Interaction Effects of BDNF and COMT Genes on Resting-State Brain Activity and Working Memory

Catechol-O-methyltransferase (COMT) and brain-derived neurotrophic factor (BDNF) genes have been found to interactively influence working memory (WM) as well as brain activation during WM tasks. However, whether the two genes have interactive effects on resting-state activities of the brain and whether these spontaneous activations correlate with WM are still unknown. This study included behavioral data from WM tasks and genetic data (COMT rs4680 and BDNF Val66Met) from 417 healthy Chinese adults and resting-state fMRI data from 298 of them. Significant interactive effects of BDNF and COMT were found for WM performance as well as for resting-state regional homogeneity (ReHo) in WM-related brain areas, including the left medial frontal gyrus (lMeFG), left superior frontal gyrus (lSFG), right superior and medial frontal gyrus (rSMFG), right medial orbitofrontal gyrus (rMOFG), right middle frontal gyrus (rMFG), precuneus, bilateral superior temporal gyrus, left superior occipital gyrus, right middle occipital gyrus, and right inferior parietal lobule. Simple effects analyses showed that compared to other genotypes, subjects with COMT-VV/BDNF-VV had higher WM and lower ReHo in all five frontal brain areas. The results supported the hypothesis that COMT and BDNF polymorphisms influence WM performance and spontaneous brain activity (i.e., ReHo).


