How functional network connectivity changes as a result of lesion and recovery: An investigation of the network phenotype of stroke

Yuan Tao; Brenda Rapp

doi:10.1016/j.cortex.2020.06.011

. Author manuscript; available in PMC: 2022 May 10.

Published in final edited form as: Cortex. 2020 Jul 15;131:17–41. doi: 10.1016/j.cortex.2020.06.011

How functional network connectivity changes as a result of lesion and recovery: An investigation of the network phenotype of stroke

Yuan Tao ^a,^*, Brenda Rapp ^a,^b,^c

PMCID: PMC9088558 NIHMSID: NIHMS1612427 PMID: 32781259

Abstract

This study, through a series of univariate and multivariate (classification) analyses, investigated fMRI task-based functional connectivity (FC) at pre- and post-treatment time-points in 18 individuals with chronic post-stroke dysgraphia. The investigation examined the effects of lesion and treatment-based recovery on functional organization, focusing on both inter-hemispheric (homotopic) and intra-hemispheric connectivity. The work confirmed, in the chronic stage, the “network phenotype of stroke injury” proposed by Siegel et al. (2016) consisting of abnormally low inter-hemispheric connectivity as well as abnormally high intra-hemispheric (ipsilesional) connectivity. In terms of recovery-based changes in FC, this study found overall hyper-normalization of these abnormal inter and intra-hemispheric connectivity patterns, suggestive of over-correction. Specifically, treatment-related homotopic FC increases were observed between left and right dorsal frontal-parietal regions. With regard to intra-hemispheric connections, recovery was dominated by increased ipsilateral connectivity between frontal and parietal regions along with decreased connectivity between the frontal regions and posterior parietal-occipital-temporal areas. Both inter and intra-hemispheric changes were associated with treatment-driven improvements in spelling performance. We suggest an interpretation according to which, with treatment, as posterior orthographic processing areas become more effective, executive control from frontal-parietal networks becomes less necessary.

Keywords: Stroke recovery, Neuroplasticity, Language deficit, Functional connectivity, Functional network

1. Introduction

The over-arching questions in the study of the neural bases of post-stroke¹ language impairment have long been: How does a lesion affect the functional brain organization involved in language processing? How does this organization change in response to therapy/recovery? What is the relationship between these neural characteristics and language behaviors? To date, most of the research addressing these questions has involved the analysis of local neural activity, based largely on mean blood-oxygen-level-dependent (BOLD) signals from task-based fMRI. This body of work has primarily considered how levels of neural activation in specific brain areas differ between individuals with aphasia and healthy controls and how these activation levels change with recovery or in response to treatment (e.g., Blasi et al., 2002; Cao et al., 1999; Fridriksson et al., 2010; Turkeltaub et al., 2011, 2009; Gold & Kertesz, 2000; Hartwigsen & Saur, 2017; Heiss et al., 1999; Jarso et al., 2013; Kuest & Karbe, 2002; Saur et al., 2006). More recently, the analytic toolbox has expanded to include the analysis of functional connectivity (FC). Rather than evaluating the activation levels of specific brain areas, FC analyses evaluate the strength of the functional relationships among brain areas, with connectivity typically operationalized as the degree of synchronicity (correlation) of neural activity between areas. Importantly, this approach can evaluate the degree to which brain areas are integrated with or segregated from one another, and how these integration/segregation relationships are affected by lesion and recovery.

Siegel et al. (2016), on the basis of their investigation of functional connectivity (FC) in a large cohort of individuals in the sub-acute post-stroke stage with a range of cognitive deficits (including language), proposed a “network phenotype of stroke injury”. The proposed phenotype consists of two key features: (1) abnormally low inter-hemispheric (homotopic) FC and (2) abnormally high intra-hemispheric (ipsilesional) FC. The investigation we report on here evaluates the proposed phenotype in the chronic stage of post-stroke language impairment specifically and, furthermore, considers the effects of treatment and recovery on the phenotype features. To do so, fMRI data were collected before and after a period of targeted therapy in individuals with chronic post-stroke dysgraphia, while the participants were performing a spelling task. We consider the impact of stroke-based lesions on FC based on comparisons with healthy controls at the pre-treatment time-point and we also evaluate the nature of recovery via comparisons of the lesioned group’s FC across the pre- and post-treatment time-points. To do so, the investigation deployed a multivariate pattern analysis (MVPA) approach that involved a set of classification-based analyses of FC. The work is specifically focused on understanding the impact of lesion and recovery on patterns of functional integration and segregation within the ipsi and contralesional hemispheres and across the homotopic regions of the two hemispheres.

1.1. The impact of a stroke on functional network organization

The impact of stroke on network organization is determined via comparisons between individuals with lesions and healthy controls (at a single time-point). However, it is worth noting that while a distinction can be drawn between the neural consequences of a lesion on the one hand and the effects of spontaneous recovery/treatment on the other, in reality they are difficult to disentangle. In the acute and even sub-acute stages the effects of lesion and recovery can be more easily differentiated as there has been little time for recovery and/or treatment. However, even at those stages, the picture can be complicated by the short-lived, reversible physiological effects related to things such as hematoma, etc. In the chronic stage, brain organization necessarily reflects a combination of the consequences of a lesion and the effects of recovery (spontaneous and/or therapy-based). These caveats should be taken into account when interpreting findings regarding the consequences of stroke.

In the specific context of stroke-induced aphasia, of particular interest is the Klingbeil et al. (2017) review of 14 published studies that evaluated resting-state fMRI connectivity (RS-FC) in post-stroke aphasia. Although the studies they reviewed varied considerably in terms of patient characteristics (e.g., time post-stroke, lesion site) and analytic methods [e.g., different regions-of-interest (ROIs) and neural measures], the authors concluded that the findings supported the Siegel et al. (2016) proposal of a “network phenotype of stroke injury”. The Siegel et al. (2016) investigation of RS-FC was based a large cohort of 132 individuals with deficits affecting a range of cognitive functions (visual memory, attention, verbal memory, etc.), though only 33 of the individuals were diagnosed with acquired language deficits. The group was studied at the sub-acute post-stroke stage (<2 weeks). As indicated earlier, using whole-brain connectivity and multivariate analyses they found that, in comparison with healthy controls, the stroke group exhibited lower homotopic and higher intra-hemispheric connectivity. With regard to intra-hemispheric connectivity, the pattern was complex as they observed higher connectivity in the ipsilesional hemisphere between certain sub-networks but not others. In terms of the relationship between behavior and these network characteristics, Siegel et al. (2016) found that overall, across all deficit types, better performance was associated with stronger homotopic connectivity and weaker intra-hemispheric connectivity. Specifically with regard to the subset of individuals with language impairment [as measured by Boston Diagnostic Aphasia Examination (BDAE)], the same pattern was observed for inter-hemispheric connectivity as was reported for the larger group, although with regard to intra-hemispheric connectivity the pattern was mixed in that certain sets of strong intra-hemispheric connections were associated with lower language scores, while others were associated with stronger language scores.

These two proposed stroke phenotype features are generally consistent with findings from language-focused studies (see Klingbeil et al. (2017) for discussion). Although inter-hemispheric FC in language deficits has been especially under-researched, several studies have reported lower than normal inter-hemispheric FC (Zhu et al., 2014; New et al., 2015; Sandberg, 2017; see also Warren et al., 2009; Marcotte et al., 2012; but see Yang et al., 2017 for different findings). Specifically, New et al. (2015) found that reduced FC between bilateral premotor cortex was correlated with apraxia severity, Zhu et al. (2014) found decreased FC, relative to controls, between the left and right hemisphere regions of the language network, and Sandberg (2017) found that, compared to healthy controls, individuals with aphasia exhibited widespread hypo-connectivity, including inter-hemispheric FC. With regard to intra-hemisphere FC, most language studies have only examined sub-sets of specific regions of interest (ROIs) and have provided a mixed set of findings regarding increased or decreased connectivity relative to controls (e.g., Nair et al., 2015; Sharp et al., 2010; van Hees et al., 2014; Zhu et al., 2014). However, the small literature evaluating FC with graph-theoretic methods is also relevant. Some of these studies examined modularity (Newman, 2006) which is a measure of the degree of system segregation quantified across the whole brain or hemisphere (Gratton et al., 2012; Siegel et al., 2018; Tao & Rapp, 2019). Brain modularity can be related to intra-hemispheric FC in that higher overall intra-hemispheric FC can give rise to lower modularity, due to reduced segregation of the hemisphere into modules. For example, Gratton et al. (2012) found that individuals with lesions had lower modularity than healthy controls in both the lesioned and non-lesioned hemispheres, and individuals with particularly low modularity seemed to exhibit more between-module FC within either hemisphere (Fig. 8c in Gratton et al., 2012). In our previous study (Tao & Rapp, 2019), we also found that lower modularity was driven by increased between-module connectivity (see also Siegel et al., 2018). In summary, with regard to intra-hemispheric connectivity in post-stroke aphasia, studies on the whole-brain scale are generally consistent with the proposed phenotype feature, although studies only examining specific language regions have yielded mixed results.

1.2. Functional connectivity and the recovery of language functions

In the literature, the term “recovery” has been used to refer to a wide range of situations, including simply the chronic stage. However, as we noted earlier, the chronic stage represents a mixture of the effects of the lesion and recovery (spontaneous and/or therapy-based), and data from a single time-point cannot disentangle them. Therefore, we will use the term “recovery” to refer only to studies that involve comparisons of different time-points as only these provide a strong basis for inferences about neural and cognitive changes. These across time-point comparisons can be carried out via longitudinal or cross-sectional approaches. These investigations can examine the effects of spontaneous recovery (natural history) or pre-to post-treatment time-points can be compared. The latter is the approach taken in this investigation as it allows for a more direct test of the effects of recovery.

In the context of recovery, a question that has been of paramount interest is whether neural changes associated with recovery result in the restoration of normal network organization and dynamics (normalization) or in re-organization that produces a network organization that is different than normal. These determinations, of course, require comparison of FC in lesioned cases with healthy (“normal”) participants. Further, the term recovery implies behavioral improvement. However, while normalization is presumably associated with improved language performance, re-organization can be either beneficial (supporting improved performance) and thus compensatory or, alternatively, re-organization can be maladaptive, creating barriers for improvement (e.g., Cao et al., 1999; Gainotti et al., 2015; Hartwigsen & Saur, 2017; Heiss et al., 1999; Kuest & Karbe, 2002; Turkeltaub et al., 2015, 2012). The question of normalization or re-organization (compensatory or maladaptive) has been a particular focus of interest in the long-standing attempts to understand the role of the contralesional right hemisphere in language recovery. Thus, the question of normalization and re-organization can be examined not only brain-wide, but also in each of the two hemispheres separately. Within the framework of the Siegel et al. (2016) stroke phenotype proposal, the default predictions are that normalization-based recovery associated with improvements in performance should result from increases in homotopic connectivity and/or decreases in intra-hemispheric connectivity. We consider the literature in light of these predictions.

To date, there have been only a small number of studies investigating the relationship between FC and recovery of language functions. These include four longitudinal FC studies identified in Kingbeil et al. (2017) (i.e., Nair et al., 2015; Sebastian et al., 2016; Zhu et al., 2014; Siegel et al., 2018) and several studies with pre-to post-treatment designs (Duncan & Small, 2016; Gili et al., 2017; Marangolo et al., 2016; Marcotte et al., 2012; Sandberg et al., 2015; Tao & Rapp, 2019; van Hees et al., 2014). Specifically, with regard to inter-hemispheric homotopic FC, no studies have systematically examined longitudinal or pre-to post-treatment changes but, nonetheless, some do provide some relevant findings. Longitudinal studies from the acute to the chronic stage have generally reported global increase of FC towards the normal level over time, including inter-hemispheric FC (Nair et al., 2015; Sebastian et al., 2016; Zhu et al., 2014). A treatment study at the chronic stage (Marcotte et al., 2012) reported increased FC within the posterior default-mode network (though still lower than the controls after treatment), which might also have included increased homotopic connectivity of the temporal and parietal lobes. However, given the sparsity of relevant findings, at this point we simply do not know the extent to which the homotopic hypo-connectivity that has been identified as a part of the stroke phenotype is “remedied” in the course of recovery of language functions.

In terms of intra-hemispheric recovery-related FC changes, most studies only examined selective regions of interest and, at odds with the predictions of the stroke phenotype that intra-hemispheric FC should decrease, have reported increases in intra-hemispheric connectivity associated with recovery in the left (typically ipsilesional) and/or right hemisphere, although the functional role of increased right hemisphere RH remains unclear (Gili et al., 2017; Marangolo et al., 2016; Nair et al., 2015; Sandberg et al., 2015; Sebastian et al., 2016; van Hees et al., 2014; Zhu et al., 2014). However, the abovementioned studies investigating modularity (Duncan & Small, 2016; Siegel et al., 2018; Tao & Rapp, 2019) have reported findings consistent with the phenotype prediction that recovery might be driven by increases in intra-hemispheric segregation, as both Duncan and Small (2016) and Tao and Rapp (2019) found that modularity increased from before to after treatment and that these modularity changes were associated with behavioral changes. Specifically, with regard to normalization, Siegel et al. (2018) found that modularity normalized (became comparable to that of healthy controls) from the sub-acute stage to 3 months post stroke. However, Tao and Rapp (2019) found that post-treatment levels of modularity became higher than normal following treatment, indicating that the increased level of segregation resulting from treatment was likely a form of compensatory re-organization. Duncan and Small did not include the comparison with normal controls that is needed to evaluate normalization/re-organization.

1.3. Functional networks and language processing

While functional connectivity (FC) can be evaluated broadly across the brain or within and between hemispheres or other regions, specific networks or sub-networks defined by their functional properties play an increasingly prominent role in the analysis and interpretation of functional connectivity (e.g., Power et al., 2011; Smith et al., 2009). Functional networks typically correspond to sets of non-adjacent nodes/regions whose co-varying activity is thought to reflect a common functional purpose. However, it is important to note there is no consensus on how functional networks are defined, how they should be analytically identified nor even what the basic nodes of functional networks should be. Accordingly, functional networks have been defined at various scales, with network nodes ranging from individual voxels, ROIs or even networks themselves. Commonly referenced functional networks derived from the clustering or parcellation of co-variation patterns of RS-fMRI activity include: the default mode network (DMN; Raichle et al., 2001), the dorsal attention network (DAN; Fox et al., 2006), the ventral attention network (VAN; Fox et al., 2006), cingulo-opercular network (CON; Dosenbach et al., 2007), the frontoparietal network (FPN; Cole et al., 2013), and the sensorimotor network (Biswal et al., 1995).

With regard to the FC of language processing, it may be useful to consider the “core and periphery” framework proposed by Fedorenko and Thompson-Schill (2014), according to which language processing is accomplished via the interaction of highly specialized functional language networks (the core) and domain-general functional networks (the periphery). The highly specialized core components are dedicated to specific language functions (e.g., grammatical processing, spelling, etc.), while the domain-general components are flexibly recruited for different language tasks depending on changing task requirements and, especially, in demanding processing contexts. With regard to the domain-general components, Cole et al. (2013) and others (e.g., Dosenbach et al., 2007; Bassett et al., 2011; Braun et al., 2015; Bertolero et al., 2015; Yeo et al., 2014; see also; Wig, 2017) have proposed that a bilateral frontoparietal network (FPN) of brain areas serves as a domain-general “cognitive control region” that is engaged in both practiced and novel tasks and is adaptively recruited according to task demands. Others, such as Duncan (2013) refer to multiple networks that form a part of a broader domain-general “multiple demand” (executive control) system, which includes the FPN, as well as the DAN and the cingulo-opercular network (CON) – all supporting the cognitive control, selection and attentional functions that can be deployed across multiple cognitive and language domains. Regions within the multiple demand system have been identified not only as being active when language processing demands are high (e.g., Fedorenko and Thompson-Schill, 2014) but also as playing a role in recovery of language function (Geranmayeh, et al., 2017; Zhu et al., 2014). Interestingly, it has also been proposed that regions within this network are recruited as compensatory scaffolding in the face of the neurofunctional decline associated with aging (Park & Reuter-Lorenz, 2009; Reuter-Lorenz & Park, 2014). In this investigation, we will report that functional connectivity changes within and between multiple demand network regions are associated with recovery of function.

1.4. The current study

This brief review indicates that, although the issue of inter-hemispheric functional connectivity (FC) has been very sparsely investigated in post-stroke aphasia, the existing evidence is generally consistent with Siegel et al.’s (2016) observation of post-stroke hypo-connectivity between homotopic regions (New et al., 2015; Sandberg, 2017; Zhu et al., 2014). With regard to intra-hemisphere connectivity, there is more evidence, but it is mixed: Studies have reported both hyper- and hypo-connectivity within the hemispheres, and thus no clear characterization of the left and right hemisphere post-stroke FC has emerged (Nair et al., 2015; Sandberg, 2017; Sharp et al., 2010; van Hees et al., 2014; Zhu et al., 2014). The study we report on here seeks to add to the literature on network disruption in post-stroke aphasia by examining inter- and intra-hemispheric connectivity by considering the chronic stage and also by evaluating the effects of recovery on patterns of inter and intra-hemispheric connectivity. It differs from the previous literature in three respects: the type of FC data used, the language deficit examined, and the analytic approach adopted.

