Persons on adequate medication for hypertension, diabetes or hypercholesterolemia were not excluded. Participants were not allowed to consume nicotine or caffeine during the test period or in the lab premises, but were not required to abstain from these substances prior to attendance. There were 195 females (72%) and the average total years of education was 13.9 (s.d. 3.0). The Vocabulary and Matrix Reasoning subscales of the Wechsler Abbreviated Scale of Intelligence (WASI, [27]) were used to estimate IQ, and the California Verbal Learning Test II (CVLT-II, [28]) to test verbal episodic memory.
The Lothian Birth Cohort 1921 (LBC1921). are surviving participants of the Scottish Mental Survey of 1932 (SMS1932). They were recruited in Edinburgh and the surrounding areas either through the Community Health Index -a list of individuals registered with a General Practitioner -or as volunteers replying to media calls [29]. The first wave of follow-up ran from 1999-2001, and 550 individuals (234 men and 316 women) were tested individually at the Wellcome Trust Clinical Research Facility (WTCRF) at the Western General Hospital, Edinburgh. The mean age of the LBC1921 participants was 10.9 years (s.d. 0.3) at the time of the SMS1932 and 79.1 years (s.d. 0.6) at wave 1 assessment.
In the SMS1932, all participants completed the Moray House Test (MHT) Number 12 which was re-administered at age 79. The Moray House Test is a well-validated IQ-type test with a predominance of verbal reasoning items, though there are also some other types of item. The raw MHT scores were corrected for age in days at the time of testing and then converted into IQ scores. At wave 1, a further battery of cognitive tests was also completed, including the Raven's Progressive Matrices to assess non-verbal reasoning [30], Verbal Fluency to assess executive function [31], and Logical Memory to assess verbal declarative memory [32]. The Mini-Mental State Examination as a brief screen for dementia [33] and the Hospital Anxiety and Depression Scale [34] were also administered. Further description can be found in Deary et al. (2004) [29].
In the present study, only the association with IQ score, logical memory (immediate, delayed and total), verbal fluency and Raven's matrices were analyzed as these variables were the most comparable to the variables analyzed in the NCNG sample.
The  [35]). They were recruited in Edinburgh and the surrounding areas either through the Community Health Index -a list of individuals registered with a General Practitioner-or as volunteers replying to media calls [36]. All participants lived independently in the community and were able to travel to the clinical research facility for testing. They undertook medical and cognitive testing at age 70 (mean age of 69.6 years, s.d. 0.8) as reported in detail previously [36], including the same MHT test as they had taken at age 11. The raw MHT scores were corrected for age in days at the time of testing and then converted into IQ scores. The Mini-Mental State Examination (MMSE) was used to screen for possible dementia [33]. The battery of cognitive tests sampled a variety of specific cognitive abilities, with an emphasis on memory and processing speed. Memory domains were assessed by the following subtests of the Wechsler Memory Scale-IIIUK (WMS-IIIUK; [37]): Logical Memory I (immediate verbal declarative memory), Logical Memory II (delayed verbal declarative memory), Verbal paired associates (immediate and delayed verbal memory and learning) and Spatial span (non-verbal, spatial memory). The information processing speed battery comprised two psychometric tests from the WAIS-IIIUK (Digit symbol coding and Symbol search) and two elementary cognitive tasks, Reaction Time (simple and choice conditions) and Inspection Time. Other cognitive tests which tapped diverse abilities included: Backward digit span (working memory) from the Wechsler Memory Scale-IIIUK; and Letternumber sequencing (working memory), Matrix reasoning (nonverbal reasoning) and Block design (constructional ability) from the WAIS-IIIUK [37]. The Verbal fluency test provided a measure of executive function [31].The g factor was calculated via principal components analysis of the following Wechsler tests: Backward digit span, Letter-number sequencing, Matrix reasoning, Block design, Digit Symbol and Symbol Search subtests. A full description of these tests can be found in the Lothian Birth Cohort 1936 protocol article [36]. The final sample for analysis (i.e., those who also had genotype data) was N = 1077, and included 535 females and 542 males.