INTRODUCTION
Working memory (WM) refers to a limited capacity system allowing the temporary storage and manipulation of information (Baddeley and Hitch, 1974) and plays a crucial role in a wide range of cognitive tasks such as reading, problem solving, and learning (Baddeley and Della Sala, 1983;Conway et al., 2003). Researchers have recently examined the genetic underpinnings of WM and identified two candidate genes, Catechol-O-methyltransferase (COMT) and brain-derived neurotrophic factor (BDNF; Egan et al., 2001;Numata et al., 2007;Nagel et al., 2008;. COMT has been found to be associated with dopamine concentration in the prefrontal cortex (PFC; Lachman et al., 1996b;Akil et al., 2003;Chen et al., 2004), which is in turn associated with WM (Egan et al., 2001). The COMT Met allele has also been found to be associated with attention shifting, which plays an important role in the updating of WM (Nolan et al., 2004) and efficient physiological responses during the two-back WM task (Egan et al., 2001). Interestingly, studies have found ethnic differences in the effects of COMT. Most studies that used Caucasian samples (Bruder et al., 2005;Aguilera et al., 2008) found that compared with the Val allele, the Met allele was correlated with better cognitive abilities, especially WM. On the other hand, in Asian samples, the Met allele was instead associated with higher incidence of major depressive disorder (MDD; Wang et al., 2016), lower IQ (Qian et al., 2009), lower scores on competency tests (Yeh et al., 2009), and lower WM .
BDNF plays an important role in neuronal survival and differentiation and synaptic plasticity. Even though not all studies are consistent (for exceptions, see Egan et al., 2003;Hansell et al., 2007), there is a growing number of studies showing associations between BDNF (and its product BDNF) and cognitive ability. On the biological level, several animal studies have shown that performance on WM tasks can be affected by both BDNF overexpression (Papaleo et al., 2011) and under-expression in mice (Li et al., 2010). Higher BDNF levels in the frontal cortex have been linked to fewer WM errors in aged rats  and Ts65D mice . In human studies, BDNF rs6265 Val/Met polymorphisms have been correlated with WM. A study of healthy Han Chinese found that BDNF rs6265 Val homozygotes performed better in digital WM tasks than did Met homozygotes (Gong et al., 2009). In studies of bipolar patients, the Met variant of BDNF (rs6265) has been associated with altered hippocampus and PFC formation, which are in turn linked to poorer WM performance. Egan et al. (2003) found that a common BDNF (val66met) polymorphism was associated with episodic memory, although not with WM as mentioned earlier.
It should be mentioned that due to genetic pleiotropy, both COMT and BDNF have been found to be associated with several other aspects of cognition beyond WM, including reaction time , cognitive flexibility (Kang et al., 2013), agingrelated cognitive degradation (Ghisletta et al., 2014;Voelcker-Rehage et al., 2015), executive function (Leckie et al., 2014;Sapkota et al., 2015), and neurocognitive deficits in people with schizophrenia (Ahmed et al., 2015) or Parkinson's disease (Khalil et al., 2016). Nevertheless, WM has been the main focus of the previous literature in this field.
In terms of specific interaction effects of the two genes on WM, however, there was only one study that found that the effect of COMT on WM was modulated by BDNF in a sample of aging Germans (Nagel et al., 2008). Other studies suggested such an effect indirectly by demonstrating similar interaction effects on the anatomy of the hippocampus (Hajek et al., 2012) and PFC (Gothelf et al., 2005;Ho et al., 2007;Takahashi et al., 2008), brain cortical plasticity and frontostriatal circuit brain activities (Wang et al., 2015), all of which could affect memory performance. Interactions between COMT and BDNF polymorphisms exist for WM likely because both genes are widely expressed in the cortex and white matter (in humans and mice, see The Human Protein Atlas, http://www.proteinatlas.org/, and Allen Brain Atlas, http:// www.brainatlas.org) (Su et al., 2004). They both have been found to influence dopamine levels in the midbrain (Meyer-Lindenberg et al., 2005;Berton et al., 2006) and the PFC (Lachman et al., 1996a;Gogos et al., 1998;Malhotra et al., 2002;Chen et al., 2004).
The current study aimed to explore the potential interactions between BDNF Val66Met and COMT Val158Met on resting brain regional homogeneity (ReHo) as well as WM in a sample of Chinese subjects. Resting-state fMRI has increasingly been used to investigate neural basis of cognitive ability including WM (Christoff and Gabrieli, 2000;Stark and Squire, 2001) both in patients and in healthy people. For example, a previous study found that resting-state functional connectivity was positively correlated with WM performance (Hampson et al., 2006). Zou et al. (2013) further found that the amplitude of low-frequency fluctuation (ALFF) of certain brain areas such as the precuneus was correlated with WM performance (Zou et al., 2013). ReHo, calculated by using Kendall's coefficient concordance (KCC) to measure the degree of synchronization between the timeseries of a voxel and its nearest voxels, is believed to reflect the regional spontaneous blood oxygenation level dependent (BOLD) fluctuations of the whole brain during rest (Zang et al., 2004). Many studies have found abnormal ReHo associated with various brain diseases, such as depression (Guo et al., 2011;Peng et al., 2011), Parkinson's disease (Wu et al., 2009), autism spectrum disorder (Paakki et al., 2010), and Alzheimer's disease and mild cognitive impairment . Compared to resting-state functional connectivity (RSFC), which focuses on the long-distance interregional temporal correlations of BOLD signals, ReHo focuses on the functions within each region, in the case of this study, the frontal lobe. Although, no study has examined the effects of COMT and BDNF on ReHo of restingstate brain activity, the aforementioned literature review led us to hypothesize that COMT and BDNF would have main and interactive effects on both WM and ReHo.

EXPERIMENTAL PROCEDURES Participants
Participants were Chinese college students recruited from Beijing Normal University. Genetic data (BDNF and COMT) and WM data were available for 428 participants, but 11 participants were excluded from further analysis because they were outliers in their WM performance (outside of 3 standard deviations of the group mean). Of the remaining 417 subjects (57% female, mean age = 21.31 years), resting state fMRI data were available for 298 participants ( Table 1). All subjects were Han Chinese and reported no history of psychiatric diseases, head injuries, or stroke/seizure. Moreover, scores on the Beck Depression Inventory (Beck et al., 1996) and Beck Anxiety Inventory (Beck and Steer, 1990) showed that none of the participants met the criteria for either major depression or anxiety disorder. No participants met the criterion for nicotine dependence according to the Fagerström Test for Nicotine Dependence (Heatherton et al., 1991), but four participants in the larger sample reported using a large amount of alcohol based on the Alcohol Use Disorders Identification Test (Saunders et al., 1993).    form was obtained from each participant after a full explanation of the study procedure. Participants were compensated for their participation.