While the vast majority of previous studies that have evaluated FC have used resting-state FC, the current study evaluated FC using task-related BOLD signal in individuals performing a spelling task (also see Marcotte et al., 2012; Sandberg et al., 2015; Warren et al., 2009 that examined task-based FC). We specifically analyzed what Norman-Haignere et al. (2011) and Al-Aidroos et al. (2012) have referred to as “background connectivity” and what others (Cole, et al., 2014, 2013; Gratton et al., 2016) have referred to simply as “functional connectivity”. Background connectivity corresponds to the FC calculated with the residual time-course of task-based fMRI (rather than the time-course of RS-fMRI) after removing task-related responses. Various studies have shown that background connectivity is very similar to the functional connectivity observed at rest, but that it includes additional and significant task-related components (Fair et al., 2007; Cole et al., 2014; Gratton et al., 2016, 2018). In this regard, Cole et al. (2014), compared connectivity matrices for: resting state fMRI, multi-task background connectivity (combining 64 tasks) and individual task background connectivity. They concluded that task-based background connectivity consists of three components (in order of magnitude) – a primary component corresponding to the intrinsic connectivity seen during rest, a task-general component corresponding to the connectivity patterns shared across multiple tasks and a task-specific component reflecting the connectivity specific to individual tasks. On this basis, we assume that the background connectivity measure may be more likely than RS-FC to reveal behavioral effects associated with specific behavioral treatments. However, to be clear, it is not a goal of the investigation we report on here to evaluate the Cole et al. (2014) characterization of background connectivity, here we simply note its potentially informative features.

Regarding deficit type, all previous FC studies (except Tao & Rapp, 2019) have considered deficits affecting some aspect of spoken language processing, although these studies have varied widely in terms of the deficit types investigated, including verbal fluency (Nair et al., 2015), language comprehension (Zhu et al., 2014), spoken naming (Duncan & Small, 2016; Sebastian et al., 2016; van Hees et al., 2014), repetition (Marangolo et al., 2016) and discourse (Gili et al., 2017). In this study, we examine individuals with post-stroke dysgraphia who received an average of 25.6 weeks of behavioral treatment. Dysgraphia has been shown to be a common and persistent consequence of stroke, impacting personal and professional success and it is reported to be among the most disruptive difficulties faced by stroke survivors two years post stroke (Hillis & Tippett, 2014). Across a range of studies and meta-analyses (Purcell et al., 2011; Planton et al., 2013) spelling has been shown to be highly left lateralized, recruiting areas including the IFG, posterior parietal areas and ventral occipito-temporal cortex.

In terms of analytic approaches, in the work we report on here, we deploy both univariate and multivariate (classification) approaches as Siegel et al. (2016) did. Given the large number of connections in whole-brain FC datasets, univariate approaches typically either consider only a few connections among selected ROIs, or consider averaged values calculated across a large number of connections (e.g., the average connectivity strength of all homotopic connections), and those values are compared between subject groups or time-points (e.g., whether average homotopic connectivity values differ between patient and control groups). Whereas averaging across connections is useful in reducing the dimensionality of the data, one drawback is that specific information can be lost. Multivariate approaches, on the other hand, simultaneously take into account multiple features (in the context of FC, these correspond to multiple individual connections) without averaging, and hence can be more sensitive than univariate approaches (e.g., Haxby, 2012; Norman et al., 2006). This may be especially useful in studying clinical populations, as their FC patterns can be very complex. The weakness of the multivariate approaches, however, is that it is can be difficult to succinctly characterize the observed FC patterns.

The current study, therefore, deployed both univariate and multivariate (classification) approaches to evaluate background functional connectivity data collected during task-based fMRI at pre- and post-treatment time-points in individuals with post-stroke dysgraphia. The study was designed to address the following questions: (1) In chronic stroke, is there evidence consistent with the Siegel et al. (2016) functional connectivity phenotype regarding patterns of homotopic and/or intra-hemispheric connectivity? (2) Are there recovery-related changes in FC that affect homotopic and/or intra-hemispheric connectivity? (3) If so, to what extent do the recovery-related changes in FC result in normalization versus re-organization of homotopic and/or intra-hemispheric connectivity? And, more specifically, are these changes similar or different for each hemisphere?

To address these questions, our general analytic approach was as follows. We first computed whole-brain FC for all lesioned participants at pre- and post-treatment time-points and for healthy controls at a single time-point. The healthy control data was used for the parcellation of brain regions (thru hierarchical clustering) into 8 networks (i.e., functional clusters) to be used in subsequent data analysis and for characterizing various findings. Following this, we carried out a multi-stage analysis approach to compare participant groups and time-points. First, to evaluate network phenotype features, we conducted a set of univariate analyses comparing the average values of various sets of connections between the lesioned participants (before treatment) and the healthy controls. Second, we carried out a set of multivariate classification analyses that considered individual connections to identify pre- to post-treatment changes as well as the direction of those changes (increases/decreases; normalization/re-organization). Finally, the results of the classification analyses were related to the behavioral changes that were due to treatment. This approach allowed us to further our understanding of the effects of lesion and recovery on the functional interactions within and between hemispheres.

2. Methods

We report how we determined our sample size, all data exclusions (if any), all data inclusion/exclusion criteria, whether inclusion/exclusion criteria were established prior to data analysis, all manipulations, and all measures in the study.

2.1. Participants

Eighteen individuals (5 females) with no history of reading/spelling disabilities who suffered acquired dysgraphia due to a single left hemisphere stroke (>1 yr post stroke, mean 56 months, SD = 33) were included in the study. The mean age of the participants was 60 (SD = 11) and years of education ranged from 12 to 19. Inclusion criteria established prior to data analysis were: having a single ipsilateral stroke at least 6 months ago, no contraindication for MRI, no other neurological diseases or history of developmental dyslexia/dysgraphia, suffering from dysgraphia resulting from the stroke, being able to complete the behavioral treatment. All but three participants self-reported being pre-morbidly right-handed (including one ambidextrous) as assessed by the Edinburgh Handedness Inventory (Oldfield, 1971). Fig. 2 depicts a distribution of the lesions, indicating greatest density of damage in the left insula, inferior frontal gyrus opercularis (opIFG), precentral gyrus, and the left superior longitudinal fasciculus. The participants completed an average of 25.6 biweekly dysgraphia treatment sessions and functional and structural MRI scanning before and after the treatment period. See Table 1 for further details.

Fig. 2 – — Lesion overlap of the 18 participants in the Lesion Group. The gray-scale indicates the number of participants with lesion. The greatest density of damage is in the left insula, inferior frontal gyrus opercularis (opIFG), precentral gyrus, and the left superior longitudinal fasciculus.

Table 1 –

Characteristics of participants in the Lesion Group. “Spelling severity” corresponds to the percentage accuracy on the spelling-to-dictation test PALPA 40 (Kay et al., 1992). “Improvement” corresponds to the pre- to post-treatment changes in Trained Items accuracy estimated with linear-mixed-effects models (See section 2.2 and Supplementary Materials 1).

Subject ID	Sex	Age	Education (years)	Handedness	Month post-stroke	Lesion vol. (mm³)	# Treatment sessions	Spelling severity (accuracy)	Improvement (beta-value)	Spelling task in the scanner
1 – ABS	M	58	18	R	97	198,480	19	70%	6.29	auditory
2 – AEF	F	55	16	R	101	280,768	24	70%	1.98	auditory
3 – DSK	M	67	16	R	59	207,472	11	47.5%	1.2	auditory
4 – DTE	F	80	18	R	14	111,968	48	27.5%	1.83	auditory
5 – ESG	M	62	16	L	38	155,232	27	10%	3.97	auditory
6 – FCE	M	64	12	R	119	68,456	19	27.5%	4.63	auditory
7 – JGL	F	72	16	R	32	74,840	48	42.5%	2.14	auditory
8 – KMN	M	55	15	R	28	96,192	48	22.5%	2.72	auditory
9 – KST	M	61	14	R + L	46	46,400	29	65%	2.47	auditory
10 – MSO	M	45	18	R	103	217,440	30	42.5%	2.94	auditory
11 – PQS	M	54	18	R	17	143,128	17	52.5%	4.15	auditory
12 – RFZ	M	60	18	R	46	98,984	16	82.5%	3.79	auditory
13 – RHH	M	45	16	R	82	145,368	18	92.5%	1.69	auditory
14 – RHN	F	75	19	L	27	17,712	16	62.5%	3.21	auditory
15 – TCK	M	69	16	R	68	41,096	26	45%	4.55	auditory
16 – CCN	M	40	18	R	38	116.136	23	20%	2.57	pictorial
17 – TTR	F	46	16	R	21	154,848	21	7.5%	2.38	pictorial
18 – THD	M	67	16	R	76	274,824	21	15%	5.08	pictorial

Open in a new tab

Two groups of age-matched, neurologically healthy control (HC) individuals with no history of reading/spelling disabilities participated in the MRI scanning. Control Group 1 consisted of 10 participants (8 females, mean age = 60.7, SD = 12, years of education: 12–18 years, all were right-handed) and Control group 2 consisted of 13 participants (10 females, mean age = 57, SD = 7.39, years of education: 12–18 years, three were left-handed). Control Group 1 was administered the same spelling task protocol during fMRI scanning as administered to all the participants in the Lesion Group except three who had auditory comprehension deficits and who were administered a picture version of the task; Control Group 2 performed the picture version of the spelling task which was administered to the three Lesion Group participants with auditory comprehension deficits. Both tasks are described below. All comparisons between the Lesion and the Control Group (described in Section 2.5.3 and 2.6) were conducted using both Control Groups and referred to as the HC Group (n = 23). See Supplementary Fig. 2 for results indicating no significant differences between the two control groups when analyzed separately. All participants provided informed consent following procedures approved by Johns Hopkins University Institutional Review Board.

2.2. Cognitive/Language assessments and dysgraphia treatment

In addition to a spelling evaluation (JHU Dysgraphia Battery, Goodman & Caramazza, 1985; PALPA 40, Kay et al., 1992), the participants were administered an extensive battery of cognitive and language tests before and after dysgraphia treatment. Test results are reported for the following cognitive and language domains (Section 3.1): oral reading of single words (PALPA 35; Kay et al., 1992), written single-word comprehension (PALPA 51; Kay et al., 1992); auditory single-word comprehension (Northwestern Naming Battery; Thompson et al., 2012); visual recognition memory (Doors and People Test; Baddeley et al., 1995); semantic comprehension (Pyramid and Palm Trees; Howard & Patterson, 1992); spoken picture naming (Northwestern Naming Battery; Thompson et al., 2012).

A set of 40 Training Words was developed for each individual in the Lesion Group for use in the dysgraphia treatment and the fMRI experiment. These words were selected such that, across two behavioral assessments, letter accuracy on each word was between 25% and 80%. In addition, for each participant, a set of 30 Known Words was identified on which they performed with 100% accuracy on two assessments. For the dysgraphia treatment, a spell-study-spell technique (Rapp & Kane, 2002) was used during approximately 60–80 min sessions, typically 2×/week. Treatment sessions continued until participants achieved 90% or greater letter accuracy on two consecutive sessions or if performance remained constant over 6 sessions. The number of sessions for each participant is reported in Table 1. For the purposes of subsequent analyses, the severity of the spelling impairment of each individual was quantified as performance accuracy on the PALPA 40 spelling sub-test (Kay et al., 1992. Table 1). The pre-post treatment improvement measure corresponds to the improvement in letter accuracy on each individual’s Training Words, quantified as a beta value derived from linear mixed-effects models (Table 1), details of which are reported in Supplementary Materials 1. Accuracy on Trained Words was also evaluated three months after the end of treatment.

2.3. Neuroimaging experiment

Individuals in the Lesion Group participated in two scanning sessions at each of the pre- and post-treatment time-points. The two within time-point scanning sessions occurred an average of 6.1 days apart. At each session, participants performed several tasks during fMRI scanning and were evaluated with multiple structural scanning protocols, including the T1-weighted imaging we report on here. With regard to functional imaging, this study focuses on a spelling task administered at each of the two scanning sessions, at each of the two time-points. Each session included two 7.7-min runs of the spelling task, generating a total 30.8 min of data for each participant at each time-point (Fig. S1).

The spelling task administered during scanning included a spelling probe task and a case-verification control task (Rapp and Lipka, 2010) with event-related designs. Each spelling probe trial consisted of the following sequence of events: (1) task prompt (“Is the letter in the word?”) presented both visually and auditorily for 1800 msec; (2) a 500 msec central fixation cross, (3) an auditory target word plus a variable period of silence depending on the word length, for a total duration of 1500 msec during which the fixation cross remained on the screen, (4) A response period consisting of a single visually presented probe letter presented for 1500 msec, followed by a 1700 msec fixation cross. During this time period, participants responded whether or not the letter was in the spelling of the target word with a button press using their left index or middle finger. There was a random inter-trial interval (2–7.5 sec) between trials. The case-verification trials were identical, except that the task prompt was “Is the letter uppercase?” and participants were instructed to ignore the auditory stimulus and judge the case of the letter with a button press to indicate if the letter was upper or lower case. As indicated above, three participants in the Lesion Group and all those in Control Group 2 were administered a modified spelling task that was the same as the spelling probe task described above, except that the auditorily presented target words were replaced with drawings.

Each 7.7-min scanning run consisted of 45 trials composed of the following stimuli: 15 Known Word spelling-probe trials, 15 Training Word spelling-probe trials, and 15 case-verification trials with words that were matched in length and frequency to those used in the spelling trials. In this way for each participant, each scanning session included presentation of a total of 30 Known Words and 30 Training Words which were randomly selected from the individual’s 40-word training list. The same stimuli were presented at all scanning sessions but their order varied within a run. In order to minimize task-switching costs, 3 to 6 trials of each task were presented consecutively in “mini-blocks”. Individuals in both Healthy Control Groups participated in a single scanning session that included four runs of the spelling task. For these participants, the target words ranged in frequency from .27 to 557.12 with a median = 8.3 (Balota et al., 2007) and were 4 or 7 letters in length.

2.4. Imaging data acquisition

All MRI data were collected using a Phillips 3T scanner at the F.M. Kirby Research Center for Functional Brain Imaging (Baltimore, MD, USA). The acquisition parameters for all participant groups were as follows: TR = 1500 msec, TE = 30 msec, FOV = 216*120*240 mm (ap, fh, rl), flip angle = 65°, voxel dimension = 1.875*1.875*3 mm (1.5 mm interslice gap), data matrix = 128*128*27. One run contained 308 TRs (7.7 min). The T1-weighted structural MRI acquisition parameters were as follows: TR = 6 msec, TE = 2.91 msec, FOV = 256*256*176 mm (ap, fh, rl), flip angle = 9°, voxel dimension = 1*1*1 mm, data matrix = 256*256*176.

2.5. fMRI data preprocessing, model-based analyses, and functional connectivity estimation

2.5.1. Pre-processing

Functional data were preprocessed with FEAT in FSL 5.0.9 (Jenkinson et al., 2012) with the following pre-processing steps: removal of the first 4 volumes of each run, motion correction with MCFLIRT, slice-timing correction, non-brain tissue removal with BET, spatial smoothing using a 5 mm FWHM Gaussian kernel, grand-mean intensity normalization, high-pass temporal filtering (.01 Hz). Registration was carried out with FSL’s two-step method: functional images were first registered to the subject’s T1 image with the boundary-based registration (BBR) method, then to standard MNI space (spatial resolution 2 mm, Jenkinson et al., 2002; Greve & Fischl, 2009). A lesion mask for each participant was manually drawn on the original T1 image with MRICron (http://people.cas.sc.edu/rorden/mricron/index.html). The masks were then aligned to standard MNI space using the transformation calculated with the T1 image.

2.5.2. GLM analysis

For each scanning run of each individual participant in the HC Group and the Lesion Group at the pre- and post-treatment time-points, univariate general linear models (GLM) were evaluated using FEAT in FSL (Jenkinson et al., 2012). The models, in addition to six motion parameters, included the following regressors of interest: task prompt, Known Word trials, Training Word trials, and Case-Verification trials (for the healthy controls, there were only Spelling and Case-Verification trials); these trial-type regressors modeled the 3 sec time period that included target word presentation, visual letter probe and response. The fixation periods and inter-trial intervals were left unmodelled. The stimulus regressors were convolved with a double-gamma hemodynamic response function provided in FSL. As described below, the individual GLM analysis formed the basis for evaluating group-level mean BOLD as well as calculating FC.

