Right Hemisphere Functional Connectivity Increases in Proportion to Left Hemisphere Lesion Size, but Not Within the Language Network

ABSTRACT

Left hemisphere (LH) strokes often disrupt the language network (LN), causing aphasia. The right hemisphere (RH) has been proposed to contribute to post‐stroke aphasia outcomes, but existing findings on RH functional connectivity (FC) changes after LH stroke are inconsistent. This study systematically investigates whether and how the RH FC differs in chronic stroke survivors compared to healthy controls, and whether these differences relate to lesion characteristics and language outcomes. Using a naturalistic movie‐watching fMRI paradigm, we compared the RH FC in a large sample of chronic LH stroke survivors (n = 85) to neurologically healthy controls (n = 71). We first used a validated semantic decision fMRI task to identify a core LN and a non‐canonical LN (labeled Semantic Network (SN)). We then mapped group FC differences within and between these LH networks and their RH homotopes. Compared to controls, stroke survivors had higher overall within‐RH FC and lower within‐LH and interhemispheric connectivity. These differences were greater with larger lesions. However, few of the RH FC differences were observed within the RH LN. Rather, most changes were between networks and were distributed widely in the RH. Greater RH connectivity was not associated with lesions or reduced FC in homotopic LH regions, although lesions of LH superior and middle temporal gyri were associated with increased FC between RH SN and LN. RH FC was not associated with behavioral outcomes. These findings show that LH lesions lead to increased RH FC, but the pattern of results suggests that these changes may not reflect meaningful functional reorganization that contributes to language function after stroke.

Trial Registration

clinicaltrials.gov NCT04991519

Left‐hemisphere stroke was associated with widespread increases in right‐hemisphere functional connectivity; however, these changes were mostly outside the right‐hemisphere ‘language’ network and did not relate to behavior. Thus, these functional connectivity changes may reflect broad network disruption rather than a meaningful functional reorganization that contributes to aphasia recovery.

1. Introduction

Stroke is one of the leading causes of long‐term disability, with approximately 795,000 strokes occurring in the United States every year (Martin et al. 2025). About one third of strokes result in aphasia, a disorder impacting language use and comprehension (Gottesman and Hillis 2010). Aphasia affects many aspects of daily life, including the ability to communicate with family and friends, return to work, and participate in social life (Bullier et al. 2020; Hilari et al. 2012). With such substantial effects on the quality of life, understanding mechanisms supporting recovery from aphasia is critically important.

Aphasia outcomes depend not only on the nature of the stroke (e.g., size and location), but also on the function of brain structures that are not directly injured by the stroke (Turkeltaub 2019). The right hemisphere (RH) has received particular attention for its potential role in stroke outcomes, though its contribution remains debated, with evidence supporting both compensatory and maladaptive roles. Evidence for the RH's compensatory role comes from stroke and neuroimaging studies. For example, case reports of sequential left hemisphere (LH) and RH strokes have demonstrated that language recovery after LH stroke may, in part, rely on the RH (Barlow 1877; Basso et al. 1989; Turkeltaub et al. 2012). Consistent with this, imaging studies have shown increased activity in RH language homotopes, at least in some regions and for some individuals (Turkeltaub et al. 2011, 2026). Notably, the increased RH activity has been associated with better language outcomes (Skipper‐Kallal et al. 2017a, 2017b; Turkeltaub et al. 2026). In contrast, other lines of evidence suggest that inhibition of the RH inferior frontal gyrus (IFG) using transcranial magnetic stimulation improves language performance, suggesting that RH involvement can be maladaptive (Ding et al. 2022).

Recruitment of RH regions homotopic to lesioned LH language nodes is often explained on the basis of network reweighting (Chang and Lambon Ralph 2020). In this framework, some RH regions homotopic to the LH language network (LN) form a “RH LN” that is weakly involved in language in typical adults, possibly as a developmental remnant of a bilateral LN from infancy (Martin et al. 2022). Lesions to LH language nodes result in reweighting of the network through synaptic plasticity upregulating their counterpart RH nodes that may perform similar computations. A recent task‐related fMRI study using a subset of the participants examined in the present study found support for this mechanism of RH recruitment, specifically in right ventral IFG and mid‐anterior temporal lobe language regions (Turkeltaub et al. 2026). Specifically, this study found that these RH regions were weakly engaged in language in neurologically healthy controls, that their activation levels correlated with their LH counterparts in both controls and stroke survivors, and that their activity was upregulated by lesions affecting their LH counterparts. Another theoretical framework invoked to explain changes in RH regions symmetric to the stroke lesion is the interhemispheric disinhibition hypothesis (Kinsbourne 1974). This framework proposes that the two hemispheres exert mutual transcallosal inhibition on one another, such that the LH normally suppresses the homotopic RH areas. After a LH stroke, this inhibition is disrupted, which releases RH homotopic regions from suppression and leads to upregulation in those regions (Heiss 2009; Martin et al. 2004). The hypothesis therefore predicts that the RH changes after LH stroke should be concentrated in regions homotopic to the lesioned LH areas. However, direct support for this prediction is limited (Turkeltaub et al. 2026).

Some frameworks also propose that the involvement of the RH in aphasia recovery is more prominent when LH damage is severe (e.g., large lesions), but empirical evidence supporting this theory is limited (Anglade et al. 2014; Hamilton et al. 2011; Heiss and Thiel 2006). For example, although RH activation during picture naming has been linked to lesion size, similar effects were observed in the bilateral visual cortex, suggesting that these effects were related to greater effort and longer looking times in people with severe aphasia (Skipper‐Kallal et al. 2017b). More broadly, interpretation of task‐based fMRI signals is often confounded by differences in task performance (Turkeltaub 2019; Wilson et al. 2018). Studies that control for these factors report a more nuanced picture. In a recent fMRI study of chronic post‐stroke aphasia that addressed task performance confounds using an adaptive fMRI task, no relationship between lesion size and RH activation was found, but different lesion locations were associated with different patterns of RH activity (Turkeltaub et al. 2026). While this study found overall greater language activity in the RH for people with aphasia compared to healthy controls, especially in younger, highly educated, left‐handed individuals, another similar study using the same adaptive fMRI task did not (Schneck et al. 2026). Using other measures to assess RH reorganization after LH stroke while minimizing task performance confounds might help to clarify the nature of RH plasticity in aphasia.

Task‐free functional connectivity (FC) provides a suitable approach for addressing the above‐mentioned challenges by examining the intrinsic network organization independent of task performance. FC measures the correlation between fMRI‐measured blood oxygenation level dependent (BOLD) signal across different brain regions, facilitating the examination of large‐scale networks that support complex behaviors such as language (Fox and Greicius 2010; Tomasi and Volkow 2012; Turken and Dronkers 2011). This method is widely used to identify large‐scale resting‐state networks (RSN) that include visual (VIS), sensorimotor (SMN), dorsal attention (DAN), ventral attention (VAN), limbic (LIM), frontoparietal (FPN), and default mode (DMN) networks, as well as more specialized networks, including language (Tomasi and Volkow 2012; Yeo et al. 2011). Importantly, FC analysis can help examine how interareal communication is disrupted after a stroke, and how compensation or reorganization might occur during recovery (Klingbeil et al. 2019).