Material and Cognitive Tasks
WM performance was assessed by a two-back WM paradigm with three subtasks (semantic, phonological, and morphological judgment). Each task included four blocks of 10 trials each. Participants were asked to make continuous judgments between the word currently presented and the word presented two trials earlier in terms of their semantic (whether they belonged to the same semantic category), phonological (whether they rhymed with each other), or morphological relations (whether they were the same Tibetan character; Xue et al., 2004). For more details on the specific procedure, please see Li et al. (2011). Following the previous studies (Fry and Hale, 1996;Rosenheck et al., 2006;Zhu et al., 2010;Li et al., 2011;Wang and Gathercole, 2013), the average accuracy scores of the three WM subtasks were used in order to obtain a more stable and reliable estimate of WM. We also measured the participants' IQ with the Wechsler Adult Intelligence Scale. IQ was used as a control variable.
BDNF was genotyped through the polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP) method. The forward primer was ITS-1: (5 ′ -ACTCTGGAGAGCGTGAAT-3 ′ ) and the reverse primer was ITS-2: (5 ′ -ATACTGTCACACACGCTC-3 ′ ). The PCR amplification reactions were performed in a 15 µl reaction mixture containing 2 µl template DNA, 1.5 µl MgCl2, 0.3 µl of each primer, 0.3 µl dNTP, 0.12 µl Taq polymerase, 1.5 µl 10 × buffer, 3 µl PCR Optimizer, and water up to a total volume of 15 µl. The thermocycler conditions were as follows: 95 • C for 5 min, 35 cycles of 95 • C for 30 s, 62 • C for 30 s, 72 • C for 30 s, and a final extension at 72 • C for 10 min. Amplified products of 308 bp was electrophoresed in a 2% agarose gel and visualized with ethidium bromide. The polymorphism was genotyped via restriction digestion with 5 µl of the PCR product, 2U Eco72I, 0.8 µl buffer, 0.08 µl BSA, and water up to the total volume of 8 µl. The digested products were electrophoresed on a 2% agarose gel and visualized with ethidium bromide staining.
Ambiguous or unidentifiable results were re-amplified and rescored. A subset of the samples (10%) were randomly selected and tested twice for confirmation. The detection rates of COMT and BDNF were 97.9 and 99.5%, respectively, in the original larger sample, but only subjects with good genotype data were later enrolled in the WM and brain imaging study.

Brain Imaging Data Collection and Preprocessing
fMRI Data Acquisition MR images were collected using a SIEMENS TRIO 3-Tesla scanner in the Brain Imaging Center of Beijing Normal University. Participants lay supine with their heads snugly fixed by a belt and foam pads to minimize head motion. Each participant underwent an 8 min resting state fMRI (RS-fMRI) scanning session and a 3D anatomic session. During the RS-fMRI session, the participants were instructed to keep their eyes closed, be as still as possible, and not to think about anything in particular. Images were obtained with the following parameters: 33 axial slices, thickness/gap = 3/0.6 mm, matrix = 64 * 64, repetition time (

Image Preprocessing
Data Processing Assistant for Resting-State fMRI (DPARSF, http://rfmri.org/DPARSF; Yan and Zang, 2010) was used to preprocess the RS-fMRI data. Steps included: (1) discarding the first 10 volumes; (2) correcting for within-scan acquisition time differences between slices and head motions; (3) coregistering the T1 image to the mean functional image using a linear transformation; (4) segmenting the coregistered T1 images into gray matter, white matter and cerebrospinal fluid; (5) normalizing the head motion corrected functional images to standard template using the transformation matrix estimated from T1 segmentation and reslicing them to 3 mm isotropic resolution; (6) linear detrending and temporal band-pass filtering (0.01∼0.08 Hz) to reduce the effects of low-frequency drift and high-frequency physiological noise; and (7) regressing out the nuisance signals including the six head motion profiles, global mean signal, cerebrospinal fluid signal, and white matter signal. ReHo was calculated from these preprocessed images using Kendall coefficient of concordance (KCC; Kendall and Gibbons, 1990) based on 27 nearest neighboring voxels. Then the ReHo maps were spatially smoothed (FWHM = 6 mm) and standardized by dividing the maps by the mean ReHo of the entire brain (Zang et al., 2004).

