Expression of TCN1 in Blood is Negatively Associated with Verbal Declarative Memory Performance

Memory is indispensable for normal cognitive functioning, and the ability to store and retrieve information is central to mental health and disease. The molecular mechanisms underlying complex memory functions are largely unknown, but multiple genome-wide association studies suggest that gene regulation may play a role in memory dysfunction. We performed a global gene expression analysis using a large and balanced case-control sample (n = 754) consisting of healthy controls and schizophrenia and bipolar disorder patients. Our aim was to discover genes that are differentially expressed in relation to memory performance. Gene expression in blood was measured using Illumina HumanHT-12 v4 Expression BeadChip and memory performance was assessed with the updated California Verbal Learning Test (CVLT-II). We found that elevated expression of the vitamin B12-related gene TCN1 (haptocorrin) was significantly associated with poorer memory performance after correcting for multiple testing (β = −1.50, p = 3.75e-08). This finding was validated by quantitative real-time PCR and followed up with additional analyses adjusting for confounding variables. We also attempted to replicate the finding in an independent case-control sample (n = 578). The relationship between TCN1 expression and memory impairment was comparable to that of important determinants of memory function such as age and sex, suggesting that TCN1 could be a clinically relevant marker of memory performance. Thus, we identify TCN1 as a novel genetic finding associated with poor memory function. This finding may have important implications for the diagnosis and treatment of vitamin B12-related conditions.


Supplementary
-4.43 -6.30, -2.55 -0.38 4.42e-06* The first learning trial of CVLT was used as a marker of working memory. Memory consolidation was calculated by dividing the long delay free recall score (CVLT2) by the last learning trial of CVLT. Recognition memory was assessed with a separate measure within the CVLT test. All analyses were adjusted for age and sex. Std β: Standardized regression coefficients. * p < 0.05.

Analysis of TCN1 Expression and Memory Stratified by Diagnostic Status
We performed additional regression analyses in which we looked at the effect of TCN1 expression on memory performance within each diagnostic category. In these stratified analyses, we adjusted for age and sex as we did in the initial genome-wide screening. In order to keep the individual sample sizes as big as possible, and thereby avoid losing statistical power, we merged the discovery and replication samples together. Both CVLT and HVLT scores were used as the memory performance metric after transforming raw memory scores to z-scores. The association between TCN1 expression and memory was nominally significant (p<0.05) in SCH and BD groups, but not in the CTRL group (Supplementary Table S4).
However, the direction of the effect was as expected in all groups, i.e. a negative correlation between TCN1 and memory (Supplementary Table S4 and Supplementary Figure S3).
Moreover, the TCN1 effect size was still comparable in magnitude to the effect sizes of age and sex, indicating that there is a real and negative effect of TCN1 expression on memory performance, and that this effect does not seem to be explained by the psychiatric condition.
The fact that the association is not significant in the CTRL group could be due to the large proportion of subjects that were tested with HVLT rather than CVLT in the CTRL group compared with the SCH and BD groups (Supplementary Table S5). As discussed in the main text, HVLT may be less sensitive to detect memory decline because of the shorter word list in HVLT compared to CVLT.

RNA microarray preprocessing and quality control
Multidimensional scaling and hierarchical clustering were used for regular quality control and removal of multiple batch effects (RNA extraction batch, RNA extraction method, DNase treatment batch, cRNA labelling batch, and chip hybridization). The "detectOutlier" function in the R package lumi was used to detect outlier samples based on distance to the cluster center (defined as the average of all samples after removing 10 percent of the samples farthest away from the center). This function detects a sample as outlier when its distance to the center is larger than a certain threshold (we used the default threshold of 2*median distance to the center). In general, the samples that were removed by this QC step had lower numbers of detected transcripts and low signal-noise ratio. In total, 84 samples were identified as outliers and removed from the data file. Since the reproducibility of Illumina gene expression chips is generally high, and since the technology implies that several beads on the array carry the same probe as a built-in technical replicate, we did not perform any separate analysis of technical replicates. The R package illuminaHumanv4.db was used to map Illumina probe identifiers to gene symbols. Mappings were based on data provided by Entrez Gene (ftp://ftp.ncbi.nlm.nih.gov/gene/DATA). Further, to make sure that the significantly associated probes were correctly annotated, we mapped the probe sequences provided by Illumina directly to the human reference using the BLAT tool available at the University of California Santa Cruz Genome Browser website (http://genome.ucsc.edu/).