2.5.3. Evaluation of group-level mean BOLD activation

The group-level statistics of mean BOLD signal were calculated with the multi-level framework FLAME by FEAT in FSL5.0 (Woolrich et al., 2004). First, for each HC participant, the estimated coefficients of the contrast spelling > case-verification of the four scanning runs were averaged (with the fixed effects option in FEAT), and the results for each participant was entered into the higher-level mixed effects estimation (FLAME1) to calculate the group (HC Group) activation with one-sample t-test. For calculating the pre-post treatment difference in mean BOLD for the Lesion Group, for each participant (as for the HC group), the estimated coefficients of the contrast training-word-spelling > case-verification of the four scanning runs at either pre- and post-treatment were first averaged with the fixed effect option, then the results of each participant at each time-point were entered into the higher-level mixed effects estimation (FLAME1) and a paired t-test was conducted in order to compare the pre- and post-treatment time-points. For all analyses, the resulting statistical maps were thresholded using the cluster-based approach implemented in FEAT, with voxelwise threshold p < .05 (Z > 1.6) and clusterwise threshold p < .05.

2.5.4. Background functional connectivity estimation

We estimated functional connectivity using the residual time-series of the GLM (described in 2.5.2), a measure referred to as “background connectivity” in Norman-Haignere et al. (2011) and Ai-Aidroos et al. (2012) (also see Cole et al., 2014: See Introduction for further discussion). Estimation was performed using Nilearn, a Python package for neuroimaging data analysis (Abraham et al., 2014). All analyses used the 264 locations identified in Power et al. (2011), excluding 29 in subcortical areas as defined by the FSL segmentation template. For the remaining 235 locations, we extracted the averaged time-series from 5 mm-radius spheres (estimated to contain 125 2*2*2 mm voxels), referred to as “nodes”, centered at each location (Fig. 1). Prior to calculating pairwise-connectivity, we additionally regressed out the first derivatives of the 6 motion parameters and the averaged time-series of the brainstem, the cerebellum, the basal ganglia, and the ventricles defined by the segmentation template in FSL. For each of the four runs (collected both at pre- and post-treatment time-points), we calculated pairwise Pearson correlations which were Fisher’s z-transformed and converted to absolute values.² Each set of four correlation matrices was averaged to yield two 235-by-235 connectivity matrices for each participant, one for each pre- or post-treatment time-point (Supplementary Fig. 1). Each of the Control Group participants had a single average matrix (also combining 4 runs) corresponding to the single data acquisition time-point. Prior to the univariate and multivariate FC analyses (Section 2.6), the connectivity matrix of each participant (and time-point) was z-score normalized.

Fig. 1 – — The functional clusters identified with the healthy control connectome. The functional clusters were computed from background connectivity measured from a spelling task administered during fMRI. Hierarchical clustering was conducted with the mean pairwise correlation matrix, based on 235 nodes, of the Healthy Control Group 1 (n = 10). Nodes were grouped into 8 bilateral clusters (numbered 1–8, visualized on the right). Information regarding the 8 clusters is reported in Table S1.

For each individual participant in the Lesion Group, nodes with more than 25% (32) lesioned voxels (falling within the lesion mask) were considered to be damaged nodes, and all the connections of the damaged nodes were excluded from the univariate analysis, and for the multivariate classification analysis, zero values were assigned to those connections as the classification analysis requires the same set of features (i.e., connections) for all samples (i.e., participants), and in this way, no treatment-related changes could be attributed to damaged connections. The total percentage of damaged nodes per participant ranged from 2% to 22% (mean = 10%, SD = 6%), and there was a high correspondence between the percentage of damaged nodes and lesion volume (Pearson r = .91).

2.5.5. Identifying the functional clusters

Nodes were grouped into clusters using hierarchical clustering (agglomerative) with Ward’s criterion implemented in SciPy (https://scipy.org). The clustering was carried out with the averaged correlation matrix of Control Group 1. The resulting clusters of nodes are conceptually comparable to the “resting-state networks” identified with resting-state data (e.g., Power et al., 2011; Smith et al., 2009) in the sense that, in both cases, functional networks are identified as sets of nodes with highly correlated time-series. However, the term “network” has also been used to refer to the distribution of non-adjacent brain regions that support a specific cognitive function, such as “the spelling network”, “the face processing network”, and so on. Thus to avoid confusion, here we refer to the groups of nodes identified from hierarchical clustering (based on background connectivity from the spelling task) simply as functional clusters.

2.6. Functional connectivity analyses

2.6.1. Univariate functional connectivity analyses

In the analyses that we refer to as “univariate”, we compared participant groups and time-points in terms of average values for various sets of connections. First, to determine whether the Lesion Group exhibited the Siegel et al. (2016) stroke network phenotype, for each of three data sets (homotopic, left hemisphere and right hemisphere), we compared the average strength of all the connections of the Lesion Group (N = 18) at the pre-treatment time-point to the HC Group’s values using unpaired two-tailed t-tests. Also, to evaluate the relationship between connectivity and spelling behavior, the average connectivity strength values of the Lesion Group were correlated with their spelling severity (PALPA 40, Kay et al., 1992).

Second, we also compared the connectivity strength of the Lesion versus HC Groups by cluster (identified as indicated in 2.5.5). For homotopic connectivity, we grouped the connections into the 8 clusters and conducted the same between group comparisons, correcting for multiple comparisons with false discovery rate (FDR) correction (Benjamini-Hochberg procedure, alpha = .05, n of comparisons = 8). For the intra-hemispheric connectivity, we separately considered the within-cluster and between-cluster connections for each cluster within each hemisphere. Given 8 clusters, there were 8 sets of within-cluster connections and 28 sets (pairwise connectivities between the 8 clusters) of between-cluster connections, for each hemisphere. The results were FDR corrected for multiple comparisons (alpha = .05, n of comparisons = 8 and 28 respectively).

Third, to examine pre- to post-treatment changes in the Lesion Group, we compared pre- and post-treatment values with paired two-tailed t-tests for the same sets of connections described above. That is, we evaluated both the pre- to post-treatment changes in average strength of all the homotopic, within-LH, and within-RH connections, and also the changes in cluster-specific connections (8 within-cluster and 28 between-cluster per hemisphere).

2.6.2. Classification (multivariate) analyses: Evaluating changes in pre to post-treatment connectivity

First, we identified the changes in neural connectivity from pre- to post-treatment via multivariate classification analysis. Unlike in the univariate analysis, in the classification analyses, individual connection strengths were entered into a model, each as independent variables, without being averaged. Using this classification approach, we evaluated pre- to post-treatment changes for both (1) homotopic and, (2) intra-hemisphere connectivity. For inter-hemispheric connectivity, all homotopic connections were considered and for intra-hemispheric connectivity, the connections were analyzed by hemisphere and cluster (8-cluster * 2-hemisphere sets in total).

Classification analyses to identify connections that changed significantly from pre- to post-treatment were conducted using Linear Discriminant Analysis (LDA) implemented in Scikit-learn (Pedregosa et al., 2011). Each of the classification analyses followed the same procedure described here. We employed a between-subject classification approach with a leave-one-subject-out cross-validation (CV) scheme such that during each iteration, the classifier attempted to learn a boundary between the pre- and post-treatment connectivity patterns of 17 participants (i.e., the training dataset). This boundary was then applied to the held-out participant’s data (i.e., the test dataset). This procedure was repeated once for every participant (for a total of 18 iterations). For computational ease and noise reduction, as is standard in these approaches, we used univariate feature selection in conjunction with CV to reduce the feature space (i.e., number of connections). Specifically, in each CV iteration, the N most informative connections were identified using one-way-ANOVA based on the training data only: The N most informative connections were identified as the ones that differed the most across the classes (which were pre- and post-treatment here) as ranked by the f-statistic. As there is no standard rule for the value of N, we examined a range of values (N = 100, 200, 300, 500, 1000, 1500). Statistical significance of classification accuracies was evaluated using a one-tailed binomial test against the chance level .5.

As reported in the Results (Section 3.5.1 and 3.5.2), the classification accuracies were consistent across the range of N and, for reasons of efficiency, we used N = 200 for subsequent analyses (the significance of the classification results of N = 200 were further evaluated with Monte–Carlo tests with 2000 permutations). Given a statistically significant classification result, we identified the connections that were selected in all the 18 CV iterations across participants (using N = 200) and we refer to these are the “informative connections”. In this way, we identified the connections that most strongly contributed to significant classification of pre- versus post-treatment data and, thus, were those that most consistently differed between time-points across the participants.

Second, using the most informative connections identified for each significant classification, we examined the relationship between the pre- to post-treatment connectivity strengths of these connections and behavioral changes (i.e., treatment outcomes) using linear mixed-effects regression carried out with the LMER function in R (Bates et al., 2015). Specifically, the dependent variable corresponded to the difference between the pre- and post-treatment connectivity strengths of the informative connections of all the participants (e.g., if 10 most informative connections were identified, the model included 10 by 18-participant observations in total). Furthermore, those informative connections were separated according to whether they increased or decreased in strength from pre- to post-treatment and evaluated in two separate regression models. We did so because the directionality of connectivity changes might have a different role vis-à-vis behavioral change. For consistency, the dependent variables of both connectivity increase and decrease models always corresponded to the magnitude of the changes, (i.e., larger positive values indicated larger changes in connectivity, which were either increases or decreases), thus a positive correlation means larger neural changes (either increase or decrease) correlated with larger behavior changes in both models. All models had the same predictors: The key predictor variable was the pre-to post-treatment improvement of spelling accuracy for the Training Words (Table 1) and additional fixed-effect variables included: the pre-treatment connectivity strength for each connection, spelling severity, age, years of education, motion during fMRI scanning (RMS mean displacement), and lesion volume; Two random-effects terms were included: A random intercept for the subjects and a random intercept for the connections.

2.6.3. Evaluating normalization vs re-organization

For the most reliably informative connections identified in the classification analyses (i.e., connections that most consistently changed from pre to post-treatment, Section 2.6.2), one key question was whether they became more similar to the healthy controls (i.e., normalization) or more different (re-organization). To answer this question, for each set of the reliably informative connections, we compared the Lesion and HC groups at pre- and post-treatment time-points (Lesion Group mean minus HC Group mean), allowing us to evaluate whether and how the Lesion Group differed from the Healthy Controls at each time-point. Moreover, in order to evaluate the topological distribution of the differences, we grouped the informative connections based on their location (by cluster), and then, for each cluster, assessed whether the HC-Lesion differences were statistically higher or lower than zero using a two-tailed sign test [a non-parametric test for evaluating whether the median differs from zero (Conover, 1999)]. We used this non-parametric test because it is less affected by small sample sizes and violations of the normality assumption.

3. Results

3.1. Spelling and language/cognitive assessments

After dysgraphia treatment, the Lesion Group showed significant improvement for Training Words (β = 3.64, p < .001) and all 18 individuals also improved significantly. It is worth underscoring the robustness of the treatment as evidenced by the findings that Trained Words were still more accurately spelled at the three months follow-up than before treatment (p < .001), and that the treatment also resulted in generalization to untrained words that exhibited a significant pre to post-treatment improvement in accuracy (p < .001). More details regarding the treatment outcome assessment are described in Supplementary Materials 1 (also see Wiley & Rapp, 2019). In contrast, performance on other language and cognitive domains did not show significant changes, indicating that any neural changes we observed from pre- to post-treatment were likely to be specific to the recovery of spelling functions, rather than to generalized cognitive changes [Oral reading of single words: t(17) = 1.73, p = .10; Written single-word comprehension: t(17) = 1.47, p = .16; Auditory single-word comprehension: t(17) = .59, p = .56; Visual recognition memory t(17) = 1.78, p = .09; Pyramid and Palm Trees: t(17) = 1.38, p = .18; Spoken picture naming: t(17) = 1.88, p = .08].

3.2. In-scanner behavioral performance

Participants performed with comparable (and high) accuracy across pre- and post-treatment time-points for case-verification trials [mean accuracy: pre = .93, post = .95, t(17) = .59] and spelling probe trials with the Known Words [pre = .81, post = .84, t(17) = 1.06], indicating that, not only were participants actively engaged during the experiment, but neither the spelling treatment nor the repeated scanning produced generalized improvements in scanner performance. Furthermore, consistent with spelling performance outside the scanner, the Trained Words spelling condition showed lower in-scanner accuracy than the Known Words condition at both time-points, but there was a significant improvement from pre to post-treatment for the Training words only [pre = .68, post = .77, t(17) = 3.37, p = .0036] and the in-scanner improvement for the Trained Words was marginally larger than that for the Known Words: t(17) = 1.86, p = .08.

3.3. The functional clusters of spelling

Hierarchical clustering analysis of the connectivity matrices of Control Group 1 clustered the nodes into 8 bilateral, generally symmetrical clusters (i.e., networks) of 15–43 nodes each (Fig. 1). This clustering organization was validated with Control Group 2 (Supplementary Fig. 2). In neuroanatomical terms, the 8 clusters can be labeled as follows (see Table S1 in terms of the specific anatomical areas associated with each): (1) temporal, (2) ventromedial prefrontal, (3) dorsal medial occipital, (4) ventral occipitotemporal (VOT), (5) perisylvian, (6) posterior parietal, (7) dorsal frontoparietal, and (8) dorsal prefrontal. In terms of damage, the left hemisphere perisylvian cluster was the most affected, with mean percent damage (in terms of nodes) of 14.4%, in line with the lesion distribution typically seen in middle cerebral artery (MCA) infarction (see Table S1).

To situate this functional connectivity-based clustering structure within the context of the typical “spelling network” identified in previous activation-based studies carried out with healthy individuals (e.g., meta-analysis carried out by Purcell et al., 2011), we calculated the HC’s whole-brain activation using the GLM-based contrast spelling > case-verification (Section 2.3). As shown in Fig. 9a, spelling activated a left-lateralized network including the inferior frontal, posterior parietal and the ventral occipital temporal regions consistent with the typical activation pattern that has been previously reported for spelling (Planton et al., 2013; Purcell et al., 2011). The clusters derived from the hierarchical clustering analysis that overlapped the most with these activation regions were the: dorsal medial occipital, VOT, posterior parietal, and dorsal prefrontal clusters.

Fig. 9 – — Group-level mean BOLD activation. a) The activation (gray scale) of spelling (contrasted with letter case-verification) for healthy controls (HC) is overlaid on the clustering results reported in Fig. 1. Spelling activates a left-lateralized network including the inferior frontal, posterior parietal and the ventral occipital temporal regions. In terms of the clusters, the activation especially overlaps with the occipital (blue), VOT (azure), the posterior parietal (orange), and the dorsal prefrontal (magenta) clusters. b) Pre- to post-treatment changes in mean BOLD activation. Differences (post > pre) in mean BOLD activation (in gray) are overlaid on the FC clustering results reported in Fig. 1. After treatment, the bilateral precuneus and the right posterior parietal lobe show stronger mean BOLD than before treatment. This area overlaps primarily with the posterior parietal cluster (orange). The mean BOLD analyses were carried out with a standard voxel-wise GLM approach. Statistical maps were corrected by cluster-based corrections (voxelwise p < .01, clusterwise p<. 1).

3.4. Univariate functional connectivity analyses

3.4.1. Pre-treatment homotopic connectivity

The analysis revealed that the participants in the Lesion Group had lower overall homotopic connectivity values than did the healthy control (HC) Group [Lesion mean = 1.11, SD = .17; HC mean = 1.18, SD = .11, t(39) = 2.77, p = .0085, Fig. 3a]. Furthermore, there was a significant correlation between spelling severity measured at the pre-treatment time-point (PALPA40, Kay et al., 1992) and homotopic connectivity, such that participants with higher overall levels of homotopic connectivity showed better spelling performance (Pearson r = .47, p = .02 by 10 K permutation test. Fig. 4). These findings are consistent with the Siegel et al. (2016) findings that lesions significantly disrupt homotopic connectivity and that the extent of disruption may be associated with behavior.

Fig. 4 – — Relationship between homotopic connectivity and spelling performance. The correlation of the mean homotopic connectivity values of the Lesion Group and their pre-treatment spelling accuracy scores reveals that higher homotopic connectivity is associated with higher scores on a standardized spelling-to-dictation test PALPA40 (r = .47, p = .02 by 10 K permutation test).

When homotopic connectivity was examined by cluster (Fig. 3c, diagonal line of the top right/bottom left quadrant), all but one cluster (posterior parietal, #6) showed numerically lower connectivity in the Lesioned Group than in the HC Group. The largest and the only significant difference between the groups involved the homotopic connections of the perisylvian cluster [Lesion mean = .11, SD = .23; HC mean = .19, SD = .25, t(39) = 3.78, two-tailed p = .0005, FDR corrected].

3.4.2. Pre-treatment intra-hemispheric connectivity

The individuals in the Lesion Group showed overall higher intra-hemispheric connectivity than the HC in both the ipsilesional [LH: Lesion mean = .08, SD = .09; HC mean = .03, SD = .04, t(39) = −2.30, p = .0272] and contralesional hemispheres [RH: Lesion mean = .05, SD = .07; HC mean = .02, SD = .04, t(39) = −1.86, p = .0704, Fig. 3b]. This finding is also generally consistent with the Siegel et al. (2016) finding of increased intra-hemispheric connectivity between specific networks. However, we did not find significant correlations between spelling severity and overall intra-hemispheric connectivity for either hemisphere (LH: r = .21; RH: r = .06).

