Elsevier

NeuroImage

Volume 36, Issue 3, 1 July 2007, Pages 736-745
NeuroImage

A validated network of effective amygdala connectivity

https://doi.org/10.1016/j.neuroimage.2007.03.022Get rights and content

Abstract

Regulatory interactions with the amygdala are thought to be critical for emotional processing in the extended limbic system. Structural equation modeling (path analysis) is a widely used method to quantify interactions among brain regions based on connectivity models, but is often limited by lack of precise anatomical and functional constraints. To address this issue, we developed an automated elaborative path analysis procedure guided by known anatomical connectivity in the macaque. We applied this technique to a large human fMRI data set acquired during perceptual processing of angry or fearful facial stimuli. The derived models were inferentially validated using a bootstrapping split-half approach in pairs of 500 independent groups. Significant paths across the groups were used to form a rigorously validated and consistent path model. We confirm and extend previous observations of amygdala regulation by an extended prefrontal network encompassing cingulate, orbitofrontal, insular, and dorsolateral prefrontal cortex, as well as strong interactions between amygdala and parahippocampal gyrus. This validated model can be used to study neurocognitive correlates as well as genotype or disease-related alterations of functional interactions in the limbic system.

Introduction

The limbic system is essential to emotional processing (LeDoux, 2000). The amygdala is a critical node in this network, and is necessary for imbuing percepts with affective significance, especially for fearful or dangerous environmental stimuli (Amaral and Price, 1984, LeDoux, 2000). In agreement with this, studies of patients with lesions in the amygdala (Adolphs et al., 1994) have consistently shown deficits in the ability to identify the affect of faces. Functional neuroimaging studies have also implicated amygdala activity in fear response (Morris et al., 1996, Hariri et al., 2000) with highest amygdala response to fearful or angry faces.

However, while the amygdala is a central part of the neural circuitry for emotion, it does not operate in isolation. In fact, the importance of functional interconnectedness between component structures of the “limbic lobe” was stressed even in 1937, when Papez emphasized their “harmonious mechanism” as the basis of emotional regulation (Papez, 1995). Functional interactions between component structures in the limbic system are wide-ranging. Tracer studies in macaque monkeys show that the amygdala is extensively anatomically connected with cortical areas including cingulate, prefrontal cortex, parahippocampal gyrus, and insula (Barbas and De Olmos, 1990, Amaral and Price, 1984, Ghashghaei and Barbas, 2002, Stefanacci et al., 1996) and subcortical areas including the hippocampus (Saunders et al., 1988). Therefore, defining the functional limbic network is expected to provide a better characterization and classification of healthy and abnormal function related to emotional cognition and regulation. Network definition requires neuroimaging methodology which takes functional interconnections into specific account. Here, we use structural equation modeling (SEM) or path analysis, a procedure to identify directional interaction among regions given the pairwise correlations of their time series, in a large data set of 83 healthy humans scanned during a well-validated task involving perceptual judgment on angry and fearful faces (Hariri et al., 2002).

SEM has been applied in functional neuroimaging for over a decade (McIntosh and Gonzalez-Lima, 1994, Bullmore et al., 2000, Steele et al., 2004). Despite its wide acceptance in the field, applying SEM to neuroimaging data is not straightforward. One important difficulty comes from the fact that results from SEM as it is commonly used are only inferential insofar as they support or do not support an a priori model of connectivity. This poses a problem since our current anatomical knowledge often does not constrain modeling of interactions to a sufficient degree in the highly interconnected limbic system.

To overcome this obstacle, we used an elaborative approach: we started with a “nuclear model” specifying only a small number of very well validated connections (Ghashghaei and Barbas, 2002, Paus, 2001, Phillips et al., 2003) and then used a data-driven search algorithm (Bullmore et al., 2000, Sorbom, 1989) to iteratively add paths, constrained by known anatomical connectivity (Kötter, 2004), until a parsimonious model was formed. Since the final model selected was thus guided by the data, it is possible that derived paths might reflect noise in the sample and not the inherent causal structure of the data. To guard against this eventuality of over-fitting and to provide a stringent test of the derived model, a bootstrapping approach was used in which the subject pool was combined into pairs of 500 independent groups, with each group containing 20 subjects, in order to verify the models. For each pair, a path model was independently derived on the first group of the data and then forced upon the second group for inferential validation. Significant paths across the first group of each pair were used to form a rigorously validated and consistent path model.

