Neural correlates of formal thought disorder: An activation likelihood estimation meta‐analysis

Abstract

Formal thought disorder (FTD) refers to a psychopathological dimension characterized by disorganized and incoherent speech. Whether symptoms of FTD arise from aberrant processing in language‐related regions or more general cognitive networks, however, remains debated. Here, we addressed this question by a quantitative meta‐analysis of published functional neuroimaging studies on FTD. The revised Activation Likelihood Estimation (ALE) algorithm was used to test for convergent aberrant activation changes in 18 studies (30 experiments) investigating FTD, of which 17 studies comprised schizophrenia patients and one study healthy subjects administered to S‐ketamine. Additionally, we analyzed task‐dependent and task‐independent (resting‐state) functional connectivity (FC) of brain regions showing convergence in activation changes. Subsequent functional characterization was performed for the initial clusters and the delineated connectivity networks by reference to the BrainMap database. Consistent activation changes were found in the left superior temporal gyrus (STG) and two regions within the left posterior middle temporal gyrus (p‐MTG), ventrally (vp‐MTG) and dorsally (dp‐MTG). Functional characterization revealed a prominent functional association of ensuing clusters from our ALE meta‐analysis with language and speech processing, as well as auditory perception in STG and with social cognition in dp‐MTG. FC analysis identified task‐dependent and task‐independent networks for all three seed regions, which were mainly related to language and speech processing, but showed additional involvement in higher order cognitive functions. Our findings suggest that FTD is mainly characterized by abnormal activation in brain regions of the left hemisphere that are associated with language and speech processing, but also extend to higher order cognitive functions. Hum Brain Mapp 38:4946–4965, 2017. © 2017 Wiley Periodicals, Inc.

INTRODUCTION

Formal thought disorder (FTD) is characterized by symptoms of disorganized and incoherent speech [Andreasen, 1979, 1986]. The first conceptualization of FTD dates back to the late 19th and 20th century [Berrios, 1996]. Later in his seminal work, Bleuler reported a common “loosening of associations” in psychiatric patients, which he described as disturbances of thought and speech production [Bleuler, 1911, 1950]. Although FTD is mainly investigated in schizophrenia and considered one of its core symptoms, it is diagnostically not limited to one particular psychiatric disorder [International Classification of Diseases, ICD; World Health Organization, 1992; Diagnostic and Statistical Manual of Mental Disorders, DSM, American Psychiatric Association, 2013]. Indeed, symptoms of FTD are frequently observed in various psychiatric disorders [Andreasen, 1979; Eussen et al., 2015; Nagels et al., 2016 ; Sax et al., 1995] and can also be pharmacologically induced in healthy individuals in a dose‐dependent manner by administration of the NDMA‐antagonist ketamine [Adler et al., 1999; Nagels et al., 2012].

Based on behavioral empirical findings, FTD is mainly associated with deficits in higher order cognition, such as executive function, on the one hand (dysexecutive hypothesis), and language and speech on the other (dyssemantic hypothesis) [Barrera et al., 2005]. For instance, FTD symptoms might arise from difficulties in planning, monitoring and maintaining a specific set of goals [McGrath, 1991; McGrath et al., 1997], pointing towards an executive processing dysfunction in FTD. Other studies found enhanced semantic priming [Safadi et al., 2013] and semantic task performance [McKay et al., 1996] in schizophrenia patients with FTD, which rather supports the dyssemantic hypothesis. Recent studies, however, suggest a more integrative view on deficits in executive function and language [Remberk et al., 2012; Stirling et al., 2006], although findings in this regard are far from conclusive.

Most functional neuroimaging studies so far have assessed FTD focusing on aspects of language and speech processing. Most studies point towards a deficit in speech production and semantic processing networks that might be pronounced in left‐hemispheric activation changes of superior and middle temporal, as well as superior and inferior frontal brain regions [Horn et al., 2012; Kircher et al., 2001a,2008; Sass et al., 2014]. However, FTD might be also associated with processing deficits in domains other than language and speech, especially higher order cognitive functions, such as object recall and selective attention [Remberk et al., 2012; Stirling et al., 2006]. Given these findings and the close relationship between higher order cognitive functions and language [Baddeley, 2003], deficits in the former or the latter domain might not constitute distinct phenomena in independent brain regions, but could rather be indicative of a common neural correlate in FTD.

However, most studies rely on comparatively small sample sizes [Button et al., 2013] and the operationalization of FTD in functional neuroimaging is highly heterogeneous due to the specificity of research questions associating certain features of FTD (such as symptom severity) with changes in brain activation. This hampers a broader view on a potential common underlying neurobiological deficit in FTD. Yet, FTD represents a highly debilitating individual burden regardless of its diagnostic context, emphasizing an urgent need for the integration of published research on FTD on the functional neuroimaging level. To address these issues, we performed a quantitative meta‐analysis synthesizing previously published functional neuroimaging studies on activation changes in FTD. Additionally, we aimed to identify if the ensuing brain regions might be associated with neural networks in healthy individuals using task‐dependent and task‐independent functional connectivity analyses.

MATERIALS AND METHODS

Literature Search and Inclusion Criteria

We conducted a PubMed database (http://www.ncbi.nlm.nih.gov/pubmed) search using search string combinations related to formal thought disorder (see Supporting Information Table 1 for a detailed list of all search terms), together with either “fMRI,” “PET,” or “neuroimaging.” Additional studies were determined by reference tracing of already identified studies and qualitative reviews. Studies were included if they (1) were published in a peer‐reviewed journal, (2) assessed symptoms of FTD irrespective of a clinical diagnosis (using questionnaires, interviews, rating scales), (3) performed whole‐brain analysis of task‐based functional MRI or PET neuroimaging data, and (4) reported results in standard stereotactic space [Montreal Neurological Institute, MNI; Holmes et al., 1998; Talairach & Tournoux, TAL; Talairach and Tournoux, 1988]. Further, we only included studies that reported activation maxima ensuing from whole‐brain analyses which were associated with FTD based on (A) group contrasts between subjects with symptoms of FTD and a control group without FTD symptoms or (B) a correlation analysis of FTD scores and BOLD signal change within an FTD group. This procedure yielded a total of 18 original studies with 55 experiments, including 34 experiments on group contrasts between subjects with and without symptoms of FTD (i.e., FTD > controls, FTD < controls), and 21 experiments on correlations between FTD scores and BOLD signal change (i.e., positive and negative correlations). Of these 18 studies, activation maxima related to FTD were reported for schizophrenia patients in 17 studies and for healthy individuals administered to S‐ketamine in one study (Tables 1 and 2).

Table 1.

Overview of study samples included in the ALE meta‐analysis

Reference	N FTD		N controls		Mean age (FTD)	FTD handedness (right/left)	FTD assessment	FTD diagnosis	No. foci	Space
	SCZ	Healthy	SCZ	Healthy
Arcuri et al. [2012]	9	—	9	10	Median: 33.0	9/0	SAPS	9 SCZ	10	TAL
Assaf et al. [2006]	16	—	—	16	38.6	13/3	TDI	14 SCZ, 2 schizoaffective	16	TAL
Borofsky et al. [ 2010]	14	—	—	14	13.3	14/0	K‐FTDS	14 COS	29	TAL
Erkwoh et al. [ 2002] a	10	—	10	—	30.4	10/0	PANSS	10 SCZ	5	MNI
Han et al. [2007]	12	—	—	12	39.8	12/0	SAPS	12 SCZ	6	MNI
Kircher et al. [ 2001a] b	6	—	6	7	34.3	6/0	SAPS	6 SCZ	16	TAL
Kircher et al. [ 2001b] b	6	—	—	—	34.3	6/0	SAPS	6 SCZ	7	TAL
Kircher et al. [ 2002] b	6	—	—	6	34.3	6/0	SAPS	6 SCZ	29	TAL
Kircher et al. [ 2003] b	6	—	—	—	34.3	6/0	TLI	6 SCZ	13	TAL
Kircher et al. [ 2005] b	6	—	—	6	34.3	6/0	SAPS	6 SCZ	9	TAL
Kircher et al. [ 2008]	12	—	—	12	26.8	12/0	SAPS/SANS	12 SCZ	8	TAL
Matsumoto et al. [ 2013] b	6	—	—	6	34.3	6/0	TLI	6 SCZ	2	TAL
McGuire et al. [1998] a	6	—	—	—	34.0	6/0	TLCI	6 SCZ	9	TAL
Nagels et al. [ 2012]	—	15	—	15	27.9	15/0	PANSS	—	7	MNI
Ragland et al. [2008]	13	—	—	14	36.2	11/2	CDI	13 SCZ	33	MNIc
Sass et al. [ 2014]	14	—	—	14	36.4	14/0	SAPS	14 SCZ	8	MNI
Tagamets et al. [ 2014]	11	—	—	—	40.0	11/0	CSA/PTC	11 SCZ	19	MNI
Weinstein et al. [2007]	12	—	—	—	35.9	12/0	TLI	12 SCZ	1	MNIc

^aIncluded PET‐studies.

^bComprising multiple studies, which were modeled as one experiment.

^cTransformed reference space from TAL to MNI.

CDI, Communication Disorders Index [Docherty et al., 1996]; CSA, Computational Semantic Analysis; K‐FTDS, Kiddie Formal Thought Disorder Rating Scale [Caplan et al., 1989]; PTC, Personal Thematic Coherence; TDI, Thought Disorder Index [Solovay et al., 1986]; TLC, Scale for the Assessment of Thought, Language and Communication [Andreasen, 1986]; TLCI, Thought Language and Communication Index [Liddle, 1995]; TLI, Thought and Language Index [Liddle et al., 2002]. COS, childhood‐onset schizophrenia; FTD, formal thought disorder; SCZ, schizophrenia. TAL, Talairach; MNI, Montreal Neurological Institute.

Table 2.

Overview of experiments and functional paradigms included in the ALE meta‐analysis

Reference	Functional paradigm	Contrast(s)	Original no. of experiments	Pooled no. of experiments (included in analysis)	Source
Arcuri et al. [2012]	Semantic decision sentence task	Congruent vs. incongruent	3 Experiments: 1 FTD Patients > Controls Patients. 1 FTD Patients < Controls Patients. 1 FTD Patients < Controls Healthy	Three experiments	Table III; (c)
Assaf et al. [2006]	Verbal object recall	Recall vs. no recall	3 Experiments: 1 FTD Patients > Controls Healthy. 1 FTD Patients < Controls Healthy. 1 pos. correlation with Symptom severity	Three experiments	Group Contrasts: Table IV. Correlation Analyses: p. 455
Borofsky et al. [ 2010]	Semantic judgment	(1) Semantic judgment vs. baseline rest. (2) Syntactic judgment vs. baseline rest	4 Experiments: 1 FTD Patients < Controls Healthy, Syntactic vs. baseline. 1 FTD Patients < Controls Healthy, Semantic vs. baseline. 1 neg. correlation with Symptom severity, Semantic. 1 neg. correlation with Symptom severity, Syntactic	One experiment	Group contrasts: Table VI. Correlation analyses: Table VII (upper panel)
Erkwoh et al. [ 2002] a	Go/NoGo	Go/No‐Go vs. sustained attention	2 Experiments: 1 FTD Patients > Controls Healthy. 1 FTD Patients < Controls Healthy	Two experiments	Table IV (–FTD > +FTD and +FTD > –FTD)
Han et al. [2007]	Semantic word priming	Unrelated vs. low vs. high	4 Experiments: 1 FTD Patients < Controls Healthy. 1 pos. correlation with Distractive Speech. 1 pos. correlation with Illogicality. 1 pos. correlation with Incoherence	Two experiments	Group contrasts: p. 276. Correlation analyses: p. 277
Kircher et al. [ 2001a] b	Sentence completion task	(1) Word generation vs. reading. (2) Word generation vs. decision. (3) Decision vs. reading	7 Experiments: 1 FTD Patients > Control Patients, Generation vs. reading. 1 FTD Patients < Control Patients Generation vs. reading. 1 FTD Patients > Control Healthy, Generation vs. reading. 1 FTD Patients < Control Healthy, Generation vs. reading. 1 FTD Patients > Control Healthy, Generation vs. decision.1 FTD Patients < Control Healthy, Generation vs. decision.1 FTD Patients < Controls Healthy, Decision vs. reading	Eight experiments	Generation vs. reading: Table V. Generation vs. decision and decision vs. reading: p. 32
Kircher et al. [ 2001b] b	Free verbal association	Verbal Association vs. baseline rest	2 Experiments: 1 pos. correlation with Symptom severity. 1 neg. correlation with Symptom severity	See Kircher et al. [2001a]	Table II
Kircher et al. [ 2002] b	Free verbal association	Verbal Association vs. baseline rest	4 Experiments: 1 FTD Patients > Control Healthy. 1 FTD Patients < Control Healthy. 1 pos. correlation with Rate of articulation. 1 neg. correlation with Rate of articulation	See Kircher et al. [2001a]	Correlation analyses: Table III. Group contrasts: Table IV
Kircher et al. [ 2003] b	Free verbal association	Verbal Association vs. baseline rest	2 Experiments: 1 pos. correlation with Symptom severity. 1 neg. correlation with Symptom severity	See Kircher et al. [2001a]	Table II
Kircher et al. [ 2005] b	Free verbal association	(1) Simple sentences vs. baseline rest. (2) Complex sentences vs. baseline rest	4 Experiments: 1 FTD Patients > Controls Healthy, Simple vs. baseline. 1 FTD Patients < Controls Healthy, Simple vs. baseline. 1 FTD Patients > Controls Healthy, Complex vs. baseline. 1 FTD Patients < Controls Healthy, Complex vs. baseline	See Kircher et al. [2001a]	Simple sentences: Table III. Complex sentences: Table IV
Kircher et al. [ 2008]	Free verbal association,	(1) Free association vs. word reading. (2) Semantic verbal fluency vs. word reading	2 Experiments: 1 FTD Patients < Controls Healthy, Free vs. reading. 1 FTD Patients < Controls Healthy, Semantic vs. reading	One experiment	Table IV
Matsumoto et al. [ 2013] b	Free verbal association	(1) Between‐clause pauses vs. continuous speech. (2) Within‐clause pauses vs. continuous speech	2 Experiments: 1 FTD Patients < Controls Healthy, Between vs. continuous. 1 FTD Patients < Controls Healthy, Within vs. continuous	see Kircher et al. [2001a]	Table III (between‐clause pauses, controls > FTD and within‐clause pauses, controls > FTD)
McGuire et al. [1998] a	Free verbal association	Verbal Association vs. baseline rest	2 Experiments: 1 pos. correlation with Symptom severity. 1 neg. correlation with Symptom severity	Two experiments	Table I
Nagels et al. [ 2012]	Overt word generation	Phonological verbal fluency vs. baseline rest	3 Experiments: 1 pos. correlation with Conceptual disorganization. 1 pos. correlation with Difficulties in abstract thinking. 1 pos. correlation with Flow of conversation	One experiment	Table II
Ragland et al. [2008]	Semantic fluency	(1) Semantic category vs. baseline word generation. (2) Semantic category vs. overlearned sequence. (3) Overlearned sequence switch vs. overlearned sequence no switch. (4) Semantic category switch vs. semantic category no switch	5 Experiments: 1 FTD Patients > Control Patients, Semantic vs. baseline. 1 FTD Patients < Control Patients Semantic vs. baseline. 1 FTD Patients > Control Healthy, Semantic vs. sequence. 1 FTD Patients > Control Healthy, Sequence switch vs. no switch. 1 FTD Patients > Control Healthy, Semantic switch vs. no switch	Three experiments	Tables III–VI
Sass et al. [ 2014]	Semantic priming	(1) Related vs. unrelated. (2) Unimodal vs. crossmodal	2 Experiments: 1 FTD Patients > Controls Healthy, Related vs. unrelated. 1 FTD Patients < Controls Healthy. Unimodel vs. crossmodal	Two experiments	Table III; (A) and (B)
Tagamets et al. [ 2014]	One‐back	(1) Words vs. baseline fixation. (2) Homographs vs. baseline fixation. (3) Homophones vs. baseline fixation	3 Experiments: 1 pos. correlation with Symptom severity, Words. 1 pos. correlation with Symptom severity, Homographs. 1 pos. correlation with Symptom severity, Homophones	One experiment	Table V
Weinstein et al. [2007]	Passive listening	(1) English vs. baseline rest. (2) Reversed English vs. baseline rest. (3) Mandarin vs. baseline rest	1 Experiment: 1 pos. correlation with Symptom severity, English	One experiment	p. 191

^aIncluded PET studies.

^bComprising multiple studies, which were modeled as one experiment.

pos., positive; neg, negative; FTD, formal thought disorder.

Within each study, we summarized experiments reflecting activation maxima in FTD with respect to hyperactivation (i.e., FTD > controls and positive correlations) and hypoactivation (i.e., FTD < controls and negative correlations). We then pooled our analysis across hyper‐ and hypoactivation, since we were mainly interested in general aberrant activation changes associated with FTD. Moreover, it remains unclear to which extent results of included experiments arise from group‐by‐condition interaction effects, which might no longer be associated with activation maxima reflecting proper hyper‐ and hypoactivation [Muller et al., 2017]. Put differently, differential activations with respect to tasks and analysis approaches across original studies (and hence, assumed neurobiological mechanism) might be detected as hyperactivation in one study and hypoactivation in the other.

All included activation maxima were conveyed into the same stereotactic reference space (MNI) using linear transformation [Lancaster et al., 2007] to enable ALE analysis. Results of experiments using the software packages SPM (http://www.fil.ion.ucl.ac.uk/spm) or FSL (http://fsl.fmrib.ox.ac.uk/fsl) were considered as coordinates in (proper) MNI space when no explicit transformation of reference spaces was reported [see Brett et al., 2001, 2002; Lancaster et al., 2007], as the standard space in SPM and FSL is MNI. Six identified studies [Kircher et al., 2003, 2001a,b, 2002, 2005; Matsumoto et al., 2013] used the same sample of subjects to analyze different aspects of FTD in schizophrenia. We thus modeled the results of those six studies as one experiment (separately for hyper‐ and hypoactivation, as well as group contrasts and correlation analyses) to avoid that one particular FTD group might systematically bias our results [Turkeltaub et al., 2012]. Based on these criteria of combining experiments to reduce effects of the same sample, the original 55 experiments were reduced to 30 experiments, with 15 experiments reflecting hyperactivation and 15 experiments hypoactivation.

ALE Meta‐Analysis

Activation likelihood estimation

In the current meta‐analysis, the revised version of the activation likelihood estimation (ALE) algorithm [Eickhoff et al., 2009] was used. The aim of the ALE approach lies in identifying convergent clusters of foci in which activation changes are significantly higher than under a random spatial null‐distribution. The main concept of ALE is to treat coordinates not as single points, but as peaks of random Gaussian 3D distributions to account for the spatial uncertainty underlying functional neuroimaging data. The probabilities of all activation foci for a given experiment are then combined for each voxel in a modeled activation (MA) map [Turkeltaub et al., 2012]. The union of the individual MA maps across experiments results in respective ALE scores reflecting the convergence of foci for each location in the brain. These ALE scores are then compared to a null‐distribution of random association between experiments to infer whether convergence between experiments occurs on an above‐chance level [Eickhoff et al., 2012]. Statistical parametric maps were thresholded using cluster‐level family‐wise error (FWE) correction at P < 0.05 (cluster‐forming threshold at voxel‐level P < 0.001 [Eickhoff et al., 2016]).

Follow‐up: Analyses on ensuing clusters

In order to identify functional connectivity (FC) networks of brain regions showing convergent activation changes associated with FTD, we conducted subsequent meta‐analytic connectivity modeling (MACM) and resting‐state connectivity analysis [Muller et al., 2013; Nickl‐Jockschat et al., 2015]. To assess whether networks were independent of either modality, that is, task‐dependent (MACM) and task‐independent (resting‐state) FC, we performed conjunction analyses across both FC approaches resulting in a “consensus connectivity network” (CCN). Ensuing CCNs were then functionally characterized. This approach allows for inference about the involvement of clusters ensuing from our ALE meta‐analysis in larger‐scale cortical networks with reference to healthy populations. However, it does not consider differential activation patterns between subjects with and without FTD. Our basic aim with this approach was the identification of neural networks to further characterize the physiological functions of the respective brain regions derived from the ALE meta‐analysis.

Task‐based functional connectivity: Meta‐analytic connectivity modeling (MACM)

For meta‐analytic connectivity modeling (MACM) [Eickhoff et al., 2011], each brain region derived from the ALE‐meta analysis was considered as an individual seed. In this approach, we used the BrainMap database [http://www.brainmap.org; Laird et al., 2011, 2009] to identify all studies that report at least one focus of activation within the respective seed. Subsequently, all co‐activations reported by the original study are extracted and quantitative meta‐analysis was employed to test for convergence across these foci. This enables the identification of task‐dependent functional connectivity networks of the seed region.

Only studies reporting group analyses of functional mapping experiments of healthy subjects were included, while experiments dealing with disease or drug effects were excluded. Subsequently, we performed a coordinate‐based meta‐analysis to identify consistent co‐activations across experiments using ALE. Results were thresholded at P < 0.05 corrected for multiple comparisons applying FWE correction on the cluster‐level (cluster‐forming threshold at voxel‐level of P < 0.001).

Task‐independent functional connectivity: Resting‐state fMRI

For resting‐state connectivity analysis, the same seeds as for MACM were used. Similar to MACM, we also focused on only healthy subjects in the resting‐state analyses. In this approach, we assessed connectivity profiles of the seed regions ensuing from the ALE meta‐analysis in a well‐described sample at rest (see Supporting Information for technical details on sample characteristics, scanning parameters and imaging preprocessing/processing steps). Resting‐state images were denoised using FMRIBs ICA‐based Xnoiseifier (FIX) [Salimi‐Khorshidi et al., 2014] to control for artifactual fluctuations not related to neural activity in the resting‐state data. Processing of resting‐state images included (1) alignment, (2) spatial normalization, (3) deformation and smoothing, and (4) band‐pass filtering, following standard procedures [e.g., Poeppl et al., 2015; Reetz et al., 2012; Rottschy et al., 2013]. For each subject, time‐courses of all voxels within a seed region were extracted as their first eigenvariate. Resting‐state FC was assessed by computing linear Pearson correlation coefficients between the time series of each seed region and those of all other gray matter voxels across the whole brain. The ensuing correlation coefficients were transformed into Fisher's z‐scores and used in a second‐level analysis of variance (ANOVA), including phenotypical information on age, gender, and handedness as covariates of no interest. Consistent with the MACM analysis, results were thresholded at P < 0.05 with cluster‐level FWE correction (cluster‐forming threshold P < 0.001).

Functional connectivity: Conjunction analyses

We aimed to identify patterns of FC independent of modality, that is, irrespective of the subjects' mental state (i.e., task or task‐free). We thus performed a conjunction analysis between MACM and resting‐state connectivity networks identified for each of the three seed regions separately using the minimum statistics [Nichols et al., 2005] resulting in a CCN of each region. We additionally conducted a conjunction analysis across the three CCNs yielded by conjunction of the MACM and resting‐state analyses to assess whether these regions could be regarded as part of a single network that might be involved in aberrant activation associated with FTD. To avoid incidental overlap between connectivity networks, we applied an additional cluster threshold of k > 50 voxels to all conjunctions.

Anatomical labeling

Labeling of anatomical brain regions (macroanatomy) and cytoarchitectonic areas (microanatomy) was performed using the SPM Anatomy Toolbox Version 2.2 [Eickhoff et al., 2005]. For macroanatomical brain regions not labeled in the Anatomy Toolbox, we referred to the probabilistic Harvard‐Oxford atlas [Desikan et al., 2006] as provided by FSLView v4.0.1 (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki). References regarding the labels of assigned probabilistic cytoarchitectonic areas are provided in the captions of each results table.

Functional characterization

To further characterize brain regions derived from our ALE on FTD and their CCNs, we assessed whether we could find significant functional associations with behavioral domains and paradigm classes across experiments indexed in the BrainMap database [http://www.brainmap.org; Laird et al., 2011, 2009]. In this regard, behavioral domains describe mental processes, which are defined by contrasts (divided in the broad subcategories of action, cognition, emotion, interoception, and perception), whereas paradigm classes represent specific tasks (e.g., n‐back, Go/No‐Go, Stroop; for a detailed BrainMap taxonomy, see http://www.brainmap.org/taxonomy). We performed forward and reverse inference. While the former denotes the probability of a particular task activating a brain region, the latter refers to the probability of a psychological process given activation in a specific brain region. In other words, in the forward inference approach we assessed whether the probability of neural activation given a particular mental process [P(activation|task)] was higher than activation [P(activation)] at baseline, that is, finding a (random) activation from BrainMap in the respective regions. Significance was assessed using a binominal test (P < 0.05), corrected for multiple comparisons using false discovery rate (FDR). In the reverse inference approach, a region's functional profile was determined by identifying the most likely behavioral domains and paradigm classes in case of activation in a network. Here, significance was assessed by means of a chi‐square test (P < 0.05; FDR corrected).

Follow‐up: Additional ALE analyses on subgroups

Of note, two of the studies included in the initial ALE meta‐analysis comprised samples that differed from the remaining with respect to diagnostic status [non‐schizophrenic, but ketamine‐administered; Nagels et al., 2012] and age [non‐adults; Borofsky et al., 2010] (Table 1). We thus performed additional separate ALE meta‐analyses while excluding the studies in question, respectively. Accordingly, the ALE subanalysis on schizophrenia patients only comprised 29 experiments, 14 on hyperactivation, and 15 on hypoactivation. In the ALE subanalysis on adult study samples only, 29 experiments (15 on hyperactivation and 14 on hypoactivation) were included.

However, since we were mainly interested in neural correlates of FTD irrespective of any a priori assumptions with respect to subjects' clinical status, subsequent analyses were performed for all initially included studies only.

RESULTS

ALE Meta‐Analysis

Across all experiments (including hyper‐ and hypoactivation, as well as group contrasts and correlation analyses), we identified three clusters showing convergent aberrant activation changes related to FTD. Clusters were located in the temporal lobe of the left hemisphere, including the left superior temporal gyrus (STG; −54, −28, 4; k = 110; seven experiments contributing) and two regions in the posterior medial temporal gyrus (p‐MTG), one located more ventrally (vp‐MTG; −46, −50, −2; k = 71; six experiments contributing), and one more dorsally (dp‐MTG; −56, −56, 12; k = 109, seven experiments contributing) (Fig. 1, Table 3).

Figure 1.

ALE meta‐analysis results. Significant clusters of convergent aberrant activation related to FTD were found in the left ventral posterior middle temporal gyrus (vp‐MTG), left superior temporal gyrus (STG) and the left dorsal posterior middle temporal gyrus (dp‐MTG). Representations of resulting brain regions on a three‐dimensional‐rendered single subject brain template (A) and in coronal cross‐sections (B). Statistical significance was assessed at P < 0.05 (cluster‐level FWE corrected for multiple comparisons, cluster‐forming threshold P < 0.001 at voxel level). [Color figure can be viewed at http://wileyonlinelibrary.com]

Table 3.

Clusters showing convergent activation maxima in the ALE meta‐analysis

Cluster	k (voxels)	Anatomical location	MNI‐coordinates			t‐value	Contributions of experiments to cluster
		Macroanatomy	x	y	z
Cluster 1	71	L middle temporal gyrus	−46	−50	−2	4.03	Six of 30 experiments
Cluster 2	110	L superior temporal gyrus	−54	−28	4	4.80	Seven of 30 experiments
Cluster 3	109	L middle temporal gyrus	−56	−56	12	4.03	Seven of 30 experiments

To explore whether obtained clusters were mainly related to hyper‐ or hypoactivation of the respective brain regions, we analyzed the maxima contributing to each cluster [Tahmasian et al., 2016]. Our analysis revealed that convergence in the vp‐MTG was highly driven by experiments showing hyperactivation in FTD (97.20%). In contrast, both hyper‐ and hypoactivation maxima contributed to clusters in the left STG (58.10% hypoactivation) and dp‐MTG (62.88% hyperactivation).

Follow‐Up: Analyses on Ensuing Clusters

Functional connectivity: MACM and resting‐state fMRI

For details on functional connectivity networks ensuing from MACM and resting‐state connectivity analyses individually, please refer to the Results section of the Supporting Information.

Functional connectivity: Conjunction analyses

We performed conjunction analyses between task‐based and task‐independent FC networks for the three seed regions separately to delineate a CCN of the three seed regions ensuing from the ALE meta‐analysis that might be irrespective of modality (i.e., MACM and resting‐state fMRI).

Left vp‐MTG seed

The CCN of the left vp‐MTG seed included the seed region and the surrounding left MTG area and the left inferior frontal gyrus (IFG) including the pars triangularis and pars opercularis extending into the left precentral gyrus.

Left STG seed

For the left STG seed, consistent functional connectivity patterns encompassed the bilateral STG, with the left cluster extending into the pars triangularis, pars opercularis, and pars orbitalis of the left IFG, the left temporal pole, and the right cluster also including the rolandic operculum. In addition, the left STG was found to be functionally connected to the left postcentral gyrus and the left posterior‐medial frontal cortex.

Left dp‐MTG seed

The left dp‐MTG seed showed functional connectivity with the bilateral MTG extending into the STG on the right, pars triangularis, pars opercularis, and pars orbitalis of the bilateral IFG, as well as the bilateral precentral gyrus (Fig. 2, Table 4).

Figure 2.

Results of conjunction analysis of FC for each individual cluster. Clusters of overlap between MACM and RS‐connectivity analyses for the left vp‐MTG (A), left STG (B), and left dp‐MTG (C) seed region. Statistical significance was assessed at P < 0.05 (cluster‐level FWE corrected for multiple comparisons, cluster‐forming threshold P < 0.001 at voxel level, k > 50). [Color figure can be viewed at http://wileyonlinelibrary.com]

Table 4.

Brain regions and activation peaks for “consensus connectivity networks” ensuing from conjunction analyses across MACM‐ and RS‐connectivity networks

Seed region	Cluster	k (voxels)	MNI coordinates			Macroanatomy	Assigned cytoarchitecture
			x	y	z
vp‐MTG	Cluster 1	886	−50	−48	−4	L middle temporal gyrus
	Cluster 2	320	−50	14	16	L IFG (p. opercularis)	Left	Area 44
			−46	34	8	L IFG (p. triangularis)
			−48	12	34	L precentral gyrus
			−48	28	16	L IFG (p. triangularis)	Left	Area 45
			−54	24	4	L IFG (p. triangularis)	Left	Area 45
STG	Cluster 1	3525	−54	−28	6	L superior temporal gyrus
			−58	−12	−2	L superior temporal gyrus	Left	Area TE 3
			−46	22	20	L IFG (p. triangularis)
			−48	16	18	L IFG (p. opercularis)
			−50	14	16	L IFG (p. opercularis)	Left	Area 44
			−38	26	2	L IFG (p. triangularis)
			−40	24	−6	L IFG (p. orbitalis)
			−46	30	8	L IFG (p. triangularis)
			−48	28	0	L IFG (p. triangularis)
			−50	10	−6	L temporal pole
			−50	12	−2	L temporal pole	Left	Area 44
	Cluster 2	1954	60	−16	2	R superior temporal gyrus
			58	−24	4	R superior temporal gyrus
			60	−10	10	R rolandic operculum	Right	Area OP4 [PV]
			54	−14	12	R rolandic operculum	Right	Area OP4 [PV]
			52	−20	14	R rolandic operculum	Right	Area OP1 [SII]
	Cluster 3	218	−48	−6	40	L postcentral gyrus
	Cluster 4	169	−4	10	52	L posterior‐medial frontal
			−4	0	64	L posterior‐medial frontal
dp‐MTG	Cluster 1	1454	−56	−54	10	L middle temporal gyrus
			−40	−70	14	L middle temporal gyrus
	Cluster 2	872	−46	26	12	L IFG (p. triangularis)
			−50	12	16	L IFG (p. opercularis)	Left	Area 44
			−44	28	−10	L IFG (p. orbitalis)
			−54	26	−2	L IFG (p. triangularis)	Left	Area 45
			−42	24	22	L IFG (p. triangularis)
			−54	28	4	L IFG (p. triangularis)	Left	Area 45
			−54	14	4	L IFG (p. opercularis)	Left	Area 44
	Cluster 3	841	52	−56	10	R middle temporal gyrus
			50	−68	0	R middle temporal gyrus	Right	Area hOc4la
			62	−48	0	R middle temporal gyrus
			60	−44	12	R superior temporal gyrus
			54	−44	12	R superior temporal gyrus
			56	−30	2	R superior temporal gyrus
	Cluster 4	210	−46	2	48	L precentral gyrus
	Cluster 5	106	48	24	−4	R IFG (p. orbitalis)
			50	28	−2	R IFG (p. triangularis)	Right	Area 45
	Cluster 6	53	44	4	46	R precentral gyrus
vp‐MTG ∩ STG ∩ dp‐MTG	Cluster 1	144	−48	24	16	L IFG (p. triangularis)
			−52	16	16	L IFG (p. opercularis)	Left	Area 44
			−48	26	10	L IFG (p. triangularis)
			−50	24	12	L IFG (p. triangularis)	Left	Area 45
			−48	26	−2	L IFG (p. triangularis)
	Cluster 2	120	−54	−44	6	L middle temporal gyrus

Peak coordinates (x, y, z) are reported in standard space of the Montreal Neurological Institute (MNI) (cluster‐level FWE‐corrected P < 0.05, k > 50). For detailed information on assigned probabilistic cytoarchitectonic areas, please refer to Areas 44 and 45 Amunts et al. [1999]; Area PFcm: Caspers et al. [2006]; Areas hIP1 and hIP3 Choi et al. [2006]; Scheperjans et al. [2008]; Lubule VI: Diedrichsen et al. [2009]; Areas OP1 and OP4: Eickhoff et al. [2006]; Area Ig2: Kurth et al. [2010]; Area FG3: Lorenz et al. [2015]; Area hOc4Ia: Malikovic et al. [2016].

k, number of voxels; vp‐MTG, ventral posterior middle temporal gyrus; STG, superior temporal gyrus; dp‐MTG, dorsal posterior middle temporal gyrus; IFG, inferior frontal gyrus.

Finally, we conducted a conjunction analysis across CCNs of all three seed regions, revealing FC of all three seed regions with a cluster in the left IFG and a cluster in the MTG, located anterio‐dorsally to the vp‐MTG cluster and anterior to the dp‐MTG cluster (Fig. 3, Table 4). Of note, there was no overlap between the MTG cluster resulting from this conjunction and the ALE results used as seeds for FC analyses.

Figure 3.

Results of conjunction analysis of FC across all clusters. Clusters of overlap between consensus connectivity networks were identified in the left IFG and left MTG. Representations of resulting brain regions on a three‐dimensional‐rendered single subject brain template (A) and in coronal cross‐sections (B). Those clusters showed no overlap with the initial seed regions from the ALE. Statistical significance was assessed at P < 0.05 (cluster‐level FWE corrected for multiple comparisons, cluster‐forming threshold P < 0.001 at voxel level, k > 50). [Color figure can be viewed at http://wileyonlinelibrary.com]

Functional Characterization

Further, we functionally characterized the three ALE seed regions, the CCNs associated with these seeds and the overall CCN using the BrainMap database.

Left vp‐MTG seed

Functional characterization revealed that the left vp‐MTG was solely significantly associated with the behavioral domain of semantic cognition and paradigm class of semantic monitoring/discrimination (Fig. 4). The modality independent FC network of the left vp‐MTG seed, in turn, was strongly associated with several behavioral domains regarding language, especially with syntax, phonology, orthography and speech cognition, but also with higher cognitive functions (i.e., working memory). Additional paradigm classes associated with the CCN of the vp‐MTG included phonological, orthographic and pain discrimination, covert reading, overt and covert word generation, reward, delayed match‐to‐sample and passive viewing with language processing (Fig. 5).

Figure 4.

Functional characterization of ALE brain regions. Behavioral domains and paradigm classes from the BrainMap database significantly associated with the three seeds. Axis labeling indicates likelihood ratio values (left column) and probability values (right column) for forward‐ and reverse‐inference, respectively. FDR‐corrected for multiple comparisons at P < 0.05. [Color figure can be viewed at http://wileyonlinelibrary.com]

Figure 5.

Functional characterization of consensus connectivity networks. Additional behavioral domains and paradigm classes from the BrainMap database significantly associated with each seed's “consensus connectivity network.” Axis labeling indicates likelihood ratio values (left column) and probability values (right column) for forward‐ and reverse‐inference, respectively. FDR‐corrected for multiple comparisons at P < 0.05. [Color figure can be viewed at http://wileyonlinelibrary.com]

Left STG seed

The left STG seed region was significantly associated with behavioral domains of auditory perception, but also language‐related cognition and speech execution. Associated paradigm classes involved phonological, pitch, and tone discrimination, as well as passive listening and overt reading (Fig. 4). The left STG CCN was significantly linked to language‐related processing and auditory perception but also other sensory processing (i.e., pain perception). Similar to the vp‐MTG seed, additional paradigm classes associated with the STG network showed a heterogeneous profile that extended from mainly language‐ and auditory functions to a variety of higher order cognitive tasks (e.g., word generation, recitation/repetition, music comprehension/production) (Fig. 5).

Left dp‐MTG seed

For the left MTG, we found significant association with behavioral domains of social cognition and semantic language processing. Paradigm class analysis revealed an association with semantic discrimination, and theory of mind (Fig. 4). The CCN related to the left dp‐MTG was also significantly associated with social cognition, as well as the language related processes (i.e., syntax, semantics, phonology, speech, orthography). Analogous to the vp‐MTG and STG CCNs, paradigm classes associated with the dp‐MTG network were again wide‐ranged, spanning from speech‐related tasks (e.g., phonological discrimination, reading) to higher cognitive social tasks (e.g., reward processing, theory of mind) (Fig. 5).

Left vp‐MTG ∩ STG ∩ dp‐MTG

Combined functional decoding of the two clusters ensuing from the conjunction analysis across all three CCNs yielded significant associations with the behavioral domains of syntactic, phonological, semantic, orthographic, and speech processing in the language domain, interoception, as well as higher order cognition (i.e., social cognition). Significantly associated paradigm classes comprised phonological and semantic discrimination, overt and covert word generation, covert reading and reward processing (Fig. 6).

Figure 6.

Functional characterization of the overall consensus connectivity network. Behavioral domains and paradigm classes from the BrainMap database significantly associated with the overall “consensus connectivity network.” Axis labeling indicates likelihood ratio values (left column) and probability values (right column) for forward‐ and reverse‐inference, respectively. FDR‐corrected for multiple comparisons at P < 0.05. [Color figure can be viewed at http://wileyonlinelibrary.com]

Follow‐Up: Additional ALE Analyses on Subgroups

Across experiments assessing schizophrenia patients, we identified two clusters of convergent activation maxima in the left vp‐MTG (–46, −54, 3; k = 72; six experiments contributing) and left STG (–54, −32, 9; k = 114; seven experiments contributing) only, but not the dp‐MTG (see Fig. 7A, Table 5). We further analyzed the maxima contributing to each cluster with respect to hyper‐ and hypoactivation, revealing that the cluster in the left vp‐MTG was mostly driven by experiments reporting hyperactivation with respect to FTD (97.44%). Experiments contributing to the cluster in the left STG reported both hyper‐ and hypoactivation regarding FTD (58.15% hyperactivation).

Figure 7.

Additional ALE meta‐analysis results. For schizophrenia patients (A), significant clusters of convergent aberrant activation related to FTD were found in the left superior temporal gyrus (STG) and the left ventral posterior middle temporal gyrus (vp‐MTG), but no longer in the left dorsal middle temporal gyrus (dp‐MTG). In adult study samples (B), significant clusters of convergence were found in the left ventral posterior middle temporal gyrus (vp‐MTG) and the left dorsal posterior middle temporal gyrus (dp‐MTG), but no longer in the left superior temporal gyrus (STG) Representations of resulting brain regions on a three‐dimensional‐rendered single subject brain template and in coronal cross sections. Statistical significance was assessed at P < 0.05 (cluster‐level FWE corrected for multiple comparisons, cluster‐forming threshold P < 0.001 at voxel level). [Color figure can be viewed at http://wileyonlinelibrary.com]

Table 5.

Brain regions ensuing from ALE meta‐analysis on study samples with a primary diagnosis of schizophrenia only and adult study samples only

Cluster	k (voxels)	Anatomical location	MNI‐coordinates			t‐value	Contributions of experiments to cluster
		Macroanatomy	x	y	z
Schizophrenia patients
Cluster 1	114	L superior temporal gyrus	−54	−32	9	4.82	Seven of 29 experiments
Cluster 2	72	L middle temporal gyrus	−46	−54	3	4.05	Six of 29 experiments
Adult study samples
Cluster 1	85	L middle temporal gyrus	−46	−50	−2	4.15	Six of 29 experiments
Cluster 2	72	L middle temporal gyrus	−56	−58	12	3.88	Six of 29 experiments

In turn, inclusion of experiments with samples of adult age revealed convergent activation maxima in the left vp‐MTG (–46, −50, −2; k = 85; six experiments contributing) and dp‐MTG (–56, −58, 12; k = 72; six experiments contributing), but no longer in the STG (see Fig. 7B, Table 5). Analyzing maxima contributing to each cluster in terms of hyper‐ and hypoactivation showed that both the left vp‐MTG and dp‐MTG were mainly associated with hyperactivation (vp‐MTG: 97.36%; dp‐MTG: 73.68%).

DISCUSSION

Auditory Perception and Semantic Language Processing Dysfunction in FTD

Our ALE meta‐analysis revealed one cluster located in the left STG on the one hand and two clusters in the left posterior MTG (vp‐MTG and dp‐MTG) on the other. While the former was associated with the behavioral domains of speech and auditory processing (based on our functional characterization using the BrainMap database), the latter two were involved in semantic processing of language‐related information. Of note, the more dorsally located cluster of the MTG also showed functional association with social cognition and theory of mind. These findings are consistent with previous research on the functional roles of these regions in the processing of language and speech [Price, 2012]. Given that the cluster in the left posterior STG was located in close proximity to the primary auditory cortex, its functional characterization is well in line with other studies designating this brain region to auditory perception and speech production [Celsis et al., 1999; Warburton et al., 1996]. In contrast, the two clusters located in the left posterior MTG showed more specific processing of language as opposed to basic auditory perception. While the vp‐MTG was solely associated with language processing on a semantic level, the dp‐MTG showed an additional involvement in social cognition. Previous studies indicate that the posterior MTG is involved in both controlled semantic retrieval of conceptual information [Davey et al., 2015], as well as the processing of implicit and explicit social signals [Sugiura et al., 2014], potentially integrating semantic information into a more conversation‐directed and, hence, more “social” aspect of language. This rather hierarchical functional domain dissociation between the vp‐MTG and dp‐MTG might be reflective of increases in language processing task demands [Price, 2012].

Language‐Related Dysfunctional Brain Regions and Associated Networks in FTD

The brain regions identified by ALE were imbedded in larger‐scale cortical networks, enrolling frontal, temporal and parietal brain regions mainly in the ipsilateral, but also the contralateral hemisphere. These networks were primarily involved in language‐ and speech processing tasks, but also in additional functional domains of higher order cognitive processing and social cognition (i.e., working memory, time perception, reward processing). While our findings are to a considerable degree in line with the early emerging idea that FTD might foremost reflect deficits in semantic language processing [Goldberg et al., 1998], the CCNs' functional profiles also tackle various other aspects of language processing (e.g., phonology, syntax, speech, auditory perception), as well as higher cognitive demands (working memory) and social cognition. It stands to reason that in individuals with FTD, dysfunction of a network involved in adequate processing of phonological input, context‐specific information processing, and the retrieval and potential integration of language‐related information from memory might lead to most of the observed symptoms that are later subsumed under the concept of FTD. Intuitively, the integration of basic language and higher order cognitive processing is essential to produce coherent and comprehensive speech. Most evidence in this respect is based on neuroimaging and lesion studies, highlighting that cognitive processing steps necessary for speech production are hampered by disturbances of higher order brain regions. For instance, disruptions in brain regions such as the left IFG, which is involved in working memory functions as well as resolving of conflicting information, might subsequently result in disorganized or distorted speech output [Baddeley, 1996, 2003; Szczepanski and Knight, 2014]. Previous studies in schizophrenia patients point towards a close relationship between FTD in these patients and deficits in semantic language and higher order cognitive processing. More specifically, severity of specific FTD symptoms, such as poverty of speech, derailment or perseverations, was correlated with performance on standard neuropsychological tests requiring recruitment of language processing and executive functioning [McGrath et al., 1997]. Importantly, schizophrenia patients with FTD appear to perform worse on these sorts of tests compared to patients without FTD [Stirling et al., 2006; Tan and Rossell, 2014].

Methodological Considerations: ALE Subanalyses

We need to acknowledge that our analytic approach at hand raises a few methodological constraints and open questions, which should be considered in the context of the present discussion.

Two of the studies included in the initial ALE meta‐analysis comprised samples that differed from the remaining with respect to diagnostic status [non‐schizophrenic, but ketamine‐administered; Nagels et al., 2012], as well as age [non‐adults; Borofsky et al., 2010] and. We thus performed two additional separate ALE meta‐analyses while excluding the studies in question, respectively. Results revealed that including only experiments investigating schizophrenia patients resulted in convergent activation maxima in the vp‐MTG and STG only, but not the dp‐MTG. Inclusion of experiments with samples of adult age revealed convergent activation maxima in the vp‐MTG dp‐MTG, but no longer in the STG.

Given these alterations in results based on the exclusion of specific study samples, the most robust cluster of aberrant activation changes in FTD was located in the left vp‐MTG, while the remaining clusters in the left STG and dp‐MTG appeared more prone to be driven by one particular experiment of interest. Nonetheless, we deliberately refrained from excluding those experiments from our main analyses. Our reasoning in this decision was two‐fold: first, whether subsamples with unique demographic or phenotypical characteristics, such as children and adolescents, show similar or diverging neural correlates associated with FTD as adults or if there might be age‐related changes during brain development in this regard remains unknown. Hence, the inclusion of under‐age groups—especially in the context of such a broadly defined psychiatric dimension as FTD—appeared reasonable to us and might shed light on alterations in neurobiological underpinnings of FTD irrespective of differences in demographics or phenotype. This strategy has proven valuable in other ALE studies assessing psychiatric disorders, such as autism spectrum disorders [Nickl‐Jockschat et al., 2012].

Secondly, as already mentioned previously, FTD is a psychopathological dimension, which is also prevalent in diagnostic groups other than schizophrenia patients [Eussen et al., 2015], and might be induced by means of administration of pharmacological agents in healthy individuals [Adler et al., 1999]. Most evidence on the neural underpinnings of FTD is based on studies in schizophrenia patients [e.g., Kircher et al., 2008; Shenton et al., 1992] and FTD represents one of the cardinal symptoms within this disorder [Andreasen, 1979]. Thus, it remains an open question, whether neural correlates of FTD in other disorders might differ from our findings presented here. It is quite possible that an increase in the number of future studies meeting our inclusion criteria for a coordinate‐based meta‐analysis on the functional underpinnings of FTD might either strengthen or weaken our initial results. Indeed, some results of studies not meeting our inclusion criteria provide hints regarding the impact of FTD on study samples other than schizophrenia. For instance, in healthy individuals, FTD‐like symptoms induced by administration of ketamine were correlated with activation changes in the left STG, MTG, and IFG [Honey et al., 2008].

Methodological Considerations: Language and the Assessment of Formal Thought Disorder

During fMRI or PET scanning sessions, FTD is mainly assessed in terms of spoken language paradigms. This consequently increases the likelihood that our findings foremost represent brain regions that are functionally associated with the domains of language and speech. Hence, assessment of FTD by means of paradigms using language and speech resulting in brain regions associated with respective functional domains comprises the peril of circular argumentation. In other words, assessment of disordered thought is constraint to the level of observed speech and language, which, in turn, might obliquely point towards disordered thought processes. This point has already been made quite early in the course of FTD research [Rochester and Martin JR, 1979]. Yet, only a limited number of neuroimaging studies aimed to circumvent this issue by means of language‐independent experimental paradigms [Erkwoh et al., 2002; Tagamets et al., 2014]. Hence, brain regions associated with higher cognitive domains might ensue if more FTD neuroimaging studies focus on the assessment of cognitive and executive paradigms rather than aberrant language‐ and speech functions.

Furthermore, neuroimaging research on FTD (not only, but particularly) is challenged by a phenotypical heterogeneity of this psychopathological dimension on the one hand (e.g., positive and negative FTD) and vast variations in employed functional neuroimaging paradigms on the other. Since these variations might very likely be reflected in (partially) differing neurobiological correlates [Kircher et al., 2003, 2008], it seems worthwhile to address these questions with specifically designed research questions. However, the reasons to refrain from sub‐analyses in this regard were twofold: first, given the large variations in experimental designs and phenotype in view of an already rather small number of included studies, additional ALE meta‐analyses tackling these questions would be unfeasible for the sake of statistical reliability and robustness [Eickhoff et al., 2016]. Secondly, we deliberately pooled studies included in our analysis across different experimental paradigms and phenotypical markers. Indeed, one of the conceptual strengths of ALE lies in the possibility to synthesize results over studies of differing methodological approaches, potentially revealing a common neural correlate of FTD. This, in turn, does not mean that particular sub‐syndromes, such as FTD related to positive or negative symptoms, might not result in partially or generally differential neural patterns.

Distinct Neural Correlates of FTD and Aphasia

The two clusters that emerged from the conjunction of all three CCNs were located in the left IFG and left MTG and showed no overlap with the initial seeds, indicating that the affected regions resulting from ALE analysis (vp‐MTG, STG, and dp‐MTG) show no systematic modality‐independent functional connectivity among each other [Tomasi and Volkow, 2011]. Dysfunction of left IFG and MTG ensuing from conjunction analysis has been associated with speech‐ and language‐related deficits, reflected by expressive aphasia (“Broca's aphasia”) and receptive aphasia (“Wernicke aphasia”) [Bates et al., 2003; Rosen et al., 2000; Weiller et al., 1995]. Expressive aphasia is mainly caused by damage to Brodmann Areas (BA) 44 and 45 (which corresponds to our left IFG cluster resulting from the FC conjunction) and characterized by agrammatism, usually manifesting itself as telegraphic speech. However, in severe cases, expressive aphasia might subsequently result in total inability of speech production [Mohr et al., 1978]. Conversely, receptive aphasia results from damage to BA 39 and 40 (corresponding to left MTG), and patients with receptive aphasia often display difficulties in auditory speech comprehension up to complete loss of speech [Davis, 2007]. However, neither the former nor the latter are usually considered as a proper part of FTD symptomatology [Kircher et al., 2014; Liddle et al., 2002]. Our findings are well in line with these notions, indicating no immediate relationship between FTD and the complex of Broca's and Wernicke's aphasia.

Nonetheless, since the CCNs of all three seed regions were functionally connected to the left IFG and MTG, there might be a rather indirect association between FTD on the one hand and aphasia on the other. Direct comparisons between individuals with aphasia and FTD are indeed relatively sparse and timeworn. Early studies on schizophrenia patients indicate that aphasia and FTD might be differentiated based on the quality of verbal production. However, findings are contradictory and both clinical syndromes might show substantial overlapping language‐ and speech‐related deficits [Faber et al., 1983; Gerson et al., 1977]. Some studies thus concluded that FTD might foremost be considered a disorder of thought compared to aphasia as a disorder of language [Gerson et al., 1977]. However, this simplified view does not hold in light of apparent working memory deficits in aphasia [Caspari et al., 1998] and frequently observed semantic language dysfunction in FTD [Dwyer et al., 2014; Kiefer et al., 2009; Tan and Rossell, 2014]. Consequently, our findings of FTD as a disorder of mainly language‐related neural networks urge for a better clinical characterization of differential symptoms in patients with FTD and aphasia.

CONCLUSIONS

Synthesizing the results of previous functional neuroimaging studies, we were able to gain insight on the underlying neural correlates associated with FTD. There is still ongoing discussion on whether FTD might be considered a deficit of solely language and speech processing on the one hand or executive functioning on the other. Our findings highlight that FTD is foremost characterized by aberrant activation in temporal brain regions, associated with auditory perception, language processing and social cognition. These regions, in turn, are imbedded in cortical networks, which are functionally associated with language‐related processing, but also extend to higher order cognitive functions, which may lead to most symptoms observed in this pathological dimension. Furthermore, FTD might be differentiated from other language‐related pathological phenomena, such as aphasia, on a neurobiological basis. Despite heterogeneous operationalization across functional neuroimaging experiments investigating FDT, our findings point towards a common neural correlate of FTD.

Supporting information

Supporting Information