Functional and anatomical connectivity abnormalities in left inferior frontal gyrus in schizophrenia

Abstract

Functional studies in schizophrenia demonstrate prominent abnormalities within the left inferior frontal gyrus (IFG) and also suggest the functional connectivity abnormalities in language network including left IFG and superior temporal gyrus during semantic processing. White matter connections between regions involved in the semantic network have also been indicated in schizophrenia. However, an association between functional and anatomical connectivity disruptions within the semantic network in schizophrenia has not been established. Functional (using levels of processing paradigm) as well as diffusion tensor imaging data from 10 controls and 10 chronic schizophrenics were acquired and analyzed. First, semantic encoding specific activation was estimated, showing decreased activation within the left IFG in schizophrenia. Second, functional time series were extracted from this area, and left IFG specific functional connectivity maps were produced for each subject. In an independent analysis, tract‐based spatial statistics (TBSS) was used to compare fractional anisotropy (FA) values between groups, and to correlate these values with functional connectivity maps. Schizophrenia patients showed weaker functional connectivity within the language network that includes left IFG and left superior temporal sulcus/middle temporal gyrus. FA was reduced in several white matter regions including left inferior frontal and left internal capsule. Finally, left inferior frontal white matter FA was positively correlated with connectivity measures of the semantic network in schizophrenics, but not in controls. Our results indicate an association between anatomical and functional connectivity abnormalities within the semantic network in schizophrenia, suggesting further that the functional abnormalities observed in this disorder might be directly related to white matter disruptions. Hum Brain Mapp, 2009. © 2009 Wiley‐Liss, Inc.

INTRODUCTION

Wernicke [ 1894] was the first to posit that abnormal connections among brain regions may play a critical role in the etiology of schizophrenia. His early hypotheses have remained, however, only speculative, until very recent developments in functional neuroimaging. More specifically, correlational analysis exploring the relationship between various cortical areas activated during functional MRI experiments [Friston, 1996; Friston and Büchel, 2007] has opened new windows into understanding the way brains are wired. One measure, derived from fMRI, is a measure of “functional connectivity,” which estimates the strength of temporal correlations between brain sites. Several studies (using multiple modalities to detect brain function, such as, functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and magnetoencephalography (MEG)), in fact, suggest that such connectivity may be weakened, or disrupted in schizophrenia [Meyer‐Lindenberg et al., 2005; Michelogiannis et al., 1991; Morrison‐Stewart et al., 1991]. These studies further point to fronto‐temporal connectivity [Breakspear et al., 2003; Friston et al., 1996; Meyer‐Lindenberg et al., 2005; Michelogiannis et al., 1991; Morrison‐Stewart et al., 1991], and the relationship between fronto‐temporal connectivity disruptions and clinical and cognitive symptoms of schizophrenia such as verbal hallucinations [Ford et al., 2002; Lawrie et al., 2002; Norman et al., 1997], auditory hallucinations, distractive speech, illogicality, incoherence, and semantic priming abnormalities [Han et al., 2007]. Since language (semantic) disturbances have been consistently reported in schizophrenia [Ceccherini‐Nelli et al., 2007; Condray, 2005; Kircher et al., 2007; Phillips et al., 2004], and have demonstrated superior diagnostic specificity over Schneider's first rank symptoms in schizophrenia diagnosis [Ceccherini‐Nelli and Crow, 2003; Ceccherini‐Nelli et al., 2007], it is important to further understand the relationship between language and functional connectivity in schizophrenia.

Interestingly, language‐related abnormalities in schizophrenia are not reported in the functional domain only. That is, in‐vivo structural MRI studies, also frequently report volumetric deficits in gray matter brain regions traditionally viewed as language‐related (including inferior frontal gyrus, superior temporal gyrus, and inferior parietal lobule) [Kwon et al., 1999; Niznikiewicz et al., 2000; Torrey, 2007; Venkatasubramanian et al., 2008]. Also, local white matter volume reduction in prefrontal [Breier et al., 1992; Buchanan et al., 1998; Hulshoff Pol et al., 2004; Wible et al., 2001] and temporal white matter [Okugawa et al., 2002; Spalletta et al., 2003] have been reported in schizophrenia. White matter myelinated axons provide communication between gray matter sites, and thus disruptions in their integrity, as suggested by white matter volume reductions, might underlie some of the functional deficits observed in schizophrenia. This hypothesis is further reinforced by recent post mortem, as well as genetic studies, suggesting myelin‐related abnormalities within white matter in schizophrenia [Karoutzou et al., 2008; Lipska et al., 2006], and refueled by recently developed diffusion tensor imaging (DTI) [Kubicki et al., 2005a], a method sensitive to white matter fiber tract integrity. Of note, DTI is a method that has consistently demonstrated disruptions in white matter integrity schizophrenic subjects within several fronto‐temporal tracts, including the cingulum bundle [Kubicki et al., 2003b; Manoach et al., 2007; Nestor et al., 2007], uncinate fasciculus [Price et al., 2007; Szeszko et al., 2008], and arcuate fasciculus [Burns et al., 2003; Douaud et al., 2007; Kubicki et al., 2005b]. However its relationship to clinical, cognitive, and functional abnormalities observed in schizophrenia is still unclear.