Our results show that a well-fitting model of limbic circuitry can be derived and statistically validated from functional MRI. The properties of the model confirm and extend current knowledge about functional interactions in the human limbic system and provide a framework of effective connectivity that can be used to study genetic and disease-related variation across individuals.

Section snippets

Materials and methods

The automated path analysis procedure used on the first group of each pair is shown graphically in Fig. 1 and largely follows Bullmore et al. (2000) with exceptions in the use of modification index (Sorbom, 1989), estimate of effective degrees of freedom (Kruggel et al., 2002), model build based on a nuclear model, and correlations derived from residual activity as discussed below.

Results

A model was independently derived from the first of each group of 500 pairs. These models reached an average parsimonious fit index of ρ = 0.75 ± 0.07, and all 500 derived models survived the P > 0.05 threshold. The maximum of the parsimonious fit index across all 500 groups is shown in Fig. 3.

For validation, the connections derived automatically from the first group in each pair were subsequently forced on the observed correlation matrix derived independently from the second group in that pair. 499

Discussion

In the present work, we used a data-driven approach to construct a parsimonious model of effective connectivity during neural processing of fearful stimuli that was validated using a bootstrapping approach in a large data set of healthy participants. Our data show that iterative search algorithms guided by known neuroanatomy are a feasible approach to the characterization of neural interactions in the human brain. The derived model confirms and extends previous results on human amygdala

Acknowledgments

We would like to thank Dr. Douglas Steele for consultation on structural equation modeling. This research was supported by the Intramural Program of the National Institute of Mental Health.

References (60)

  • J.L. McGaugh et al.

    Role of adrenal stress hormones in forming lasting memories in the brain

    Curr. Opin. Neurobiol.

    (2002)
  • A. Meyer-Lindenberg et al.

    Neural basis of genetically determined visuospatial construction deficit in Williams syndrome

    Neuron

    (2004)
  • W.D. Penny et al.

    Modeling functional integration: a comparison of structural equation and dynamic causal models

    NeuroImage

    (2004)
  • E.A. Phelps

    Human emotion and memory: interactions of the amygdala and hippocampal complex

    Curr. Opin. Neurobiol.

    (2004)
  • M.L. Phillips et al.

    Neurobiology of emotion perception I: the neural basis of normal emotion perception

    Biol. Psychiatry

    (2003)
  • G.J. Quirk et al.

    Prefrontal mechanisms in extinction of conditioned fear

    Biol. Psychiatry

    (2006)
  • S.L. Rauch et al.

    Neurocircuitry models of posttraumatic stress disorder and extinction: human neuroimaging research-past, present, and future

    Biol. Psychiatry

    (2006)
  • A.P. Smith et al.

    Task and content modulate amygdala–hippocampal connectivity in emotional retrieval

    Neuron

    (2006)
  • J.D. Steele et al.

    Neural predictive error signal correlates with depressive illness severity in a game paradigm

    NeuroImage

    (2004)
  • R. Adolphs et al.

    Impaired recognition of emotion in facial expressions following bilateral damage to the human amygdala

    Nature

    (1994)
  • D.G. Amaral et al.

    Amygdalo-cortical projections in the monkey (Macaca fascicularis)

    J. Comp. Neurol.

    (1984)
  • H. Barbas et al.

    Projections from the amygdala to basoventral and mediodorsal prefrontal regions in the rhesus monkey

    J. Comp. Neurol.

    (1990)
  • H. Barbas et al.

    Architecture and intrinsic connections of the prefrontal cortex in the rhesus monkey

    J. Comp. Neurol.

    (1989)
  • K.A. Bollen

    Structural Equations with Latent Variables

    (1988)
  • C. Büchel et al.

    Modulation of connectivity in visual pathways by attention: cortical interactions evaluated with structural equation modelling and fMRI

    Cereb. Cortex

    (1997)
  • A.J. Calder et al.

    Neuropsychology of fear and loathing

    Nat. Rev., Neurosci.

    (2001)
  • C. Cavada et al.

    The anatomical connections of the macaque monkey orbitofrontal cortex. A review

    Cereb. Cortex

    (2000)
  • R. Cudeck et al.

    A simple Gauss-Newton procedure for covariance structure analysis with high-level computer languages

    Psychometrika

    (1993)
  • E.M. Drabant et al.

    Catechol O-methyltransferase Val158Met genotype and neural mechanisms related to affective arousal and regulation

    Arch. Gen. Psychiatry

    (2006)
  • P. Ekman et al.

    Pictures of Facial Affect

    (1976)
  • Cited by (347)

    View all citing articles on Scopus
    View full text