Since FC can be measured in the absence of any explicit task, it is easy to complete and free of task performance confounds even for individuals with severe strokes (Hartwigsen and Saur 2019). In recent years, movie‐watching has gained attention as an advantageous task‐free paradigm for examining FC. Although movies involve visual and auditory processing, they do not require discrete stimulus–response coupling and impose no explicit task demands on participants. Movie‐watching provides an ecologically valid context that enhances participant engagement, improves arousal levels, and reduces head motion compared to resting‐state scans (Eickhoff et al. 2020; Sonkusare et al. 2019). Moreover, FC measured during movie watching shows higher test–retest reliability and better prediction of individual behavioral traits, while preserving both cross‐subject consistency in network organization and individual variations in FC patterns (Finn and Bandettini 2021; Vanderwal et al. 2017). Importantly, movie‐watching FC has been applied in stroke populations to examine post‐stroke FC changes (Ketchabaw et al. 2022; Tao et al. 2022; Tao and Rapp 2020). For these reasons, the current study uses movie‐watching to examine FC of the RH following LH stroke.

Findings from prior studies examining RH FC after LH stroke have been inconsistent. Several studies have found decreased RH FC in stroke survivors relative to controls. Zhu et al. (2014) reported decreased FC between the RH IFG and middle frontal gyrus (MFG) and RH medial frontal cortex in subacute stroke participants (Zhu et al. 2014). Similarly, Sandberg (2017) found reduced RH FC within the executive control network (RH angular gyrus to RH paracingulate gyrus) in chronic stroke. However, the direction of RH FC change was not uniform within this study; in the semantic network (SN), which included the frontal, temporal and parietal regions, an approximately equal number of connections was increased and decreased relative to controls (Sandberg 2017). Other studies, however, have reported increased RH FC after LH stroke. Tao et al. (2022) found that in chronic stroke participants 32 RH connections had increased FC whereas only 7 showed decreases compared to controls (Tao et al. 2022). Similarly, Yang et al. (2016) reported increased FC between the contralesional hippocampal/parahippocampal area and the fusiform gyrus in stroke subjects (Yang et al. 2016).

In addition to the direction of FC change, the association between the RH FC and LH lesion size also remains unclear. Yourganov et al. (2021) found greater RH FC with larger LH lesion size in chronic stroke survivors. However, FC was not compared to a control group to determine whether connectivity was increased in individuals with large lesions, decreased in individuals with small lesions, or both. In contrast, Sandberg (2017) found that larger LH lesion size was related to lower FC in LH and RH connections in several networks, specifically in DMN, executive function, SMN, and semantic (anterior superior temporal gyrus (STG) and temporo‐occipital middle temporal gyrus (MTG) regions) networks (Sandberg 2017).

Changes in RH FC have also been shown to relate to behavior, although the direction of effects is variable. Teki et al. (2013) observed that greater effective connectivity (i.e., directional FC) between the RH STG and the RH auditory cortex was associated with better semantic skills (Teki et al. 2013). However, several studies have reported a negative association between the RH FC and behavioral performance. Yang et al. (2016) found that FC in the RH hippocampus and parahippocampus negatively correlated with construction scores from the Aphasia Battery of Chinese (Yang et al. 2016). Ramage et al. (2020) reported that stronger FC between the RH posterior MTG and the RH IFG was associated with worse performance on Western Aphasia Battery (WAB) and Naming and Word Finding tasks (Ramage et al. 2020). Additionally, Vicentini et al. (2021) found that cognitive performance was related to weaker contralesional SN FC at the subacute stage (Vicentini et al. 2021). However, this study did not control for lesion size.

Overall, results of prior studies on RH FC changes after LH stroke have been highly inconsistent. This is likely due to several key limitations of prior work, including small sample sizes, short scanning times, and participant groups at mixed stages of recovery, any of which may account for variable results across studies. Additionally, some studies failed to exclude lesioned voxels, which could lead to misinterpretations of the results in the LH. Finally, studies have often examined RH FC without a priori delineation of RH LN homotopes, which the task‐related fMRI literature indicates are the key RH regions undergoing plastic change in post‐stroke aphasia (Turkeltaub et al. 2011, 2026).

In sum, the task‐related fMRI literature suggests RH LN plasticity may play a role in aphasia recovery, and examining task‐free RH FC might contribute to this area of inquiry. However, gaps and limitations of prior studies of RH FC in aphasia have left several key questions parallel to those addressed in the task‐related fMRI literature unanswered: Is RH FC different after LH stroke compared to typical adults, and if so, do demographic factors relate to these differences? Do RH FC differences relate to lesion size, with differences primarily occurring in individuals with large lesions? Do the differences primarily occur in the homotopic RH LN as the task‐related fMRI literature might suggest, or between the homotopic LN and other regions, or outside the RH LN entirely? Do changes in FC occur as a result of diminished FC in homotopic regions of the LH? If not, do lesions to specific LH structures result in changes in RH FC? Does RH FC relate to language abilities in aphasia? Addressing these questions in a large sample using rigorous analysis methods might help to elucidate the mechanisms underlying differences in RH FC of individuals with LH stroke. This in turn might help clarify the broader role of RH LN plasticity in aphasia recovery.

Here, we aim to address several limitations of prior studies to answer the questions raised above about RH FC in LH stroke survivors with aphasia. Compared to prior studies, we use a larger sample size, longer scan duration using movie‐watching, and restrict our participants to those in the chronic period of recovery. Importantly, we use an fMRI language localizer task that is well‐validated in people with aphasia (Wilson et al. 2018) to identify LH LN nodes and their RH homotopes. First, we determine whether RH FC in chronic LH stroke survivors differs from that of healthy controls, also describing differences in within‐LH FC and interhemispheric FC to provide context for the RH differences. We then determine if demographic factors and LH lesion size relate to differences in RH FC. We examine the effect of the LH lesion size both continuously and by dividing the stroke group into small and large lesion groups using a median split. This approach allows us to characterize the pattern of RH FC changes, specifically in individuals whose lesions are expected to cause major disruptions to brain networks (i.e., large lesions). We then determine if differences in RH FC primarily occur within the homotopic RH LN, outside this network, or between the RH homotopic LN and other RH regions. To further characterize the distribution of FC group differences, we also map the connections in relation to the canonical RSNs and anatomical regions, which provides a more comprehensive picture of which systems are more affected outside the LN. We then examine whether lesion location influences the pattern of RH FC differences, as lesions to different LH LN nodes may affect RH FC differently. Finally, we test whether RH FC relates to language outcomes after LH stroke. This series of analyses provides perhaps the most comprehensive examination to date of RH FC in individuals with chronic LH stroke.

2. Methods

2.1. Participants

We recruited 99 LH stroke and 71 demographically matched control participants for this study (Table 1). Participants were recruited as part of a larger observational study of aphasia recovery (clinicaltrials.gov NCT04991519) through several sources: Author P.E.T.'s Aphasia Clinic and affiliated Speech Language Pathologist networks, community referrals, social networks, community flyers and online platforms (clinicaltrials.gov, ResearchMatch, aphasia.org). All stroke survivors were at the chronic stage of recovery (> 6 months post‐stroke). Stroke participants were included if they were aged 18 years or older, had a stroke in the LH, with or without aphasia, and if they had learned English at or before age 8. Controls were included if they were aged 18 years or older, had no history of brain injury from stroke, trauma, infection or tumor, and had also learned English at or before age 8. Exclusion criteria included a history of other brain conditions that could impact the interpretation of results (e.g., multiple sclerosis, premorbid dementia), severe psychiatric conditions that could interfere with study participation, or a history of a learning disability that could affect the results. Additionally, participants were excluded if they had contraindications to the MRI scan. The study was approved by the Institutional Review Board, and all participants provided written informed consent prior to enrollment.

TABLE 1.

Participant information.