Selection, genotyping and analysis of genetic markers
Markers selection, genotyping and analysis in the NCNG sample. Haplotype tagging markers (single nucleotide polymorphisms -SNP) were selected using the Phase I (16c.1, June05, based on NCBI B34 assembly, dbSNP b124) version of HapMap (http://www.hapmap.org/cgi-perl/gbrowse/gbrowse/ hapmap_phaseI/). [38], and according to the protocol described in Christoforou et al. [39]. Briefly, Hapmap data for the CEU trios were downloaded and analysed in Haploview v2.5, using the following criteria: pair-wise comparisons of markers more than 500 kb apart were ignored, minor allele frequency $0.10, a Hardy-Weinberg (HW) P-value $0.001, genotyping success rate $0.75. The haplotype blocks were defined using the solid spine of LD approach, using Haploview's internal tagging program. Haplotype tagging SNPs were selected on a block-by-block basis to represent haplotypes of frequencies higher than or equal to 0.10. The markers were genotyped on a Sequenom Massarray platformTM (http://www.sequenom.com/, Sequenom Inc., San Diego, CA, USA) at CIGENE, Center for Integrative Genetics (Universitetet for miljø-og biovitenskap, Å s, Norway, http:// www.umb.no/), which is the national FUGE platform for genotyping (www.fuge.no), supported by the Research Council of Norway.
Markers selection, genotyping and analysis in the LBC1921 and LBC1936 samples. In the replication sample we decided to concentrate only on the DCLK1 gene. Seven markers that showed association in the NCNG sample (p-value ,0.01: rs4591003, rs1926467, rs943220, rs10507435, rs7323560, rs7334245, rs9315383) were selected for replication in the LBC samples. In addition, we chose to include another eight markers for genotyping in these Scottish samples, since they show association to psychiatric disorders in preliminary studies (unpublished information): rs12430800, rs2296645, rs10492555, rs872060, rs9545332, rs7989245, rs7989807 and rs9315390. The marker rs4391923 was also included for its potential functional relevance.
Assays were developed for the ABI PRISMH 7900HT Sequence Detection System using TaqMan technology (Applied Biosystems, CA, USA), with genotyping at the Welcome Trust Clinical Research Facility, Genetics Core (http://www.wtcrf.ed.ac.uk/ genetics/default%20genetics.htm, University of Edinburgh, UK).
Genotypes were quality controlled with the following criteria: individual samples with genotype call rate ,90% and markers with call rate ,96% or Hardy Weinberg P-value ,0.001 were excluded from analysis (see supplementary material -SOM, for summary of locations, markers tested and analysed).
Genotyping data were analyzed using the Helix Tree software for linear regression, genotypic association and haplotype trend regression of 2-and 3-markers sliding windows, similarly to the analysis described in [20].
CNV considerations. Copy Number Variants (CNVs) were not considered at the time of the markers selection (few data were available at this time, September 2005). Retrospectively, the Toronto database (http://projects.tcag.ca/variation/) was screened for CNVs in the genes investigated. Markers in the NPTX2, BDNF and DCLK1 genes are not affected by any known CNVs. ARC is located within the rare variation_30296 seen in 1 in 1086 chromosomes by Jakobsson et al. [41], KLF10 is located within the CNV region of variation_7659 described by de Smith et al. [42] observed in 1 in 100 chromosomes, and NRN1 is located on the same clone as the variation_0072 seen in 1 in 110 chromosomes by Iarfate et al. [43] which has not been further refined. Considering their relatively low frequency, and the Hardy Weinberg equilibrium observed for genetic variants of this region in our genotyping, we consider that none of these known CNVs are likely to affect the associations reported in this study.
The specificity of all RT-PCR assays was verified by DNAsequencing of the amplified PCR products.