As discussed above, while fMRI studies provide evidence for functional connectivity, disruptions within the semantic network in schizophrenia, and anatomical studies, especially DTI, provide evidence for disturbances in anatomical connectivity in this disorder, it is important to combine these two methodologies to better understand the relationship between anatomy and function, and the role of anatomical and functional abnormalities in the psychophysiology of schizophrenia. To date, however, only one study has combined fMRI and DTI to explore connectivity within the semantic network in healthy controls [Powell et al., 2006]. This study reports a correlation between functional activations related to verbal fluency (left frontal gyrus), and reading comprehension (left supramarginal gyrus), and integrity of superior longitudinal fasciculus [Powell et al., 2006]. No such studies exist in schizophrenia.

The goal of the current study is to investigate both functional and anatomical connectivity deficits within the semantic language network, as well as their relationship to schizophrenia.

METHODS

Subjects

Ten male patients with schizophrenia and ten male healthy control subjects participated in this study. Patients were recruited from the VA Boston Healthcare System, Brockton Division. Male control subjects were recruited through newspaper advertisements. Subjects were matched on handedness, parental socioeconomic status (PSES) [Hollingshead, 1965], and age. Table I shows demographic and clinical characteristics of these two groups. Verbal IQ was significantly lower in the schizophrenia group, compared with control subjects. Since verbal IQ could affect measures of semantic processing [Elvevag et al., 2001] and white matter integrity [Deary et al., 2006; Mabbott et al., 2006], we used it as a covariate in both group comparisons as well as in correlation analyses with white matter integrity. All patients were diagnosed with chronic schizophrenia, by trained interviewers, using DSM‐IV criteria based on the structured clinical interview patient edition (SCID‐P) [First et al., 1998b] and a review of the medical records. Control subjects were screened using the structured clinical interview non‐patient edition (SCID) [First et al., 1998a] by the same trained interviewers. No control subjects had an Axis‐I psychiatric disorder or a first‐degree relative with an Axis‐I psychiatric disorder. The study was approved by the VA Boston Healthcare System Human Subjects Committee and by the Brigham and Women's Institutional Review Board. All subjects gave written informed consent prior to participation in the study, and all were compensated for their time.

Table I.

Sample characteristics

	Schizophrenic subjects (n = 10)	Control subjects (n = 10)
Age	39.6 ± 7.0	44.1 ± 5.4
Education	11.8 ± 1.5a	16.1 ± 2.3
Socioeconomic status (SES)	4.4 ± 0.7b	2.0 ± 1.1
Parental SES	2.7 ± 1.2	2.9 ± 1.3
Handedness	0.86 ± 0.2	0.71 ± 0.2
Verbal IQ	84.9 ± 13.4c	109.0 ± 9.5
WRAT3 Reading scores	91.9 ± 17.9	102.3 ± 8.2
Age of onset	19.9 ± 2.8	—
Chlorpromazine equivalent of neuroleptic dose	637.9 ± 353.7	—

WRAT, the wide range achievement test.

^at(18) = 5.0, P < 0.001.

^bt(18) = −6.0, P < 0.001.

^ct(18) = 4.7, P < 0.001.

The Levels‐of‐Processing Paradigm (Semantic Paradigm)

The main purpose of this study was to further investigate the semantic encoding network in schizophrenia, by combining fMRI and DTI data. In the current analysis, we used semantic processing fMRI data published previously [Kubicki et al., 2003a]. Full details regarding data acquisition and behavioral paradigm can be found in original paper [Kubicki et al., 2003a], thus here we describe them only briefly. During the fMRI session, patients and controls performed both semantic (“press the button if the word presented on the screen is abstract/concrete”) and perceptual (“press the button if the presented word is in lower/upper case”) encoding tasks. Stimuli were presented in 30‐s blocks, separated by 30‐s “resting” baseline condition [for details see Kubicki et al., 2003a]. The words were presented with 2.5‐s stimulus onset asynchrony (SOA), to both patients and controls in two 9‐min runs, and subjects' responses (judgment e.g., ABSTRACT or UPPERCASE) were collected.

Data Acquisition

All subjects underwent both fMRI and DTI procedures on 1.5‐T whole body MRI Echospeed system (General Electric Medical Systems, Milwaukee, WI). For the fMRI acquisition, the total of 174 EPI BOLD scans (24 oblique coronal slices, 6‐mm thick, TR 3 s; TE 40 ms; flip angle 90°; 64 × 64 matrix) were acquired perpendicular to the long axis of the hippocampus. For the DTI, line‐scan diffusion tensor scans (LSDI) [Maier et al., 1998] were collected in the coronal oblique plane, with the following parameters: 35 slices, 4‐mm thick, 1‐mm gap, TR 3 s; TE 64 ms; B 1,000 and 5 μm s⁻¹; six gradient directions (scanning procedures for DTI are described in detail elsewhere [Kubicki et al., 2004]).