	LH Stroke Survivors Mean (SD)	Controls Mean (SD)
N	85	71
Age (Years)	61 (11)	61 (11)
Sex (F/M)	34/51	36/35
Handedness (L/R)	9/76	5/66
Race
African‐American	35	23
Caucasian	49	48
More than one race	1	—
Education (years)	16 (3)	17 (3)
Time Post‐Stroke (months)	45 (52)	—
Median	26
IQR	41
Lesion Size (cm3)	94 (86)	—
Median	78
IQR	134
Pyramids and Palm Trees (%)	86 (12)	97 (4) a
Range	45–100	83–100
Picture Naming (%)	66 (31)	95 (4) b
Range	0–98	82–100
Pseudoword Repetition (%)	57 (29)	90 (6) c
Range	0–100	73–98
WAB AQ Auditory Verbal Comprehension (/10)	8.7 (1.6)	—
Range	4.4–10
WAB Naming and Word Finding (/10)	7.6 (2.6)
Range	0–10
WAB Repetition (/10)	7.7 (2.5)	—
Range	0.4–10
WAB Spontaneous Speech (/20)	15.3 (5)	—
Range	2–20
WAB AQ (/100)	79 (22.3)	—
Range	14–100

Abbreviations: IQR, interquartile range; SD, standard deviation; WAB, western aphasia battery—Revised (score < 93.8 is impaired).

^a Based on 41 controls.

^b Based on 36 controls.

^c Based on 31 controls.

Fourteen participants were excluded from the study. One participant was excluded due to severe motor speech impairment that significantly impacted performance on the behavioral measures of interest (e.g., picture naming, pseudoword repetition). Four participants were excluded due to absence of lesion tracing or FC data. The remaining nine participants were excluded due to the presence of lesions or severe white matter disease outside the LH.

2.2. Scan Acquisition

All the MRI scans were conducted on a Siemens Prisma 3 T scanner with a 20‐channel head coil. High‐resolution anatomical images were acquired with a T1‐weighted MPRAGE sequence with the following parameters: TR = 1900 ms, TE = 2.98 ms, 176 slices, slice thickness = 1 mm (sagittal orientation), field of view = 256 mm, matrix = 256 × 256, and flip angle = 9°. Additionally, a T2‐weighted fluid‐attenuated inversion recovery (FLAIR) scan was obtained with TR = 5000 ms, TE = 38.2 ms, 192 slices, slice thickness = 1 mm (sagittal orientation), field of view = 256 mm, matrix = 256 × 256, and flip angle = 120°. The BOLD T2*‐weighted scans were conducted with the following parameters: TR = 794 ms, multiband factor = 4, 48 axial slices, slice thickness = 2.6 mm with a 10% slice gap, field of view = 211 mm, flip angle = 50°, matrix = 74 × 74, voxel size = 2.9 × 2.9 mm. The functional scans included a task‐free scan during movie watching and a semantic decision task (Wilson et al. 2018). Task‐free scan was collected while participants watched The Gruffalo movie (2009) (1096 volumes, duration = 14.5 min). During the semantic decision task, participants viewed word pairs and pairs of pseudofonts and had to press a button if the words were semantically related (e.g., calendar‐date), or if the pseudofonts were identical (504 volumes, duration = 6.7 min) (Wilson et al. 2018). The difficulty of the task was adjusted based on participants' accuracy (Wilson et al. 2018).

Lesions were manually traced on FLAIR and T1‐weighted images by author P.E.T., a board‐certified neurologist, using ITK‐SNAP software (Yushkevich et al. 2016).

2.3. Preprocessing

Preprocessing was conducted with a pipeline leveraging several software packages: AFNI (Cox 1996), FSL (Jenkinson et al. 2012), and SPM12 Friston et al. (1994). Files were reconstructed with dcm2niix (Li et al. 2016). Fieldmaps were applied via FSL's applytopup (version 6.0.5.2). Volumes with excessive striping artifacts which reflect intra‐TR motion were flagged as bad and replaced by interpolating data from the nearest neighbor volume. Further steps included slice timing correction, motion correction, despiking, susceptibility distortion correction, detrending, scaling to a global mean of 10,000, and Gaussian smoothing with a 6 mm FWHM kernel.

For the semantic decision task data, we applied a temporal high‐pass filter at 0.008 Hz. Then, a general linear model (GLM) was fit using the FMRISTAT fmrilm function (Worsley et al. 2002). The task condition was modeled as a boxcar function convolved with the hemodynamic response function. The covariates of no interest included 6 head‐motion parameters, signals from white matter, cerebrospinal fluid, non‐brain global signal, and the first derivative of each confounder. The contrast of interest included semantic decision over pseudofont control task [1–1]. For the task‐free scan during movie watching, we applied a bandpass filter (0.008–0.09 Hz) after a GLM that included the same non‐task‐related covariates as in the semantic decision data preprocessing. Connectomes were derived from the residualized timeseries data.

After preprocessing, we used the Brainnetome atlas (Fan et al. 2016) to parcellate each participant's brain into 246 regions, with 123 parcels in each hemisphere. Lesioned parcels retaining fewer than 25 unlesioned voxels were excluded from the analyses on a per‐subject basis before computing parcel timecourses. Thus, only unlesioned voxels contributed to each parcel's mean BOLD signal. Since lesion locations vary across participants, no parcel was excluded for all stroke survivors. The parcel with the highest exclusion rate was excluded for 23 out of 85 stroke survivors. Only LH parcels were excluded; no RH parcel was excluded for any analysis. The exclusion of lesioned parcels reduces statistical power for edges that are associated with extensively lesioned LH areas; however, it does not affect the within‐RH analyses, which are the focus of this paper.

For each parcel, we computed the mean timecourse of the BOLD signal measured during the task‐free scan. Further, we calculated FC as pairwise Pearson correlations between the timeseries of each brain parcel and the timeseries of every other parcel across the entire brain. The resulting values were Fisher Z‐transformed to normalize the distribution of the correlation coefficients.

In this study, we excluded edges with a Fisher Z‐transformed correlation coefficient below 0.2 in either the control or stroke group. Currently, there is no consensus on how to interpret the negative correlations (Chai et al. 2012; Murphy et al. 2009). Additionally, the interpretation of alterations in anticorrelations is uncertain. We chose a 0.2 threshold to focus on the analyses of relatively stronger connections to capture changes that reflect robust functional reorganization.

2.4. Network Definition

The 246 Brainnetome‐defined parcels were further grouped into LN and non‐LN. We grouped the parcels to independently assess changes within the LN compared to those in non‐LN regions. We aimed to determine whether disruptions within the LH LN drive change specifically in RH homotopic regions. For the LN localization, we used a validated semantic decision task (Wilson et al. 2018). We identified parcels that overlapped with the areas of significant activation in controls in response to the semantic decision task and their RH homotopic parcels. As in our other recent work, the task‐activated parcels were further divided into LN and SN based on anatomical location and previous literature defining the canonical language areas (Friederici 2011; Price 2012; Turkeltaub et al. 2026). Parcels in lateral inferior frontal and temporoparietal areas, corresponding to classical perisylvian language areas, were included in LN, while parcels activated during the semantic decision task but located outside of canonical language regions were included in SN. This distinction highlights the difference between core language processing areas and the broader semantic network that is recruited by the language localizer task. All the parcels that were not included in the LN or SN were grouped and labeled as Rest of the Brain (RoB) (Figure 1, Table S2).

1. Language (green): 19 parcels in the superior and middle temporal gyri, posterior superior temporal sulci, inferior parietal lobules, and inferior frontal gyri.