We also examined intra-hemispheric connectivity by cluster in each hemisphere (Fig. 3c, top left and bottom right quadrants). The intra-hemispheric connections can further be divided into within-cluster and between-cluster. In terms of within-cluster connectivity, the Lesion Group showed significantly higher average values than the HC in the left perisylvian cluster [Lesion mean = .72, SD = .49; HC mean = .32, SD = .27, t(39) = 3.23, p = .0025] and in both left and right hemisphere posterior parietal clusters [LH: Lesion mean = 1.73, SD = .54; HC mean = 1.08, SD = .42, t(39) = −4.22, p = .0001. RH: Lesion mean = 1.88, SD = .53; HC mean = 1.32, SD = .57, t(39) = −3.13, p = .0033]. p values were FDR corrected Fig. 3c, diagonal line. With regard to between-cluster intra-hemispheric connectivity, we did not find any significant group-differences after correcting for multiple comparisons. However, as can be seen in the LH–LH and RH–RH quadrants in Fig. 3c, the left-hemisphere of the Lesion Group exhibited more abnormalities than did the right-hemisphere, with both higher and lower levels of between-cluster connectivity compared to the controls. These results demonstrate that while, overall, the lesioned participants had higher than normal intra-hemisphere average connectivity in both hemispheres, the pattern was more complex when examined at the individual cluster level.

3.4.3. Pre- to post-treatment changes in homotopic and intra-hemisphere connectivity

For the homotopic and intra-hemisphere FC reported just above, we also carried out univariate evaluations of pre- to post-treatment connectivity changes for the Lesion Group. With regard to homotopic connectivity, the lesioned participants showed no significant overall pre to post-treatment difference [pre mean = 1.05, SD = .17, post mean = 1.05, SD = .16, t(17) = .03, p = .97] and their FC values remained lower than those of the HC after treatment [t(39) = −2.87, p = .0066]. Likewise, the overall intra-hemisphere connectivity for each hemisphere did not change from pre- to post-treatment [LH: pre mean = .08, SD = .09; post mean = .08, SD = .08, t(17) = −.06, p = .96. RH: pre mean = .05, SD = .07; post mean = .05, SD = .06, t(17) = −.05, p = .96] and remained higher for the Lesion Group than for the HC after treatment [Post-treatment compared with HC: LH: t(39) = −2.43, p = .0200, RH: t(39) = −1.97, p = .0557]. Finally, when examined by cluster, there were no statistically significant pre-post-treatment differences in terms of either homotopic or intra-hemisphere connectivity after FDR correction for multiple comparisons (Supplementary Fig. S3). This absence of statistical significance motivated the use of the more sensitive multi-variate analysis approach, the results of which are reported in the subsequent sections.

3.5. Classification (multivariate) analyses: Evaluating pre- to post-treatment connectivity changes

3.5.1. Changes in pre- to post-treatment homotopic connectivity

We used classification analysis to examine treatment-related FC changes given that, unlike in the univariate analysis, individual connection values are taken into account simultaneously by the classifier, potentially increasing the sensitivity of the approach. As indicated in the analysis Section (2.6.2), we carried out feature selection in conjunction with classification in order to reduce the feature space (i.e., the number of connections evaluated in subsequent analyses). A range of values for the number of selected features (N) was evaluated (from 100 to 1500). As reported in Fig. 5a, the classification accuracy for homotopic connections was generally significantly above chance across these different values of N. For instance, the mean classification accuracy was .61 (binomial p = .07) for N = 200 and remained stable as N increased. For efficiency, the subsequent analyses that examined the informative connections were based on N = 200.

Fig. 5 – — Pre- to post-treatment classification of homotopic connections. (a) Pre- to post-treatment classification accuracy for homotopic connections based on different numbers of selected connections (N = 100–1500, see text Section 2.6.2 for details on the methods for feature selection). The means and standard errors of the classification accuracies are calculated using a *leave-one-subject-out* cross-validation (CV) procedure. The red dashed lines indicate the binomial p-values of .1, .05, .01. (b) With N = 200, 62 connections are identified as most reliable on the basis of being selected across all the leave-one-out CV iterations, and are visualized in a glass brain. The results of paired t-test evaluations of the pre-post connectivity differences are indicated by the color scale [green: pre- to post-treatment increase (t > 0); pink: decrease (t < 0)]. The vast majority of the reliably informative homotopic connections increased in strength from pre- to post-treatment (52/62). (c) The relationship between homotopic pre-post FC changes and behavioral improvement was estimated by two linear-mixed-effects regression models. The dependent variables are the FC changes (y-axis) of the increased (green in b) and decreased connections (pink in b) respectively. The individual behavioral improvements in spelling accuracy for Trained Words (Beta values from LMEM described in Supplementary Material 1) are plotted on the x-axis. The magnitude of the changes for the connections that exhibited pre- to post-treatment FC increases are significantly (negatively) correlated with the participants’ behavioral improvements (left, p = 4.94e-06).No significant correlation with behavioral improvement was seen for the connections that exhibited decreases in connectivity strength (right).

Using the parameter N = 200, we identified 62 reliably informative homotopic connections that contributed to the successful pre- to post-treatment classification of homotopic connections across participants. As depicted in Fig. 5b, most of them (52 out of 62) increased in strength from pre- to post-treatment. We also examined the locations of those connections in terms of their cluster membership. While they were found in all 8 clusters, the vast majority of the significantly changing connections were found in the dorsal frontoparietal and the dorsal prefrontal clusters (numbers of connections identified for each cluster, from #1 to #8 respectively: 7, 3, 4, 5, 9, 1, 13, 20. Fig. 5b and Supplementary Fig. 4).

3.5.2. Changes in pre- to post-treatment intra-hemispheric connectivity

Classification analyses were carried out evaluating the intra-hemispheric connections for each cluster and in each hemisphere separately (2*8 = 16 sets of connections in total). Specifically, for any given cluster, the features used for classification included both connections within the cluster (within-cluster) and the connections between the cluster and every other cluster (between-cluster). As with the homotopic connections (section 3.5.1), a range of values for the number of selected features (N) was evaluated (from 100 to 1500) and the classification accuracy was generally consistent across the different values of N for all the clusters (Supplementary Fig. 5). Across different levels of N, three clusters consistently showed significant pre- to post-treatment classification accuracies: the left-hemisphere perisylvian cluster and the bilateral dorsal prefrontal clusters and therefore, as for the homotopic FC analyses, for efficiency, we based the subsequent analyses using N = 200. The pre- to post-treatment classification accuracies for intra-hemispheric connections, by cluster (with N = 200) are reported in Fig. 6. (#5 mean accuracy = .69, Monte–Carlo p = .0005; #8 LH: mean accuracy = .67, Monte–Carlo p = .03; RH: mean accuracy = .69, Monte–Carlo p = .03).

Fig. 6 – — Pre- to post-treatment classification of intra-hemispheric connections. Classification analysis of pre- and post-treatment FC values was carried out for intra-hemispheric connections of each cluster, according to hemisphere. The means and standard errors were calculated from the leave-one-subject-out cross-validation procedures. The black dashed line indicates the chance level (50%) and the red dashed line indicates the binomial p-value = .05. Indicated by an asterisk and circled below are the three clusters whose connections exhibited significant pre-post change as evaluated by permutation tests: LH perisylvian (#5 purple) and left and right dorsal prefrontal clusters (#8 magenta).

For each of these three sets of intra-hemispheric connections, the most reliably informative features were identified (left dorsal prefrontal n = 78; right dorsal prefrontal n = 78, left perisylvian: n = 65. Total = 221). Of the total number of 221 reliably informative connections, approximately half of them (120) showed decreased pre- to post-treatment connectivity strength. This reflects the rather complex distribution of pre- to post-treatment increases and decreases in connectivity strengths depicted in Fig. 7a. The left perisylvian cluster (#5) exhibited a mixed pattern with overall increased connectivity with the temporal and the dorsal frontoparietal regions and decreased connectivity with posterior parietal and occipital areas. On the other hand, the left and the right hemisphere dorsal prefrontal clusters (#8) presented very similar patterns such that both showed increased connectivity with their adjacent ipsilateral dorsal frontoparietal areas, and mainly decreased connectivity with all other areas. The exact distributions by cluster of those connections are reported in Supplementary Fig. 6. Interestingly, the distribution of the informative connections was overwhelmingly concentrated in between-cluster connections, with only a small number of informative within-cluster connections. This indicates that the changes identified with this classification analysis approach largely involved the interactions between clusters rather than within individual clusters.

Fig. 7 – — FC characteristics for the clusters that exhibited significant pre- to post-treatment changes in intra-hemispheric connectivity. a) The yellow ellipses indicate the three clusters depicted in Fig. 6, from top to bottom: left perisylvian, left dorsal prefrontal, and right dorsal prefrontal. Their respective intra-hemispheric connections are visualized in the glass brain, with their pre- to post-treatment connectivity differences (quantified by paired t-tests) indicated by the color scale (green: increase from pre- to post-treatment (t > 0); pink: decrease). b) The relationship between pre-post changes in FC and improvement in spelling accuracy examined by linear-mixed-effects models in which the dependent variables were the magnitudes of the pre-post FC changes evaluated by cluster. Separate models are calculated for connections that exhibited pre-post FC increase or decrease. The y-axes indicate the magnitudes of FC changes while the x-axes plot the behavioral improvements. The pre- to post-treatment FC changes for the perisylvian cluster were not significantly associated with behavioral improvement (#top row), whereas the FC changes in the left and the right dorsal prefrontal clusters (middle and bottom row) exhibited a similar relationship with behavior. Specifically, the magnitude of the FC increases was negatively correlated with behavioral improvements whereas the magnitude of the FC decreases exhibited a positive relationship with behavior (***p < .001, **p < .01, p < .05, ~ p < .1).

3.5.3. The relationship between treatment-related neural and behavioral changes

Regression analyses were conducted to examine the relationship between connectivity changes and treatment outcomes. First, for the homotopic connections, multiple regression analysis (Fig. 5c) found a significant negative correlation such that larger connectivity increases were associated with smaller improvements in spelling accuracy for Trained words (t = −3.11, p = .0019). In contrast, there was no significant relationship between the magnitude of the decreased connectivity changes and behavioral changes (t = −.51, p = .61). Second, for the intra-hemispheric connections, we constructed a total six models for the set of increased or decreased connection strengths for each of the three significant clusters (Fig. 7b). No significant effects were found for the perisylvian cluster. However, the right and left hemisphere dorsal pre-frontal clusters again showed similar patterns, such that for both clusters: larger connectivity decreases were associated with larger behavioral improvements (LH: t = 2.0, p = .046; RH: t = 3.40, p = .0007) while larger connectivity increases were associated with smaller behavioral improvements (LH: t = −1.78, p = .0756; RH: t = −5.59, p = 3.49e-08).

As reported just above, for both homotopic connectivity and for the right and left dorsal pre-frontal clusters (#8), we found a negative correlation between FC increases and behavioral improvement, such that larger connectivity increases were associated with smaller behavioral improvements. At least as concerns the homotopic connectivity, this seems counter-intuitive given that higher homotopic connectivity was associated with better performance at pre-treatment and was closer to the Healthy Control group pattern. Accordingly, we considered whether higher connectivity after treatment was actually maladaptive (e.g., those nodes might have become hyper-connected, hindering behavioral improvement). We reasoned that if higher connectivity strength after treatment was maladaptive, then higher post-treatment connectivity levels should be associated with lower post-treatment spelling accuracy. To test this, we evaluated similar linear mixed effects models as the ones reported above, but instead, the dependent variable corresponded to post-treatment connectivity and we added the post-treatment accuracy of the Trained Words as a key predictor. Results showed that, contrary to the maladaptivity hypothesis, for both the homotopic and the bilateral dorsal prefrontal connections, we found either no correlation or a positive correlation such that higher post-treatment connectivity was associated with higher post-treatment accuracy (homotopic: t = 2.38, p = .0176, intra-L#8: t = .84, p = .4, intra-R#8: t = 3.39, p = .0007). An alternative to the maladaptive hypothesis, is that the higher FC values are beneficial but that, in terms of the connectivity between the regions that experienced FC increases, some participants only needed to make small FC changes to achieve large behavioral benefits (see Purcell et al., 2019 for a similar findings and interpretation, albeit for a different neural measure). Presumably, those individuals with already healthy connectivity before treatment might reap large behavioral benefits from relatively small FC adjustments and, likewise, individuals with more disrupted connective before treatment might reap only modest behavioral benefits despite large treatment-driven connectivity changes. Consistent with this account, in the regression models (described in 2.6.2) in which the pre-treatment FC was also a predictor of FC changes, we found that participants with higher FC values prior to treatment had smaller pre-post treatment FC increases (homotopic:t = −9.85, p < 2e-16, intra-L#8: t = −7.0, p = 1.05e-11, intra-R#8: t = −9.78, p < 2e-16). Together, these results, represent a challenge to a maladaptivity explanation of the negative relationship between FC and behavioral changes.

3.5.4. Normalization/re-organization

We compared the connectivity strengths of the identified reliably informative connections at pre- and post-treatment time-point with those of the HC group to determine whether those changing connections became more comparable to controls (normalization) or more deviant (re-organization). Fig. 8 reports the differences between FC values of the Lesion and Healthy Control groups before treatment (left panel) and after treatment (right panel). Lesion versus HC differences are reported for homotopic connections and for the three clusters that showed significant pre to post treatment connectivity changes. The significant findings are as follows.

With regard to the homotopic connectivity, as can be seen in Fig. 8a, for the 62 most informative homotopic connections the results indicate that at pre-treatment the preponderance of connections were weaker than normal (below 0), while after treatment, the majority of those connections became similar to normal (around 0) or stronger. In particular, homotopic connectivity between left and right hemisphere prefrontal and frontal-parietal clusters became significantly stronger than normal after treatment (green: 13/13 above zero, p = .0002; magenta: 19/20 above zero, p = 4.0e-5). This indicates that the treatment not only “remedied” the overall pre-treatment homotopic hypo-connectivity that we have reported, but that it resulted in what might be considered to be “over-correction” with regard to the homotopic FC of the prefrontal and frontal-parietal regions.

With regard to the intra-hemispheric connections, the results (depicted in Fig. 8b–d) were as follows. First the perisylvian cluster did not show a clear pattern or normalization/re-organization (Fig. 8b): Although individual connections exhibited changes from pre- to post-treatment, we did not observe consistent changes with others areas. In contrast, the left and right dorsal prefrontal clusters showed very similar patterns of pre- to post-treatment connectivity changes (Fig. 8c and d). In both cases, there was the development of striking hyper-connectivity with their ipsilateral dorsal frontoparietal clusters at post-treatment (green. LH: 17/17 above 0, p = 1.5e-5; RH: 19/20 above 0, p = 4.0e-5). This contrasted with the changes with other areas that either moved from generally hyper-connected to either normal or significantly weaker than normal connectivity (hypo-connectivity) at post-treatment. This was especially evident for the posterior regions [dorsal medial occipital (blue); ventral occipitotemporal (azure); posterior parietal (orange)]. Specifically, the left dorsal prefrontal cluster (Fig. 8c) at pre-treatment was significantly hyper-connected with the ventral occipitotemporal cluster (#4, azure 15/17 connections > 0, p = .0023), while at post-treatment the connectivity decreased to below normal levels (4/17 connections > 0, p = .049), along with similar changes in connectivity with the adjacent posterior parietal cluster (#6, orange. 1/14 connections >0, p = .0018). The right dorsal prefrontal cluster also showed similar pattern though the effects were smaller (Fig. 8d). Taken together, the patterns indicate that, as a result of treatment, the left and right dorsal prefrontal areas increased their integration with ipsilateral dorsal frontoparietal regions and enhanced their segregation from posterior parietal-occipitotemporal areas (See Supplementary Fig. 6 for the pre-post FC changes for these clusters).

3.6. Pre- to post-treatment changes in mean BOLD activation

We also examined the pre- to post-treatment changes in mean BOLD activation with whole-brain GLM analysis. Specifically, the activation associated with the Trained Words was compared across the time-points. As shown in Fig. 9b the bilateral precuneus and right posterior parietal region (overlapping with parietal cluster, #6, orange) was found to show greater activation at post- than pre-treatment, while no areas were found to be significantly less active at post-treatment.