Data Processing and Analysis

FMRI

FMRI data was processed and reanalyzed with newer statistical parametric mapping package (SPM5, http://www.fil.ion.ucl.ac.uk/spm) than in the original paper reporting fMRI findings in this sample [Kubicki et al., 2003a]. In addition, additional preprocessing, also not carried in original analysis, that included artifacts removal, was performed using FSL software (http://www.fmrib.ox.ac.uk/fsl/melodic/index.html)—MELODIC (multivariate exploratory linear optimized decomposition into independent components). The first four scans of each run were discarded. The remaining 280 images (140 images per run) were spatially realigned to the first volume of the first run, and realignment parameters were saved to be used as covariates in the subsequent statistical analysis. In addition, both scanner‐related and physiological artifacts were removed using MELODIC. Time‐courses of all slices of each volume were adjusted to the time course of the 12th slice, to correct for the acquisition time delay between different slices. Next, the realigned images were spatially normalized to the Montreal Neurological Institute (MNI) EPI template, resampled to 2 mm³ voxel and smoothed with an 8‐mm FWHM Gaussian filter. The general linear model was used for further reducing the effects of head motion and regressing out the linear drift [Oakes et al., 2005].

Activation maps for the contrast between semantic and nonsemantic encoding conditions were constructed separately for each subject using the first session of data, and fixed effect analysis was performed using the general linear model in each group. To reduce nonwhite noise, an autoregression method for estimating serial autocorrelations in the model residuals was applied in first‐level group analysis [Smith et al., 2007a]. Fixed effect t tests were used for group comparison. Finally, P‐values were subjected to cluster‐based permutation test, which generated the null distribution of the maximum t statistic for both functional connectivity and the fMRI‐DTI correlational analyses. One reason for using nonparametric permutation test rather than parametric test was the fact that the noise in the data might not follow a Gaussian distribution [Nichols and Holmes, 2002]. The other was that some statistical assumptions (linearity, homogeneity of slope) of analysis of covariance could not be met for parametric test when the removal of a confounding effect of IQ was needed for the between‐group comparison.

Functional Connectivity Analysis (Interregional Pearson's Correlations Analysis)

Before performing functional connectivity analysis, fMRI time series were preprocessed using high‐pass filter (cut‐off frequency: 1/128 s) to remove low‐frequency drift and signal fluctuation. FMRI analysis demonstrated decreased activity in left inferior frontal gyrus during the semantic encoding in schizophrenia subjects, compared to control subjects (see results for more details). Thus, for the purpose of functional connectivity analysis, mean time‐series of this fMRI activation (left IFG) was extracted from the cubes of 27 (3 × 3 × 3) voxels around the maximal intensity (highest t values) pixels for each subject separately [Koshino et al., 2007]. The Pearson's correlation coefficient was calculated between the left IFG time‐series and the time‐series of each voxel in the whole‐brain to create an r‐map in each subject. For statistical analysis, the r‐map was transformed into the z‐map using Fisher's r‐to‐z transformation of z = 0.5 × log[(1 + r)/(1 − r)]. The z‐map of each subject was further fed into second‐level analyses using a one‐sample t‐test to examine significant functional connectivity in each group. Two‐sample t tests were performed to examine significant differences in functional connectivity between groups after removing confounding factor of verbal IQ. Cluster‐size statistical threshold (P < 0.005, which corresponds to z value of 2.81) was used to correct for multiple comparisons, by using the null distribution of the maximum test statistics (over 5,000 permutations). Further, to identify significantly correlated brain regions, the clusters were thresholded at a level of P < 0.05, corrected for multiple comparisons and for a spatial extent of at least 50 voxels in both within group as well as between‐group comparisons. To further test for between‐group differences in correlation patterns, we applied threshold method followed by the Fisher exact test that was used previously in correlational analyses of fMRI studies [Auer et al., 2008; Baudewig et al., 2003]. To define the threshold for the regions showing a correlation between IFG and whole brain in each subject, first, a group histogram of correlation coefficient maps of all 20 subjects was produced. Then, a Gaussian curve was fitted to the central portion of the group histogram. A true group distribution of correlation coefficients was rescaled into percentile ranks of the noise distribution (represented in the low correlation coefficient values in the distribution). Finally, the region of true correlation was defined as being a correlation coefficient above 0.374, which matched the 99.99 percentile rank of the noise distribution, and chosen as the statistical threshold in each subject [Auer et al., 2008; Baudewig et al., 2003]. The number of subjects with true correlation was compared between two groups using Fisher's exact test in each region.