2. Semantic (orange): 26 parcels in the middle, superior, and orbital frontal, inferior temporal and fusiform, parahippocampal, posterior cingulate gyri, amygdala, and hippocampus.

FIGURE 1.

Glass brain showing nodes included in language (green) and semantic (orange) subnetworks. The nodes are shown on the RH. R, right hemisphere.

For additional edge localization, we classified each parcel according to the resting‐state networks (RSN) as defined by Yeo et al. (2011) and by anatomical regions. This classification was used as a secondary descriptive approach to characterize which networks showed the greatest FC changes, complementing the primary LN, SN, and RoB analyses. The RSNs included the DMN, DAN, VAN, FPN, SMN, LIM, and VIS networks (Yeo et al. 2011). The assignment of parcels to RSNs was as suggested by the Brainnetome atlas. Parcels that were not assigned to any RSN included mainly subcortical parcels and were combined into a network labeled Subcortical. Anatomical divisions included frontal, temporal, parietal, occipital, insular, subcortical, and limbic, also following the Brainnetome atlas definition.

2.5. Behavioral Measures

This study included four behavioral tasks that measure orthogonal aspects of impaired language processing based on a factor analysis (Lacey et al. 2017). Lacey et al. (2017) administered 25 behavioral tests to 38 chronic LH stroke survivors and used principal component analysis to extract 4 factors that accounted for 81% of variance in the test scores. The factors corresponded to Word Finding/Fluency, Comprehension, Phonology/Working Memory Capacity, and Executive Function. We selected one measure per factor, thus ensuring broader coverage of the cognitive domains that are frequently affected by LH stroke. One behavioral score was selected for each component based on the factor loading scores from the Lacey et al. analysis: the WAB Auditory Verbal Comprehension (WAB AVComp) subscore (Kertesz 2007), Pseudoword Repetition (Fama, Henderson, et al. 2019), the Pyramids and Palm Trees test (PPT, Howard and Patterson 1992), and Picture Naming (Fama, Snider, et al. 2019). The WAB AVComp requires participants to respond to various types of verbal questions or instructions from the examiner. Specifically, participants answer yes/no questions, point to pictures that correspond to verbally presented words, and execute multi‐part verbal commands given by the examiner. The WAB AVComp tests the participant's ability to process and understand spoken language at multiple levels, from single words to more complex sentences. In the Pseudoword Repetition task, participants hear recordings of pseudowords that are 1, 2, or 3 syllables in length and are asked to repeat each pseudoword they hear out loud. The task evaluates phonological processing. PPT is a semantic association task that assesses participants' ability to understand the meanings of objects and their relationships with one another. In this task, participants are shown a target picture and two other pictures underneath it. They must choose the picture that is most closely related in meaning to the target above (Howard and Patterson 1992). In the Picture Naming task, participants are shown a series of images on a computer screen, one at a time, and are asked to name aloud each object as quickly and accurately as possible (Fama, Snider, et al. 2019). The descriptive statistics for the four behavioral measures of interest are presented in Table 1 and Figure S1.

2.6. Statistical Analysis

The analyses were conducted in MATLAB (The MathWorks Inc. 2023) and JASP (JASP Team 2025), an open‐source statistical software package. First, we conducted edgewise, independent‐sample two‐tailed t‐tests to evaluate differences in FC strength between the entire LH stroke group and the neurologically healthy control group.

Next, we analyzed whether the number of stronger RH connections in stroke survivors compared to controls relates to demographic factors (age, sex, handedness, education) and lesion features (lesion size and time post‐stroke). We conducted a regression analysis using these variables to predict the difference in the number of stronger connections per participant with strength above 2 SD and strength below 2 SD from the control mean. The analysis was conducted in JASP (JASP Team, 2025).

To further assess the effect of LH lesion size on FC changes post‐stroke, we divided the LH stroke group into two subgroups based on the median lesion size (78 cm³): a large stroke subgroup (n = 42, lesion size: mean = 179 cm³, median = 153 cm³, IQR = 119 cm³) and a small stroke subgroup (n = 43, lesion size: mean = 31 cm³, median = 25 cm³, IQR = 37 cm³) (Figure 2). We divided the stroke group into subgroups using the median split. Lesion size is typically right‐skewed, which makes the median a more robust division criterion than the mean. Additionally, this approach avoids the need to define an arbitrary cutoff and provides approximately equal group sizes. The median split approach has also been used in prior studies (e.g., Idesis et al. 2022, 2024). It should be noted that the right‐skewed distribution leads to greater variability in lesion sizes in the large lesion subgroup, which should be considered when interpreting the results.

FIGURE 2.

Lesion overlap maps for the small (left) and large lesion (right) groups.

We then conducted edgewise independent‐sample t‐tests to compare FC differences between each stroke subgroup and the controls, and between the large and small lesion groups. A chi‐squared test was used to analyze the difference in the proportion of FC change between the small and large lesion groups.

Next, we aimed to examine whether RH FC changes were distributed evenly across the whole hemisphere or concentrated in a specific area. To address this question, we mapped the edges showing significant changes (i.e., significant t‐statistic values as compared to controls after multiple comparison correction) in each of the two stroke groups, based on our predefined networks (LN, SN, RoB, see Network definition section for details). We calculated the proportion of stronger, weaker, and unchanged edges for intrahemispheric within‐ and between‐network connections, as well as interhemispheric homotopic connections in stroke survivors relative to controls. Further, we conducted similar localization analyses based on the RSNs and on anatomical regions (see Network definition section for details). We computed these metrics separately for small lesion vs. controls and large lesion vs. controls.

To analyze whether specific LH lesion location relates to the changes in RH FC, we conducted a support vector regression‐lesion symptom mapping analysis (SVR‐LSM) (Zhang et al. 2014) using a MATLAB‐based toolbox (DeMarco and Turkeltaub 2018). SVR‐LSM uses a machine learning algorithm to identify the relationship between the LH damage and the FC measures (i.e., difference between the number of connections above and below 2 SD from the control mean). We tested the hypothesis that there is a relationship between the LH lesion location and the RH FC. Lesion size was added as a covariate. The resulting SVR‐β were thresholded at p < 0.005 and underwent cluster correction at p < 0.05. Thresholds were determined based on 10,000 permutations. We estimated six models testing the relationship between the lesion location and the FC measure within each network (LN, SN, RoB) and between the networks (LN‐SN, LN‐RoB, SN‐RoB).

Finally, we aimed to determine whether the observed alterations in FC were associated with better language outcomes suggesting a compensatory pattern, or worse language outcomes suggesting a maladaptive pattern. To assess the relationship between FC and behavioral outcomes, we computed the difference for each stroke participant between the number of edges that are 2 SD above the control mean and the number of edges that are 2 SD below the control mean. The resulting scores reflect the net direction of deviation from the normative distribution, with a positive value showing that a participant has more hyper‐ than hypo‐connected edges within a given connectivity type (intrahemispheric within‐network, intrahemispheric between‐network, or interhemispheric homotopic) compared to controls. This difference was calculated separately for intrahemispheric within‐network FC (within LH and RH LN, SN, and RoB), for intrahemispheric between‐network FC (between LN and SN, SN and RoB, LN and RoB within LH and RH), and interhemispheric homotopic FC (LH to RH LN, SN, and RoB). Then, we computed partial correlations between each FC measure (i.e., number of edges above 2 SD minus number of edges below 2 SD from the control mean for each between‐ or within‐network relationship) and behavioral test scores (i.e., WAB AVComp subscore, Pseudoword Repetition, PPT, and Picture Naming), controlling for age, education, handedness, lesion size and time post‐stroke. For all the tests, we corrected for multiple comparisons using false discovery rate (FDR correction with a significance threshold of α = 0.05) (Benjamini and Hochberg 1995).

