The Hierarchy of Brain Networks Is Related to Insulin Growth Factor-1 in a Large,Middle-Aged,Healthy Cohort: An Exploratory Magnetoencephalography Study

Abstract

Recently, a large study demonstrated that lower serum levels of insulin growth factor-1 (IGF-1) relate to brain atrophy and to a greater risk for developing Alzheimer's disease in a healthy elderly population. We set out to test if functional brain networks relate to IGF-1 levels in the middle aged. Hence, we studied the association between IGF-1 and magnetoencephalography-based functional network characteristics in a middle-aged population. The functional connections between brain areas were estimated for six frequency bands (delta, theta, alpha1, alpha2, beta, gamma) using the phase lag index. Subsequently, the topology of the frequency-specific functional networks was characterized using the minimum spanning tree. Our results showed that lower levels of serum IGF-1 relate to a globally less integrated functional network in the beta and theta band. The associations remained significant when correcting for gender and systemic effects of IGF-1 that might indirectly affect the brain. The value of this exploratory study is the demonstration that lower levels of IGF-1 are associated with brain network topology in the middle aged.

Introduction

The structural and functional networks in the healthy human brain have shown to contain aspects of scale-free (Chialvo, 2010) and small-world networks (Sporns and Zwi, 2004), as well as hierarchical modularity (Meunier et al., 2010). That is, the configuration of brain networks may reflect the outcome of an optimization process tuned to obtain an efficient communication between areas while avoiding that individual regions become overloaded (Boersma et al., 2013). During normal aging, changes in global functional integration (Stam, 2014) might provoke cognitive decline, often referred to as the “disconnection” hypothesis (Geschwind, 1965a,b; Hof and Morrison, 1994; Morrison et al., 1990; O'Sullivan et al., 2001). Consistently, network analyses have shown that alterations in the topology of functional brain networks occur with aging. These changes show globally a less efficient network, with increased local clustering and reduced centrality of hub regions (Achard and Bullmore, 2007; Otte et al., 2015; Wang et al., 2012). Furthermore, it has been proposed that overload of specific areas and cascading effects across the network might explain vulnerability to neurodegeneration (de Haan et al., 2012), constituting a common pathophysiological mechanism shared by many different etiologies of neurodegeneration (Stam, 2014). In addition, it has been demonstrated that within the Amsterdam Growth and Health Longitudinal Study (AGHLS) cohort, global functional interaction in the network relates positively to physical fitness and intelligence quotient (Douw et al., 2014).

Insulin growth factor-1 (IGF-1) is a protein primarily secreted by the liver, although it is also synthesized to a lesser extent in the brain (Torres-Aleman, 2010). IGF-1 plays an important role in the human brain (Adamo et al., 1989); both in vitro and animal studies have shown that IGF-1 affects a number of molecules that are key players in regulating neuronal homeostasis and thereby regulates neuronal survival (Barber et al., 2001; Barthwal et al., 2003; Dupraz et al., 2013; Finkbeiner, 2000; Hui et al., 2005; Jones et al., 2003; Li et al., 2008; Rommel et al., 2001). Furthermore, IGF-1 associates with the functioning of neuronal circuitry and networks, relating to both presynaptic (Xing et al., 2007) and postsynaptic (Huang et al., 2010) transmissions in the hippocampus.

Numerous studies in humans have investigated the levels of IGF-1 in Alzheimer's disease (AD), Parkinson's disease (PD), and other neurodegenerative disorders [for a review, see Bassil et al. (2014)]. However, despite the fact that IGF-1 is invariably found to be altered in these patients, there is no agreement on whether protection is granted by high levels or low levels of IGF-1 (Bassil et al., 2014).

Within the AGHLS, which is a relatively young and healthy population, the association between IGF-1 serum levels and cognitive functions was examined. No significant associations were found between cognitive measures and IGF-1 levels (Licht et al., 2014). However, Tumati et al. (2016) showed that higher IGF-1 serum levels associated with a worse cognitive profile in middle-aged males.

Recently, the association of IGF-1 on brain volume was assessed in a healthy elderly population within the Framingham cohort, showing that lower serum levels of IGF-1 are related to smaller brain volumes, as well as to a higher risk for developing AD (Westwood et al., 2014). Hence, we hypothesize that IGF-1 might be linked to functional connectivity in a relatively young and healthy cohort. Furthermore, besides the “disconnection hypothesis,” that is, alterations at the level of individual connections, it has recently been shown that also at the level of global network characteristics, functional brain networks are less integrated in dementias (Yu et al., 2016). In this line of thinking, one might expect that lower serum levels of IGF-1 would also associate with a network that is less integrated. To test whether functional networks are associated to IGF-1 levels, we used data from the AGHLS, a middle-aged healthy cohort, in which resting-state magnetoencephalographic signals were recorded (Wijnstok et al., 2013).

Functional brain networks were subsequently reconstructed using an atlas-based beamforming approach (Hillebrand et al., 2012, 2016). Since many network metrics are affected by network size or (arbitrary) thresholding or edge density (van Wijk et al., 2010), we chose to reconstruct the minimum spanning tree (MST). The MST is a loopless subgraph of the original network that connects all the nodes of the original graph and describes the backbone of the network, yet retaining most of the information about the original network (Tewarie et al., 2015). Furthermore, the MST allows to obtain network metrics that are statistically comparable across groups. Finally, to test whether lower serum IGF-1 relates to a less integrated network, we looked for associations between serum IGF-1 levels and MST parameters.

Materials and Methods

Participants

All subjects participated in the AGHLS, which is an observational cohort study started in 1976, initially consisting of 600 boys and girls at the age of 13 (Wijnstok et al., 2013). In the most recent round of measurement, in 2006, at the age of 42, magnetoencephalography (MEG) was added to the test battery (Wijnstok et al., 2013). The Ethics Committee of the VU University Medical Center approved the study and all participants provided written informed consent before participation.