4. General discussion

Brain areas vary in the degree to which their activity is synchronized, and these patterns of functional connectivity (FC) result in an overall modular functional organization with regions varying in the extent to which they are integrated with or segregated from one another. There has been increasing interest in understanding how these functional connectivity patterns are affected by lesions (e.g., Gratton et al., 2012; Siegel et al., 2016, 2018) and how they are changed by treatment to support recovery of function (e.g., Duncan & Small, 2016; Gili et al., 2017; Marangolo et al., 2016; Marcotte et al., 2012; Sandberg et al., 2015; Tao & Rapp, 2019; van Hees et al., 2014). This study, through a series of univariate and multivariate (classification) analyses, investigated task-based FC in 18 individuals with chronic post-stroke dysgraphia who received an average of 25.6 biweekly sessions of behavioral treatment. Specifically, background FC (Al-Aidroos et al., 2012; Norman-Haignere et al., 2011) was obtained from these individuals based on a spelling task administered during scanning before and after the treatment, as well as from age-matched healthy controls performing the same task (HC, N = 23). The investigation was designed to understand the effects of lesion and treatment-based recovery on functional organization, focusing on both inter-hemispheric (homotopic) and intra-hemispheric connectivity [within the left (ipsilesional) and right hemispheres respectively]. The key findings are as follows: (1) Prior to treatment, inter-hemispheric (homotopic) connectivity was reduced in the Lesion Group compared to the HC Group (Fig. 3a) and the levels of inter-hemispheric connectivity were positively correlated with spelling performance (Fig. 4), (2) Prior to treatment, intra-hemispheric connectivity overall (and within each hemisphere individually) was higher for individuals with lesions compared to healthy controls (Fig. 3b), (3) Specific inter-hemispheric (homotopic) connections changed significantly from pre- to post-treatment, with most of the changes consisting of increases in FC between left and right dorsal frontal and parietal cortices (Fig. 5b, Supplementary Fig. 4), (4) Three regions underwent significant changes in their intra-hemispheric connectivity from pre- to post-treatment: left and right dorsal prefrontal cortex and left perisylvian cortex: In both left and right hemispheres, the dorsal prefrontal areas increased their FC with ipsilateral dorsal frontoparietal regions and decreased their FC with ipsilateral posterior parietal-occipito-temporal brain regions (Fig. 7, Supplementary Fig. 6), and (5) The neurotopography of the FC changes from pre- to post-treatment was quite distinct from that of the local mean BOLD activation changes (Fig. 9b). In the following sections, we discuss these findings and their implications for the questions that we set out to address with this investigation.

4.1. The network phenotype of stroke: Converging evidence

As indicated in the Introduction, a key question of this investigation was to determine if the Siegel et al. (2016) proposal of a “network phenotype of stroke injury”, which was based on evidence from a range of deficit types (including language deficits) in sub-acute stroke, was also present in chronic stroke. According to the stroke phenotype proposal, post-stroke functional connectivity (FC) is affected such that 1) inter-hemispheric FC is reduced and 2) intra-hemispheric FC is increased. To evaluate this hypothesis, we examined FC at the pre-treatment time-point, comparing Lesion and Healthy Control groups. Importantly, our investigation found evidence consistent with the Siegel et al. proposal in individuals with acquired dysgraphia who were an average of 56 months post-stroke.

With regard to inter-hemispheric FC at the pre-treatment time-point, we found that overall inter-hemispheric FC was significantly lower in the Lesion Group compared to the HC group (Fig. 3a), consistent with the phenotype. Specifically, most individual clusters showed numerically lower FC with their homotopic counterparts in the Lesion compared to the HC group, although differences were significant only for the perisylvian cluster (Fig. 3c). In terms of FC-behavior relationship, our findings were also consistent with the FC stroke phenotype in that levels of homotopic FC were significantly and positively correlated with spelling performance (Fig. 4) such that stronger and, hence, more normal FC levels were associated with more normal spelling accuracy.

With regard to intra-hemispheric FC, the picture was more complex. While, consistent with the phenotype, we found higher than normal FC in both hemispheres (Fig. 3b), the situation was more complex than the overall values indicated. Thus, analysis at the cluster level revealed a complex picture (see Fig. 3c) such that while some clusters exhibited significantly higher than normal within-cluster connectivity (left perisylvian and bilateral posterior parietal clusters) for the between-cluster FC values there were numerical differences in both directions. However, it was also the case that the Siegel et al. (2016) findings regarding intra-hemispheric FC were also quite mixed: They did not actually find that overall intra-hemispheric FC was higher for the Lesion compared to the HC Group, but only that it was higher than normal between a specific pair of networks – the ipsilesional DAN and DMN (dorsal attentional and default mode networks). The fact that the Siegel et al. study included individuals with a wide range of deficit types could have contributed to their complex set of ipsilesional intra-hemispheric FC findings, but even with our more homogeneous group, we also found that intra-hemispheric FC patterns could not be simply characterized as being overall higher or lower than normal.

With regard to the relationship between intra-hemispheric FC and behavior at the pre-treatment time-point, we did not find significant overall correlations with spelling performance, for either hemisphere. However, here again, Siegel et al. also reported a more complex picture than was found for inter-hemispheric connectivity, especially for their subset of participants with language deficits. They found that better language performance was associated with both higher and lower than normal levels of intra-hemispheric FC in both left and right hemisphere.

In sum, we found that the FC stroke phenotype was confirmed in chronic stroke in terms of the general characteristics of both inter- and intra-hemispheric connectivity. We found that, as Siegel et al. reported, the largest and most consistent findings concern the reduction in inter-hemispheric connectivity, resulting in generally lower integration across the hemispheres. However, intra-hemispheric connectivity changes due to stroke, could not be characterized in a straightforward fashion, neither in the original Siegel et al. (2016) report nor in our investigation, indicating that the effects of lesions on FC are likely to vary across the different functional clusters/networks within each hemisphere.

4.2. Treatment-based changes in the FC stroke phenotype

A second key objective of the study was to examine the pre- to post-treatment changes in both inter- and intra-hemispheric FC and their relationships with changes in spelling accuracy. Univariate analyses – based on average FC values per hemisphere/cluster – showed no significant pre- to post-treatment changes, for either inter- or intra-hemispheric FC (Section 3.4.3, Supplementary Fig. 3). Given these findings, we under-took multivariate classification analyses that were based on individual connection strengths (rather than the cluster-based averages as in the univariate analyses) that allowed for a more sensitive analysis.

With regard to inter-hemispheric (homotopic) FC changes, most of the connections that exhibited consistent pre- to post-treatment change in connectivity strength (as identified via classification analysis) exhibited increases in connectivity (52/62). While these increases in FC were distributed across all of the clusters, they were most pronounced in the connections between left and right dorsal prefrontal and dorsal frontoparietal areas (Fig. 5b and Supplementary Fig. 4). These inter-hemispheric FC changes were negatively correlated with behavioral changes such that smaller FC increases were associated with larger improvements in spelling accuracy (Fig. 5c). We discuss this negative FC-behavior relationship below.

With regard to intra-hemispheric FC changes, three clusters were identified as exhibiting statistically significant pre-to post-treatment FC changes: the left and right dorsal pre-frontal clusters and the left perisylvian cluster (Fig. 6). Across the three clusters, there were approximately equal numbers of connections that increased or decreased in connectivity strength from pre- to post-treatment, and these almost entirely involved connections with other clusters rather than connections within the clusters themselves (Fig. 7a and Supplementary Fig. 6). For the left perisylvian cluster, connections with increased and decreased FC were distributed across a number of left hemisphere clusters with no clear pattern. However, the two dorsal prefrontal clusters showed a very similar pattern of connectivity changes, such that connections with their nearby frontoparietal clusters increased in strength, while connections with distant posterior areas (ventral occipito-temporal and posterior parietal) decreased in strength (see Fig. 7a and Supplementary Fig. 6). In terms of the relationship between the FC and behavioral changes, the perisylvian pre-post FC changes were not systematically associated with changes in spelling accuracy. However, the left and right dorsal prefrontal clusters showed a very similar pattern: larger FC increases (primarily involving connections with the nearby frontoparietal clusters) were associated with smaller improvements in spelling accuracy, while larger FC decreases (primarily to posterior regions) were associated with larger improvements in spelling accuracy (Fig. 7b). We discuss these patterns in the next section.

In sum, for the first time, treatment-related inter-hemispheric connectivity changes in post-stroke language impairment have been documented and these primarily involved increasing connectivity between bilateral dorsal prefrontal and frontoparietal regions. In addition, significant treatment-related changes in intra-hemisphere connectivity were observed, such that frontal-parietal regions became more strongly connected with one another (within both hemispheres), while ipsilateral frontal and posterior regions became more disconnected with one another. These treatment-related FC changes were associated with changes in spelling performance.

4.3. Do treatment-related changes result in normalization or re-organization?

Given the pre- to post-treatment findings described above, one natural next question is whether or not these changes represent FC normalization or, instead, re-organization to new FC patterns. Addressing this question requires a comparison with healthy controls. Accordingly, Fig. 8 depicts the FC differences between the Lesion and HC groups at pre- and post-treatment, for the homotopic and the three intra-hemispheric clusters whose FC changed significantly from pre- to post-treatment (Figs. 5b and 7a, also see Supplementary Fig. 4 and 6). A comparison of the two sets of graphs helps in understanding the direction and extent of pre to post-treatment changes vis-à-vis the normal pattern.

With regard to inter-hemispheric (homotopic) FC (Fig. 8a), the results indicate that, prior to treatment, the majority of FC values were below normal levels (HC minus Lesion < 0). The pattern was markedly different after treatment, at which time most of connections had FC values that were higher than normal. This reflects a general trend towards normalization of homotopic, inter-hemispheric FC. However, the fact that the connectivity between homotopic frontal and ipsilateral frontal-parietal regions actually became significantly stronger than normal may represent an over-correction resulting in “hyper-normalization”.

There has long been great interest in understanding the role of the right hemisphere in recovery of language function (e.g., Hartwigsen & Saur, 2017; Saur et al., 2006; Turkeltaub et al., 2011) and this investigation may also be relevant to this issue. In this regard, the first thing to note is that at pre-treatment the right hemisphere had fewer anomalous clusters than the left hemisphere. This can be seen in Fig. 3c simply by comparing the color intensities of the RH–RH quadrant with those of the LH–LH quadrant. In addition, classification analysis identified only one right hemisphere cluster with significant pre-post treatment FC changes (the right dorsal prefrontal cluster) while two left hemisphere clusters (the left perisylvian and dorsal prefrontal cluster) were identified. Overall, this indicates that the ipsilesional left hemisphere experienced both greater disruption and treatment-related changes than the right. Interestingly, the fact that both left and right prefrontal clusters exhibited significant pre-post changes allows us to specifically consider how their recovery patterns were similar/different to one another. While there were some differences between them, the similarities in terms of pre-post changes are striking. Both left and right dorsal prefrontal regions, exhibited increased FC with their ipsilateral dorsal frontoparietal regions, mirroring what was observed inter-hemispherically (Fig. 7a). Furthermore, these ipsilateral changes resulted in higher than normal FC – another possible over-correction. Also, for both hemispheres these FC increases occurred along with decreased FC with almost all other regions, resulting in lower than normal connectivity with posterior parietal temporo-occipital regions (Fig. 8c and d). Thus, while for both hemispheres the treatment moved the intra-hemispheric FC pattern away from its pre-treatment hyper-connectivity, the pattern of changes that occurred was more complex and potentially more interesting than this simple characterization would indicate. In both hemispheres, treatment resulted in regions that became more strongly integrated with one another, while others became more strongly segregated. In sum, with regard to the role of the right hemisphere in recovery, the results indicate that although the right hemisphere underwent less overall change then did the left hemisphere, the two hemispheres exhibited some highly similar patterns of FC changes that supported recovery of function.

With regard to the relationship between FC changes and spelling performance, because all individuals significantly improved their spelling performance, one could say that performance “normalized”. However, as the results indicate, the specific relationships between FC and spelling accuracy changes also indicate a complex picture. First, we found that, in both hemispheres, the decreases in connectivity between the dorsal pre-frontal regions and their ipsilateral posterior regions were significantly and positively associated with gains in spelling accuracy, an indication that these changes were beneficial. Second, we found that increases in both ipsilateral and homotopic FC between dorsal prefrontal and frontoparietal regions were negatively related to changes in spelling accuracy (Figs., 5c and 7b), such that those individuals with the smallest FC increases improved the most. While this may suggest that these FC increases were maladaptive, as reported in Section 3.5.3, we did not find evidence consistent with a maladaptivity hypothesis. Specifically, we assumed that if higher post-treatment FC values were interfering with behavioral recovery, then higher post-treatment FC for these connections should be associated with lower post-treatment spelling accuracy. However, we did not find such a relationship, casting doubt on the maladaptivity hypothesis. An alternative to the maladaptivity hypothesis is that the higher FC values were indeed beneficial but that some individuals only needed to make small FC changes to achieve large behavioral benefits. This explanation was referred to by Purcell et al. (2019) as the “recovery efficiency hypothesis”. Consistent with this account, we found that participants with the smallest pre-post treatment FC increases on these connections had higher FC values on them prior to treatment. This plausible account underscores that the relationship between neural and behavioral change is complex. Thus, we suggest that, in the context of overall improvements in the behavior targeted by the treatment, negative correlations between pre to post-treatment neural changes and behavioral changes were likely to due to poorly understood non-linearities in the neural-behavioral relationships.

To summarize, we found that in response to treatment, on average both homotopic and intra-hemispheric FC changes moved away from abnormal pre-treatment FC patterns. However, treatment-related FC changes, both within and across hemispheres, often appeared to be over-corrections resulting in “hyper-normalization” that was manifested in higher or lower than normal connectivity after treatment. It would be interesting to know if, with time, these FC patterns would eventually move towards more normal values (as might be expected if the changes we observed represented temporary “overcorrections”) or if, instead, these values reflected an optimized re-organization of the networks.

4.4. Functional connectivity changes may support differentiation of executive and core processes

As reported just above, the treatment-related FC changes that we observed served to: a) integrate dorsal prefrontal and frontoparietal regions (shown by increased homotopic and intra-hemispheric FC across these regions), and b) segregate the dorsal frontoparietal regions from posterior areas (shown by the decreased FC between the frontoparietal and the posterior parietal and occipital regions). One way of understanding these changes is to consider them within the context of the “core and periphery” framework of language processing (e.g., Fedorenko & Thompson-Schill, 2014) reviewed in the Introduction.

Functional imaging meta-analyses of the brain areas normally recruited during spelling (e.g., Purcell et al., 2011) as well as analyses reported in this paper (see Fig. 9a) reveal a left-lateralized network that includes dorsal frontal, inferior frontal, posterior parietal and ventral occipito-temporal (VOT) areas. Among those regions, the VOT, inferior frontal, and the posterior parietal areas have been associated with specific functions relating to orthographic long-term and working memory by lesion-based and fMRI studies directed at understanding the neural bases of spelling (see Rapp & Dufor, 2011; Rapp et al., 2016; Rapp & Lipka, 2011). Generally speaking, within a core and periphery framework, the bilateral dorsal frontoparietal areas in which we observed significant FC changes would correspond to periphery functions while the posterior areas would correspond to core spelling areas. As we have described, the treatment-related changes involved both increased FC synchronization of bilateral and intra-hemispheric activity patterns within the frontoparietal network and also decreased FC synchronization with posterior areas. The decreased FC/synchronization and, hence, increased segregation, of posterior core areas from the frontoparietal control network may very well indicate that treatment served to reduce the need for cognitive control over core processes, reflecting the increasing independence and automaticity of the core processes. This interpretation is consistent with findings from other analyses carried out with this dataset. Tao and Rapp (2019) reported that modularity (Newman’s Q, a graph-theoretic measure that quantifies system segregation. Newman, 2006) increased from pre- to post-treatment, indicating increased segregation of clusters within the hemisphere. Moreover, Tao and Rapp (2019) specifically identified the posterior ventral occipital-temporal (VOT) cluster as undergoing the most within-cluster connectivity changes that enhanced the internal connectivity (segregation) of the region. This is another sign of the increased autonomy of the region that would be consistent with increased functional automaticity. Additionally, Purcell et al. (2019) reported that significant treatment-related local differentiation of cross-voxel patterns occurred within this core, left ventral occipital-temporal region. These neural changes reflect a return to the more normal differentiation of neural representations that supports processing of well-learned orthographic representations (Purcell & Rapp, 2018).

All together these findings suggest that the FC changes identified in this investigation reflect treatment-related changes that serve to strengthen core regions of the spelling network, allowing them to begin to reduce their reliance on the frontoparietal control system that can be flexibly deployed depending on task needs and whose interaction with core processes was reduced as these became more autonomous. Certainly, more research needs to be done to increase our confidence in this account, but it is consistent with the facts to date and serves as a useful framework for future research on recovery of function.