All the glass brain figures showing FC were visualized in BrainNet Viewer (Xia et al. 2013); lesion overlap maps were created in Mango (https://mangoviewer.com/mango.html).

3. Results

3.1. RH FC Is Higher in the LH Stroke Group Compared to Controls

First, we examined differences in RH FC across the entire group of stroke survivors vs. controls (Figure 3). Edgewise independent‐sample t‐tests comparing RH FC between the two groups indicated that the LH stroke group exhibited significantly higher FC in some of the RH edges. Of the 2738 RH connections analyzed, 565 connections (21%) showed significantly greater FC strength in the stroke group, whereas only 153 RH edges (6%) showed significantly lower FC compared to controls.

FIGURE 3.

Whole‐brain FC differences between the entire LH stroke and control groups. (A) Matrix with significant t‐statistic values for each analyzed connection. Green cells indicate edges that were examined but were not significant. White cells indicate edges that were not examined because their connectivity levels did not meet the inclusion threshold in either group. (B) Glass brains showing significant FC differences: Top—interhemispheric homotopic connections, only weaker connections are shown; bottom—within‐hemisphere weaker connections (left) and stronger connections (right). Warmer colors indicate greater FC, and cooler colors indicate lower FC in the LH stroke group relative to controls. Only significant FDR‐corrected results (α = 0.05) are shown. L, left hemisphere; R, right hemisphere.

For context, analyses of within‐LH and interhemispheric connectivity revealed widespread lower FC in the stroke group compared to controls. Specifically, 53% of LH edges showed weaker FC and 4% showed stronger FC. For interhemispheric edges, 51% were weaker and 4% were stronger in FC (Figure 3, see Supporting Information).

3.2. Relationship Between Demographic and Stroke Factors and RH FC

Next, we examined factors that relate to individual differences in RH FC using a multiple regression analysis (Table 2). LH lesion size was a significant predictor (p = 0.002), confirming that larger lesions were associated with a higher number of RH edges with higher FC in stroke survivors relative to controls (Figure 4). Age, sex, education, handedness, and stroke chronicity were not significant predictors of differences in RH connectivity.

TABLE 2.

Results of regression predicting RH FC in LH stroke group.

Effect	Estimate	SE	t	p	95% CI
					LL	UL
Intercept	−10.98	7.69	−1.43	0.16	−26.29	4.33
Sex (male)	−0.44	2.07	−0.21	0.83	−4.57	3.69
Age	−0.04	0.09	−0.39	0.70	−0.22	0.15
Education	0.42	0.36	1.19	0.24	−0.29	1.13
Handedness	0.03	0.02	1.67	0.10	−0.01	0.07
Stroke chronicity	0.05	0.95	0.05	0.96	−1.84	1.95
Lesion size	0.68	0.21	3.16	0.002	0.25	1.10

Note: R ² = 0.2, adjusted R ² = 0.12, F(6, 78) = 3.1, p = 0.01, RMSE = 8.8.

FIGURE 4.

Relationship between residuals of LH lesion size and the count of RH connections that were stronger than controls. The red line is a fitted regression line; blue dashed lines represent 95% confidence intervals. Each point represents an individual participant.

3.3. Stronger RH FC Primarily Occurs With Large LH Lesions

To facilitate the analyses below examining localization of FC differences after stroke, we also examined the effect of lesion size by dividing the stroke group into large versus small lesion subgroups. Edgewise independent‐sample t‐tests indicated a prevalence of stronger RH connections in the large lesion group compared to controls (Figure 5). Specifically, the large lesion group showed significantly greater FC in 914 out of 2738 RH edges (33%), while the small lesion group showed greater FC in only 102 edges (3.7%). In contrast, the large lesion group exhibited 83 significantly weaker edges (3%) compared to 100 weaker edges (3.7%) in the small lesion group, both relative to controls. A chi‐squared test revealed a significant difference in the proportion of stronger RH connections between the small and large lesion groups (χ ² (1) = 794.83, p < 0.001). The difference in the proportion of weaker connections in the RH was not significant (χ ² (1) = 1.45, p = 0.23).

FIGURE 5.

Difference in FC between the small lesion group and controls (A, B), the large lesion group and controls (C, D), and the small and large lesion groups (E, F). The matrices (A, C, E) represent t‐statistic values; warmer colors indicate stronger FC and cooler colors show weaker FC in the stroke groups vs. controls, or in the large lesion group compared to the small lesion group. Only significant results that survived FDR correction are displayed (FDR, α = 0.05). Green cells indicate edges that were examined but were not significant. White cells indicate edges that were not examined because their connectivity levels did not meet the inclusion threshold. The corresponding glass brains (B, D, F) show significant differences in homotopic (top) and within‐hemisphere connections (bottom). L, left hemisphere; R, right hemisphere.

Edgewise independent‐sample t‐tests between the large and small stroke groups showed significantly greater FC in the large lesion group in 132 out of 2738 RH edges (4.8%) compared to the small lesion group after correction for multiple comparisons (FDR, α = 0.05).

For context, we also examined within‐LH FC and interhemispheric FC in the lesion size subgroups. As expected, differences in within‐LH and interhemispheric homotopic FC were predominantly characterized by decreases in FC, with more pronounced decreases in the large lesion group (Figure 5, see Supporting Information).

3.4. Localization of FC Changes

3.4.1. Weaker LH FC and Stronger RH FC Are Infrequently Homotopic

Regarding the localization of FC changes, we first considered whether the weaker LH and stronger RH FC in stroke survivors relative to controls occurred in homotopic edges, which might be expected if stronger RH FC reflected either release of transcallosal inhibition or compensatory changes relating to dysfunction in LH networks. To examine this, we analyzed how many of the RH edges with greater FC than controls were homotopic to the LH edges with lower FC than controls. Of all the edges that had greater RH FC in the large lesion group compared to controls, only 14.5% had a corresponding homotopic LH edge that showed significantly lower FC than controls. In the small lesion group, none of the edges were homotopic (Figure 6).

FIGURE 6.

RH edges with greater FC than controls in the large stroke group that were homotopic to the LH edges with lower FC than controls. (A) RH FC matrix, the edges that showed both significantly greater FC in the RH and lower FC in the LH compared to controls are shown in red. (B) The glass brain displays the anatomical location of these RH connections. Left side—top view, top right side—lateral view, bottom right side—medial view. R, right hemisphere.

These results did not depend on the parcellation level. When we repeated this analysis using a gyrus‐level parcellation scale (24 regions per hemisphere), the number of RH strong connections that had a homotopic LH weaker connection was not significantly different from chance (observed value = 78 (49.7%), null mean = 82.9 (52.8%), 95% CI [76, 90], p = 0.18; see Supporting Information).

3.4.2. Localization of Within‐RH FC Changes

Next, we considered whether the differences in RH FC in stroke survivors vs. controls primarily occurred within the LN or beyond it. In the small lesion group, differences in RH FC were minimal; less than 6% of the total number of connections within‐ or between‐networks showed either greater or lower FC compared to controls (Figure 7A, Table S3). In the large lesion group (Figure 7B, Table S3), relatively few edges within the LN (8%) had greater FC than controls, compared to 21% of edges within the SN and 31% of edges within the RoB network. Overall, a larger proportion of between‐ than within‐network edges had greater FC in the large stroke group than controls (52% of the LN‐to‐RoB edges, 25% of LN‐to‐SN edges, and 35% of SN‐to‐RoB edges). These results suggest that differences in RH FC in individuals with aphasia due to large LH strokes are widely distributed and are not specific to the LN.