Semi-quantitative RT-PCR of BDNF-induced expression of Dclk1 in vivo
The expression of Dclk1 transcripts in the rat hippocampus were measured by semi-quantitative RT-PCR on cDNA samples obtained from our previous study on BDNF-mediated induction of Long-Term Potentiation in the rat hippocampus [25]. PCR were performed as described above with the following primer sets: Dclk1 long: 59-GGTGTGGTGAAGCGTCTGTAC, 59-CAAAA AAGTCCTGAAGGCACATC;Dclk1 short: 59-ACACTAAGAC-TGTGTCCATGTTAGAACTC, 5-GGCCATCGTTCTCATC-CATT. The primer sets for Carp and Neuritin has been described elsewhere [25].

Sequencing of intron 5 of human DCLK1
Genomic DNAs of 23 individuals selected for their genotypes at the markers rs943220 and rs10507435 from the NCNG sample were amplified to sequence the DCLK1 intron 5. Nineteen pairs of primers for PCR amplification (sequences can be made available upon request) of the intron 5 were designed, using the Primer3 tool [44] (http://fokker.wi.mit.edu/primer3/input.htm). The PCR amplicons were designed to have a sufficient (100-150 bp) overlap to obtain contiguous sequences. PCR fragments were amplified using AmpliTaq Gold (Applied Biosystems, Foster City, CA, USA) according to the manufacturer's instructions with 25 cycles: 94uC for 10 s, 55uC for 30 s, 72uC for 30 s (initial denaturation 94uC for 10 min). The PCR products were sequenced with BigDye v3.1 (Applied Biosystems) according to the manufacturer's instructions and the sequences were aligned with phredPhrap program and read in Consed39 [45] (http://bozeman.mbt.washington.edu/ consed/consed.html#documentation).
The Neuronal Network Promoter Prediction program was used to rank candidate TATA-boxes in intron 5 (http://www.fruitfly. org/seq_tools/promoter.html). Probability scores of clustering between promoter elements and TATA-boxes were investigated by the Cister-program and the TF-search engine.
Three haplotypes of the intron 5 cis2 element (hap1a, hap1b and hap2, see section 6 and supplementary online material -SOM) were PCR amplified from genomic DNA (from 3 individuals homozygous for the three haplotypes identified by sequencing), using the forward primer 59-GTCACGGATCCTTGGAAACTCAAGAA-GATAGGC and the reverse primer 59-GATCAGTCGACCCA-CAGGAAACAAAGCAACC (BamHI and Sal1 sites introduced, underlined). Amplified DNA was cloned into the BamH1/Sal1 sites of pGL4.11-Cprom, pGL4.11-Tprom and pGL4.11[luc2P], generating haplotype specific cis2-C/Tprom luciferase reporters and promoter-less control vectors. In the final plasmids, the Cis2 element locates 2216 bp upstream of the promoter-region, the two being separated by a synthetic poly(A) sequence for the reduction of background signals. All plasmid constructions were verified by DNA sequencing. See supplementary material -SOM, for a list of commercial and constructed plasmids used in the study.
Cell cultures, differentiation and plasmid transfection.
Plasmid transfections were performed over-night using Meta-fecteneTM Pro (Biontex, Munchen, Germany) under conditions optimized according to the manufacturer's guidelines. Neuronal differentiation was initiated the day after transfection by replacing the transfection media with fresh media containing BDNF (50 ng/ml)/RA (10 mM all-trans retinoic acid) (Sigma-Aldrich, St Louis, MO, USA), as described by Holback et al. [46]. Control cells (non-differentiated) were treated similarly but exposed to media with vehicle only (0.01% DMSO). Cell viability was monitored using WST-1 (F. Hoffmann-La Roche Ltd, Germany). All transfections were performed in quadruplicates on 96 multiwell plates. Each experiment was controlled for transfection efficiency by transfecting a green fluorescent protein (GFP) expressing vector (pSIREN-RetroQ-ZsGreen) into a separate set of cells. GFP expression was analyzed two days post transfection on a FACSCalibur TM Flow Cytometer (BD systems, BD Europe, Belgium).