DTI

DTI was preprocessed using the FMRIB Software Library (FSL, Oxford), including skull stripping and eddy current correction. Fractional anisotropy (FA) images of each subject were created using FSL. Next, tract‐based spatial statistics (TBSS) was used for the data analysis [Smith et al., 2007b]. TBSS has been previously reported to be hypothesis‐free, automatic, and more precise than conventional voxel‐based approach (VBM) with respect to defining and aligning anatomical white matter structures (tracts) between subjects [Smith et al., 2006]. In addition, this method does not require smoothing, which has been shown previously to skew voxel‐based morphometry (VBM) results [Jones et al., 2005]. Here, TBSS was used to calculate tract‐based differences in FA values between the schizophrenia group and the control group, and these differences were later further subjected to correlational analyses with the fMRI results. The whole preprocessing procedure is automatic, and has been described elsewhere [Smith et al., 2007b]. Resulting FA maps were analyzed using a General Linear Model where verbal IQ was used as a covariate. False discovery rate multiple comparison correction with q < 0.05 [Nichols and Hayasaka, 2003], was applied to statistical results to minimize Type I errors when the significant region was not identified at permutation testing using cluster‐based thresholding.

Relationship Between Anatomical and Functional Connectivity

From the group comparison maps, we extracted areas within the left inferior frontal white matter (IFWM), where schizophrenics had significantly lower FA values than control subjects in TBSS analysis (see Results section for more detail). Then, to examine the relationship between anatomical and functional connectivity, the correlation coefficients were computed between correlation map created by functional connectivity analysis (represented by Pearson correlation coefficient r values) and the averaged FA values of left IFWM separately for controls and schizophrenia patients. Only voxels above 2.81 of z value (P < 0.005) in each z‐map of each subject were selected. The significant correlated regions were identified using permutation‐based testing on cluster size (P < 0.05 corrected) in each group.

RESULTS

There were no group differences in age, handedness, or parental socioeconomic status. Groups differed in verbal IQ, which was then used as a covariate in the data analysis (Table I).

Semantic Encoding FMRI

There were no statistical differences between groups in task accuracy during encoding (t(16) = 1.4, P = 0.18) experiment [Kubicki et al., 2003a], with all subjects reaching at least 75% accuracy in judgments [Kubicki et al., 2003a].

When contrasting deep and shallow encoding conditions, control subjects demonstrated (P < 0.005 at cluster level) activation in pars triangularis of left IFG and the right middle orbital gyrus (see Table II and Fig. 1). For the same contrast, schizophrenic subjects showed activation in pars triangularis of right IFG, pars orbitalis of right IFG, right medial temporal pole, right inferior temporal gyrus, and left superior temporal gyrus. When contrasting groups, control subjects showed significantly (uncorrected P < 0.005 at cluster level) more activation in pars triangularis of left IFG and right thalamus, while schizophrenic subjects showed significantly more activation in the left inferior temporal gyrus, pars opercularis of right IFG, right middle frontal gyrus and right precentral gyrus. (Table II and Fig. 1: Right side regions were not displayed in Fig. 1.)

Table II.

One sample and two sample t test for semantic encoding condition

Region	BA	Cluster size	Peak voxel coordinate			Z‐score
			x	y	z
Control group
L. IFG, triangularis	46	289	−50	30	20	3.73
	46		−44	24	16	3.19
	45		−58	20	4	2.86
R. middle frontal orbital G.	11	121	40	40	−16	3.52*
			30	38	−16	2.87*
Schizophrenia group
R. medial temporal pole/R. inferior temporal G.	20	534	46	4	−34	4.96
R. IFG, triangularis	45	404	60	34	0	4.32
Orbitalis	47		56	40	−10	3.79
	9		58	22	24	3.64
L. Superior Temporal G.	22	127	−56	−4	−10	3.59*
			−48	−4	−4	2.84*
Control > Schizophrenia group
R. Thalamus	–	69	8	−6	6	3.38*
			16	−2	10	3.23*
L. IFG, triangularis	46	58	−48	28	12	2.99*
	45		−42	20	16	2.93*
	45		−50	30	4	2.65*
Schizophrenia > Control group
L. Inferior Temporal G.	20	1,154	−34	−4	−40	7.02
R. IFG opercularis	44	72	62	14	26	3.56*
R. Middle Frontal G.	9		52	28	32	3.13*
R. Precentral G.	6	80	28	−14	58	3.36*
	4		32	−22	66	3.09*
	6		26	−10	66	2.8*

Notes: BA = Brodman's area, L = left, R = right, G = gyrus, IFG = inferior frontal gyrus. The coordinates of maximally activated voxels are given in MNI space. All activations identified at cluster‐level significance of P < 0.005 (corrected) for a spatial extent of at least 50 voxels, except at *, identified at cluster‐level significance of P < 0.005 (uncorrected).

Figure 1.