FIGURE 7.

The proportion of differences in RH FC for the small lesion group (A) and large lesion group (B) compared to controls. The outer arcs represent within‐network FC differences, the connections within the circle represent the between‐network FC differences. Red sections indicate the proportion of edges with greater FC in the stroke group than controls, blue sections indicate the proportion of edges with lower FC, and gray sections indicate the proportion of edges with no significant difference. The total width of each ribbon refers to the maximum number of connections above 0.2 either between‐ or within‐networks. The glass brains illustrate the corresponding networks: Language (green), Semantic (orange), and RoB (magenta). L, left hemisphere; R, right hemisphere.

Because many differences in RH FC occurred beyond the LN and SN defined by our localizer task, we further characterized the localization of FC differences by reorganizing the edges based on canonical RSNs (Yeo et al. 2011). This analysis again showed that the RH between‐network connections were more affected than within‐network connections. Specifically, 15% of all the within‐RSN connections and 25% of all the between‐network connections were stronger in the large lesion group compared to controls (test for difference of these proportions: χ ² (1) = 30.7, p < 0.001). The three networks with the largest proportion of stronger within‐network connections relative to controls—DAN (31%), Subcortical (28%) and VIS (24%)—also exhibited the largest proportion of stronger between‐network connections (i.e., the connections between DAN, Subcortical, or VIS to all the other RH networks) with 32%, 32%, 37% of edges showing greater FC in the large lesion group compared to controls, respectively (Figure S4A).

Since the networks examined above span distributed anatomical regions but strokes tend to affect contiguous structures, we considered whether differences in RH FC after stroke might occur mainly in certain anatomical regions. To address this, we reorganized the edges based on lobes and subcortical structures. As with the findings above showing that differences in FC tended to occur between networks, here we found that differences in FC tended to occur between anatomical regions rather than within them (χ ² (1) = 28.1, p < 0.001). The highest proportions of stronger RH between‐areal connections relative to controls were found for the subcortical areas (33%), temporal (33%), and occipital (36%) lobes. Subcortical areas (28%) and occipital lobe (34%) also had the highest proportion of stronger within‐areal connections (Figure S4B). These results suggest that differences in RH FC after large LH strokes tend to occur in long‐range posterior between‐network connections.

Decreases in within‐LH and interhemispheric connections were localized primarily in the LN and RoB in both small and large lesion groups relative to controls (Figure 7, see Supporting Information).

3.5. Relationship Between LH Lesion Location and RH FC Change

Next, we considered whether the differences in RH FC observed above were related to lesion location in addition to lesion size. Of the six SVR‐LSM models tested, one revealed a significant relationship between LH lesion location and changes in RH FC. Specifically, lesions located in the LH temporal region, primarily involving the MTG and STG, were associated with stronger FC than controls between the RH LN and SN (cluster p = 0.004; Bonferroni‐corrected α‐level across analyses was 0.008; Table 3, Figure 8). Overlaying the SVR‐LSM results with the Brainnetome Atlas revealed that most of the parcels where lesions were associated with greater RH FC were located within the LH LN.

TABLE 3.

LH parcels overlapping with lesion regions linked to stronger RH LN to SN connectivity in the stroke group.

Parcel label	Network	MNI coordinates			Overlap (%)
Temporal middle aSTS	LN	58	−16	−10	53
Temporal middle A21c	RoB	65	−29	−13	46
Temporal superior A22r	LN	56	−12	−5	23
Temporal middle A21r	RoB	51	6	−32	22
Temporal posteriorSTS rpSTS	LN	53	−37	3	19
Temporal superior A22c	LN	66	−20	6	18
Temporal superior A38l	LN	47	12	−20	17
Temporal inferior A20cl	RoB	61	−40	−17	15

Abbreviations: MNI, Montreal neurological institute; STS, superior temporal sulcus.

FIGURE 8.

LH regions where the lesion is significantly associated with stronger FC between RH LN and SN in stroke survivors relative to controls.

3.6. Association With Behavior

Finally, we tested whether the RH FC after stroke was related to aphasia outcomes after accounting for lesion size and demographic factors. None of the 24 analyses examining the relationship between the RH FC measures and behavioral scores returned significant results (Table S7).

By comparison, analyses of the LH FC measures revealed positive associations between LH RoB FC and Picture Naming (ρ = 0.32, p < 0.001, corrected p = 0.03) such that greater connectivity related to better naming performance (Figure 9A).

FIGURE 9.

Scatter plots showing the relationship between picture naming and LH RoB (A) and interhemispheric RoB (B).

Similarly, interhemispheric RoB FC showed significant positive associations with Picture Naming (ρ = 0.37, p < 0.001, corrected p = 0.01), indicating that greater interhemispheric RoB connectivity is associated with better Picture Naming (Figure 9B).

4. Discussion

This study investigated whether and how RH FC differs in LH stroke survivors and whether these alterations relate to language outcomes in individuals with post‐stroke aphasia. We found that extensive damage to the LH was associated with supranormal FC in the RH. However, the observed differences were primarily outside the RH regions that are homotopic to the LH LN, clustering in long‐range between‐network connections. Moreover, FC strength did not relate to language outcomes.

Below, we interpret between‐group differences in LH stroke survivors compared to controls as increases or decreases in FC from pre‐stroke levels, under the assumption that, if not for the stroke, both groups would have similar FC. There are several mechanisms that could help explain the observed changes after stroke. The most straightforward possible interpretation of higher RH FC after LH stroke is that it reflects neuroplastic changes triggered by a large brain injury. Neuroplasticity is the ability of the nervous system to change and adapt its activity by remodeling its structural and functional organization, as well as connections, in response to different internal or external stimuli (Cramer and Riley 2008). Such neuroplasticity can operate through two main types of cellular mechanisms that support recovery, homeostatic and Hebbian plasticity (Billot and Kiran 2024; Carmichael 2010). Homeostatic plasticity is associated with changes in synaptic scaling that maintain overall neural activity within an optimal range. When neural activity is reduced or increased for a prolonged period of time, synaptic strength is upregulated or downregulated to restore balance (Citri and Malenka 2008). Hebbian plasticity is based on the principle that synapses are strengthened when presynaptic and postsynaptic neurons are repeatedly co‐activated within a short time window and weakened when their activity is out of sync. This mechanism underlies long‐term potentiation and long‐term depression, which are important for learning and adaptation (Murphy and Corbett 2009). Thus, either of these mechanisms could contribute to the reconfiguration of synaptic communication that could manifest as increased FC in the contralesional hemisphere.

Changes at the cellular and synaptic levels that can affect the large‐scale networks are not confined to the lesioned hemisphere (Turkeltaub 2019). Animal models have shown that stroke‐induced cellular changes such as axonal sprouting, dendritic branching, synaptogenesis, gliogenesis, and hyperexcitability also occur in the contralesional hemisphere (for review, see Johansson 2000; Ward 2017). These findings suggest that recovery often involves both local and distributed structural remodeling, which can affect post‐stroke functional architecture. Thus, the observed differences in functional coactivation at rest could in part stem from these anatomical changes.

