Abstract
Polymorphisms in the gene encoding catenin-β-like 1 (CTNNBL1) were recently reported to be associated with verbal episodic memory performance—in particular, delayed verbal free recall assessed between 5 and 30 min after encoding—in a genome-wide association study on healthy young adults. To further examine the genetic effects of CTNNBL1, we tested for association between 455 single-nucleotide polymorphisms (SNPs) in or near CTNNBL1 and 14 measures of episodic memory performance from three different tasks in 1743 individuals. Probands were part of a population-based study of mentally healthy adult men and women, who were between 20 and 70 years old and were recruited as participants for the Berlin Aging Study II. Associations were assessed using linear regression analysis. Despite having sufficient power to detect the previously reported effect sizes, we found no evidence for statistically significant associations between the tested CTNNBL1 SNPs and any of the 14 measures of episodic memory. The previously reported effects of genetic polymorphisms in CTNNBL1 on episodic memory performance do not generalize to the broad range of tasks assessed in our cohort. If not altogether spurious, the effects may be limited to a very narrow phenotypic domain (that is, verbal delayed free recall between 5 and 30 min). More studies are needed to further clarify the role of CTNNBL1 in human memory.
Similar content being viewed by others
Introduction
Papassotiropoulos et al.1 recently reported that variants in the gene CTNNBL1 (encoding catenin-β-like 1, located on chromosome 20 q11.23-q12) are associated with verbal episodic memory performance. The authors reached this conclusion based on the data from a genome-wide association study on 1073 healthy young adults from Switzerland; they found that the minor allele of single-nucleotide polymorphism (SNP) rs16986890 was significantly (P=7E−08) associated with better verbal episodic memory performance. This result was corroborated in a second and independent cohort of young adults from Serbia (n=524, P=0.003), although a different, nontypical experimental paradigm was used to assess episodic memory performance in that data set. In addition to these genetic association results, the authors also provided evidence from both gene expression data and functional magnetic resonance imaging experiments to support the notion that rs16986890 in CTNNBL1 may account for genotype-dependent differences in memory-related brain functions.
Although the results reported by Papassotiropoulos et al.1 were consistent for young and highly educated adults, no information was provided as to whether or not they can be generalized to adults with a wider range of ages and more diverse educational backgrounds. Moreover, their findings were based mainly on relatively simple verbal recall tasks. Because the concept of episodic memory has a number of other aspects beyond verbal recall, a more comprehensive assessment of this general phenotype is needed for a better understanding of the potential role(s) of genetic determinants in episodic memory.
The present study represents the first independent attempt to replicate the role of rs16986890 in human episodic memory performance following the initial report. In addition, we substantially extend the analyses of Papassotiropoulos et al.1 by investigating the potential effects of 454 other SNPs in or near CTNNBL1, and by testing a broad range of verbal and nonverbal episodic memory tasks in both young and old adults. Assessments were performed in a large and independent genome-wide association study sample from Germany (n=1743), collected as part of the Berlin Aging Study II. Despite having sufficient power to replicate findings of the original report, we found no evidence for a genetic link between CTNNBL1 and episodic memory performance in our cohort.
Materials and methods
Participants
Our sample was recruited as participants in the Berlin Aging Study II,2 a multidisciplinary project aiming to identify and characterize genetic, psychological, medical and socioeconomic factors relevant to human aging in residents of Berlin, Germany. The sample currently comprises 1946 genetically unrelated, mentally healthy individuals, all of whom were of self-reported Caucasian ancestry. The study was approved by the local Institutional Review Board, and all participants gave signed informed consent before participation. Among the 1743 probands available for analysis in our study, 425 were young adults aged between 20–30 years and the remaining 1318 were old adults aged between 60–70 years.