Data Analysis
ANCOVA was used to examine the effects of COMT and BDNF on WM performance with gender, age, and IQ as covariates. With the larger (N = 417) sample with genetic and WM data, we ran a COMT(3 groups) * BDNF(3 groups) ANCOVA.
Genetic effects on resting brain state were analyzed using SPM8 (http://www.fil.ion.ucl.ac.uk/spm) implemented in Matlab 7.8 (The Mathworks, Massachusetts, US) for Windows. Considering the smaller sample size with imaging data, we followed previous research (Li et al.'s, 2016) and combined M/V with M/M to form the Met-carrier group. Subjects were divided into four groups, ordered from high to low in presumed dopamine signaling (see Li et al.'s, 2016): COMT-M+/BDNF-VV, COMT-M+/BDNF-M+, COMT-VV/BDNF-VV, and COMT-VV/BDNF-M+. The design matrix was specified with four regressors defined as "1" (subject in the group) or "0" (subject not in the group). A random effects general linear model was fitted onto ReHo maps with age, gender, and IQ as covariates. Genetic main and interaction effects maps were obtained by setting up contrasts, and Region of Interest (ROI) analysis was further conducted to explore interactions by defining the significant clusters located in the frontal lobe as ROIs and averaging the ReHo in each ROI for each subject. We focused on the frontal lobe for two main reasons. First, the frontal lobe has been consistently found to be a vital brain area correlated with WM in lesion patients and healthy populations (Owen et al., 1990;Curtis and D'Esposito, 2003;Barbey et al., 2013). Most neuroimaging studies have targeted the frontal cortex to investigate the neural basis of N-back WM (Brunoni and Vanderhasselt, 2014). Second, despite the widely distributed  expression of COMT and BDNF gene in the brain, studies have found that WM was associated with dopamine concentration in the PFC (Lachman et al., 1996b;Akil et al., 2003;Chen et al., 2004) and was affected by COMT and BDNF levels in the frontal cortex .Then WM was correlated with the averaged ReHo in each ROI. We also regressed WM to whole-brain ReHo to confirm the relationship between ReHo and WM. Furthermore, as the global brain signal regression (GSReg) may lead to artificial results Saad et al., 2013), we reanalyzed our data without GSReg to confirm the stability of the findings.
For this study, the significance threshold was set at 0.05 for the omnibus F-test in ANCOVA. When the omnibus Fvalue was significant, it was followed by LDS post-hoc test, p < 0.05. The statistical results of the whole-brain analysis of fMRI data were corrected for multiple comparisons using the "AlphaSim" implemented in REST (http://www.restfmri. net; Song et al., 2011). This function is based on the Monte Carlo simulation in AFNI (https://afni.nimh.nih.gov/pub/dist/ doc/manual/AlphaSim.pdf). Voxels with p < 0.01 and cluster Size > 459 mm 3 (17 voxels) were considered significant, which corresponds to corrected p < 0.05.

RESULTS
The mean IQ of the sample was 125.92 for the 417 sample and 125.94 for the 298 sample.

Genotype Distribution
For the COMT gene, 237 subjects were V/V, 157 were V/M, and 23 were M/M. The sample was in Hardy-Weinberg Equilibrium (HWE, χ 2 = 0.11, p > 0.05). The Met minor allele frequency in our sample was 0.24, which is comparable to the HapMap data for Han Chinese (0.29 for Han Chinese in Beijing and 0.26 for Chinese in Metropolitan, Denver, Colorado, according to the HapMap Genome Browser release #28).
For the BDNF gene, the genotype frequencies (113 V/V, 201 V/M, and 103 M/M) were also in HWE (χ 2 = 0.31, p > 0.05). The frequency of the BDNF Met allele (0.49) in our sample is similar to other Asian data, 0.41 in a Japanese sample (Shimizu et al., 2004), 0.47 in a Chinese Han population (Yi et al., 2011), and 0.47 in a Korean sample (Kim et al., 2007).
subsample ( Table 1). They were nevertheless used as covariates in the main analyses.  Figure 1B), partly due to the latter group's small sample size. Post-hoc analyses of the BDNF main effect did not reveal significant contrasts (see Figure 1C). Finally, the interaction effect of the two genes was due to significant simple effects of COMT for the two groups of individuals who were homozygous for BDNF (i.e., V/V and M/M, p = 0.029 and 0.016, respectively), but not for the BDNF-VM individuals. After combining the VM and MM, the COMT-VV was still higher than COMT-M+ for BDNF-VV [F (1, 410) = 8.42, p = 0.004; see Figure S1].