Another account of the neuroplastic changes at the large‐scale network level would predict that damage to critical nodes of the LH LN triggers upregulation of their RH homotopes, that is, spared areas that are capable of supporting language function but typically play a limited role in language (Chang and Lambon Ralph 2020; Stefaniak et al. 2020). This pattern has been reported in activation studies that find increased activation in homotopic regions following LH lesions (Blank et al. 2003; Skipper‐Kallal et al. 2017a, 2017b; Turkeltaub et al. 2011, 2026). If the same logic applied to FC, we would expect the largest FC increases to be localized to the RH homotopes of the damaged LH language areas. However, this is not what we found in the present study. We observed increases in between‐network connections distributed broadly across the RH, which suggest that post‐stroke network reorganization may involve more diffuse changes, rather than a focused homotopic shift within the LN. It is not necessarily expected that changes should occur predominantly within the homotopic RH LN since the strokes themselves are not limited to language regions in the LH, and connectivity was measured during movie‐watching, not during performance of a language task. Still, irrespective of their localization in relation to the LN, few of the increases in RH connectivity occurred in edges that were homotopic to those that decreased in the LH due to the lesions. In fact, of the LH networks, the LN showed the largest number of decreased edges, reflecting the localization of the strokes in the middle cerebral artery territory, while the RH LN showed the smallest number of increased edges.

This asymmetry is inconsistent with prevailing ideas about RH LN recruitment in aphasia. The theory that recruitment of RH processors results from transcallosal disinhibition predicts that damage to LH frontal, temporal, and parietal regions should disinhibit their RH homotopic areas and increase their FC with other RH areas, but this was not generally the case in our findings. This asymmetry also suggests that increased RH FC is not broadly related to recalibration of weights in bilateral networks due to unilateral LH lesions (Chang and Lambon Ralph 2020). Instead, large lesions were associated with FC increases in contralesional areas with more posterior distribution. In relation to the large‐scale RSNs, the DAN and VIS networks were the most affected both in terms of within‐ and between‐network connectivity. This pattern could reflect a potentiation of synaptic activity related to visual and attentional processing. Such disproportionate involvement could reflect increased reliance on visual and attentional resources when language processing is impaired, which is consistent with behavioral evidence showing beneficial effects of visual support for people with aphasia (Dietz et al. 2009; Grechuta et al. 2017; Waller and Darley 1978).

Within the broadly distributed pattern of changes described above, we did find that damage to LH STG and MTG regions was associated with increased FC between the LN and SN in the RH. This extends the earlier observation that changes in RH connectivity are not limited to homotopic regions since even this relatively circumscribed damage within the LH is not associated with focal increases in RH homotopic areas, but rather with increased between‐network connectivity in the RH. The LH lesion primarily affected the STG and MTG, areas that mostly fall within the LN and are associated with lexical‐semantic processing (Friederici 2011). A recent task‐related fMRI study in a subset of the cohort examined here found that lesions to similar left temporal regions resulted in upregulation of homotopic RH temporal lobe language nodes, and that greater RH activity, particularly in the mid‐STS region, related to better performance on a task involving lexical retrieval (Turkeltaub et al. 2026). The pattern of results in that study suggested a network reweighting mechanism by which RH language nodes that were already weakly engaged in language processing were upregulated by lesions damaging their LH counterparts. The present findings may suggest that loss of key LH temporal lobe hubs for lexical‐semantic processing may lead to a broader shift in network dynamics in the RH, linking language and semantic regions. However, in contrast to the task‐related activation findings, we did not find that greater RH LN‐to‐SN connectivity related to better performance on lexical‐semantic tasks. Thus, we cannot draw strong conclusions from the current findings, and further investigation is warranted to confirm and extend the lesion‐FC finding observed here.

Although the primary and secondary auditory regions involved in early spectrotemporal analysis of speech were not directly implicated by the LSM analysis, an alternative explanation for the relationship between LH temporal lesions and greater RH LN‐to‐SN connectivity is that the STG constitutes a part of a bilateral auditory speech processing network (Hickok and Poeppel 2007; Turkeltaub and Coslett 2010). The LH and RH STG areas work in concert to support normal function. Thus, damage to the LH STG disrupts a part of a bilateral system, and the RH STG may respond to the loss of its interhemispheric partner by reorganizing its coupling with other RH networks, which could manifest as increased RH LN‐to‐SN FC. Such increased coupling may reflect network‐level rebalancing connected to the loss of interhemispheric auditory input, rather than reorganization specific to language networks. Overall, we favor the explanation related to network reweighting of lexical‐semantic processors above given that the LSM findings are inferior to regions typically implicated in spectrotemporal processing of speech.

Another possibility is that this shift involves recruitment of domain‐general regions in the RH which have previously been connected to aphasia recovery (Brownsett et al. 2014; Geranmayeh et al. 2017; Jiang et al. 2025). A recent study by Jiang et al. (2025) showed that language recovery is associated with facilitation from domain‐general regions to the LN, both in the cases of frontal and temporo‐parietal lesions (Jiang et al. 2025). Specifically, they found that the RH multiple‐demand network, including the dorsolateral prefrontal cortex (DLPFC) modulated activity of the language areas after stroke. DLPFC partially overlaps with our defined SN, so it is possible that LH temporal damage leads to increased coupling between language‐specific and domain‐general systems in the RH. Why this should predominantly occur with temporal lobe lesions will require further investigation, examining both network reorganization hypotheses and behavioral hypotheses (e.g., do these lesions cause specific deficits that necessitate the recruitment of extra cognitive resources for language processing?).

Another interpretation is that the observed association between left temporal lesions and RH LN‐to‐SN connectivity is spurious and that the broad non‐specific increases in RH FC after LH stroke reflect a failure to reorganize systematically. Rather than reflecting compensatory reorganization, the widespread RH FC increases may reflect noise or non‐specific upregulation that has no functional effect. The most direct way to assess whether the RH FC increase reflects compensatory reorganization is its relationship to behavioral outcomes. We found no association between the RH FC and behavioral language outcomes after accounting for lesion size and demographic variables. The lack of association is not entirely surprising, since the FC increases in the RH were diffusely distributed across the networks, rather than concentrated in the LN, so they would not be expected to predict language outcomes. Here, we examined language measures reflecting the principal components from a comprehensive language battery (Lacey et al. 2017), and we acknowledge that these tasks do not target all aspects of language or other cognitive functions. Perhaps examining a broader range of behaviors would reveal additional relationships that would inform the nature of the widespread increased RH FC observed after large LH strokes.

Alongside explanations that assume FC differences reflect real changes in neural communication after stroke, we should also consider the possibility that the observed differences between the stroke and control groups are artifactual. This possibility warrants consideration especially because our recent task‐related fMRI study in a subset of the participants examined here found only modestly greater RH language activity in a minority of LH stroke survivors compared to controls, rather than the widespread large differences in RH FC observed here (Turkeltaub et al. 2026). First, the observed between‐group differences in RH FC could stem from a mathematical artifact causing inflated correlations rather than true differences in connectivity. Specifically, the loss of LH nodes and interhemispheric connections may change the general strength of timeseries correlations in the spared RH due to the unmixing of signals that contribute to the BOLD timecourse at each RH node. A given node's activity is related to the activity of all other nodes to which it is connected. As a result, the timecourse of BOLD activity at a given node is a mixture of the timecourses of all the connected nodes. Since FC is measured as the correlation between the BOLD timecourses of two nodes, the influence of other nodes outside that pair limits the degree of FC that can be observed. Effectively, to the extent that each node in the pair is connected to different nodes outside the pair, the influence of those other nodes modulates the measured FC between the pair. Thus, if nodes are removed by a lesion, then the measured connectivity between remaining nodes may increase, not because the underlying connections have changed, but because the lesioned nodes no longer influence the BOLD timecourses of the remaining nodes. In our data, the loss of interhemispheric connectivity from the LH might be expected to result in increased connectivity within the RH due to this unmixing effect. This effect should elevate FC values not just in the RH but also in unlesioned parts of the LH. Indeed, this seems to be the case in the large lesion group, in whom medial, anterior, and posterior within‐LH edges outside the lesion distribution increased along with the RH edges (Figure 5D).