Luciferase Reporter Assays. The transcriptional activity of each reporter plasmid was measured two days post transfection using the Dual-Luciferase Reporter Assay system (Promega Biotech AB, Madison, WI, USA), according to manufacturer's protocol. Luciferase signal intensities were recorded on a Chameleon plate reader (Reactionlab Sverige AB, Sweden). All plasmid transfections and luciferase measurements were performed in 96 multiwell plates, with quadruplicates of each sample. Each experiment was repeated 5 times for the pGL4-C/Tprom plasmids, and 7 times for plasmids harboring cis2 elements. All experiments were performed blind to the vectors haplotypes. Luciferase signal intensities were normalized for transfection efficiency on the basis of renilla luciferase signal intensities from the co-transfected pGL4-73[hRluc/SV40] plasmid. The transcriptional activity of a given reporter plasmid is listed relative to its respective promoter-less control vector. All cis2 promoter-less control vectors showed similar background activity (data not shown).

Association to verbal memory and general cognition in the NCNG sample
In the original study design, seven genes were selected: BDNF,ARC, ARL4L, NEURITIN, DCLK1, KLF10 and NPTX2. During quality check of the genotyping results, we noticed that the official annotation of the rat Arl4l had been updated to Arf4l and that the human homolog we selected (ARL4L) was erroneous and should have been ARL4D. We therefore decided to exclude the ARL4L markers from further analysis as this gene is apparently not up-regulated during LTP in rat brain.
We excluded DNA samples with a genotyping success below 0.9, and markers with call rate ,96% or Hardy Weinberg P-value ,0.001 from analysis. For the 6 remaining genes studied, a total of 48 markers were genotyped in the Norwegian Cognitive Neuro-Genetics -NCNG -sample. This sample consists of 271 individuals (mean age: 62.6 years, range 50-75), which had been recruited via media advertisements and who were subjected to cognitive testing, i.e. verbal memory and general cognition (IQ score). In this sample, we found strong associations between several markers in the DCLK1 gene (doublecortin-and calmodulin kinase like 1, a.k.a. DCAMKL1) and aspects of verbal memory function and IQ score, which resisted Bonferroni correction (see Table 1, Table 2 and SOM for all results of markers tested). The most significant associations were observed for intron 5 markers m5.1 (rs10507435) and m5.2 (rs943220), where the less frequent genotypes were significantly associated with reduced verbal memory performance (see Table 1). Marker m5.1 was also associated with IQ score, like several other DCLK1 SNPs (see Table 2).
For the other genes tested we found some significant effects, but no strong association (p,0.01) or associations across several cognitive traits (see SOM). However, in a 2-loci analysis for genegene interaction effects, there were significant interactions between DCLK1 intron 5 markers (m5.1 and m5.2) and markers tagging BDNF and ARC on the association with verbal memory (pvalue = 0.0320.003, after Bonferroni correction, see SOM). At the genetic level, this finding is consistent with the co-upregulation of Dclk1 transcription by Arc and Bdnf, as observed in a rat hippocampal model of synapse consolidation.

Replication in the Lothian Birth Cohorts
Sixteen DCLK1 SNPs were selected for replication testing in two independent samples (see SOM). The Scottish Lothian Birth Cohort (LBC) studies of individuals born in 1921 and in 1936 (LBC1921 and LBC1936) are two independent cohorts of individuals who underwent an IQ test at age 11 (IQ11), and who at the age of 79 years for the LBC1921, and 70 years for the LBC1936, participated in follow-up testing with examination of IQ (IQ79 and IQ70, respectively) and other cognitive functions [29,36].
In the LBC1921, we found an association between verbal memory and m11, an exon 11 synonymous marker (see Table 1). For IQ79, especially when regressed for IQ11, we replicated and further strengthened the association to markers in the intron 15 and intron 19 (which are located in the same haplotype block), and to intron 19 3-markers haplotype (resisting permutation testing, see Table 2).