Genetic Polymorphisms and ReHo
The main effects of the COMT gene on ReHo were significant in several regions. COMT Val homozygotes had significant higher ReHo than the COMT Met carriers in the bilateral cerebellum anterior lobe, left cerebellum posterior lobe, left putamen, right hippocampus, fusiform gyrus, and middle occipital gyrus, but lower ReHo in right superior and inferior frontal gyrus, left middle frontal gyrus, bilateral superior temporal gyrus, right postcentral gyrus, bilateral precentral gyrus (Figures 2A,C and Table S1). In terms of the main effects of the BDNF gene, Val homozygotes had significant higher ReHo than the BDNF Met carriers in the bilateral lingual gyrus, left limbic lobe, right precentral and postcentral gyrus, right precuneus, right superior frontal gyrus, but lower ReHo in left middle and inferior temporal gyrus, right superior and medial frontal gyrus and left inferior frontal gyrus, bilateral parahippocampal gyrus, left paracentral lobule, and bilateral cerebellum posterior lobe (Figures 2B,D and Table S2).
To plot the interaction effects, we focused on the five clusters located in the frontal lobe. The mean ReHo for each ROI was extracted for each subject. Simple effects analyses showed that all five ROIs consistently showed a U-shaped pattern across the four groups arranged along the presumed strength of dopamine signaling. COMT-M+/BDNF-VV and COMT-VV/BDNF-M+ had significantly higher ReHo than COMT-M+/BDNF-M+ and COMT-VV/BDNF-VV; Figure 4). Finally, we correlated the average ReHo of each of the five ROIs with WM performance. Results showed that ReHo of rSMFG had a modest negative correlation with WM performance (r = −0.122, p = 0.036). Moreover, the regression analysis with whole-brain images showed that the ReHo of rSMFG showed a negative correlation with WM performance (see Figure 5 and Table 4). The results of the reanalysis without GSReg showed that, although the number of clusters showing significant COMT and BDNF gene effect on ReHo decreased from 26 to 19, the rSMFG was still included (Table S3) and showed the same pattern of results as with GSReg ( Figure S2). Furthermore, we regressed WM to whole-brain ReHo and confirmed the negative relationship between ReHo and WM (Table S4).