4.5. Different measures: background connectivity, mean local BOLD and RS-fMRI

The focus of this paper is on the impact of lesion and recovery on functional connectivity (FC). However, one important question concerns how FC changes relate to local BOLD activation changes. Fig. 9b reports a very different neurotopography of treatment-related change for local BOLD activation levels compared to what was observed for FC. This figure depicts the areas of significant pre- to post-treatment BOLD increase, which very clearly correspond to the bilateral precuneus and right posterior parietal regions that overlaps almost entirely with the FC-defined cluster #6 and which includes: superior parietal lobule, posterior supramarginal gyrus, angular gyrus, posterior cingulate cortex, superior lateral occipital cortex and precuneus. As can be seen in Fig. 9, this region of local BOLD activation change overlaps with a similar region in the spelling network as identified in HC [and consistent with the neuroimaging meta-analyses of spelling referred to earlier (Purcell et al., 2011)]. However, the FC analyses indicated that the connectivity of this region with others was not significantly changed by treatment (see Fig. 6) and, therefore, the pre-post treatment changes in local BOLD activation presumably reflect cluster-internal changes. Furthermore, this region corresponds to one of the posterior regions that becomes segregated from the frontoparietal network as a result of treatment. Although it is beyond the scope of this paper to go into more detail on the various changes that this region underwent, at this point it is sufficient to highlight that while analysis of local mean BOLD activation levels un-doubtedly provides information relevant to understanding recovery-related changes, mean BOLD changes seem to tap into different processes than those revealed by the FC analyses.

The vast majority of FC studies have analyzed resting-state data while task-based fMRI has been used either for analyses of local BOLD activation or for connectivity analyses focused on pre-defined ROIs and specific networks. In this study we used task-based fMRI to analyze “background connectivity” (the residual of the task-based GLM; see Norman-Haignere et al., 2011; Al-Aidroos et al., 2012, also more recently Cordova et al., 2016 and Tompary et al., 2018) to evaluate brain-wide FC. As we have indicated earlier, studies examining the relationship between RS-FC and task-based FC have concluded that background connectivity has many of the properties observed in RS-FC but also includes task-specific components (Fair et al., 2007; Cole et al., 2014; Gratton et al., 2016 see also, Norman-Haignere et al., 2011; Al-Aidroos et al., 2012). In the context of our study, the finding of treatment-based changes in background connectivity indicate that behavioral improvement is supported by interactions among distributed brain regions, independent of local mean BOLD activation. It remains an important question for future research to understand the extent to which background connectivity and RS-FC reveal similar or different aspects of FC and how they are both affected by lesion and recovery.

4.6. Limitations

There are several limitations of this investigation. First, although all the participants had a single left-hemisphere stroke, there was still considerable heterogeneity in their lesion distribution. Lesion heterogeneity poses a challenge to studying neural mechanisms of recovery of function. For example, perilesional tissue is often argued to play an important role in recovery but this can be difficult to assess in a set of heterogeneous lesions. Second, there is also the issue of deficit heterogeneity. Although the fact that all participants suffered from chronic acquired dysgraphia provides a homogeneity of deficit type absent in many studies, it is still the case that the participants suffered from different sub-types of dysgraphia and various degrees of other language and cognitive deficits. Third, although multivariate pattern classification has been shown to be a powerful method for analyzing neuroimaging data, one limitation of this approach is that it is difficult to pinpoint the exact source of the information that gives rise to the successful classification, as the classification decisions are made on the basis of the whole set of features collectively (see Pereira et al. (2009) for an overview). One approach that has been developed to partially address this issue is the “searchlight” approach (Kriegeskorte, Goebel, & Bandettini, 2006), such that the same multivariate analysis (e.g., classification) is repeatedly carried out with small sets of spatially adjacent voxels (i.e., searchlights). By this means, one can more precisely localize the source of the relevant information. However, the searchlight approach cannot be straightforwardly translated to FC analysis as it is unclear how to define “spatially adjacent connections”. In the current study, we grouped the connections according to clusters in order to achieve greater specificity. The drawback of this approach is that the analysis is heavily dependent on the cluster definition that was derived from the hierarchical clustering analysis. Although we explored different clustering solutions within this approach and they produced similar results, there are other approaches to identifying these basic units of analysis and these may yield different findings. Fourth, the pre- to post-treatment changes were only examined in participants who received treatment, and no untreated comparison group was included in this study. Future research could benefit from including a no-treatment group or longitudinal data from healthy controls in order to rule out placebo or repeated testing effects.

Finally, while this is not necessarily a limitation, it is worth noting that we chose to characterize FC in terms of the absolute value of the connections. This was motivated primarily by the fact that there is no clear consensus on how to deal with negative correlations in large-scale FC analysis. Thus, while most studies implicitly or explicitly exclude negative correlations (e.g., in the course of thresholding steps), if we assume that these connections carry valuable information then this would not seem to be the best course of action. Given the fact that averaging positive and negative correlations can lead to misleading conclusions of absence of relationships as average values approach zero, the use of absolute values allows for the preservation of information carried by negative correlations. However, we believe that this complex issue merits further and more detailed attention in future work that examines the consequences of the various approaches to the analysis of negative FC correlations.

5. Conclusions

This study confirmed that the general post-stroke network phenotype observed by Siegel et al. (2016) in the sub-acute stage was also present in chronic post-stroke dysgraphia. This study went beyond this previous work in examining treatment-related changes with respect to the phenotype. In this regard, we found that while, overall, treatment-related FC changes moved to “correct” the patterns of hyper- and hypo-connectivity of the stroke network phenotype, in many areas these corrective changes resulted in a complex pattern of ‘hyper-normalization”, possible due to over-correction. These treatment-related FC changes primarily involved increases in the integration of dorsal frontoparietal areas both within and across the hemispheres along with a decrease in the integration of these frontoparietal areas with ipsilateral posterior regions. We suggested an interpretation according to which as core orthographic processes become more effective as a result of treatment, they also become more automatic and independent, requiring less executive control from the frontoparietal network. In future work, it will be important to determine whether or not and, if so, how this overall pattern is affected by specific lesion locations and types of language impairment.

Supplementary Material

mmc2

NIHMS1612427-supplement-mmc2.docx^{(20.8KB, docx)}

mmc1

NIHMS1612427-supplement-mmc1.docx^{(924.1KB, docx)}

Acknowledgments

We are grateful to NIH support for its support for this research (DC012283) and we thank Jennifer Shea, Jeremy Purcell, Donna Gotsch and Robert Wiley for their very many valuable contributions to data collection and analysis.

Funding

This work was supported by the National Institute on Deafness and Other Communication Disorders (grant number DC012283) to BR that is part of a P50 award supporting a multi-site project examining the neurobiology of language recovery in aphasia.

Footnotes

Credit author statement

Yuan Tao: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Validation, Visualization, Writing - original draft, Writing - review & editing.

Brenda Rapp: Conceptualization, Methodology, Resources, Supervision, Validation, Writing - review & editing.

Open practices

The study in this article earned an Open Materials a badge for transparent practices. Materials and data for the study are available at https://osf.io/bqt59/?view_only=b0f0cd2bb2ab448e836e14e1c3f5e066.

Declaration of Competing Interest

The authors claim no conflict of interest.

Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.cortex.2020.06.011.

We specifically refer to stroke because this is the most commonly studied type of focal lesion, but focal lesions with other etiologies (e.g., surgery) can usually be grouped with stroke.

There is no consensus in the FC literature regarding the analysis of negative correlation values. Negative FC values are often explicitly or implicitly discarded. Importantly, negative correlations are unlike zero correlations in that they represent a relationship between brain areas. However, if both positive and negative values are included, then, analyses that involve averaging combinations of positive and negative correlations may incorrectly suggest an overall absence of correlation as the average values approach zero. Therefore, the use of absolute values preserves the relationship information represented by the negative correlations.

Data availability

The analysis scripts and data to reproduce the results have been made available for review on the Open Science Framework website (https://osf.io/bqt59). No part of the study procedures or analyses were preregistered prior to the research being undertaken. The raw imaging data have not been made publicly available to comply with the privacy policy of Johns Hopkins Institutional Review Boards. Researchers who are interested in accessing the raw data should contact the corresponding author in order to complete the proper procedure for data transfer. Data sharing agreement: 1) a commitment to using the data for research purposes only and not to identify any individual participant; 2) a commitment to securing and properly handling the data without unauthorized re-sharing; 3) Any reuse of the data should make proper attribution to the source.