In the LBC1936, we observed associations between an intron 5 marker (m5.3) and several memory-specific and general cognition traits at age 70 (see Table 1, and 2). Also in this sample, the association with intron 5 was even stronger (resisting permutation testing), for both memory and general cognition traits, at the haplotype level (3-markers haplotype covering the intron 5, see Table 2). In this sample, the intron 15 -intron 19 markers (except for m19.1, rs12430800) were associated with childhood cognitive variables (IQ11). Further association testing with other cognitive abilities (see SOM) showed an effect of markers in both the intron 5 and intron 15-19, but also an effect of the age at testing. This age-dependent association is especially pronounced for the general cognitive ability factor, g factor, which was associated with both intron 5 (m5.3, rs4391923) and intron 15 -intron 19, but only the intron 5 association remained significant when IQ11 was adjusted for in the regression model.

Expression of rat Dclk1 variants after BDNF treatment in the dentate gyrus
The rodent Dclk1 gene is expressed as several transcripts, e.g., long (exons 1-20 except 6 and 8), short (exons 6-20 except 8) and Carp (exons 6-8; see figure 1A), which vary in expression (spatial and developmental) and in function (see discussion). In rat, we previously demonstrated that the Carp variant is induced by BDNF  exposure, but at this time we did not further look at the other transcripts. Further examination of the samples from that study, show that all short DCLK1 variants are induced by BDNF but not the long variants (see figure 1B).

Expression of DCLK1 variants in human brain tissues
In the rodent, the short variants show higher expression in adult brains whereas long variants dominate during embryonic stages in rat [47]. In humans, the expression of DCLK1 variants has not been fully documented, but expression of long DCLK1 has been shown in both embryonic and adult brain tissues [48]. Using transcript-specific PCR assays, we found expression of both long and short transcripts of DCLK1 in the fetal brain (26-40 weeks) as well as in specific regions of the adult human brain (see Figure 1C). The expression of short variants was especially high in regions involved in memory performance (hippocampus, occipital pole, frontal lobe and temporal lobe). The expression of CARP was low in all human brain tissues, consistent with previous observations in rodents of low basal CARP expression and robust induction by specific treatments [25,49].

In silico characterisation of regulatory elements and identification of additional genetic variants in the intron 5 of DCLK1
Considering that one of the main signals of association was observed for markers in the intron 5 of DCLK1 and that short transcripts of DCLK1 all start from the exon 6, we hypothesised that the intron 5 might harbour alternative promoter and regulatory regions. We searched for transcriptional cis-regulatory elements and promoters in the human intron 5 with the aim of analyzing the effect of these in luciferase reporter assays. The search predicted a TATA-box promoter and three specific cis-regulatory elements (cis1-3, see Figure 2A). To identify additional sequence variants that could affect the regulatory elements, we sequenced the entire intron 5 from genomic DNA of 23 individuals from the NCNG sample. In the putative alternative promoter region in intron 5, the marker m5.3 (C/T variant) was located 2 bp from the potential transcription start. For the adjacent regulatory elements we observed several variants in the cis2 that could affect the regulatory properties, defining three haplotypes: the related hap1a and hap1b, and the hap2 (see SOM for details).

Assessment of the functional effect of SNPs in potential regulatory regions on the expression of DCLK1 short variants
Luciferase reporter assays were constructed by cloning the promoter C (C-prom) or T allele (T-prom) of m5.3, from genomic DNAs of NCNG individuals. Additional reporter assays were constructed to mimic in vitro the potential effect of the 3 cis2 haplotypes on the promoter. The assays were analysed for expression of the reporter luciferase protein under control conditions and under BDNF/RA differentiation (retinoic acid, protocol for BDNF induced differentiation as described by Holback et al. [46]).