DISCUSSION
The present study was designed to examine COMT and BDNF gene interactions on WM and resting state fMRI (i.e., ReHo). As hypothesized, the COMT and BDNF had main and interaction effects on WM. The positive effect of the COMT Val allele on cognitive abilities is consistent with previous research with Asian samples (Qian et al., 2009;Yeh et al., 2009;Wang et al., 2016), but opposite of previous findings from Western samples that COMT-Met carriers showed better cognitive performance in both patients (Bilder et al., 2002;Joober et al., 2002) and healthy individuals (Caldú et al., 2007;Sheldrick et al., 2008). The main effect of BDNF on WM was modest, yielding no significant post hoc contrasts. Indeed, previous research has reported mixed findings, with some studies linking the Val allele to better performance on learning (Soliman et al., 2010), episodic memory (Egan et al., 2003), and WM tasks (Echeverria et al., 2005;Rybakowski et al., 2006), whereas others showed conflicting findings (Hansell et al., 2007;Schofield et al., 2009). Some of these conflicting results may well have been due to potential interactions between BDNF and COMT as found in our study on WM and ReHo, as well as in previous studies on pairedassociates learning, stimulation-induced cortical plasticity (Witte et al., 2012), gray matter volumes (Mortby et al., 2014), and functional connectivity (Wang et al., 2015;Li et al.'s, 2016).
Most relevant to our work, Li et al.'s (2016) found that COMT-M+/BDNF-M+ and COMT-VV/BDNF-VV groups showed higher global functional connectivity density than the other two groups (COMT-VV/BDNF-M+ and COMT-M+/BDNF-VV). These results have been explained by interactive effects of BDNF and COMT on dopamine levels. Both COMT and BDNF influence the cytoarchitecture and connectivity among the subregions. Specifically, BDNF can affect the axonal and dendritic plasticity of neurons and their natural death (Ghosh  (Okazawa et al., 1992;Nishio et al., 1998;Venero et al., 2000). In terms of their interaction, BDNF can stimulate dopamine activity (Blöchl and Sirrenberg, 1996;Murer et al., 2001). Previous biochemical research (Mannisto and Kaakkola, 1999;Winterer and Weinberger, 2004;Savitz et al., 2006;Peciña et al., 2014) showed that the COMT Met allele and the BDNF Val allele were linked to higher dopamine levels than their respective counterparts. Consequently the combination of COMT-M+ and BDNF-VV is expected to lead to a high level of dopamine, and the combination of COMT-VV and BDNF-M+ a low level of dopamine. In contrast, the other two combinations (COMT-VV and BDNF-VV; COMT-M+ and BDNF-M+) are expected to show moderate levels of dopamine. Li et al. (2016) finding that the latter two groups (with presumed medium levels of dopamine signaling) showed higher global functional connectivity density than the other two groups (either low or high levels of dopamine signaling) fits the commonly accepted inverted U-shaped association between dopamine level and outcomes (see Wang et al., 2015;Li et al.'s, 2016 for detailed discussions). Mapping onto Li et al. (2016) findings, we found consistent U-shaped associations between presumed dopamine level and ReHo of five regions in the frontal lobe. These associations were statistically significant in terms of both interactions and posthoc contrasts. As an index of a specific voxel's synchronization with its neighbors, ReHo reflects functional specificity of a brain region (Zang et al., 2004). Altered ReHo has been widely considered as an early indicator of brain dysfunction in patients (Wu et al., 2009Chen et al., 2012;Cui et al., 2016). For example, a few studies have documented increased ReHo in the right dorsolateral PFC of patients with schizophrenia (Cui et al., 2016) and in the right medial frontal gyrus of patients with Parkinson's disease (Wu et al., 2009). Wu et al. (2009) further showed that when patients with Parkinson's disease were treated with levodopa, their right frontal gyrus showed decreased ReHo. Moreover, lower ReHo in a given region may indicate that it is more likely to be modulated by other regions (Yu et al., 2014), which may enhance the updating function in the WM process (Cools and Robbins, 2004).
Moreover, for the first time, the current study found the COMT and BDNF had interaction effect on ReHo. Moreover, ReHo in one of the regions (rSMFG) had a modest negative correlation with WM, which was confirmed by whole-brain analysis. The rSMFG has been consistently related to WM in previous lesion studies (Jacobsen and Nissen, 1937;Milner, 1963), electrophysiological studies (Fuster and Alexander, 1971;Funahashi et al., 1989Funahashi et al., , 1993, rTMS studies (Mottaghy et al., 2002), and fMRI studies (D'Esposito et al., 2000;Burgess and Braver, 2010;Marklund and Persson, 2012).
Consistent with the imaging data showing the importance of a moderate level of dopamine, individuals with the combination of COMT-VV and BDNF-VV (a combination expected to produce a medium amount of dopamine) showed top performance on the WM task. Surprisingly, however, the other combination (COMT-MM and BDNF-MM) that is also expected to produce a medium amount of dopamine showed low scores on the WM task. It should be noted, of course, that this group of minor allele homozygotes had the smallest sample size. A subsequent reanalysis of the data with Met carriers showed that COMT-M+/BDNF-M+ did not differ from COMT-M+/BDNF-VV and COMT-VV/BDNF-M+ (see Figure S1). Future research needs to include a larger sample in order to understand the group of COMT-MM and BDNF-MM.
Finally, although our results were generally consistent with other studies of the interaction effects of COMT and BDNF on other brain functions, our results need to be interpreted with caution due to several limitations of the present study. First, as mentioned earlier, our sample included too few COMT minor allele homozygotes to allow for a detailed analysis of the subjects. Second, the correlation between ReHo of rSMFG and WM was modest. It would not have survived a correction for multiple comparisons. Third, we did not collect any WM task-related fMRI data. Fourth, we did not administer a post-scan questionnaire to assess whether subjects followed the instructions not to fall asleep and not to think of anything in particular during scanning. Finally, given the ethnic differences in the effects of COMT on WM, it is imperative to directly compare these effects between Asian and non-Asian populations.