The pattern of findings does not perfectly fit this unmixing explanation, however. Unmixing predicts that, after stroke, FC should increase most in the RH nodes that were strongly connected to damaged LH regions, because those nodes would be affected the most by the removal of interhemispheric variance. The densest loss of interhemispheric edges was in frontal, temporal, and parietal regions within the middle cerebral artery territory, so the largest between‐group differences in RH FC would be expected in homotopic areas, but this was not the case. Instead, the edges with greater connectivity in the stroke group were broadly distributed with few differences observed in the homotopic RH regions.

It is also possible that the observed changes in FC reflect artifactual hemodynamic processes triggered by large strokes. Neuronal activity is coupled with local blood flow, and increases in blood flow in areas of neuronal activity lead to changes in blood oxygenation, which is the basis of the BOLD signal measured by fMRI (Drew 2019). FC is inferred from spontaneous fluctuation of the BOLD signal at rest or during movie watching. However, the exact relationships between neural activity and the BOLD signal, as well as the ways in which stroke disrupts hemodynamic processes, especially in the intact hemisphere and at the chronic stage of recovery, are unclear. Some studies show that perfusion is not different from controls in the intact hemisphere (Ivanova et al. 2023), while others report hyperperfusion (Thompson et al. 2017). Yet others report a delay in hemodynamic response function in the intact hemisphere (Bonakdarpour et al. 2007). However, Siegel et al. (2016b) found that the effects of the stroke on the hemodynamic response occur primarily in the vascular distribution of the stroke and are minimal in the intact hemisphere (Siegel et al. 2016b). This suggests that possible alterations in the hemodynamics after stroke are unlikely to affect the FC measures in the RH.

In addition to examining the RH FC, we also examined the changes in LH and interhemispheric homotopic FC. We found that both were reduced after LH stroke, consistent with previous findings (New et al. 2015; Sandberg 2017; Siegel et al. 2016a; Tao et al. 2022; Tao and Rapp 2020). These changes are likely a consequence of structural damage to LH tissue, which leads to a loss or weakening of structural and functional connections within the LH and between homotopic regions. Importantly, we also found that decreased LH as well as interhemispheric homotopic FC in the RoB network related to picture naming deficits. This is possibly the most meaningful result from the clinical perspective, since anomia is the most prevalent and persistent symptom in aphasia (Crinion and Leff 2007). This FC‐behavior association is notable because RoB included the areas that are outside the canonical and non‐canonical language areas as defined in this study, specifically, RoB included medial and lateral occipital cortex, subcortical areas, cingulate and insular cortices, precuneus and postcentral gyrus, intraparietal sulcus, superior parietal lobule, parts of frontal and temporal cortices that were not included in LN or SN. From the perspective of picture naming, RoB includes a wide range of areas that have been implicated in subprocesses of picture naming, specifically including early visual perception and object recognition, as well as speech initiation and articulation (Whatmough and Chertkow 2013). A recent meta‐analysis of lesion‐symptom mapping studies identified four clusters that were significantly associated with picture naming. They included areas in the left anterior temporal lobe, posterior temporal lobe extending into inferior parietal regions, pre‐ and postcentral gyri, and MFG (Piai and Eikelboom 2023). Around half of the regions identified in this meta‐analysis were categorized as part of the RoB network in our study. Taken together, this suggests that picture naming relies not only on the integrity of the classical perisylvian language network but also on the coordinated functioning of a larger set of regions that support subprocesses of picture naming, for example, visual object recognition or articulation. The role of functional connections within and between RoB areas in anomia may be underappreciated and require further investigation. The presented behavioral findings suggest that future rehabilitation research should consider targeting connectivity within this larger network (i.e., RoB), for example, by including approaches that engage visual or sensorimotor pathways.

5. Limitations

This study includes several limitations. First, although the use of the movie‐watching fMRI scan is a strength of this study, this approach differs from the true resting‐state scan that other studies have used (Sandberg 2017; Tao et al. 2022; Yourganov et al. 2021; Zhu et al. 2014). This methodological difference may partly account for the differences in findings compared to previous studies, for example, if individuals with severe aphasia attend less to verbal elements of the movie and more to visual elements compared to controls. Second, our analyses linking FC to behavior relied on summary measures within broad networks that included many regions. This was done to reduce the number of tests performed, but it limited our ability to identify specific areas within or outside the LN that might relate to behavioral outcomes. Future work could use a more fine‐grained network definition or edgewise analyses to better understand the role of FC in aphasia recovery. Third, this study is cross‐sectional and includes data only from chronic stroke participants at a single time point. Thus, the findings may not generalize to the acute and subacute stages of recovery and cannot be used to infer longitudinal changes in FC. Future research should investigate in detail the alterations of FC across multiple time points and their association with behavioral outcomes. Fourth, the exclusion of lesioned LH parcels, while necessary to conduct interpretable analyses of LH FC, resulted in the exclusion of some LH parcels from analyses in some participants, more so in the large lesion group than in the small lesion group. Thus, the tests that included LH and interhemispheric FC varied in sample size and should be interpreted with caution. However, since no RH parcel was excluded, this does not affect our primary findings regarding RH FC. Finally, the lesion‐size distribution was right‐skewed, such that the large lesion group was more heterogeneous, which should be considered when interpreting the between‐group results.

6. Conclusions

Here, we have demonstrated that RH FC increases after damage to the LH in proportion to lesion size. These changes are broadly distributed across the RH, are located mostly outside of the homotopic RH LN areas, and have no relationship to the language outcomes examined here. The pattern of findings could reflect neuroplastic changes in brain networks after stroke, as suggested by a relationship between LH temporal lobe lesions and greater RH LN‐to‐SN connectivity, but also raises a question of whether at least some changes in FC in spared areas might reflect other effects unrelated to aphasia recovery. This possibility should be considered when examining FC changes in populations with brain lesions.

Funding

This work was supported by the National Institute on Deafness and other Communication Disorders grant R01DC014960 to PET.

Ethics Statement

The study was approved by the Institutional Review Board, and all participants provided written informed consent prior to enrollment.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Violin plot illustrating behavioral performance and stroke size. (A) Accuracy (%) on four behavioral measures: Western Aphasia Battery Auditory Verbal Comprehension (WAB AVComp) subscore, Pseudoword Repetition, the Pyramids and Palm Trees test (PPT), and Picture Naming. (B) Distribution of lesion sizes (cm³).

Figure S4: RH FC differences in the small (left) and large (right) groups compared to controls, grouped by resting‐state networks (A) and anatomical regions (B). The outer arcs represent within‐network FC changes, the connections within the circle represent the between‐network FC changes. Red sections indicate the proportion of edges with stronger FC, blue sections indicate the proportion of edges with weaker FC, and gray sections indicate the proportion of edges with no significant difference compared to controls. The total width of each ribbon refers to the maximum number of connections above 0.2 either between‐ or within‐networks.

Figure S8: Number of participants included in the analysis at each parcel, shown separately for the large and small lesion groups.

Table S2: Brainnetome parcels included in the Language and Semantic networks.

Table S3: Ratios of significant changes in the RH connections.

Table S5: Ratios of significant changes in the LH connections.

Table S6: Ratios of significant changes in homotopic interhemispheric connections.

Table S7: Partial correlations between FC measures and behavioral scores.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.