Assessment of episodic memory performance
To assess the role of CTNNBL1 in human memory, we used 14 quantitative measures of episodic memory derived from: (a) the forward and backward serial recall paradigms;3,4 (b) associative memory tasks that assessed item memory as well as item–pair recognition;5 and (c) an image recognition test at retention intervals of 2.5 and 1 h.6 A detailed description of the tasks and traits analyzed can be found in Supplementary Table 1. For a description on how multicollinearity among traits is dealt with, see description of the statistical analysis procedures below.
Forward and backward serial recall task
In this task, participants were presented with six different lists of 12 words each. After the presentation of the last item in each list, participants were asked to recall each word at its correct position. Word lists 1–3 were recalled in forward order (from the first word to the last word of the list), whereas word lists 4–6 were recalled in backward order. Recall was self-paced. For each list, responses were scored using a strict serial recall criterion: an accurate response required that both the word and its serial position were correct.
Item and pair associative episodic memory task
The item and pair associative episodic memory task (henceforth labeled as the ‘item–pair’ task) had four conditions that were tested sequentially in one session. During an initial study phase, participants were visually presented with pairs of unrelated words and were instructed to study each pair under two conditions: either as two single words (item instruction) or as a pair of words (pair instruction).
The study phase of each condition contained 30 pairs of semantically unrelated words. In the test phrase, subjects in one condition (item recognition) were asked whether they had seen the presented word during the study phase. Half of the words were old (target items), and the other half were new (distractor items). In the second condition (associative recognition), subjects had to decide whether a presented pair of words had been presented during the study phase. Half of the presented word pairs were old (target pairs), and the others were formed by recombining words in the previously studied lists (rearranged distractor pairs). Taken together, by crossing over the two instruction conditions with the two test conditions, the task resulted in four conditions that assessed item memory (item–item and pair–item tests) and associative memory (pair–pair and item–pair tests). Recognition memory performance was measured as hits minus false alarms to minimize the effects of potential individual differences in response bias.
Image recognition memory with retention
Performance in the image recognition memory task was assessed at two retention intervals: 2.5 h and 1 week. At the beginning of the first session, participants were presented with 48 complex, colored images of scenes of neutral emotional valence; all were derived from the International Affective Picture System.7
The images were encoded incidentally: during the study phase, participants were required to determine whether the scene was ‘indoor’ or ‘outdoor’—there were 24 scenes in each category—without explicit requirement of memorization. During retrieval, participants viewed each image for 3 sec and were asked to determine whether each scene had been presented (‘old’) or not (‘new’) during encoding. In each retrieval test, 24 unique old scenes and 24 unique new scenes (lures) were presented. Taking response bias into account, memory performance was measured as hits minus false alarms.
Genotyping, SNP imputation and quality control procedures
DNA of each participant was extracted from whole blood using standard procedures, and it was then subject to microarray-based SNP genotyping using the Affymetrix ‘Genome-Wide Human SNP Array 6.0’. Before imputation, SNPs violating Hardy–Weinberg equilibrium at P⩽1E−06 and those with a call rate <98%—two commonly used quality control filters—were excluded. This resulted in 829 344 autosomal SNPs in 1946 participants.
Among these individuals, 214 were excluded from subsequent analysis. Each of them had at least one of the following conditions: (a) missing information on age or years of education; (b) <95% call rate; (c) evidence for sample duplication, relatedness or contamination; (d) inconsistency between recorded and genotypic sex; (e) excessive heterozygosity; and (f) population outlier, which was determined by the EIGENSOFT program;8 specifically, because all participants were of self-reported Caucasian descents, we excluded ethnic outliers using the Eigenstrat function in EIGENSOFT with iterative outlier removal. After the above filtering steps, principal components (PCs) were computed again for the 1743 remaining samples. On the basis of the examination of the scree plot, the first four PCs were retained and used as covariates for the subsequent association analysis, to adjust for potential residual population stratification.
Genome-wide imputation of unobserved genotypes was carried out on the ‘cleaned’ data set using the IMPUTE v2.3.2 software,9, 10 on the basis of the precompiled ‘1000 Genomes Phase I Integrated Variant Set’ reference panels from the IMPUTE website (March 2012 release). As suggested by Southam et al,11 we also applied post-imputation quality control filtering, including only SNPs with IMPUTE- information thresholds ⩾0.8 and minor allele frequencies at or above 5%. After this post-imputation filtering, a total of 455 high-quality SNPs, 71 genotyped and 384 imputed, within a ±50 kb window surrounding the CTNNBL1 locus (that is, between start bp 36 272 434 and end bp 36 550 520; hg19 human reference genome assembly) were retained for subsequent statistical analyses.
Association analyses
The phenotypes we used for evaluating participants’ episodic memory performance are functionally related and statistically correlated. The 14 phenotypes are correlated with an average correlation coefficient (r) of 0.46. One strategy to analyze these data is to test each SNP against each of the 14 phenotypes individually. Although this approach is straightforward, it is limited by not incorporating potentially useful information from the structure of multiple (and partially correlated) phenotypes. To address this issue, we also used a second approach that can test several correlated phenotypes simultaneously. As suggested by a recent review,12 we first applied principal component analysis (PCA) to condense information in the phenotypes by extracting a small number of orthogonal variables (that is, the PCs) that were weighted linear combinations of the original phenotypes. The extracted variables, which were the first three PCs from PCA, were then used for association analyses in place of the original phenotypes. As expected, the three components were correlated to the three different episodic memory tasks, and altogether they could explain 80% of the phenotypic variance.
Association analyses were carried out using the episodic memory measures (that is, each trait individually and the PCA variables) as quantitative traits in an additive linear model, adjusted for age, gender, years of education, as well as the four PCs to account for potential population stratification. All analyses were performed separately in the ‘old’ and ‘young’ strata, to avoid stratification problems. Association tests were performed using SNPTEST v.1.3,13 which can account for the uncertainty of imputed genotype calls via missing data likelihood tests.
Power calculations
The power of our study was assessed in the young and old subgroups separately and was based on the reported effect sizes from the original study.1 Monte Carlo simulations were performed with 1000 runs for empirical power calculations. For the one-trait-at-a-time approach, we used the SimpleM software14, 15 to account for the correlations among the 14 test items and the correlations among SNPs due to linkage disequilibrium.16 The results of this analysis, that is, the effective number of independent tests, were then used for Bonferroni-correction to account for multiple testing.16 The estimated effective number of independent tests was 11 and 60 across the 14 phenotypes (or traits) and 455 SNPs, respectively. Hence, the experiment-wide corrected alpha level was set to 4.55E−03 (that is, 0.05/11), for testing the association between at least one of the traits and the previously reported significant SNP rs16986890; and it was 7.58E−05 (that is, 0.05/(11 × 60)) for testing the association between at least one of the traits and any SNP in the CTNNBL1 gene region.
Our power to detect an association between at least one of the traits and rs16986890 at the originally reported effect size was between 93% and 100% for the ‘young’ stratum and 100% for the ‘old’ stratum (see Table 1). When we extended our search to the whole CTNNBL1 gene region, the power to detect association between at least one of the traits and any of the 455 SNPs in the CTNNBL1 gene region was between 61% and 97% for the ‘young’ stratum and 100% for the ‘old’ stratum (see Table 2). With the PCA approach, our power to detect association between at least one of the three PCs (that is, the extracted variables from PCA of the 14 traits) and rs16986890 was between 90% and 100% for the ‘young’ stratum and 100% for the ‘old’ stratum; and it was between 56% and 96% for the ‘young’ stratum and 100% for the ‘old’ stratum when testing possible associations between at least one of the PCs and any SNP in the CTNNBL1 gene region.
Results
After quality control and adjusting for potential population stratification, there was only minimal evidence of P-value inflation: λ ranged from 1.00–1.04 across the 14 traits tested in the ‘young’ and ‘old’ strata. The minor allele frequency of rs16986890 was 0.056 and 0.058 for the ‘young’ and the ‘old’ in our German sample, respectively, which are consistent with data (that is, 0.058) reported by the 1000 Genomes Project for European (CEU) samples.18
As seen in Table 1, there was no evidence that SNP rs16986890, which elicited the strongest signal in the original report,1 was associated with any of the 14 memory measures after correcting for multiple comparisons. Adjusted results approached experiment-wide significance (corresponding to a nominal P-value of 4.55E−03, see Materials and Methods) for two memory measures (‘delay_1_week’ (P=0.0671) and ‘PC3’ (0.0639)) in the ‘old’ stratum. However, the directions of these effects were opposite to what was reported in the original study, suggesting that worse, instead of better, memory performance was related to the minor allele of rs16986890. Moreover, as seen in the local chromosome region views in Figure 1 and the Supplementary Figures, this ‘signal’ was indistinguishable from noise. Finally, there was no evidence that any of the other 454 SNPs was significantly associated with the memory measures (all adjusted P-values >0.05 (corresponding to a nominal P-value of 7.58E−05, see Materials and Methods)). Results obtained using the PCA approach were very similar to those obtained with the one-trait-at-a-time analyses. Reanalysis of all comparisons without adjusting for years of education—which might themselves have a weak genetic component19 did not change the results appreciably (data not shown). Summaries of the full results can be found in Table 1, Table 2, Figure 1 and the Supplementary Figures.
Discussion
In this study, we comprehensively investigated the potential effects of common genetic variants in or near CTNNBL1 on a broad range of verbal and nonverbal episodic memory tasks for both young and old adults. Assessments were performed in over 1700 individuals recruited as part of the Berlin Aging Study II. In contrast to the recently reported findings by Papassotiropoulos et al.,1 our independent data provide no support for the notion that SNPs in the CTNNBL1 gene region, including rs16986890, exert any significant effects on episodic memory performance. This conclusion holds true regardless of whether we considered the various tested cognitive traits individually or in combined analyses. Likewise, we saw no evidence of association with respect to age. To the best of our knowledge, this is the first independent replication attempt since the publication of the original report. On the basis of our data, it remains highly doubtful that genetic variants in CTNNBL1 are genuinely involved in mechanisms controlling episodic memory in humans.
The reason for the observed discrepancy between our results and those from the original paper remain elusive. Although we did not apply the same memory tests (that is, verbal delayed free recall between around 5 and 30 min) as in Papassotiropoulos et al.,1 our assessments cover a wide range of related tasks making it unlikely that we have missed a strong general effect of CTNNBL1 on episodic memory. Most of the tests applied to our participants can be considered more challenging than the free recall tests used in the original study. For instance, the serial recall tasks place more demand on associative memory than free recall, which relies mainly on item memory.3,17 Likewise, the delayed image recognition test used here6 involves a much longer retention interval than 30 min as applied in the original study. Therefore, we cannot rule out the possibility that a potential effect of SNP rs16986890 (or other SNPs in the CTNNBL1 region) is limited to a very narrow phenotypic domain, that is, verbal delayed free recall between around 5 and 30 min.
In summary, our study provides considerable negative evidence against the notion that genetic variants in CTNNBL1 are associated with episodic memory performance. More independent assessments with sufficiently large sample sizes are needed to further clarify the potential role, if any, of this gene in human memory.
References
Papassotiropoulos A, Stefanova E, Vogler C, Gschwind L, Ackermann S, Spalek K et al. A genome-wide survey and functional brain imaging study identify CTNNBL1 as a memory-related gene. Mol Psychiatry 2011; 18: 255–263.
Bertram L, Böckenhoff A, Demuth I, Düzel S, Eckardt R, Li S et al. Cohort profile: the Berlin Aging Study II (BASE-II). Int J Epidemiol 2014; 43: 703–712.
Lewandowsky S, Murdock B Jr . Memory for serial order. Psychol Rev 1989; 96: 25.