Regions activated for semantic encoding condition versus perceptual encoding and group differences (fixed effect analysis). Unlike control subjects, schizophrenic subjects did not show left inferior frontal activation during semantic encoding condition (corrected cluster level P < 0.005, at least 15 voxels). Control subjects showed significantly (uncorrected P < 0.005 at cluster level, at least 15 voxels) more activation in pars triangularis of left IFG and right thalamus, while schizophrenic subjects showed significantly more activation in brain regions including the left inferior temporal gyrus temporal pole, right precentral gyrus, pars opercularis of right inferior frontal gyrus (not shown in Fig. 1, see Table II). Neurological orientation.

Functional Connectivity

Many studies [Haller et al., 2007; Shtyrov and Pulvermuller, 2007; Vigneau et al., 2006] including our previous investigation [Kubicki et al., 2003a] have reported the left IFG as a key structure that is activated during semantic versus nonsemantic encoding. Thus in this analysis, we explored functional connectivity between left IFG (pars triangularis), which showed decreased activation in schizophrenia (MNI: x = −48, y = 28, z = 12), and other regions of the brain. Results indicated that in control subjects, the mean time series of left IFG was correlated with mean time series of several regions, including left middle temporal gyrus/left superior temporal sulcus, pars triangularis of right IFG, bilateral superior frontal gyrus, left superior parietal lobule/left supramarginal gyrus, bilateral cerebellum, and left thalamus. In schizophrenic subjects, the mean time series of left IFG were correlated with bilateral superior frontal gyrus, left cuneus. In a between‐group comparison of the correlation maps, the mean time series of left IFG were significantly less correlated with the time course of several parietal, temporal, and frontal regions in schizophrenic subjects compared to control subjects. In particular, the mean time series of the left IFG were less correlated with the left middle temporal gyrus/left superior temporal sulcus, left superior parietal lobule/intraparietal sulcus/supramarginal gyrus, right superior parietal lobule/intraparietal sulcus, left globus pallidus, left thalamus, superior frontal gyrus, precentral gyrus, cerebellum, and the right IFG in patients than in control subjects (See Table III, Fig. 2: Note that both the cerebellum and the right side of the brain are not included in the view in Fig. 2). These group differences were shown in all regions described above at the level of the correction of multiple comparisons using cluster level permutation test after controlling for verbal IQ.

Table III.

Correlation between left inferior frontal gyrus (x = −48, y = 28, z = 12) and all area in semantic condition

Regions	BA	Control subjects					Schizophrenia subjects					Control–Schizophrenia subjects
		Cluster size	Peak voxel coordinate			No. of subjects (N/10)a	Cluster size	Peak voxel coordinate			No. of Subjects (N/10)a	Cluster size	Peak voxel coordinate			Fisher's Exact testb
			x	y	z			x	y	Z			x	y	z	χ2	df	P
L. IFG	44, 45	5,190	−44	22	14	10	2,102	−46	28	9	10					‐	—	—
R. IFG, triangularis	46, 47	4,370	44	26	18	10					7	2,544	50	20	2	3.53	1	.21
Bilateral superior Frontal/paracingulate G	6, 8	3,093	−2	30	40	10	578	0	40	42	9	2,690	0	26	44	1.05	1	1.00
Bilateral precentral G.	6	236	4	−18	56	6					1	655	8	−28	62	5.50	1	.06
L. supeior parietal lobule/intraparietal sulcus/supramarginal G.	2, 40	752	−40	−50	46	10					5	674	−30	−44	56	6.67	1	.03
R. superior parietal lobule	40	669	36	−50	46	9					5	408	32	−58	42	3.81	1	0.14
L. superior temporal sulcus/middle temporal G.	21	75	−56	−34	−4	8					3	67	−60	−38	−2	7.50	1	0.02
L. globus pallidus		185	−20	−10	−2	9					4	137	−24	−8	−4	5.50	1	0.06
L. thalamus		133	−10	−20	2	9					4	94	−8	−16	2	5.50	1	0.06
L. cuneus	17, 30						972	4	−70	8						2.22	1	0.48
L. cerebellum		1,787	−22	−70	−24	10					7	1,191	−16	−68	16	3.53	1	0.21
R. cerebellum		1,519	32	−64	−22	10					6	1,223	24	−72	−24	5.00	1	0.09

Notes: No suprathreshold regions identified for the contrast of Schizophrenia subjects – Control subjects. BA = Brodman's area, L = left, R = right, G = gyrus, IFG = inferior frontal gyrus.

Statistical significance was confirmed through analysis of 5,000 nonparametric permutations. Activations identified at cluster‐level significance of P < 0.05 for a spatial extent of at least 50 voxels for both within and between group analysis. The effect of verbal IQ was removed for the between group analysis.

^aNo. of Subjects: the number of subjects showing true correlation which was defined as a correlation coefficient greater than 0.374 which matched the 99.99 percentile rank of the noise distribution, at each region of interest.

^bFisher's Exact test: group comparison test of No. of Subjects showing true correlation.

Figure 2.

Functional correlation map (correlation coefficient) between left IFG – MNI: x = −48, y = 28, z = 12 (which showed decreased activation in schizophrenia) and other regions of the brain. In control subjects, compared with schizophrenia subjects, the mean time series of left IFG was significantly more correlated with several regions including adjacent left inferior frontal area, left middle temporal gyrus/left superior temporal sulcus, left superior parietal lobule/intraparietal sulcus/supramarginal gyrus and superior frontal gyrus in control subjects. The bilateral cerebellum and several right hemisphere brain regions including the right IFG, and the right superior parietal lobule (not shown in Fig. 2, see Table III) also showed a greater correlation in control subjects (compared to schizophrenics). In schizophrenic subjects, the mean time series of left IFG was not correlated with the left middle temporal gyrus/left superior temporal sulcus, or the left superior parietal lobule/intraparietal sulcus/ supramarginal gyrus. Statistical significance was confirmed through analysis of 5,000 nonparametric permutations. Activations identified at cluster‐level significance of P < 0.005 for a spatial extent of at least 50 voxels.

The Fisher's exact test further confirmed group differences in correlation patterns, showing that a significantly higher number of control subjects (compared to schizophrenic subjects) demonstrated significant functional connectivity within the left perisylvian language areas during semantic processing (Table III).

Anatomical Connectivity

TBSS analysis revealed significant (q < 0.05; false discovery rate correction for multiple comparisons) FA reduction in several white matter areas, including left anterior limb of internal capsule and left IFWM in schizophrenia, when compared with control subjects. There were no regions characterized by higher FA values in schizophrenia (see Table IV, Fig. 3A). Probabilistic tractography using FSL was then used to qualitatively determine which anatomical structures (fiber tracts) are involved in connections between regions showing less activation and less functional connectivity in schizophrenia, and at the same time pass through left IFWM of regions with lower FA values in schizophrenia. Two tracts originating from left IFG reached left middle temporal gyrus/left superior temporal sulcus and crossed the left IFWM regions with reduced FA values in schizophrenia—left arcuate fasciculus, and left inferior occipito‐frontal fasciculus and/or left inferior longitudinal fasciculus (Fig. 3B).

Table IV.

TBSS results

Nearest regions	Cluster size	Control–Schizophrenia subjects Peak voxel coordinate			Tracta
		x	Y	z
L. IFG, triangularis	11	−35	31	9	16% inferior fronto‐occiptial fasciculus, 8% uncinate, 5% anterior thalamic radiation
L. anterior limb of internal capsule	45	−22	12	15	61% anterior thalamic radiation, 3% inferior fronto‐occiptial fasciculus
L. middle frontal G	25	−30	18	44
R. rolandic operculum	11	48	−6	20	18% superior longitudinal fasciculus
R. middle temporal G	13	50	−22	−15	8% inferior longitudinal fasciculus
R. precuneus	24	10	−63	38
L. splenium of corpus callosum	24	−16	−52	19	8% forceps major
R. optic radiation (10% Pallidum, 10% Putamen)	14	29	−18	−5	3% inferior longitudinal fasciculus
L. cerebellum VI	34	−32	−63	−19
R. cerebellum crus1	12	34	−68	−32
R. cerebellum VIII	17	34	−67	−54
	17	25	−73	−55

Notes: No suprathreshold regions identified for the contrast of Schizophrenia subjects – Control subjects. L = left, R = right. All regions identified at false discovery rate of q < 0.05 for a spatial extent of at least 10 voxels. The effect of verbal IQ was removed.

^aJHU white‐matter tractography atlas (Hua et al., 2008).

Figure 3.

(A) White matter area (blue color) showed significant reduction of FA in patients with schizophrenia, compared to control subjects. The significantly reduced regions were dilated for just visualization in here (B and C). Probabilistic tracts (green color) generated from left inferior frontal gyrus (orange color: region showing decreased activation during the semantic encoding condition in schizophrenic subjects) to left superior temporal sulcal/middle temporal region (orange color in temporal lobe: region showing less functional correlation with mean time series of left inferior frontal gyrus in patients with schizophrenia, compared to control subjects). Yellow: skeletonized white matter. All images place on MNI space. Neurological orientation.

Relationship Between Anatomical and Functional Connectivity

Finally, we investigated the relationship between mean FA value in left IFWM regions showing decreased white matter integrity within the fiber tracts interconnecting between left IFG and left middle temporal gyrus/left superior temporal sulcus in schizophrenia and the correlation coefficients of correlation map created by the functional correlation analysis. We hypothesized that the white matter disruption affecting anatomical connectivity would predict functional connectivity disruptions observed in patients. As expected, schizophrenic subjects (but not controls) showed significant, positive correlation between left IFWM integrity, FA value, and the correlation coefficient of several regions including left middle temporal gyrus/left superior temporal sulcus, left supramarginal gyrus, left thalamus, right insula, and bilateral cerebellum created by the functional correlation analysis (see Table V and Fig. 4A). Specifically, we observed positive relationship between disruptions of white matter integrity of left IFWM, and the functional connectivity abnormalities between left IFG and left middle temporal gyrus during semantic encoding in schizophrenia (Fig. 4B).

Table V.

Correlation between FA value in left IFWM and functional correlation map

Regions	BA	Cluster size	Schizophrenia subjects
			Peak voxel coordinate
			x	y	z
L. middle temporal G/superior temporal S.	22/42	240	−52	−44	0
L. supramarginal G/central opercular cortex	43, 40	784	−60	−26	24
Bilateral posterior cingulate G/R. precuneus	31	394	8	−48	38
L. thalamus		159	−10	−14	12
R. insula		87	38	−14	−2
R. cerebellum		297	12	−50	−18
L. cerebellum		1405	−10	−60	−14

Notes: No suprathreshold regions identified for control group. BA = Brodman's area, L = left, R = right, G = gyrus, S = sulcus, IFG = inferior frontal gyrus.

Statistical significance was confirmed through analysis of 5,000 nonparametric permutations. Activations identified at cluster‐level significance of P < 0.05 (corrected) for a spatial extent of at least 50 voxels in each group (IQ as covariate).

Figure 4.

(A) Left superior temporal sulcus/middle temporal gyrus showed positive correlation (red color) between mean FA value of left inferior frontal white matter and correlation coefficients for functional connectivity. Activations identified at cluster‐level significance of P < 0.05 (corrected) for a spatial extent of at least 50 voxels in each group (IQ as covariate) (B). The probabilistic tract (green color) generated from left inferior frontal gyrus is passing through the left superior temporal sulcus/middle temporal gyrus.

DISCUSSION

Our study demonstrated two previously hypothesized, but never reported phenomena in schizophrenia. First, we detected deficits in functional connectivity using fMRI measures as well as anatomical connectivity using DTI measures within the left language network that involves left IFG, left middle temporal gyrus/left superior temporal sulcus and white matter tracts interconnecting these regions in schizophrenia. Second, we demonstrated that these deficits are closely related to each other. More specifically, during semantic encoding, control subjects showed a high degree of connectivity between the pars triangularis of left IFG and the left middle temporal gyrus/left superior temporal sulcus. Schizophrenic patients did not show statistically significant connectivity between these regions. Anatomical connectivity analysis using DTI measures revealed a disruption within the white matter interconnecting left hemisphere language regions in schizophrenia. Finally correlation analysis demonstrated that the functional deficits observed within the language network, might be related to white matter disruption within this network.

In terms of the functional abnormalities observed in our study, underactivation of left IFG in schizophrenia during deep encoding has been reported previously by us [Kubicki et al., 2003a], and was attributed to different encoding memory strategy used by patients [Bonner‐Jackson et al., 2005, 2007]. Ragland et al. [ 2005] demonstrated no difference in the activation of left IFG between schizophrenics and controls when performing levels‐of‐processing paradigm. There is some evidence, however, that left IFG function is abnormal in schizophrenia. Kuperberg, for example, has demonstrated that schizophrenics show an abnormal increase in the activity of left IFG during semantic priming while controls showed a decrease in activity in response to semantically related words [Kuperberg et al., 2007]. Conversely, underactivation of the left IFG has also been reported in patients with schizophrenia during various language tasks including verbal learning [Eyler et al., 2008], language comprehension [Dollfus et al., 2005] and word encoding [Ragland et al., 2004]. Also, several fMRI studies demonstrate a relationship between left inferior frontal gyrus and semantic processing [Baker et al., 2001; Marinkovic et al., 2003; Savage et al., 2001]. Finally, Bonner‐Jackson et al. [ 2005] demonstrated that when being provided with semantic processing strategy, schizophrenics show increased left prefrontal activation, including left IFG, which reached and exceeded activation levels seen in control subjects. Commenting on those findings, the authors suggested that the wider prefrontal cortex activation during deep (successful) encoding represented a potential endophenotype marker of genetic liability for schizophrenia. We would argue that intact functional connectivity between left IFG and other parts of the frontal cortex, and decreased communication between left IFG and temporal or parietal lobe (as shown by both functional and anatomical connectivity results), might underlie the described results.

Of note, functional connectivity is defined as the synchronization between the concurrent activities of different cortical regions. This synchronization is usually based on quantifying covariances/correlations among the brain activation time series. A functional connectivity group comparison can provide some insight into the neural circuits' abnormalities of the diseased brain. Previous studies with schizophrenia, for example, have demonstrated altered fronto‐temporal functional connectivity during cognitive tasks [Lawrie et al., 2002; Wolf et al., 2007]. While the studies using ROI based‐functional connectivity analysis [Cheung et al., 2007; Ford et al., 2002] are usually restricted to the network of an a priori hypothesized set of ROIs, whole brain correlation analysis, as used here, is free from such limitations. On the other hand, results of such an analysis should still be interpreted with caution. Our analysis can not, for example, infer causality, and the confounding factors, such as incompletely removed task‐related movement, can influence the results. Thus, in the effect, regions that are not part of the same functional network can still express similar time courses, and appear as functionally connected in this analysis.