Magnetoencephalography acquisition

MEG recordings were obtained for 339 participants, as previously described (Wijnstok et al., 2013), using a 151-channel whole-head MEG system (CTF Systems, Inc., Port Coquitlam, Canada), while subjects were in supine position inside a magnetically shielded room (Vacuumschmelze, Hanau, Germany). Magnetic fields were recorded during a 5-min, no-task, eyes-closed condition. A third-order software gradient was used with a recording passband of 0–150 Hz and a sampling frequency of 625 Hz (Vrba and Robinson, 2001). The head position was recorded at the start and end of each recording by passing small alternating currents through three head position coils attached to the left and right preauricular point and the nasion. Head movements of at the most 0.5 cm during acquisition were allowed. Artifact-free epochs of 4096 samples (6.552 sec) were selected based on visual inspection [P.S., D.N.]. The artifacts included system-related artifacts (SQUID jumps, noisy, broken, saturated channels), physiological artifacts (eye movements, eye blinks, muscle artifacts), or excessive environmental noise (Gross et al., 2013). Only data sets with at least 17 good epochs (Brookes et al., 2008) were selected for further analysis, resulting in 242 data sets with an average 30 epochs (range: 17–47).

Reconstruction of time series of neuronal activity

The time series of neuronal activity was reconstructed using an atlas-based beamforming technique described previously (Hillebrand et al., 2012) and here we provide a brief summary (Fig. 1). MEG data were coregistered with a template magnetic resonance imaging (MRI) through identification of the same anatomical landmarks where the head-localization coils were placed. The scalp surface was extracted from the coregistered MRI, and a multisphere head model was constructed (Huang et al., 1999), which was used for the beamformer computations. The voxels in the template MRI were labeled according to the automated anatomical labeling (AAL) atlas (Tzourio-Mazoyer et al., 2002). The 78 cortical regions-of-interest (ROIs) (Gong et al., 2009), as well as the hippocampi, were used in our analysis.

FIG. 1.

Schematic overview of the analysis pipeline. (A) Signals recorded by the MEG sensors. (B) Source reconstruction using an atlas-based beamformer. (C) Reconstructed time series for 80 atlas-based ROIs. (D) Adjacency matrix containing the functional connections between ROIs as estimated using the PLI. (E) Construction of the MST-based brain network (where each ROI is a node and each functional connection an edge). (F) Illustration of extreme MST configurations, a line-like and a star-like network. Red nodes are the leaves (nodes with only one connection). The betweenness centrality (the fraction of all shortest paths in the network passing through a given node) is highest for the dark blue nodes. Note that the star-like network has many leaves, is highly integrated, but that the central node could become overloaded because all shortest paths pass through it. In contrast, in the line-like network, the load is more distributed, but the network is not well integrated. The tree hierarchy reflects this balance between overload and integration, with the optimal configuration presumably in between these two extremes. MEG, magnetoencephalography; MST, minimum spanning tree; PLI, phase lag index; ROIs, regions-of-interest.

The voxel at the centroid of each ROI was used to obtain a single time series for each ROI (Hillebrand et al., 2016). The MEG data were band-pass filtered between 0.5 and 48 Hz and both data covariance matrix and a unity noise covariance matrix were used to reconstruct the normalized beamformer weights (Hillebrand and Barnes, 2005; Hillebrand et al., 2005) for each centroid using Synthetic Aperture Magnetometry (SAM) (Robinson and Vrba, 1999). The time series of neuronal activation was subsequently estimated by projecting the data through these weights, resulting in 80 broad-band time series of neuronal activation, that is, one for each of the 80 selected ROIs of the AAL atlas.

To avoid drowsiness, the first 10 artifact-free epochs were selected based on visual inspection to undergo further analysis using BrainWave software [CJS, version 09.152.1.23 available from http://home.kpn.nl/stam7883/brainwave.html]. These epochs were band-pass filtered into the six classical EEG frequency bands using a discrete Fast Fourier Transform: delta (0.5–4 Hz), theta (4–8 Hz), alpha1 (8–10 Hz), alpha2 (10–13 Hz), beta (13–30 Hz), and gamma (30–48 Hz).

Functional connectivity analysis

The phase lag index (PLI) was used as a measure of functional connectivity between two time series (Stam et al., 2007). PLI quantifies the asymmetry of the distribution of differences between the instantaneous phases (derived from a Hilbert transform of the time series) for two time series ( \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document}$${ \rm{ \Delta \varphi }}$$ \end{document} ) as follows: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}{ \rm PLI} = \vert < { \rm sign} [ { \rm sin} ( { \rm{ \Delta }}{{ \rm{ \varphi }}_{ \rm{t}}} ) ] > \vert \end{align*} \end{document}

where “< >” indicates the mean value, “sign” stands for the signum function, “||” denotes the absolute value, and “t” the discrete time steps. The phase difference is defined in the interval [−π, π]. This measure is insensitive to volume conduction (at the cost of discarding true zero-lag interactions) and ranges from 0 to 1, where 0 indicates completely symmetric distribution of the phase differences, or phase differences of zero (mod π), and 1 indicates perfectly asymmetric distribution of the phase differences (Stam et al., 2007). With this procedure, we obtained an 80 × 80 weighted adjacency matrix for each epoch and for each subject, in all of the frequency bands.

Network analysis

The weighted adjacency matrix can be used to reconstruct a network or complete weighted graph, where the 80 areas of the AAL atlas are represented as nodes, and the PLI values form the weighted edges.

A frequency-specific MST was calculated for each epoch by applying Kruskal's algorithm to the weighted adjacency matrix (Kruskal, 1956). We were interested in the strongest connections, and hence, for the construction of the MST, the edge weight was defined as 1/PLI. The algorithm first ranks the links in ascending order and then constructs the network by adding one link at a time, discarding links that would form a loop. The algorithm proceeds until all nodes are connected resulting in a loopless graph with N nodes and M = N − 1 links (Kruskal, 1956). We chose to use the MST to avoid some of the biases in traditional network analyses (Stam, 2014; Stam et al., 2014; Tewarie et al., 2015; van Wijk et al., 2010). Based on the MST, we then calculated the tree hierarchy (T_h) defined as given in a study (Boersma et al., 2013): \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}{ \rm T } _ { \rm h } = { \frac { \rm { L } } { { \rm { 2MB } } { { \rm { C } } _ { { \rm { max } } } } } } \end{align*} \end{document}

where L is the leaf number (number of nodes with degree = 1), BC is the betweenness centrality [fraction of all shortest paths that pass through a particular node (Freeman, 1978)], BC_max is the maximum BC across the network, and M is the number of links. The denominator is multiplied by two to ensure that T_h ranges between 0 and 1. A T_h equal to 0.5 indicates a star-like topology and a T_h equal to 0 implies a path-like topology (Fig. 1F). T_h values higher than 0.5 are assumed to reflect more optimal network configurations that provide a trade-off between node overload and efficient communication (Boersma et al., 2013; Tewarie et al., 2014). The leaf fraction was calculated as the fraction of nodes with a degree of 1 (Boersma et al., 2013), and it provides an indication of how integrated the network is (i.e., the higher the leaf fraction, the more integrated the network) (Fig. 1F). Furthermore, we calculated the BC for each region, to determine if specific regions were more affected by serum levels of IGF-1 than others.

Insulin growth factor-1

IGF-1 serum levels were obtained at the 2006 assessment using a commercially available assay (chemiluminescent immunometric, Immulite 2500; DPC, Los Angeles). Samples were stored at −80°C. The IGF-1 assay was calibrated with reference to the first WHO international standard 02/254. The intra-assay and interassay coefficients of variation were both 5% for the entire range of values (Ferreira et al., 2002).

Covariates

All covariates included in the analysis were obtained at the 2006 assessment. Serum levels of IGF-1 are related to a number of systemic factors (such as vascular functioning, inflammation status, and metabolic status) that in turn affect the brain networks. Since we were interested in the direct association of IGF-1 on brain topology, these confounders were taken into account.

Macrovascular function

An ultrasound scanner (Wall track system 2; Pie Medical, Maastricht, The Netherlands) was used to measure the carotid artery capacities. The standardized procedures to obtain artery capacities for the estimation of macrovascular function are described elsewhere in detail (Schouten et al., 2011). Intima-media thickness of the carotid artery (IMT) was assessed. Reproducibility of macrovascular parameters was tested in 2006. For carotid IMT, the coefficient of variation was 6.2% (Watanabe et al., 2005). Since an inverse relationship between IGF-1 level and carotid intima-media thickness has been described previously, it is an important factor to take into account (Córdova et al., 2015), since it might independently cause brain damage.

Microvascular function

Metabolic changes have shown an impact on microvasculature (Johnson et al., 1982; Nobili et al., 1997); therefore, we used as a covariate the microvascular function as assessed by functional nailfold capillaroscopy, as this has shown to linearly regress with cerebral circulation (Nazzaro et al., 2013; Wijnstok et al., 2012). A nailfold capillary videomicroscopy (Capiscope; KK Technologies, Devon, United Kingdom) was used to assess microvascular function as previously described (Wijnstok et al., 2012). Reproducibility of the microvascular parameters was tested using intraclass correlation coefficients. The intraobserver intraclass correlation coefficients of capillary density counts during baseline and after 4 min of arterial occlusion were 0.97 and 0.96, and the interobserver intraclass correlation coefficients were 0.86 and 0.90, respectively (Wijnstok et al., 2012).

C-reactive protein

We adjusted for C-reactive protein (CRP) as this is a measure of subclinical systemic inflammation that is linked to neurodegeneration and has shown to inversely correlate with IGF-1 (Efstratiadis et al., 2006; Spielman et al., 2014). Blood samples were collected to assess the CRP levels using a multiarray detection system based on electrochemiluminescence technology (Meso Scale Discovery; SECTOR Imager 2400). Intra- and interassay coefficients of variation for the CRP were 2.3% and 4.3%, respectively (van Bussel et al., 2011).

Body fat distribution

Central obesity was taken into account by assessing the waist-to-hip ratio, since obesity and the expression of IGF-1 are related (Efstratiadis et al., 2006; Spielman et al., 2014). Waist circumference was measured midway between the lowest rib margin and the iliac crest, and the hip circumference was measured at the widest levels over the great trochanter.

Pulmonary fitness

Finally, forced expiratory volume in 1 sec (FEV1) was taken into account as a confounder since it is known that FEV1 and cognitive function are positively associated in the middle-aged population (Anstey et al., 2004; Richards et al., 2005). Furthermore, IGF-1 is positively associated with fitness (Nindl et al., 2011).

From the 242 subjects, data regarding IGF-1, macro- and microvascular function, CRP, and body fat distribution were available for 156 subjects. Information with regard to the FEV1 was available for 152 subjects.

Statistical analysis

First, to exclude that any inadvertent bias was introduced by the data selection, we tested by a t-test or chi-square test whether the participants with complete data (n = 152) differed significantly on general characteristics compared to the rest of the cohort.

The association between IGF-1 and MST network topology was analyzed using a linear regression analysis with IGF-1 as the independent variable and the MST network topology as the dependent variable. To determine the normality of the outcome variables graphically, normal QQ-Plots were made. The outcome variables were normally distributed and IGF-1 was linearly associated with the MST network topology. Furthermore, we evaluated the homoscedasticity with a t-test with a median split on the IGF-1 values. The Levene's test for equality of variances did not show any significant results and therefore we concluded that the assumption of homoscedasticity was met.