In luciferase reporter assays, both the C-prom and T-prom constructs displayed basic promoter reporter-activity in the undifferentiated SH-SY5Y control cells. However, during BDNFinduced neuronal differentiation, only the T-prom construct demonstrated increased expression of the luciferase reporter (see Figure 2C). The addition of a single hap1a or hap1b variant of the cis2 element into the non-inducible C-prom reporter-vector rendered the C-prom inducible by BDNF (see Figure 2D), whereas  IQ  IQ79  IQ79  rIQ11  IQ11  IQ70   Markers  Code  Localisation LR  HTR3  LR  HTR3  LR  HTR3  LR  HTR3 LR HTR3 Both single marker linear regression (LR) and haplotype trend regression of 3-markers sliding windows analyses (HTR3) are presented. Only p-values below 0.05 are displayed. All analyses were performed using sex and age as covariates. The NCNG was assessed with the Wechsler Abbreviated Scale of Intelligence [27]. LBC1921 and LBC1936 were assessed for IQ with the Moray House Test (see SOM). For LBC1921, no association below p = 0.05 was observed for IQ11. In the LBC samples, all regression analyses were performed both with and without IQ11 as a covariate. P-values highlighted in bold are resisting a 10,000 permutations testing. doi:10.1371/journal.pone.0007534.t002  the hap2 variant of cis2 did not show this effect on C-prom. None of the cis2 variants had any significant additional effect on the activity and inducibility of the T-prom sequence. These reporter assays thus demonstrated that the predicted promoter in intron 5 displays an allele-specific promoter activity and inducibility which can be further influenced by regulatory elements, such as cis2, in an allelespecific mode.

Discussion
Our data show that genetic variants in DCLK1 significantly influence the performance on tests of memory and intellectual function in three independent samples. Although the size of our discovery sample is small, we detected association in this sample. We observed further associations between genetic variants in DCLK1 and cognitive abilities in the replication samples but with differences regarding the panel of markers that were associated. These differences could be due to type I errors, but also to type II errors by lack of power for either the discovery or the replication samples to detect association. However it is probable that these differences are due to allelic heterogeneity that reflects the variation between the samples. Even though all subjects have been phenotyped for similar traits (e.g. verbal memory and general cognition), they are different in the geographical origin, age, or mode of recruitment, as well as in the specific tests used to assess cognitive phenotypes.
Comparison of the associations in the three samples highlights three main regions of interest in the DCLK1 gene that could have an influence on the heritability of cognitive traits. The first region extends from the 59UTR to the intron 3 of the gene, where the markers m3.1 and m3.2 in single marker allelic analysis or certain 3markers haplotypes in haplotype trend regression are associated to verbal memory or to general cognition in the NCNG sample (see Tables 1 and 2), to IQ at age 79 in the LBC1921 (see SOM) and to several cognitive traits at the genotypic level in the LBC1936. The region spanned by these markers is large and covers the promoter region and the first exons and introns. Notably, it includes a large proportion of the wide intron 3, which contains a large number of highly conserved non-coding elements as seen in the UCSC browser (http://genome.ucsc.edu/cgi-bin/hgGateway).
The second region of association is localised in intron 5 of the gene. This area is associated to verbal memory and IQ at the single marker level in the NCNG sample (see Table 1 and 2), to verbal memory and several other cognitive traits at the single marker level (m5.3, see Table 1 and SOM) and to IQ at age 70 at the 3-markers haplotype level in the LBC1936 (see Table 2 and SOM). Several short transcripts for the DCLK1 start from the exon 6 (as further detailed below). We characterised a potential promoter in intron 5, which may be used for transcription of the short variants, as well as three regions with potential regulatory effect on this transcription. Our reporter assay studies of this putative promoter and one of the regulatory regions (chosen because it encompasses associated genetic variants) displayed that the efficiency of these regulatory elements was influenced by the alleles of the associated markers and by the treatment with BDNF. In this context, it is interesting to note that the BDNF-inducible C-prom/cis2 haplotypes (hap 1a/b), equivalent to 2-marker m5.1-m5.3_GA haplotype of intron 5, shows the strongest association to IQ score and verbal memory (p-value = 0.0027, see SOM and 0.0019, data not shown) in the LBC1936, and that the ''noninducible'' constructs corresponds to the m5.1-m5.3 GG haplotype that is the rarest, found only in 2.5% of the chromosomes.