As the results from functional connectivity analyses simply reflect the observed temporal correlations between brain regions, interpretation of such results would greatly benefit from additional anatomical data, such as DTI. Our DTI results revealed local white matter integrity disruption within the left frontal lobe in schizophrenia, which was reported previously by several investigators [Andreone et al., 2007; Buchsbaum et al., 1998, 2006]. Subsequent exploratory probabilistic tractography suggested that the disrupted white matter region might belong to the left superior longitudinal fasciculus, presumably left arcuate fasciculus, and left inferior occipito‐frontal fasciculus and/or left inferior longitudinal fasciculus, tracts that form anatomical connections between the left IFG and the left middle/superior temporal gyrus (see Fig. 3).

Traditionally, it has been suggested that the left arcuate fasciculus is the only connection within the semantic network [Catani et al., 2005]. Few recent studies, however, suggest a relationship between semantic function and left inferior longitudinal fasciculus, as well as left inferior occipito‐frontal fasciculus [Duffau et al., 2005; Mandonnet et al. 2007].

So far, three DTI studies demonstrate anatomical disruption of the frontal part of left arcuate fasciculus in schizophrenia [Burns et al., 2003; Douaud et al., 2007; Kubicki et al., 2005b], and one additional paper suggests increased integrity of the temporoparietal part of left arcuate fasciculus in schizophrenia patients experiencing auditory hallucinations [Hubl et al., 2004], while one study points to the left inferior occipito‐frontal fasciculus/inferior longitudinal fasciculus as related to schizophrenia [Ashtari et al., 2007; Cheung et al., 2007; Szeszko et al., 2008. Our study, however, is the first to investigate the relationship between these deficits and semantic abnormalities observed in schizophrenia.

The deficit in functional connectivity of left IFG during semantic encoding condition and its relationship with a disruption of interconnecting white matter tracts in schizophrenia not only suggests a strong relationship between functional and anatomical deficits observed in this disease, but it also suggests a need for such an integrating approach in other cognitive domains affected in schizophrenia, such as, for example, language lateralization [Oakes et al., 2005], visual processing [Toosy et al., 2004], and working memory [Olesen et al., 2003].

Limitations of the Study

This is one of the first studies that combines functional and anatomical connectivity analysis in schizophrenia population. As such, it is not free from limitations. First and foremost, sample size is small but we believe that the convergence of the results obtained independently with different modalities (DTI and FMRI) increases the strength of our findings. Next, only chronic, medicated males were involved in the study, thus some of the disease related connectivity abnormalities could be masked by small sensitivity, and/or medication effects, or possible gender effects. In the effect, both small sample size and inclusion of males only could limit the generalizability of our results. Moreover, specificity of diffusion findings is also limited, since white matter FA could be decreased for many different reasons, including axonal and myelin abnormalities, as well as fiber tract coherence. Also, IQ also could affect the activation of left IFG as it is related to the left inferior frontal (Broca's area) activation during semantic task in healthy children [Schmithorst and Holland, 2006]. Low IQ has been also shown to be related to reduced whole brain volume in schizophrenia, however, the same study did not show left inferior frontal gyrus volume and IQ relationship in schizophrenia subjects, but did show relationship between verbal memory and this region volume in healthy controls. Thus while low IQ might, to some extent, be related to schizophrenia itself, it may have only a limited effect on the underactivation of left IFG. IQ and its low values in patients with schizophrenia could be still another possible source of functional group differences. Even though our subjects showed no difference in WRAT3 Reading score, the relatively low average IQ of the patients might still affect the brain activation. Since many imaging studies using various modalities do show correlations between IQ and functional as well as anatomical brain changes [Choi et al., 2008; Deary et al., 2006; Thatcher et al., 2007, the low average IQ observed in our study should be controlled for. It is important not only because of the possible impact of IQ on semantic encoding and IFG activation, but also on functional connectivity in general.

In addition, there were also several limitations specific to the analytic methodology used‐TBSS methodology. More specifically, even though skeleton based registration [Smith et al., 2006] performs much better than the whole white matter registration, it is still based on the premise that it is possible to find and warp corresponding structures in every individual brain. Since this is usually not the case, the method can suffer from misregistration errors, especially since partial volume effects with 4‐mm thick slices can be significant. Small number of diffusion gradient directions DTI could also affect the accuracy of FA estimation, introducing additional noise to the analysis. In addition, TBSS does not assign abnormalities to anatomical structure, thus additional localization tools, such as tractography, need to be used in order to understand better the relationship between functional and anatomical connectivity within the semantic network.

In summary, our study demonstrates the additional analytic power in combining functional and anatomical information, and suggests association between semantic encoding related functional deficit and abnormal anatomical connectivity in schizophrenia.