For the tree hierarchy (T_h) and the leaf fraction, we performed analyses for six frequency bands. Besides noncorrected analyses, we also adjusted for CRP levels, intima-media thickness of the carotid artery, microvascular function, FEV1, waist-to-hip ratio, and gender. Standardized regression coefficients (β) were used to assess the strength of the association. Regression analyses were performed with SPSS statistical software (IBM SPSS, statistics, version 21.0), and a two-sided p-value lower than 0.05 was considered to be statistically significant. When there was a significant association between T_h or leaf fraction and serum levels of IGF-1, post hoc analyses were performed to analyze the association between BC and IGF-1 in the different frequency bands. For this, the BC was calculated for each ROI and compared between subjects with high and low serum levels of IGF-1 (median split) by permutation testing (Nichols and Holmes, 2002) followed by an FDR correction for multiple comparisons across ROIs (Benjamini and Hochberg, 1995).

Results

Tables 1 and 2 show descriptive information of the study population, while Tables 3 and 4 show the results of the linear regression analyses.

Table 1.

General Characteristics of the Study Population (n = 156)

Age, years	42 ± 0.7 (41–44)
Gender, % males	72 (46.2)
Waist-to-hip ratio	0.83 ± 0.08 (0.68–1.10)
Microvascular function
Baseline capillary density (capillaries/mm²)	43.13 ± 15.74 (9.25–85.53)
Macrovascular function
Intima-media thickness, mm	0.66 ± 0.12 (0.40–1.08)
Serum IGF-I level, nmol/L	23.6 ± 5.4 (11.2–39.3)
C-reactive protein, ng/mL, (M [IQR])	740 [347–1918]
FEV1, 0.01 L/s^a	3.88 ± 0.79 (1.54–5.64)

Continuous data are presented as mean ± SD or median [IQR]; dichotomous data are presented as percentages.

Data based on 152 participants.

FEV1, forced expiratory volume in 1 sec; IGF, insulin growth factor-1; IQR, interquartile range; SD, standard deviation.

Table 2.

Leaf Fraction and Tree Hierarchy Parameters in the Study Population

Leaf fraction	Mean ± SD (min–max)
Delta	0.0515 ± 0.0148 (0.4747–0.5596)
Theta	0.5023 ± 0.0135 (0.4607–0.5481)
Alpha1	0.5506 ± 0.0162 (0.4545–0.5506)
Alpha2	0.5007 ± 0.0147 (0.4619–0.5353)
Beta	0.5417 ± 0.0174 (0.4481–0.5417)
Gamma	0.4918 ± 0.0165 (0.4481–0.5355)

Tree hierarchy	Mean ± SD (min–max)
Delta	0.3840 ± 0.0130 (0.3545–0.4280)
Theta	0.3777 ± 0.1309 (0.3385–0.4284)
Alpha1	0.3795 ± 0.0157 (0.3409–0.4276)
Alpha2	0.3758 ± 0.0139 (0.3396–0.4040)
Beta	0.3740 ± 0.0154 (0.3324–0.4141)
Gamma	0.3708 ± 0.0138 (0.3324–0.4072)

Data are mean ± SD and minimum and maximum. Data based on 156 participants.

Table 3.

Crude and Adjusted Associations of Tree Hierarchy with Insulin Growth Factor-1 Levels

Serum IGF-levels
Crude					Adjusted
T _h	β	t	95% CI	p	β	t	95% CI	p
Delta	0.048	0.590	−0.113 to 0.209	0.556	−0.006	−0.063	−0.180 to 0.169	0.950
Theta	0.168	2.037	0.005 to 0.331	0.043^a	0.164	1.868	−0.010 to 0.337	0.064
Alpha1	−0.028	−0.688	−0.193 to 0.137	0.740	−0.061	−0.688	−0.236 to 0.114	0.492
Alpha2	−0.054	−0.638	−0.220 to 0.113	0.525	−0.033	−0.358	−0.213 to 0.148	0.721
Beta	0.211	2.591	0.050 to 0.372	0.011^a	0.202	2.310	0.029 to 0.372	0.022^a
Gamma	0.003	0.039	−0.162 to 0.169	0.969	−0.003	−0.033	−0.181 to 0.175	0.969

β, standardized regression coefficients as obtained from multiple linear regression analyses. Data refer to 156 subjects. T_h = tree hierarchy. Covariates were microvascular function, creatine levels, intima-media thickness of the carotid artery, FEV1, waist-to-hip ratio, and gender.

Refers to p < 0.05.

CI, confidence interval.

Table 4.

Crude and Adjusted Associations of Leaf Fraction with Insulin Growth Factor-1 Levels

Serum IGF-levels
Crude					Adjusted
Leaf	β	t	95% CI	p	β	t	95% CI	p
Delta	0.056	0.691	−0.104 to 0.215	0.491	0.013	0.145	−0.161 to 0.187	0.885
Theta	0.223	2.731	0.062 to 0.385	0.007^a	0.215	2.462	0.042 to 0.388	0.015^a
Alpha1	0.018	0.217	−0.146 to 0.182	0.829	0.035	0.395	−0.141 to 0.212	0.693
Alpha2	0.072	0.851	−0.095 to 0.238	0.396	0.086	0.935	−0.095 to 0.267	0.351
Beta	0.221	2.693	0.059 to 0.383	0.008^a	0.178	2.035	0.005 to 0.352	0.044^a
Gamma	0.004	0.052	−0.162 to 0.170	0.958	−0.042	−0.468	−0.221 to 0.136	0.641

β, standardized regression coefficients as obtained from multiple linear regression analyses. Data refer to 156 subjects. Leaf = leaf fraction. Covariates were microvascular function, creatine levels, intima-media thickness of the carotid artery, FEV1, waist-to-hip ratio, and gender.

Refers to p < 0.05.

The participants for whom complete data were available did not differ significantly on general characteristics compared to the rest of the cohort (not shown).

IGF-1 was found to be most strongly and positively associated with the tree hierarchy (T_h) in the beta band with a β of 0.211 and a p-value of 0.011 (Table 3 and Fig. 2A). When taking the covariates into account, the β was 0.202 and the p-value was 0.022.

FIG. 2.

Linear associations between IGF-1 and tree hierarchy (A) and leaf fraction (B) in the beta band, as well as with leaf fraction in the theta band (C). The dotted lines show the age-specific reference limits for IGF-1 in nmol/L (Elmlinger et al., 2004). IGF-1, insulin growth factor-1.