Li S, Papenberg G, Nagel I, Preuschhof C, Schröder J, Nietfeld W et al. Aging magnifies the effects of dopamine transporter and D2 receptor genes on backward serial memory. Neurobiol Aging 2013; 34: 358–1.
Preuschhof C, Heekeren H, Li S, Sander T, Lindenberger U, Bäckman L . KIBRA and CLSTN2 polymorphisms exert interactive effects on human episodic memory. Neuropsychologia 2010; 48: 402–408.
Papenberg G, Bäckman L, Chicherio C, Nagel I, Heekeren H, Lindenberger U et al. Higher intraindividual variability is associated with more forgetting and dedifferentiated memory functions in old age. Neuropsychologia 2011; 49: 1879–1888.
Lang P Bradley M Cuthbert B . International Affective Picture System (IAPS), 1st edn. Gainesville, FL, USA: NIMH, Center for the Study of Emotion & Attention, 2005.
Price A, Patterson N, Plenge R, Weinblatt M, Shadick N, Reich D . Principal components analysis corrects for stratification in genome-wide association studies. Nat Genet 2006; 38: 904–909.
Marchini J, Howie B . Genotype imputation for genome-wide association studies. Nat Rev Genet 2010; 11: 499–511.
Howie B, Donnelly P, Marchini J . A flexible and accurate genotype imputation method for the next generation of genome-wide association studies. PLoS Genet 2009; 5: 1000529.
Southam L, Panoutsopoulou K, Rayner N, Chapman K, Durrant C, Ferreira T et al. The effect of genome-wide association scan quality control on imputation outcome for common variants. Eur J Hum Genet 2011; 19: 610–614.
Ott J, Wang J . Multiple phenotypes in genome-wide genetic mapping studies. Protein Cell 2011; 2: 519–522.
Marchini J, Howie B, Myers S, McVean G, Donnelly P . A new multipoint method for genome-wide association studies by imputation of genotypes. Nat Genet 2007; 39: 906–913.
Gao X . Multiple testing corrections for imputed SNPs. Genet Epidemiol 2011; 35: 154–158.
Cheverud J . A simple correction for multiple comparisons in interval mapping genome scans. Heredity 2001; 87: 52–58.
Nyholt D . A simple correction for multiple testing for single-nucleotide polymorphisms in linkage disequilibrium with each other. Am J Hum Genet 2004; 74: 765–769.
Li S, Chicherio C, Nyberg L, von Oertzen T, Nagel I, Papenberg G et al. Ebbinghaus revisited: influences of the BDNF Val66Met polymorphism on backward serial recall are modulated by human aging. J Cogn Neurosci 2010; 22: 2164–2173.
1000 Genomes Project Consortium, Abecasis GR, Altshuler D, Auton A, Brooks LD, Durbin RM et al. A map of human genome variation from population-scale sequencing. Nature 2010; 467: 1061–1073.
Rietveld C, Medland S, Derringer J, Yang J, Esko T, Martin N et al. GWAS of 126,559 individuals identifies genetic variants associated with educational attainment. Science 2013; 340: 1467–1471.
Acknowledgements
This work is supported by the German Federal Ministry of Education and Research (BMBF (grants #16SV5536K, #16SV5537, #16SV5538 and #16SV5837; previously #01UW0808)). Another source of funding is the Max Planck Institute for Human Development, Berlin, Germany.
Author information
Authors and Affiliations
Corresponding authors
Ethics declarations
Competing interests
The authors declare no conflict of interest.
Additional information
Supplementary Information accompanies the paper on the Translational Psychiatry website
Supplementary information
Rights and permissions
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-nd/3.0/
About this article
Cite this article
Liu, T., Li, SC., Papenberg, G. et al. No association between CTNNBL1 and episodic memory performance. Transl Psychiatry 4, e454 (2014). https://doi.org/10.1038/tp.2014.93
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/tp.2014.93