The third region of association points to a haplotype block of markers covering intron 15 to intron 19. This region is associated to IQ score in the NCNG sample at the single marker and 3markers haplotype level (see Table 2), to IQ79 in the LBC1921 (single marker and three markers haplotype, see Table 2) and to IQ11 (see Table 2) and to other cognitive variables (see SOM) at the single marker level in the LBC1936. This association calls for additional studies, but it is interesting to note that the associated markers are located near exon 19 that is alternatively spliced in several transcripts, affecting the DCLK1 kinase activity [50], within a region of high inter-species conservation (as seen in the UCSC: http://genome.ucsc.edu/cgi-bin/hgGateway).
In addition, in both the LBC1921 and LBC1936, single marker association to the marker rs2296645 were observed for verbal memory and cognitive abilities. This genetic variant is located in the exon 11 but does not affect the amino acid sequence.
The thorough examination of the Lothian Birth Cohorts, especially the LBC1936, shows that the signal of association depends on the cognitive variable studied and on the age at examination, which points to similarities and differences in genetic contributions to cognitive abilities across the lifespan. Several cognitive associations in old age were affected by the integration of childhood IQ as a covariate, which can suggest that these are genetic associations to lifelong cognitive change.
In addition, we show that there might be interaction between variants in DCLK1 and variants in BDNF and ARC for an effect on verbal memory and general cognition. This finding is coherent at the genetic level with the in vivo observation that Dclk1 is coupregulated with Arc by BDNF in the rat hippocampus [25]. In addition to the known implication of BDNF in long term potentiation, the interactions with Arc are also noteworthy as this immediate early gene is required for multiple forms of synaptic plasticity and long-term memory formation, including synaptic potentiation induced by BDNF infusion [51,52,53].
Considering the conflicting results reported regarding the genetic association of BDNF with human cognition and with psychiatric disorders [5,6,7,8,9,10], we suggest that re-analysis of the possible BDNF-DCLK1 interaction might improve the interpretation of these studies. In this perspective, it is also interesting to notice that in two recent studies of functional convergent genomics for bipolar affective disorder [22] and a genome wide scan for personality traits [54], both BDNF and DCLK1 (a.k.a. DCAMKL1 in these reports) were ranked as strong candidates.
The rodent Dclk1 gene is expressed as several transcripts, e.g., long (exons 1-20 except 6 and 8), short (exons 6-20 except 8) and Carp (exons 6-8; see figure 1A), which vary in expression (spatial and developmental) and in function. The long transcripts encode an N-terminal domain similar to the lissencephalia-related doublecortin gene, sharing its microtubule-binding and -stabilizing properties [55,56]. N-terminal Dclk1 transgenic and Dcl knockdown mice develop brain abnormalities that affect the organization of hippocampal neurons, cortical neurogenesis, neuronal migration and axonal wiring [56,57]. The C-terminal domain, present in both long-and short-Dclk1, contains a domain similar to the Ca2+/calmodulin-dependent protein kinase which may phosphorylate the myelin basic protein [47,50,58]. The function of Carp remains largely unknown but it is expressed in response to diverse stimuli, and may be involved in both neuronal activityinduced strengthening of synaptic transmission as well as apoptosis [59]. We now show that the short variants are, similarly to Carp, up-regulated by BDNF treatment in the rat hippocampus, and that in human these variants are expressed in brain structures relevant to cognition.
In conclusion, in this study we report DCLK1, a gene upregulated BDNF, as being a novel candidate gene associated to cognitive traits, in three different samples, but with allelic heterogeneity between the samples. Further work will be needed to better understand the function(s) of DCLK1 in cognitive processes, and its role in synaptic plasticity, especially since other related genes (CAMK2G and CAMTA1) have also been associated to human memory performance [60,61]. Drawing on the example of this study, it will be interesting to mine ongoing genome wide studies of large samples characterised for cognitive abilities with further gene sets identified from micro-array gene expression experiments on models relevant to synaptic plasticity.

Supplementary information -SOM
with additional material and methods and complementary tables and figures, is linked to the online version of the paper (see Supporting Information Material S1).

Supporting Information
Supporting Information Material S1 Methods and Tables. Revised version of the supplementary information, as accepted after resubmission 1.