There was also a positive association between the leaf fraction in the beta band and IGF-1 (β = 0.221; p = 0.008) (Table 4), which remained significant when including the covariates (β = 0.178; p = 0.044) (Table 4 and Fig. 2B). Furthermore, the leaf fraction in the theta band was positively associated with IGF-1 (β = 0.223; p = 0.007), also when including covariates (β = 0.215; p = 0.015) (Table 4 and Fig. 2C).

Post hoc analysis of the BC in the beta and theta band did not reveal any significant regional differences between the high and low IGF-1 groups (not shown).

Discussion

In this study, we set out to test the hypothesis that IGF-1, a growth factor that plays a role in many neurodegenerative diseases (Bassil et al., 2014), would associate to a less integrated brain network topology in healthy, middle-aged subjects. We found that lower levels of serum IGF-1 associate with less integrated functional brain networks in both the theta and beta band.

The literature about the relationship between IGF-1 and brain function and cognitive performance has been contradictory so far, since both high and low levels of IGF-1 have been reported as detrimental (Bassil et al., 2014). However, these studies have been performed in the elderly, and the information in the middle aged is scarce. As far as cognitive functioning is concerned, one study showed in middle-aged males that high serum IGF-1 levels are associated with worse future cognitive function (Tumati et al., 2016). An earlier study performed in the same cohort (AGHLS) did not find any association between IGF-1 and cognition (Licht et al., 2014). It has previously been shown that lower levels of serum IGF-1 relate to brain atrophy and higher risk of dementia in the elderly healthy brain (Westwood et al., 2014). However, information about the association between functional brain activity and IGF-1 serum levels in the middle aged has never been explored to date.

To address this issue, we characterized the global topology of the functional network using the tree hierarchy, which quantifies the trade-off between having a well-integrated network and prevention of regional overload (Boersma et al., 2013), and the leaf fraction. It has previously been shown that lower levels of serum IGF-1 relate to brain atrophy and higher risk of dementia in elderly healthy brain (Westwood et al., 2014). In this study, we show that IGF-1 is positively associated with the tree hierarchy and leaf fraction. Together, these findings show that low serum levels are related to a less integrated (more line-like) network (Fig. 1F), with a suboptimal balance between integration and hub overload.

Since a network with a more line-like configuration can be regarded as less integrated, our results might also be framed within the “disconnection hypothesis” (Geschwind, 1965a,b; Hof and Morrison, 1994; Morrison et al., 1990; O'Sullivan et al., 2001), through which neurodegeneration would relate to a more disconnected brain network. For instance, it was shown with the MST applied to the default mode network (DMN) in functional MRI data that AD patients have a less integrated DMN than healthy controls (Ciftçi, 2011), in line with our finding that a more line-like network relates to lower levels of serum IGF-1. In this framework, one might speculate that the association of lower levels of serum IGF-1 with brain functional networks in young adults is an early expression of abnormal neuronal functioning, eventually culminating in observed brain atrophy and risk for dementia in the healthy elderly. Further (longitudinal) studies are needed to address this hypothesis.

We only found significant associations between IGF-1 and functional network topology in the beta and theta band, which is in line with previous literature. Indeed, changes in brain network topology in these frequency bands have been described previously for various neurological diseases. First off, it is important to note that alterations in connectivity in AD have been reported previously, especially in the beta band (Dauwan et al., 2016; Stam et al., 2005, 2009). In particular, in AD, the physiological posterior-to-anterior pattern has been shown to be specifically disrupted in the beta band (Dauwan et al., 2016).

In relapsing–remitting multiple sclerosis, it has been shown that network topology in the apha2 and beta band tends to deviate toward a line-like topology, thus reflecting a loss of integration (Tewarie et al., 2014). Furthermore, a longitudinal study of PD showed that the tree hierarchy decreases over time in the beta band (Olde Dubbelink et al., 2014). This is especially interesting since it has been proposed that oscillations in the beta band might play a key role in large-scale interactions involved in higher level cognitive functions, such as top-down attention, multisensory integration, and execution of motor plans (Siegel et al., 2012).

In addition, a recent study on the effect of vagal nerve stimulation in patients with pharmacoresistant epilepsy showed a reorganization of the MSTs in the theta band toward a more integrated network in the responder group, while there was a shift to a less integrated network in the nonresponders (Fraschini et al., 2014). Hence, overall, it seems that different pathologies cause a shift in the topology of the MST such that the functional networks become less integrated.

We also performed post hoc analysis of differences in node status between subjects with high and low levels of IGF-1, which revealed that the association between IGF-1 and global network topology was not driven by changes in the hub status of specific regions.

It must be noted that we projected the MEG data to a standard anatomical atlas, using a template MRI. Although this might result in suboptimal spatial resolution compared to the use of native MRIs, it has been shown that such approximations do not lead to gross errors at the group level (Beg et al., 2009; Holliday et al., 2003; López et al., 2014; Steinstraeter et al., 2009). We did not correct for multiple comparisons across frequency bands and metrics, since this is the first exploratory study on the association between serum IGF-1 levels and the brain network topology in the middle aged. Furthermore, since the metrics we used are partly related, correcting for multiple comparisons might introduce false negatives. Furthermore, it is important to notice that the free IGF-1 and the IGF-binding proteins were not assessed, and this should be taken into account when interpreting the data. Our findings are obtained in a fairly healthy, highly educated, and homogenous population (Wijnstok et al., 2013), therefore we might have underestimated the associations. Furthermore, our observations are of a correlation rather than causal in nature. To answer the important question whether IGF-1 is related to a higher risk of dementia, a later time point is needed.

Conclusions

We have shown that IGF-1 has an association with the topological organization of functional brain networks in the middle aged and, more specifically, that lower IGF-1 serum levels are associated with less integrated functional brain networks. Further studies should investigate how the observed associations between levels of IGF-1 and functional network topology relate to previous observations that lower serum levels of IGF-1 relate to brain atrophy in the healthy elderly and carry an increased risk for dementia.

