Studies of the BDNF gene polymorphism have reliably documented episodic memory deficits associated with the met allele (Egan et al., 2003
; Hariri et al., 2003
). Contemporary models of declarative memory function suggest that strong neural connectivity between neocortical association areas and medial temporal lobe (MTL) regions underlies successful memory encoding, long-term maintenance, and retrieval (Ranganath et al., 2005
). In this study we did not restrict differences in FC to only within-network regions; consequently, we observed reliable effects across networks. More specifically, we observed a reduction in hippocampal and parahippocampal to cortical connectivity at rest in carriers of the met allele within each of the three resting networks we examined: the default-mode, executive, and paralimbic networks (see Table ). This work suggests that differences are present between genetic groups both within networks and in the connectivity of these networks to other parts of the brain. Analyses of all three networks provided evidence of higher basal connectivity between MTL structures and neocortical association areas in val-allele homozygotes than in met-allele carriers. These differences in resting FC may result from differences in anatomical connectivity, in which met-allele carriers have less robust cortical-to-MTL projections, particularly in the network important for executive control and in the default-mode network.
Association studies have linked the BDNF gene to substance-related disorders, eating disorders, and schizophrenia; the association of the BDNF gene with major depression disorders and bipolar disorders, however, has been inconclusive (see, for example, meta-analysis by Gratacos et al., 2007
). Contemporary theory suggests that the allele (met vs. val) that confers risk may change across development, differing in trajectory for various disorders (Casey et al., 2009
). Comprised primarily of insular and cingulate cortices that are connected with subcortical limbic structures, the paralimbic network may be the neural network most broadly relevant to psychiatric disorders for the role it is expected to exert in the detection, integration, and filtering of interoceptive, autonomic, and emotional information (Seeley et al., 2007
; Habas et al., 2009
). The insula plays a critical role in the experience of emotion, interoceptive awareness, and bodily homeostasis (Critchley et al., 2004
; Pollatos et al., 2007
). In a complementary fashion, the cingulate cortex is implicated in resolution of emotional conflict, in part through mediation of MTL structures, in particular, the amygdala (Etkin et al., 2006
). Overall, the system is posited to play an important role in response selection and orientation toward action by integrating its numerous inputs. Given the unique role of the paralimbic neural network in integrating sensory stimuli and in generating responses to those stimuli, it is noteworthy that the two genotype groups differ in resting connectivity in this circuit. In the present study we demonstrate that connectivity between this network and neocortical association areas and the amygdala is enhanced in met-allele carriers. It is possible, therefore, that greater connectivity between the cortico-limbic structures underlies both the risks associated with this allele (e.g., disorders, schizophrenia) and the protective effects of this allele (e.g., reduced substance abuse, lower levels of neuroticism).
We found lower resting connectivity in widespread brain regions in met-allele carriers than in val-allele homozygotes in both the default-mode and executive networks. We emphasized the differences observed in parahippocampal and hippocampal regions because of the major role these structures play in memory formation and the known association between the met allele and poorer episodic memory (Egan et al., 2003
In this paper, we interpret the obtained genotype differences as reflecting greater coherence between the MTL and our seed regions in individuals who are homozygous for the polymorphism of BDNF that confers higher levels of that growth factor in the MTL. Based on our analysis, the relations between our seeds and the MTL peaks range by group from z
0.3 to 4.1. The finding that val-allele homozygotes have greater FC between hippocampal and both executive and salience networks is surprising given previous reports suggesting that these networks are negatively correlated with the MTL (Fox et al., 2005
). It is possible that by not performing global normalization, our results are more reliable in assessing the nature of the relation between these seeds and MTL, and this may account for the observed difference (Chang et al., 2009
; Murphy et al., 2009
; Weissenbacher et al., 2009
As we point out above, effective memory processing relies on distributed cortical and subcortical brain regions. In fact, the resting-state approach has been particularly fruitful in demarcating major memory networks by examining healthy participants and showing that regions of the precuneus, posterior cingulate, and parietal lobule that show greater activity to successfully remembered items are the same regions that are connected at rest to the hippocampal formation (Vincent et al., 2006
), and by drawing conceptual links between memory impairments in Alzheimer patients and reduced resting-state connectivity in the DMN in the hippocampus (Greicius et al., 2004
). Collectively, studies examining resting-state networks have found that levels of connectivity and integrity in these networks are related to the anomalies in the psychological and cognitive operations served by those networks (for reviews of DMN and neuropsychiatric disorders, see Greicius, 2008
; Broyd et al., 2009
). In the present study we found reduced connectivity in default-mode and executive networks in a number of regions known to be implicated in memory processing (hippocampus, precuneus, BA 31, inferior parietal lobule). We suggest, therefore, that reductions in FC in met-allele carriers observed for the DMN and EN may represent anomalous development of these cognitive networks in children and adolescents, a process that may give rise to the kinds of memory impairments previously reported in adult BDNF met-allele carriers (Egan et al., 2003
; Hariri et al., 2003
; Dempster et al., 2005
; Ho et al., 2006
There are a number of caveats concerning this work that warrant mention. First, genotype interacts with several factors, including age, pubertal status, and gender, to influence the intermediate phenotype of brain biology. As is typical of most neuroimaging studies, our sample size limited the number of analyses that we could reliably conduct. Moreover, our sample of children is relatively heterogeneous, and other factors will certainly contribute variance to the observed findings. Second, the data are not reported using multiple comparisons correction. Without this correction we have accepted more chance for Type 1 error. Our future work will attempt to replicate these findings in an independent sample. The final caveat concerns the ROI-based approach and the selection of seed regions. We have attempted to differentiate connectivity between each network and the medial temporal lobe, but have done so using seed regions that are not completely independent of one another. Indeed, investigators have documented some degree of overlap among these networks, and between these networks and sensorimotor networks (Seeley et al., 2007
; Habas et al., 2009
). Consequently, resting state networks should not be interpreted as independent networks of the brain, but instead, as having some degree of interplay.
In summary, the reduced mnemonic function, altered MTL function, reduced hippocampal and cortical volume, and disposition to affective illness attributed to the BDNF met allele, combined with the functional connectivity analysis in our study, suggest that reduced BDNF has implications for the formation of large-scale neural networks. Our results extend previous fMRI findings by identifying altered neural system-level functional cooperativity in carriers of the met-allele. It remains for future research to elucidate the nature of the associations among these measures in affecting specific behavioral phenotypes. The role of exogenously applied neurotrophins, for example, is an active area of investigation, and advances in this area will inform our understanding of how these large-scale networks are formed and maintained. In addition, this work will help to generate novel hypotheses about the neural underpinnings of phenotypes associated with specific BDNF alleles. For instance, increased FC in paralimbic networks may also characterize individuals with an eating disorder, given the association of the met allele with that phenotype. The present results highlight the importance of integrating neural and genetic data in elucidating phenotypic behaviors.