REFERENCES

Abraham A, Pedregosa F, Eickenberg M, Gervais P, … Varoquaux G (2014). Machine learning for neuroimaging with scikit-learn. Frontiers in Neuroinformatics, 8, 14. [DOI] [PMC free article] [PubMed] [Google Scholar]
Al-Aidroos N, Said CP, & Turk-Browne NB (2012). Top-down attention switches coupling between low-level and high-level areas of human visual cortex. Proceedings of the National Academy of Sciences of the United States of America, 10936, 14675–14680. [DOI] [PMC free article] [PubMed] [Google Scholar]
Baddeley AD, Wilson BA, & Kopelman MD (1995). Handbook of Memory Disorders. London UK: John Wiley and Sons Ltd. [Google Scholar]
Balota DA, Yap MJ, Hutchison KA, Cortese MJ, Kessler B, Loftis B, & Treiman R (2007). The English lexicon project. Behavior Research Methods, 39(3), 445–459. [DOI] [PubMed] [Google Scholar]
Bassett DS, Wymbs NF, Porter MA, Mucha PJ, Carlson JM, & Grafton ST (2011). Dynamic reconfiguration of human brain networks during learning. Proceedings of the National Academy of Sciences of the United States of America, 10818, 7641–7646. [DOI] [PMC free article] [PubMed] [Google Scholar]
Bates D, Maechler M, Bolker B, Walker S, Christensen RHB, Singmann H, … Rcpp L (2015). Package ‘lme4’. Convergence, 12(1). [Google Scholar]
Bertolero MA, Yeo BT, & D’Esposito M (2015). The modular and integrative functional architecture of the human brain. Proceedings of the National Academy of Sciences of the United States of America, 11249, E6798–E6807. [DOI] [PMC free article] [PubMed] [Google Scholar]
Biswal B, Zerrin Yetkin F, Haughton VM, & Hyde JS (1995). Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magnetic Resonance in medicine, 34(4), 537–541. [DOI] [PubMed] [Google Scholar]
Blasi V, Young AC, Tansy AP, Petersen SE, Snyder AZ, & Corbetta M (2002). Word retrieval learning modulates right frontal cortex in patients with left frontal damage. Neuron, 361, 159–170. [DOI] [PubMed] [Google Scholar]
Braun U, Schäfer A, Walter H, Erk S, Romanczuk-Seiferth N, Haddad L, … Meyer-Lindenberg A (2015). Dynamic reconfiguration of frontal brain networks during executive cognition in humans. Proceedings of the National Academy of Sciences of the United States of America, 11237, 11678–11683. [DOI] [PMC free article] [PubMed] [Google Scholar]
Cao Y, Vikingstad EM, George KP, Johnson AF, & Welch KM (1999). Cortical language activation in stroke patients recovering from aphasia with functional MRI. Stroke, 3011, 2331–2340. [DOI] [PubMed] [Google Scholar]
Cole MW, Bassett DS, Power JD, Braver TS, & Petersen SE (2014). Intrinsic and task-evoked network architectures of the human brain. Neuron, 831, 238–251. [DOI] [PMC free article] [PubMed] [Google Scholar]
Cole MW, Reynolds JR, Power JD, Repovs G, Anticevic A, & Braver TS (2013). Multi-task connectivity reveals flexible hubs for adaptive task control. Nature Neuroscience, 169, 1348. [DOI] [PMC free article] [PubMed] [Google Scholar]
Conover WJ (1999). Practical nonparametric statistics (3rd ed.). John Wiley & Sons. [Google Scholar]
Córdova NI, Tompary A, & Turk-Browne NB (2016). Attentional modulation of background connectivity between ventral visual cortex and the medial temporal lobe. Neurobiology of Learning and Memory, 134, 115–122. [DOI] [PMC free article] [PubMed] [Google Scholar]
Dosenbach NU, Fair DA, Miezin FM, Cohen AL, Wenger KK, Dosenbach RA, … Schlaggar BL (2007). Distinct brain networks for adaptive and stable task control in humans. Proceedings of the National Academy of Sciences of the United States of America, 10426, 11073–11078. [DOI] [PMC free article] [PubMed] [Google Scholar]
Duncan J (2013). The structure of cognition: Attentional episodes in mind and brain. Neuron, 80(1), 35–50. [DOI] [PMC free article] [PubMed] [Google Scholar]
Duncan ES, & Small SL (2016). Increased modularity of resting state networks supports improved narrative production in aphasia recovery. Brain Connectivity, 67, 524–529. [DOI] [PMC free article] [PubMed] [Google Scholar]
Fair DA, Schlaggar BL, Cohen AL, Miezin FM, Dosenbach NU, Wenger KK, … Petersen SE (2007). A method for using blocked and event-related fMRI data to study “resting state” functional connectivity. Neuroimage, 351, 396–405. [DOI] [PMC free article] [PubMed] [Google Scholar]
Fedorenko E, & Thompson-Schill SL (2014). Reworking the language network. Trends in Cognitive Sciences, 18(3), 120–126. [DOI] [PMC free article] [PubMed] [Google Scholar]
Fox MD, Corbetta M, Snyder AZ, Vincent JL, & Raichle ME (2006). Spontaneous neuronal activity distinguishes human dorsal and ventral attention systems. Proceedings of the National Academy of Sciences of the United States of America, 103(26), 10046–10051. [DOI] [PMC free article] [PubMed] [Google Scholar]
Fridriksson J (2010). Preservation and modulation of specific left hemisphere regions is vital for treated recovery from anomia in stroke. The Journal of Neuroscience: the Official Journal of the Society for Neuroscience, 3035, 11558–11564. [DOI] [PMC free article] [PubMed] [Google Scholar]
Fridriksson J, Bonilha L, Baker JM, Moser D, & Rorden C (2009). Activity in preserved left hemisphere regions predicts anomia severity in aphasia. Cerebral Cortex, 20(5), 1013–1019. [DOI] [PMC free article] [PubMed] [Google Scholar]
Gainotti G (2015). Contrasting opinions on the role of the right hemisphere in the recovery of language. A critical survey. Aphasiology, 29(9), 1020–1037. [Google Scholar]
Geranmayeh F, Chau TW, Wise RJ, Leech R, & Hampshire A (2017). Domain-general subregions of the medial prefrontal cortex contribute to recovery of language after stroke. Brain, 140(7), 1947–1958. [DOI] [PMC free article] [PubMed] [Google Scholar]
Gili T, Fiori V, De Pasquale G, Sabatini U, Caltagirone C, & Marangolo P (2017). Right sensory-motor functional networks subserve action observation therapy in aphasia. Brain Imaging and Behavior, 11(5), 1397–1411. [DOI] [PubMed] [Google Scholar]
Gold BT, & Kertesz A (2000). Right hemisphere semantic processing of visual words in an aphasic patient: An fMRI study. Brain and Language, 733, 456–465. [DOI] [PubMed] [Google Scholar]
Goodman RA, & Caramazza A (1985). The Johns Hopkins university dysgraphia battery. Baltimore, MD: Johns Hopkins University. [Google Scholar]
Gratton C, Laumann TO, Gordon EM, Adeyemo B, & Petersen SE (2016). Evidence for two independent factors that modify brain networks to meet task goals. Cell Reports, 17(5), 1276–1288. [DOI] [PMC free article] [PubMed] [Google Scholar]
Gratton C, Laumann TO, Nielsen AN, Greene DJ, Gordon EM, Gilmore AW, … Dosenbach NU (2018). Functional brain networks are dominated by stable group and individual factors, not cognitive or daily variation. Neuron, 98(2), 439–452. [DOI] [PMC free article] [PubMed] [Google Scholar]
Gratton C, Nomura EM, Pérez F, & D’Esposito M (2012). Focal brain lesions to critical locations cause widespread disruption of the modular organization of the brain. Journal of Cognitive Neuroscience, 246, 1275–1285. [DOI] [PMC free article] [PubMed] [Google Scholar]
Greve DN, & Fischl B (2009). Accurate and robust brain image alignment using boundary-based registration. Neuroimage, 481, 63–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
Hartwigsen G, & Saur D (2017). Neuroimaging of stroke recovery from aphasia–Insights into plasticity of the human language network. Neuroimage, 190, 14–31. [DOI] [PubMed] [Google Scholar]
Haxby JV (2012). Multivariate pattern analysis of fMRI: The early beginnings. Neuroimage, 62(2), 852–855. [DOI] [PMC free article] [PubMed] [Google Scholar]
Heiss WD, Kessler J, Thiel A, Ghaemi M, & Karbe H (1999). Differential capacity of left and right hemispheric areas for compensation of poststroke aphasia. Annals of Neurology: Official Journal of the American Neurological Association and the Child Neurology Society, 454, 430–438. [DOI] [PubMed] [Google Scholar]
Hillis AE, & Tippett DC (2014). Stroke recovery: Surprising influencesand residual consequences. Advances in Medicine, 2014. [DOI] [PMC free article] [PubMed] [Google Scholar]
Howard D, & Patterson KE (1992). The Pyramids and Palm Trees Test: A test of semantic access from words and pictures. Bury St Edmunds: Thames Valley Test Company. [Google Scholar]
Jarso S, Li M, Faria A, Davis C, Leigh R, Sebastian R, … Hillis AE (2013). Distinct mechanisms and timing of 1language recovery after stroke. Cognitive Neuropsychology, 307–8, 454–475. [DOI] [PMC free article] [PubMed] [Google Scholar]
Jenkinson M, Bannister PR, Brady JM, & Smith SM (2002). Improved optimisation for the robust and accurate linear registration and motion correction of brain images. Neuroimage, 172, 825–841. [DOI] [PubMed] [Google Scholar]
Jenkinson M, Beckmann CF, Behrens TE, Woolrich MW, & Smith SM (2012). FSL. NeuroImage, 62, 782–790. [DOI] [PubMed] [Google Scholar]
Kay J, Lesser R, & Coltheart M (1992). Psycholinguistic assessments of language processing in aphasia PALPA. Hove UK: Erlbaum. [Google Scholar]
Klingbeil J, Wawrzyniak M, Stockert A, & Saur D (2017). Resting-state functional connectivity: An G. [DOI] [PubMed]
Kriegeskorte N, Goebel R, & Bandettini P (2006). Information-based functional brain mapping. Proceedings of the National Academy of Sciences, 103(10), 3863–3868. [DOI] [PMC free article] [PubMed] [Google Scholar]
Kuest J, & Karbe H (2002). Cortical activation studies in aphasia. Current Neurology and Neuroscience Reports, 26, 511–515. [DOI] [PubMed] [Google Scholar]
Marangolo P, Fiori V, Sabatini U, De Pasquale G, Razzano C, Caltagirone C, & Gili T (2016). Bilateral transcranial direct current stimulation language treatment enhances functional connectivity in the left hemisphere: Preliminary data from aphasia. Journal of Cognitive Neuroscience, 285, 724–738. [DOI] [PubMed] [Google Scholar]
Marcotte K, Adrover-Roig D, Damien B, de Preaumont M, Genereux S, Hubert M, & Ansaldo AI (2012). Therapy-induced neuroplasticity in chronic aphasia. Neuropsychologia, 50(8), 1776–1786. [DOI] [PubMed] [Google Scholar]
Nair VA, Young BM, La C, Reiter P, Nadkarni TN, Song J, … Jensen MB (2015). Functional connectivity changes in the language network during stroke recovery. Annals of Clinical and Translational Neurology, 2(2), 185–195. [DOI] [PMC free article] [PubMed] [Google Scholar]
Newman ME (2006). Modularity and community structure in networks. Proceedings of the National Academy of Sciences of the United States of America, 10323, 8577–8582. [DOI] [PMC free article] [PubMed] [Google Scholar]
New AB, Robin DA, Parkinson AL, Duffy JR, McNeil MR, Piguet O, … Ballard KJ (2015). Altered resting-state network connectivity in stroke patients with and without apraxia of speech. NeuroImage: Clinical, 8, 429–439. [DOI] [PMC free article] [PubMed] [Google Scholar]
Norman-Haignere SV, McCarthy G, Chun MM, & Turk-Browne NB (2011). Category-selective background connectivity in ventral visual cortex. Cerebral Cortex, 222, 391–402. [DOI] [PMC free article] [PubMed] [Google Scholar]
Norman KA, Polyn SM, Detre GJ, & Haxby JV (2006). Beyond mind-reading: Multi-voxel pattern analysis of fMRI data. Trends in Cognitive Sciences, 10(9), 424–430. [DOI] [PubMed] [Google Scholar]
Oldfield RC (1971). The assessment and analysis of handedness: The Edinburgh inventory. Neuropsychologia, 9(1), 97–113. [DOI] [PubMed] [Google Scholar]
Park DC, & Reuter-Lorenz P (2009). The adaptive brain: Aging and neurocognitive scaffolding. Annual Review of Psychology, 60, 173–196. [DOI] [PMC free article] [PubMed] [Google Scholar]
Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, … Vanderplas J (2011). Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12(Oct), 2825–2830. [Google Scholar]
Pereira F, Mitchell T, & Botvinick M (2009). Machine learning classifiers and fMRI: A tutorial overview. Neuroimage, 45(1), S199–S209. [DOI] [PMC free article] [PubMed] [Google Scholar]
Planton S, Jucla M, Roux FE, & Demonet JF (2013). The “handwriting brain”: A meta-analysis of neuroimaging studies of motor versus orthographic processes. Cortex, 4910, 2772–2787. [DOI] [PubMed] [Google Scholar]
Power JD, Cohen AL, Nelson SM, Wig GS, Barnes KA, Church JA, … Petersen SE (2011). Functional network organization of the human brain. Neuron, 724, 665–678. [DOI] [PMC free article] [PubMed] [Google Scholar]
Purcell JJ, & Rapp B (2018). Local response heterogeneity indexes experience-based neural differentiation in reading. Neuroimage, 183, 200–211. [DOI] [PMC free article] [PubMed] [Google Scholar]
Purcell J, Turkeltaub PE, Eden GF, & Rapp B (2011). Examining the central and peripheral processes of written word production through meta-analysis. Front Psychol, 2, 239. [DOI] [PMC free article] [PubMed] [Google Scholar]
Purcell JJ, Wiley RW, & Rapp B (2019). Re-learning to be different: Increased neural differentiation supports post-stroke language recovery. Neuroimage, 202, 116145. [DOI] [PMC free article] [PubMed] [Google Scholar]
Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, & Shulman GL (2001). A default mode of brain function. Proceedings of the National Academy of Sciences, 98(2), 676–682. [DOI] [PMC free article] [PubMed] [Google Scholar]
Rapp B, & Dufor O (2011). The neurotopography of written word production: An fMRI investigation of the distribution of sensitivity to length and frequency. Journal of Cognitive Neuroscience, 23(12), 4067–4081. [DOI] [PubMed] [Google Scholar]
Rapp B, & Kane A (2002). Remediation of deficits affecting different components of the spelling process. Aphasiology, 164–6, 439–454. [Google Scholar]
Rapp B, & Lipka K (2011). The literate brain: The relationship between spelling and reading. Journal of Cognitive Neuroscience, 23(5), 1180–1197. [DOI] [PMC free article] [PubMed] [Google Scholar]
Rapp B, Purcell J, Hillis AE, Capasso R, & Miceli G (2016). Neural bases of orthographic long-term memory and working memory in dysgraphia. Brain, 1392, 588–604. [DOI] [PMC free article] [PubMed] [Google Scholar]
Reuter-Lorenz PA, & Park DC (2014). How does it STAC up? Revisiting the scaffolding theory of aging and cognition. Neuropsychology Review, 24(3), 355–370. [DOI] [PMC free article] [PubMed] [Google Scholar]
Sandberg CW (2017). Hypoconnectivity of resting-state networks in persons with aphasia compared with healthy age-matched adults. Frontiers in Human Neuroscience, 11, 91. [DOI] [PMC free article] [PubMed] [Google Scholar]
Sandberg CW, Bohland JW, & Kiran S (2015). Changes in functional connectivity related to direct training and generalization effects of a word finding treatment in chronic aphasia. Brain and Language, 150, 103–116. [DOI] [PMC free article] [PubMed] [Google Scholar]
Saur D, Lange R, Baumgaertner A, Schraknepper V, Willmes K, Rijntjes M, & Weiller C (2006). Dynamics of language re-organization after stroke. Brain, 1296, 1371–1384. [DOI] [PubMed] [Google Scholar]
Sebastian R, Long C, Purcell J, Faria AV, Lindquist M, Jarso S, … Hillis AE (2016). Imaging network level language recovery after left PCA stroke. Restorative Neurology and Neuroscience, 344, 473–489. [DOI] [PMC free article] [PubMed] [Google Scholar]
Sharp DJ, Turkheimer FE, Bose SK, Scott SK, & Wise RJ (2010). Increased frontoparietal integration after stroke and cognitive recovery. Annals of Neurology, 68(5), 753–756. [DOI] [PubMed] [Google Scholar]
Siegel JS, Ramsey LE, Snyder AZ, Metcalf NV, Chacko RV, Weinberger K, … Corbetta M (2016). Disruptions of network connectivity predict impairment in multiple behavioral domains after stroke. Proceedings of the National Academy of Sciences of the United States of America, 11330, E4367–E4376. [DOI] [PMC free article] [PubMed] [Google Scholar]
Siegel JS, Seitzman BA, Ramsey LE, Ortega M, Gordon EM, Dosenbach NU, … Corbetta M (2018). Re-emergence of modular brain networks in stroke recovery. Cortex, 1012018, 44–59. [DOI] [PMC free article] [PubMed] [Google Scholar]
Smith SM, Fox PT, Miller KL, Glahn DC, Fox PM, Mackay CE, … Beckmann CF (2009). Correspondence of the brain’s functional architecture during activation and rest. Proceedings of the National Academy of Sciences, 106(31), 13040–13045. [DOI] [PMC free article] [PubMed] [Google Scholar]
Tao Y, & Rapp B (2019). The effects of lesion and treatment-related recovery on functional network modularity in post-stroke dysgraphia. NeuroImage: Clinical, 101865. [DOI] [PMC free article] [PubMed] [Google Scholar]
Thompson CK, Lukic S, King MC, Mesulam MM, & Weintraub S (2012). Verb and noun deficits in stroke-induced and primary progressive aphasia: The Northwestern Naming Battery. Aphasiology, 265, 632–655. [DOI] [PMC free article] [PubMed] [Google Scholar]
Tompary A, Al-Aidroos N, & Turk-Browne NB (2018). Attending to what and where: Background connectivity integrates categorical and spatial attention. Journal of Cognitive Neuroscience, 30(9), 1281–1297. [DOI] [PMC free article] [PubMed] [Google Scholar]
Turkeltaub PE (2015). Brain stimulation and the role of the right hemisphere in aphasia recovery. Current Neurology and Neuroscience Reports, 15(11), 72. [DOI] [PubMed] [Google Scholar]
Turkeltaub PE, Coslett HB, Thomas AL, Faseyitan O, Benson J, Norise C, & Hamilton RH (2012). The right hemisphere is not unitary in its role in aphasia recovery. Cortex, 489, 1179–1186. [DOI] [PMC free article] [PubMed] [Google Scholar]
Turkeltaub PE, Messing S, Norise C, & Hamilton RH (2011). Are networks for residual language function and recovery consistent across aphasic patients? Neurology, 76(20), 1726–1734. [DOI] [PMC free article] [PubMed] [Google Scholar]
Van Hees S, McMahon K, Angwin A, de Zubicaray G, Read S, & Copland DA (2014). A functional MRI study of the relationship between naming treatment outcomes and resting state functional connectivity in post-stroke aphasia. Hum Brain Mapping, 358, 3919–3931. [DOI] [PMC free article] [PubMed] [Google Scholar]
Warren JE, Crinion JT, Lambon Ralph MA, & Wise RJ (2009). Anterior temporal lobe connectivity correlates with functional outcome after aphasic stroke. Brain, 13212, 3428–3442. [DOI] [PMC free article] [PubMed] [Google Scholar]
Wig GS (2017). Segregated systems of human brain networks. Trends in Cognitive Sciences, 21(12), 981–996. [DOI] [PubMed] [Google Scholar]
Wiley RW, & Rapp B (2019). Statistical analysis in small-N designs: Using linear mixed-effects modeling for evaluating intervention effectiveness. Aphasiology, 33(1), 1–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
Woolrich MW, Behrens TE, Beckmann CF, Jenkinson M, & Smith SM (2004). Multilevel linear modelling for FMRI group analysis using Bayesian inference. Neuroimage, 21(4), 1732–1747. [DOI] [PubMed] [Google Scholar]
Yang H, Bai L, Zhou Y, Kang S, Liang P, Wang L, & Zhu Y (2017). Increased inter-hemispheric resting-state functional connectivity in acute lacunar stroke patients with aphasia. Experimental Brain Research, 235(3), 941–948. [DOI] [PubMed] [Google Scholar]
Yeo BT, Krienen FM, Eickhoff SB, Yaakub SN, Fox PT, Buckner RL, … Chee MW (2014). Functional specialization and flexibility in human association cortex. Cerebral Cortex, 2510, 3654–3672. [DOI] [PMC free article] [PubMed] [Google Scholar]
Zhu D, Chang J, Freeman S, Tan Z, Xiao J, Gao Y, & Kong J (2014). Changes of functional connectivity in the left frontoparietal network following aphasic stroke. Frontiers in Behavioral Neuroscience, 8, 167. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

mmc2

NIHMS1612427-supplement-mmc2.docx^{(20.8KB, docx)}

mmc1

NIHMS1612427-supplement-mmc1.docx^{(924.1KB, docx)}

Data Availability Statement

[R1] Abraham A, Pedregosa F, Eickenberg M, Gervais P, … Varoquaux G (2014). Machine learning for neuroimaging with scikit-learn. Frontiers in Neuroinformatics, 8, 14. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R2] Al-Aidroos N, Said CP, & Turk-Browne NB (2012). Top-down attention switches coupling between low-level and high-level areas of human visual cortex. Proceedings of the National Academy of Sciences of the United States of America, 10936, 14675–14680. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R3] Baddeley AD, Wilson BA, & Kopelman MD (1995). Handbook of Memory Disorders. London UK: John Wiley and Sons Ltd. [Google Scholar]

[R4] Balota DA, Yap MJ, Hutchison KA, Cortese MJ, Kessler B, Loftis B, & Treiman R (2007). The English lexicon project. Behavior Research Methods, 39(3), 445–459. [DOI] [PubMed] [Google Scholar]

[R5] Bassett DS, Wymbs NF, Porter MA, Mucha PJ, Carlson JM, & Grafton ST (2011). Dynamic reconfiguration of human brain networks during learning. Proceedings of the National Academy of Sciences of the United States of America, 10818, 7641–7646. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R6] Bates D, Maechler M, Bolker B, Walker S, Christensen RHB, Singmann H, … Rcpp L (2015). Package ‘lme4’. Convergence, 12(1). [Google Scholar]

[R7] Bertolero MA, Yeo BT, & D’Esposito M (2015). The modular and integrative functional architecture of the human brain. Proceedings of the National Academy of Sciences of the United States of America, 11249, E6798–E6807. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R8] Biswal B, Zerrin Yetkin F, Haughton VM, & Hyde JS (1995). Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magnetic Resonance in medicine, 34(4), 537–541. [DOI] [PubMed] [Google Scholar]

[R9] Blasi V, Young AC, Tansy AP, Petersen SE, Snyder AZ, & Corbetta M (2002). Word retrieval learning modulates right frontal cortex in patients with left frontal damage. Neuron, 361, 159–170. [DOI] [PubMed] [Google Scholar]

[R10] Braun U, Schäfer A, Walter H, Erk S, Romanczuk-Seiferth N, Haddad L, … Meyer-Lindenberg A (2015). Dynamic reconfiguration of frontal brain networks during executive cognition in humans. Proceedings of the National Academy of Sciences of the United States of America, 11237, 11678–11683. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R11] Cao Y, Vikingstad EM, George KP, Johnson AF, & Welch KM (1999). Cortical language activation in stroke patients recovering from aphasia with functional MRI. Stroke, 3011, 2331–2340. [DOI] [PubMed] [Google Scholar]

[R12] Cole MW, Bassett DS, Power JD, Braver TS, & Petersen SE (2014). Intrinsic and task-evoked network architectures of the human brain. Neuron, 831, 238–251. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R13] Cole MW, Reynolds JR, Power JD, Repovs G, Anticevic A, & Braver TS (2013). Multi-task connectivity reveals flexible hubs for adaptive task control. Nature Neuroscience, 169, 1348. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R14] Conover WJ (1999). Practical nonparametric statistics (3rd ed.). John Wiley & Sons. [Google Scholar]

[R15] Córdova NI, Tompary A, & Turk-Browne NB (2016). Attentional modulation of background connectivity between ventral visual cortex and the medial temporal lobe. Neurobiology of Learning and Memory, 134, 115–122. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R16] Dosenbach NU, Fair DA, Miezin FM, Cohen AL, Wenger KK, Dosenbach RA, … Schlaggar BL (2007). Distinct brain networks for adaptive and stable task control in humans. Proceedings of the National Academy of Sciences of the United States of America, 10426, 11073–11078. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R17] Duncan J (2013). The structure of cognition: Attentional episodes in mind and brain. Neuron, 80(1), 35–50. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R18] Duncan ES, & Small SL (2016). Increased modularity of resting state networks supports improved narrative production in aphasia recovery. Brain Connectivity, 67, 524–529. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R19] Fair DA, Schlaggar BL, Cohen AL, Miezin FM, Dosenbach NU, Wenger KK, … Petersen SE (2007). A method for using blocked and event-related fMRI data to study “resting state” functional connectivity. Neuroimage, 351, 396–405. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R20] Fedorenko E, & Thompson-Schill SL (2014). Reworking the language network. Trends in Cognitive Sciences, 18(3), 120–126. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R21] Fox MD, Corbetta M, Snyder AZ, Vincent JL, & Raichle ME (2006). Spontaneous neuronal activity distinguishes human dorsal and ventral attention systems. Proceedings of the National Academy of Sciences of the United States of America, 103(26), 10046–10051. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R22] Fridriksson J (2010). Preservation and modulation of specific left hemisphere regions is vital for treated recovery from anomia in stroke. The Journal of Neuroscience: the Official Journal of the Society for Neuroscience, 3035, 11558–11564. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R23] Fridriksson J, Bonilha L, Baker JM, Moser D, & Rorden C (2009). Activity in preserved left hemisphere regions predicts anomia severity in aphasia. Cerebral Cortex, 20(5), 1013–1019. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R24] Gainotti G (2015). Contrasting opinions on the role of the right hemisphere in the recovery of language. A critical survey. Aphasiology, 29(9), 1020–1037. [Google Scholar]