Footnotes

Author Disclosure Statement

No competing financial interests exist.

References

Achard

, Bullmore

. 2007. Efficiency and cost of economical brain functional networks. PLoS Comput Biol, 3:e17.

Adamo

, Raizada

, LeRoith

. 1989. Insulin and insulin-like growth factor receptors in the nervous system. Mol Neurobiol, 3:71–100.

Anstey

, Windsor

, Jorm

, Christensen

, Rodgers

. 2004. Association of pulmonary function with cognitive performance in early, middle and late adulthood. Gerontology, 50:230–234.

Barber

, Nakamura

, Wolpert

, Reiter

, Seigel

, Antonetti

, Gardner

. 2001. Insulin rescues retinal neurons from apoptosis by a phosphatidylinositol 3-kinase/Akt-mediated mechanism that reduces the activation of caspase-3. J Biol Chem, 276:32814–32821.

Barthwal

, Sathyanarayana

, Kundu

, Rana

, Pradeep

, Sharma

, et al. 2003. Negative regulation of mixed lineage kinase 3 by protein kinase B/AKT leads to cell survival. J Biol Chem, 278:3897–3902.

Bassil

, Fernagut

, Bezard

, Meissner

. 2014. Insulin, IGF-1 and GLP-1 signaling in neurodegenerative disorders: targets for disease modification?. Prog Neurobiol, 118:1–18.

Beg

, Wong

, Khan

, Cheung

, Virji-Babul

, Herdman

. 2009. External landmark and head-shape-based functional data normalization. Comput Med Imaging Graph, 33:501–509.

Benjamini

, Hochberg

. 1995. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B, 57:289–300.

Boersma

, Smit

DJA

, Boomsma

, De Geus

EJC

, Delemarre-van de Waal

, Stam

. 2013. Growing trees in child brains: graph theoretical analysis of electroencephalography-derived minimum spanning tree in 5- and 7-year-old children reflects brain maturation. Brain Connect, 3:50–60.

10.

Brookes

, Vrba

, Robinson

, Stevenson

, Peters

, Barnes

, et al. 2008. Optimising experimental design for MEG beamformer imaging. Neuroimage, 39:1788–1802.

11.

Chialvo

. 2010. Emergent complex neural dynamics. Nat Phys, 6:744–750.

12.

Ciftçi

. 2011. Minimum spanning tree reflects the alterations of the default mode network during Alzheimer's disease. Ann Biomed Eng, 39:1493–1504.

13.

Córdova

, Boullosa

, Custódio

MRM

, Quaglia

, Santos

, Freitas

, et al. 2016. Atheroprotective properties of serum IGF-1 in the carotid and coronary territories and beneficial role on the physical fitness of the oldest old. J Gerontol A Biol Sci Med Sci, 71:1281–1288.

14.

Dauwan

, van Dellen

, van Boxtel

, van Straaten

ECW

, de Waal

, Lemstra

, et al. 2016. EEG-directed connectivity from posterior brain regions is decreased in dementia with Lewy bodies: a comparison with Alzheimer's disease and controls. Neurobiol Aging, 41:122–129.

15.

de Haan

, Mott

, van Straaten

ECW

, Scheltens

, Stam

. 2012. Activity dependent degeneration explains hub vulnerability in Alzheimer's disease. PLoS Comput Biol, 8:e1002582.

16.

Douw

, Nieboer

, van Dijk

, Stam

, Twisk

JWR

. 2014. A healthy brain in a healthy body: brain network correlates of physical and mental fitness. PLoS One, 9:e88202.

17.

Dupraz

, Grassi

, Karnas

, Nieto Guil

, Hicks

, Quiroga

. 2013. The insulin-like growth factor 1 receptor is essential for axonal regeneration in adult central nervous system neurons. PLoS One, 8:e54462.

18.

Efstratiadis

, Tsiaousis

, Athyros

, Karagianni

, Pavlitou-Tsiontsi

, Giannakou-Darda

, Manes

. 2006. Total serum insulin-like growth factor-1 and C-reactive protein in metabolic syndrome with or without diabetes. Angiology, 57:303–311.

19.

Elmlinger

, Kühnel

, Weber

, Ranke

. 2004. Reference ranges for two automated chemiluminescent assays for serum insulin-like growth factor I (IGF-I) and IGF-binding protein 3 (IGFBP-3). Clin Chem Lab Med, 42:654–664.

20.

Ferreira

, Twisk

JWR

, Van Mechelen

, Kemper

HCG

, Stehouwer

CDA

. 2002. Current and adolescent levels of cardiopulmonary fitness are related to large artery properties at age 36: the Amsterdam Growth and Health Longitudinal Study. Eur J Clin Invest, 32:723–731.

21.

Finkbeiner

. 2000. CREB couples neurotrophin signals to survival messages. Neuron, 25:11–14.

22.

Fraschini

, Demuru

, Puligheddu

, Floridia

, Polizzi

, Maleci

, et al. 2014. The re-organization of functional brain networks in pharmaco-resistant epileptic patients who respond to VNS. Neurosci Lett, 580:153–157.

23.

Freeman

. 1978. Centrality in social networks conceptual clarification. Soc Networks, 1:215–239.

24.

Geschwind

. 1965a. Disconnexion syndromes in animals and man. I. Brain, 88:237–294.

25.

Geschwind

. 1965b. Disconnexion syndromes in animals and man. II. Brain, 88:585–644.

26.

Gong

, He

, Concha

, Lebel

, Gross

, Evans

, Beaulieu

. 2009. Mapping anatomical connectivity patterns of human cerebral cortex using in vivo diffusion tensor imaging tractography. Cereb Cortex, 19:524–536.

27.

Gross

, Baillet

, Barnes

, Henson

, Hillebrand

, Jensen

, et al. 2013. Good practice for conducting and reporting MEG research. Neuroimage, 65:349–363.

28.

Hillebrand

, Barnes

. 2005. Beamformer analysis of MEG data. Int Rev Neurobiol, 68:149–171.

29.

Hillebrand

, Barnes

, Bosboom

, Berendse

, Stam

. 2012. Frequency-dependent functional connectivity within resting-state networks: an atlas-based MEG beamformer solution. Neuroimage, 59:3909–3921.

30.

Hillebrand

, Singh

, Holliday

, Furlong

, Barnes

. 2005. A new approach to neuroimaging with magnetoencephalography. Hum Brain Mapp, 25:199–211.

31.

Hillebrand

, Tewarie

, van Dellen

, Yu

, Carbo

EWS

, Douw

, et al. 2016. Direction of information flow in large-scale resting-state networks is frequency-dependent. Proc Natl Acad Sci U S A. doi: 10.1073/pnas.1515657113

32.

Hof

, Morrison

. 1994. The cellular basis of cortical disconnection in Alzheimer disease and related dementing conditions. In: Alzheimer Disease. New York, NY: Raven Press; pp. 197–230.

33.

Holliday

, Barnes

, Hillebrand

, Singh

. 2003. Accuracy and applications of group MEG studies using cortical source locations estimated from participants' scalp surfaces. Hum Brain Mapp, 20:142–147.

34.

Huang

, Mosher

, Leahy

. 1999. A sensor-weighted overlapping-sphere head model and exhaustive head model comparison for MEG. Phys Med Biol, 44:423–440.

35.

Huang

, Yang

, Huang

, Tai

, Hsu

. 2010. Pharmacological and genetic accumulation of hypoxia-inducible factor-1 enhances excitatory synaptic transmission in hippocampal neurons through the production of vascular endothelial growth factor. J Neurosci, 30:6080–6093.

36.

Hui

, Pei

D-S

, Zhang

Q-G

, Guan

Q-H

, Zhang

G-Y

. 2005. The neuroprotection of insulin on ischemic brain injury in rat hippocampus through negative regulation of JNK signaling pathway by PI3K/Akt activation. Brain Res, 1052:1–9.

37.

Johnson

, Brendel

, Meezan

. 1982. Thickened cerebral cortical capillary basement membranes in diabetics. Arch Pathol Lab Med, 106:214–217.

38.

Jones

, Tucker

, Rahimtula

, Mearow

. 2003. The synergistic effects of NGF and IGF-1 on neurite growth in adult sensory neurons: convergence on the PI 3-kinase signaling pathway. J Neurochem, 86:1116–1128.

39.

Kruskal

. 1956. On the shortest spanning subtree of a graph and the traveling salesman problem. Proc Am Math Soc, 7:48.

40.

, Xu

, McBurney

, Longo

. 2008. SirT1 inhibition reduces IGF-I/IRS-2/Ras/ERK1/2 signaling and protects neurons. Cell Metab, 8:38–48.

41.

Licht

CMM

, van Turenhout

, Deijen

, Koppes

LLJ

, van Mechelen

, Twisk

JWR

, Drent

. 2014. The Association between IGF-1 polymorphisms, IGF-1 serum levels, and cognitive functions in healthy adults: The Amsterdam Growth and Health Longitudinal Study. Int J Endocrinol, 2014:1–8.

42.

López

, Bruña

, Aurtenetxe

, Pineda-Pardo

JÁ

, Marcos

, Arrazola

, et al. 2014. Alpha-band hypersynchronization in progressive mild cognitive impairment: a magnetoencephalography study. J Neurosci, 34:14551–14559.

43.

Meunier

, Lambiotte

, Bullmore

. 2010. Modular and hierarchically modular organization of brain networks. Front Neurosci, 4:200.

44.

Morrison

, Hof

, Campbell

, De Lima

, Voigt

, Bouras

, et al. 1990. Cellular Pathology in Alzheimer's Disease: Implications for Corticocortical Disconnection and Differential Vulnerability. Berlin Heidelberg: Springer. pp. 19–40.

45.

Nazzaro

, Schirosi

, Mezzapesa

, Petruzzellis

, Pascazio

, Serio

, et al. 2013. Effect of clustering of metabolic syndrome factors on capillary and cerebrovascular impairment. Eur J Intern Med, 24:183–188.

46.

Nichols

, Holmes

. 2002. Nonparametric permutation tests for functional neuroimaging: a primer with examples. Hum Brain Mapp, 15:1–25.

47.

Nindl

, Santtila

, Vaara

, Hakkinen

, Kyrolainen

. 2011. Circulating IGF-I is associated with fitness and health outcomes in a population of 846 young healthy men. Growth Horm IGF Res, 21:124–128.

48.

Nobili

, Cutolo

, Sulli

, Castaldi

, Sardanelli

, Accardo

, et al. 1997. Impaired quantitative cerebral blood flow in scleroderma patients. J Neurol Sci, 152:63–71.

49.

O'Sullivan

, Jones

, Summers

, Morris

, Williams

, Markus

. 2001. Evidence for cortical “disconnection” as a mechanism of age-related cognitive decline. Neurology, 57:632–638.

50.

Olde Dubbelink

KTE

, Hillebrand

, Stoffers

, Deijen

, Twisk

JWR

, Stam

, Berendse

. 2014. Disrupted brain network topology in Parkinson's disease: a longitudinal magnetoencephalography study. Brain, 137:197–207.

51.

Otte

, van Diessen

, Paul

, Ramaswamy

, Subramanyam Rallabandi

, Stam

, Roy

. 2015. Aging alterations in whole-brain networks during adulthood mapped with the minimum spanning tree indices: the interplay of density, connectivity cost and life-time trajectory. Neuroimage, 109:171–189.

52.

Richards

, Strachan

, Hardy

, Kuh

, Wadsworth

. 2005. Lung function and cognitive ability in a longitudinal birth cohort study. Psychosom Med, 67:602–608.

53.

Robinson

, Vrba

. 1999. Functional Neuroimaging by Synthetic Aperture Magnetometry (SAM), Recent Advances in Biomagnetism. Sendai, Japan: Tohoku University Press.

54.

Rommel

, Bodine

, Clarke

, Rossman

, Nunez

, Stitt

, et al. 2001. Mediation of IGF-1-induced skeletal myotube hypertrophy by PI(3)K/Akt/mTOR and PI(3)K/Akt/GSK3 pathways. Nat Cell Biol, 3:1009–1013.

55.

Schouten

, Twisk

, de Boer

, Stehouwer

, Serné

, Smulders

, Ferreira

. 2011. Increases in central fat mass and decreases in peripheral fat mass are associated with accelerated arterial stiffening in healthy adults: the Amsterdam Growth and Health Longitudinal Study. Am J Clin Nutr, 94:40–48.

56.

Siegel

, Donner

, Engel

. 2012. Spectral fingerprints of large-scale neuronal interactions. Nat Rev Neurosci, 13:121–134.

57.

Spielman

, Little

, Klegeris

. 2014. Inflammation and insulin/IGF-1 resistance as the possible link between obesity and neurodegeneration. J Neuroimmunol, 273:8–21.

58.

Sporns

, Zwi

. 2004. The small world of the cerebral cortex. Neuroinformatics, 2:145–162.

59.

Stam

. 2014. Modern network science of neurological disorders. Nat Rev Neurosci, 15:683–695.

60.

Stam

, de Haan

, Daffertshofer

, Jones

, Manshanden

, van Cappellen van Walsum

, et al. 2009. Graph theoretical analysis of magnetoencephalographic functional connectivity in Alzheimer's disease. Brain, 132:213–224.

61.

Stam

, Montez

, Jones

, Rombouts

SARB

, van der Made

, Pijnenburg

YAL

, Scheltens

. 2005. Disturbed fluctuations of resting state EEG synchronization in Alzheimer's disease. Clin Neurophysiol, 116:708–715.

62.

Stam

, Nolte

, Daffertshofer

. 2007. Phase lag index: assessment of functional connectivity from multi channel EEG and MEG with diminished bias from common sources. Hum Brain Mapp, 28:1178–1193.

63.

Stam

, Tewarie

, Van Dellen

, van Straaten

ECW

, Hillebrand

, Van Mieghem

. 2014. The trees and the forest: characterization of complex brain networks with minimum spanning trees. Int J Psychophysiol, 92:129–138.

64.

Steinstraeter

, Teismann

, Wollbrink

, Suntrup

, Stoeckigt

, Dziewas

, Pantev

. 2009. Local sphere-based co-registration for SAM group analysis in subjects without individual MRI. Exp Brain Res, 193:387–396.

65.

Tewarie

, Hillebrand

, Schoonheim

, van Dijk

, Geurts

JJG

, Barkhof

, et al. 2014. Functional brain network analysis using minimum spanning trees in Multiple Sclerosis: an MEG source-space study. Neuroimage, 88:308–318.

66.

Tewarie

, van Dellen

, Hillebrand

, Stam

. 2015. The minimum spanning tree: an unbiased method for brain network analysis. Neuroimage, 104:177–188.

67.

Torres-Aleman

. 2010. Toward a comprehensive neurobiology of IGF-I. Dev Neurobiol, 70:384–396.

68.

Tumati

, Burger

, Martens

, van der Schouw

, Aleman

, Nauck

. 2016. Association between cognition and serum insulin-like growth factor-1 in middle-aged & older men: An 8 Year Follow-Up Study. PLoS One, 11:e0154450.

69.

Tzourio-Mazoyer

, Landeau

, Papathanassiou

, Crivello

, Etard

, Delcroix

, et al. 2002. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage, 15:273–289.

70.

van Bussel

BCT

, Henry

RMA

, Schalkwijk

, Ferreira

, Feskens

EJM

, Streppel

, et al. 2011. Fish consumption in healthy adults is associated with decreased circulating biomarkers of endothelial dysfunction and inflammation during a 6-year follow-up. J Nutr, 141:1719–1725.

71.

van Wijk

BCM

, Stam

, Daffertshofer

. 2010. Comparing brain networks of different size and connectivity density using graph Theory. PLoS One, 5:e13701.

72.

Vrba

, Robinson

. 2001. Signal processing in magnetoencephalography. Methods, 25:249–271.

73.

Wang

, Su

, Shen

, Hu

. 2012. Decoding lifespan changes of the human brain using resting-state functional connectivity MRI. PLoS One, 7:e44530.

74.

Watanabe

, Miyazaki

, Katagiri

, Yamamoto

, Idei

, Iguchi

. 2005. Relationship between serum insulin-like growth factor-1 levels and Alzheimer's disease and vascular dementia. J Am Geriatr Soc, 53:1748–1753.

75.

Westwood

, Beiser

, Decarli

, Harris

, Chen

, He

X-M

, et al. 2014. Insulin-like growth factor-1 and risk of Alzheimer dementia and brain atrophy. Neurology, 82:1613–1619.

76.

Wijnstok

, Hoekstra

, Eringa

, Smulders

, Twisk

, Serne

. 2012. The relationship of body fatness and body fat distribution with microvascular recruitment: The Amsterdam Growth and Health Longitudinal Study. Microcirculation, 19:273–279.

77.

Wijnstok

, Hoekstra

, Van Mechelen

, Kemper

HCG

, Twisk

JWR

. 2013. Cohort profile: the Amsterdam Growth and Health Longitudinal Study. Int J Epidemiol, 42:422–429.

78.

Xing

, Yin

, Chang

, Gong

, He

, Xie

. 2007. Effects of insulin-like growth factor 1 on synaptic excitability in cultured rat hippocampal neurons. Exp Neurol, 205:222–229.

79.

, Gouw

, Hillebrand

, Tijms

, Stam

, van Straaten

ECW

, Pijnenburg

YAL

. 2016. Different functional connectivity and network topology in behavioral variant of frontotemporal dementia and Alzheimer's disease: an EEG study. Neurobiol Aging, 42, 150–162.