[R25] Geranmayeh F, Chau TW, Wise RJ, Leech R, & Hampshire A (2017). Domain-general subregions of the medial prefrontal cortex contribute to recovery of language after stroke. Brain, 140(7), 1947–1958. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R26] Gili T, Fiori V, De Pasquale G, Sabatini U, Caltagirone C, & Marangolo P (2017). Right sensory-motor functional networks subserve action observation therapy in aphasia. Brain Imaging and Behavior, 11(5), 1397–1411. [DOI] [PubMed] [Google Scholar]

[R27] Gold BT, & Kertesz A (2000). Right hemisphere semantic processing of visual words in an aphasic patient: An fMRI study. Brain and Language, 733, 456–465. [DOI] [PubMed] [Google Scholar]

[R28] Goodman RA, & Caramazza A (1985). The Johns Hopkins university dysgraphia battery. Baltimore, MD: Johns Hopkins University. [Google Scholar]

[R29] Gratton C, Laumann TO, Gordon EM, Adeyemo B, & Petersen SE (2016). Evidence for two independent factors that modify brain networks to meet task goals. Cell Reports, 17(5), 1276–1288. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R30] Gratton C, Laumann TO, Nielsen AN, Greene DJ, Gordon EM, Gilmore AW, … Dosenbach NU (2018). Functional brain networks are dominated by stable group and individual factors, not cognitive or daily variation. Neuron, 98(2), 439–452. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R31] Gratton C, Nomura EM, Pérez F, & D’Esposito M (2012). Focal brain lesions to critical locations cause widespread disruption of the modular organization of the brain. Journal of Cognitive Neuroscience, 246, 1275–1285. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R32] Greve DN, & Fischl B (2009). Accurate and robust brain image alignment using boundary-based registration. Neuroimage, 481, 63–72. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R33] Hartwigsen G, & Saur D (2017). Neuroimaging of stroke recovery from aphasia–Insights into plasticity of the human language network. Neuroimage, 190, 14–31. [DOI] [PubMed] [Google Scholar]

[R34] Haxby JV (2012). Multivariate pattern analysis of fMRI: The early beginnings. Neuroimage, 62(2), 852–855. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R35] Heiss WD, Kessler J, Thiel A, Ghaemi M, & Karbe H (1999). Differential capacity of left and right hemispheric areas for compensation of poststroke aphasia. Annals of Neurology: Official Journal of the American Neurological Association and the Child Neurology Society, 454, 430–438. [DOI] [PubMed] [Google Scholar]

[R36] Hillis AE, & Tippett DC (2014). Stroke recovery: Surprising influencesand residual consequences. Advances in Medicine, 2014. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R37] Howard D, & Patterson KE (1992). The Pyramids and Palm Trees Test: A test of semantic access from words and pictures. Bury St Edmunds: Thames Valley Test Company. [Google Scholar]

[R38] Jarso S, Li M, Faria A, Davis C, Leigh R, Sebastian R, … Hillis AE (2013). Distinct mechanisms and timing of 1language recovery after stroke. Cognitive Neuropsychology, 307–8, 454–475. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R39] Jenkinson M, Bannister PR, Brady JM, & Smith SM (2002). Improved optimisation for the robust and accurate linear registration and motion correction of brain images. Neuroimage, 172, 825–841. [DOI] [PubMed] [Google Scholar]

[R40] Jenkinson M, Beckmann CF, Behrens TE, Woolrich MW, & Smith SM (2012). FSL. NeuroImage, 62, 782–790. [DOI] [PubMed] [Google Scholar]

[R41] Kay J, Lesser R, & Coltheart M (1992). Psycholinguistic assessments of language processing in aphasia PALPA. Hove UK: Erlbaum. [Google Scholar]

[R42] Klingbeil J, Wawrzyniak M, Stockert A, & Saur D (2017). Resting-state functional connectivity: An G. [DOI] [PubMed]

[R43] Kriegeskorte N, Goebel R, & Bandettini P (2006). Information-based functional brain mapping. Proceedings of the National Academy of Sciences, 103(10), 3863–3868. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R44] Kuest J, & Karbe H (2002). Cortical activation studies in aphasia. Current Neurology and Neuroscience Reports, 26, 511–515. [DOI] [PubMed] [Google Scholar]

[R45] Marangolo P, Fiori V, Sabatini U, De Pasquale G, Razzano C, Caltagirone C, & Gili T (2016). Bilateral transcranial direct current stimulation language treatment enhances functional connectivity in the left hemisphere: Preliminary data from aphasia. Journal of Cognitive Neuroscience, 285, 724–738. [DOI] [PubMed] [Google Scholar]

[R46] Marcotte K, Adrover-Roig D, Damien B, de Preaumont M, Genereux S, Hubert M, & Ansaldo AI (2012). Therapy-induced neuroplasticity in chronic aphasia. Neuropsychologia, 50(8), 1776–1786. [DOI] [PubMed] [Google Scholar]

[R47] Nair VA, Young BM, La C, Reiter P, Nadkarni TN, Song J, … Jensen MB (2015). Functional connectivity changes in the language network during stroke recovery. Annals of Clinical and Translational Neurology, 2(2), 185–195. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R48] Newman ME (2006). Modularity and community structure in networks. Proceedings of the National Academy of Sciences of the United States of America, 10323, 8577–8582. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R49] New AB, Robin DA, Parkinson AL, Duffy JR, McNeil MR, Piguet O, … Ballard KJ (2015). Altered resting-state network connectivity in stroke patients with and without apraxia of speech. NeuroImage: Clinical, 8, 429–439. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R50] Norman-Haignere SV, McCarthy G, Chun MM, & Turk-Browne NB (2011). Category-selective background connectivity in ventral visual cortex. Cerebral Cortex, 222, 391–402. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R51] Norman KA, Polyn SM, Detre GJ, & Haxby JV (2006). Beyond mind-reading: Multi-voxel pattern analysis of fMRI data. Trends in Cognitive Sciences, 10(9), 424–430. [DOI] [PubMed] [Google Scholar]

[R52] Oldfield RC (1971). The assessment and analysis of handedness: The Edinburgh inventory. Neuropsychologia, 9(1), 97–113. [DOI] [PubMed] [Google Scholar]

[R53] Park DC, & Reuter-Lorenz P (2009). The adaptive brain: Aging and neurocognitive scaffolding. Annual Review of Psychology, 60, 173–196. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R54] Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, … Vanderplas J (2011). Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12(Oct), 2825–2830. [Google Scholar]

[R55] Pereira F, Mitchell T, & Botvinick M (2009). Machine learning classifiers and fMRI: A tutorial overview. Neuroimage, 45(1), S199–S209. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R56] Planton S, Jucla M, Roux FE, & Demonet JF (2013). The “handwriting brain”: A meta-analysis of neuroimaging studies of motor versus orthographic processes. Cortex, 4910, 2772–2787. [DOI] [PubMed] [Google Scholar]

[R57] Power JD, Cohen AL, Nelson SM, Wig GS, Barnes KA, Church JA, … Petersen SE (2011). Functional network organization of the human brain. Neuron, 724, 665–678. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R58] Purcell JJ, & Rapp B (2018). Local response heterogeneity indexes experience-based neural differentiation in reading. Neuroimage, 183, 200–211. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R59] Purcell J, Turkeltaub PE, Eden GF, & Rapp B (2011). Examining the central and peripheral processes of written word production through meta-analysis. Front Psychol, 2, 239. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R60] Purcell JJ, Wiley RW, & Rapp B (2019). Re-learning to be different: Increased neural differentiation supports post-stroke language recovery. Neuroimage, 202, 116145. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R61] Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, & Shulman GL (2001). A default mode of brain function. Proceedings of the National Academy of Sciences, 98(2), 676–682. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R62] Rapp B, & Dufor O (2011). The neurotopography of written word production: An fMRI investigation of the distribution of sensitivity to length and frequency. Journal of Cognitive Neuroscience, 23(12), 4067–4081. [DOI] [PubMed] [Google Scholar]

[R63] Rapp B, & Kane A (2002). Remediation of deficits affecting different components of the spelling process. Aphasiology, 164–6, 439–454. [Google Scholar]

[R64] Rapp B, & Lipka K (2011). The literate brain: The relationship between spelling and reading. Journal of Cognitive Neuroscience, 23(5), 1180–1197. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R65] Rapp B, Purcell J, Hillis AE, Capasso R, & Miceli G (2016). Neural bases of orthographic long-term memory and working memory in dysgraphia. Brain, 1392, 588–604. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R66] Reuter-Lorenz PA, & Park DC (2014). How does it STAC up? Revisiting the scaffolding theory of aging and cognition. Neuropsychology Review, 24(3), 355–370. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R67] Sandberg CW (2017). Hypoconnectivity of resting-state networks in persons with aphasia compared with healthy age-matched adults. Frontiers in Human Neuroscience, 11, 91. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R68] Sandberg CW, Bohland JW, & Kiran S (2015). Changes in functional connectivity related to direct training and generalization effects of a word finding treatment in chronic aphasia. Brain and Language, 150, 103–116. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R69] Saur D, Lange R, Baumgaertner A, Schraknepper V, Willmes K, Rijntjes M, & Weiller C (2006). Dynamics of language re-organization after stroke. Brain, 1296, 1371–1384. [DOI] [PubMed] [Google Scholar]

[R70] Sebastian R, Long C, Purcell J, Faria AV, Lindquist M, Jarso S, … Hillis AE (2016). Imaging network level language recovery after left PCA stroke. Restorative Neurology and Neuroscience, 344, 473–489. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R71] Sharp DJ, Turkheimer FE, Bose SK, Scott SK, & Wise RJ (2010). Increased frontoparietal integration after stroke and cognitive recovery. Annals of Neurology, 68(5), 753–756. [DOI] [PubMed] [Google Scholar]

[R72] Siegel JS, Ramsey LE, Snyder AZ, Metcalf NV, Chacko RV, Weinberger K, … Corbetta M (2016). Disruptions of network connectivity predict impairment in multiple behavioral domains after stroke. Proceedings of the National Academy of Sciences of the United States of America, 11330, E4367–E4376. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R73] Siegel JS, Seitzman BA, Ramsey LE, Ortega M, Gordon EM, Dosenbach NU, … Corbetta M (2018). Re-emergence of modular brain networks in stroke recovery. Cortex, 1012018, 44–59. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R74] Smith SM, Fox PT, Miller KL, Glahn DC, Fox PM, Mackay CE, … Beckmann CF (2009). Correspondence of the brain’s functional architecture during activation and rest. Proceedings of the National Academy of Sciences, 106(31), 13040–13045. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R75] Tao Y, & Rapp B (2019). The effects of lesion and treatment-related recovery on functional network modularity in post-stroke dysgraphia. NeuroImage: Clinical, 101865. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R76] Thompson CK, Lukic S, King MC, Mesulam MM, & Weintraub S (2012). Verb and noun deficits in stroke-induced and primary progressive aphasia: The Northwestern Naming Battery. Aphasiology, 265, 632–655. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R77] Tompary A, Al-Aidroos N, & Turk-Browne NB (2018). Attending to what and where: Background connectivity integrates categorical and spatial attention. Journal of Cognitive Neuroscience, 30(9), 1281–1297. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R78] Turkeltaub PE (2015). Brain stimulation and the role of the right hemisphere in aphasia recovery. Current Neurology and Neuroscience Reports, 15(11), 72. [DOI] [PubMed] [Google Scholar]

[R79] Turkeltaub PE, Coslett HB, Thomas AL, Faseyitan O, Benson J, Norise C, & Hamilton RH (2012). The right hemisphere is not unitary in its role in aphasia recovery. Cortex, 489, 1179–1186. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R80] Turkeltaub PE, Messing S, Norise C, & Hamilton RH (2011). Are networks for residual language function and recovery consistent across aphasic patients? Neurology, 76(20), 1726–1734. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R81] Van Hees S, McMahon K, Angwin A, de Zubicaray G, Read S, & Copland DA (2014). A functional MRI study of the relationship between naming treatment outcomes and resting state functional connectivity in post-stroke aphasia. Hum Brain Mapping, 358, 3919–3931. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R82] Warren JE, Crinion JT, Lambon Ralph MA, & Wise RJ (2009). Anterior temporal lobe connectivity correlates with functional outcome after aphasic stroke. Brain, 13212, 3428–3442. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R83] Wig GS (2017). Segregated systems of human brain networks. Trends in Cognitive Sciences, 21(12), 981–996. [DOI] [PubMed] [Google Scholar]

[R84] Wiley RW, & Rapp B (2019). Statistical analysis in small-N designs: Using linear mixed-effects modeling for evaluating intervention effectiveness. Aphasiology, 33(1), 1–30. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R85] Woolrich MW, Behrens TE, Beckmann CF, Jenkinson M, & Smith SM (2004). Multilevel linear modelling for FMRI group analysis using Bayesian inference. Neuroimage, 21(4), 1732–1747. [DOI] [PubMed] [Google Scholar]

[R86] Yang H, Bai L, Zhou Y, Kang S, Liang P, Wang L, & Zhu Y (2017). Increased inter-hemispheric resting-state functional connectivity in acute lacunar stroke patients with aphasia. Experimental Brain Research, 235(3), 941–948. [DOI] [PubMed] [Google Scholar]

[R87] Yeo BT, Krienen FM, Eickhoff SB, Yaakub SN, Fox PT, Buckner RL, … Chee MW (2014). Functional specialization and flexibility in human association cortex. Cerebral Cortex, 2510, 3654–3672. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R88] Zhu D, Chang J, Freeman S, Tan Z, Xiao J, Gao Y, & Kong J (2014). Changes of functional connectivity in the left frontoparietal network following aphasic stroke. Frontiers in Behavioral Neuroscience, 8, 167. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

How functional network connectivity changes as a result of lesion and recovery: An investigation of the network phenotype of stroke

Yuan Tao

Brenda Rapp

Abstract

1. Introduction

1.1. The impact of a stroke on functional network organization

1.2. Functional connectivity and the recovery of language functions

1.3. Functional networks and language processing

1.4. The current study

2. Methods

2.1. Participants

Fig. 2 –

Table 1 –

2.2. Cognitive/Language assessments and dysgraphia treatment

2.3. Neuroimaging experiment

2.4. Imaging data acquisition

2.5. fMRI data preprocessing, model-based analyses, and functional connectivity estimation

2.5.1. Pre-processing

2.5.2. GLM analysis

2.5.3. Evaluation of group-level mean BOLD activation

2.5.4. Background functional connectivity estimation

Fig. 1 –

2.5.5. Identifying the functional clusters

2.6. Functional connectivity analyses

2.6.1. Univariate functional connectivity analyses

2.6.2. Classification (multivariate) analyses: Evaluating changes in pre to post-treatment connectivity

2.6.3. Evaluating normalization vs re-organization

3. Results

3.1. Spelling and language/cognitive assessments

3.2. In-scanner behavioral performance

3.3. The functional clusters of spelling

Fig. 9 –

3.4. Univariate functional connectivity analyses

3.4.1. Pre-treatment homotopic connectivity

Fig. 3 –

Fig. 4 –

3.4.2. Pre-treatment intra-hemispheric connectivity

3.4.3. Pre- to post-treatment changes in homotopic and intra-hemisphere connectivity

3.5. Classification (multivariate) analyses: Evaluating pre- to post-treatment connectivity changes

3.5.1. Changes in pre- to post-treatment homotopic connectivity

Fig. 5 –

3.5.2. Changes in pre- to post-treatment intra-hemispheric connectivity

Fig. 6 –

Fig. 7 –

3.5.3. The relationship between treatment-related neural and behavioral changes

3.5.4. Normalization/re-organization

Fig. 8 –

3.6. Pre- to post-treatment changes in mean BOLD activation

4. General discussion

4.1. The network phenotype of stroke: Converging evidence

4.2. Treatment-based changes in the FC stroke phenotype

4.3. Do treatment-related changes result in normalization or re-organization?

4.4. Functional connectivity changes may support differentiation of executive and core processes

4.5. Different measures: background connectivity, mean local BOLD and RS-fMRI

4.6. Limitations

5. Conclusions

Supplementary Material

Acknowledgments

Funding

Footnotes

Data availability

REFERENCES

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases