The right hemisphere’s capacity for language: evidence from primary progressive aphasia

Abstract

The role of the right hemisphere (RH) in core language processes is still a matter of intense debate. Most of the relevant evidence has come from studies of gray matter, with relatively little research on RH white matter (WM) connectivity. Using Diffusion Tensor Imaging-based tractography, the current work examined the role of the two hemispheres in language processing in 33 individuals with Primary Progressive Aphasia (PPA), aiming to better characterize the contribution of the RH to language processing in the context of left hemisphere (LH) damage. The findings confirm the impact of PPA on the integrity of the WM language tracts in the LH. Additionally, an examination of the relationship between tract integrity and language behaviors provides robust evidence of the involvement of the WM language tracts of both hemispheres in language processing in PPA. Importantly, this study provides novel evidence of a unique contribution of the RH to language processing (i.e. a contribution independent from that of the language-dominant LH). Finally, we provide evidence that the RH contribution is specific to language processing rather than being domain general. These findings allow us to better characterize the role of RH in language processing, particularly in the context of LH damage.

Introduction

While the relationship between the left hemisphere (LH) and language processing is well-established, the role of the right hemisphere (RH) in language processing is still a matter of intense debate. One position is that the RH only supports peripheral or paralinguistic processes such as emotional prosody (Weintraub et al. 1981; Brådvik et al. 1991; Weed and Fusaroli 2020) or perception of intonation contours (Blumstein and Cooper 1974), among others. A different position points to evidence of a role for the RH in core language functions such as semantic (Chee et al. 1999; Booth et al. 2002) and phonological (Crinion et al. 2003; Saur et al. 2006, Hartwigsen et al. 2010a, 2010b) processing. However, the evidence for this position is still quite scarce. Most evidence for the role of RH in core language functions comes from the study of RH gray matter (GM), focusing on RH activation patterns in healthy adults and the behavioral consequences of GM lesions. Additionally, the RH’s white matter (WM) tracts (the axon bundles connecting brain regions) constitute another even less researched, but potentially valuable, source of evidence regarding the RH’s role in language processing.

The present study examined characteristics of the WM language tracts in individuals with Primary Progressive Aphasia (PPA), with the goal of better characterizing the contribution of the RH to language processing in the context of primary LH damage. PPA is an age-related degenerative neurological syndrome that primarily affects the integrity of the LH (Mesulam 1982, 1987), and is characterized by a gradual deterioration of language functions (Gorno-Tempini et al. 2004; Neophytou et al. 2019; Sepelyak et al. 2011; Shim et al. 2012; Wilson et al. 2010). The majority of language research in PPA to date has focused on behavioral effects and their relationship to GM changes (Gorno-Tempini et al. 2004; Brambati et al. 2009, 2015; Rogalski et al. 2011; Sepelyak et al. 2011; Bonner and Grossman 2012; Henry et al. 2012, 2016; Shim et al. 2012; Grossman et al. 2013; Johnson et al. 2020), with more limited research examining WM properties in PPA (see for instance Agosta et al. 2010; Catani et al. 2013; Grossman et al. 2013; Luo et al. 2020; Mandelli et al. 2014; Schwindt et al. 2013; Wilson et al. 2010). Here, we specifically examined the impact of PPA on the integrity of the WM language tracts in the two hemispheres, as well as the relationship between the WM tract integrity of both hemispheres and language processing. Importantly, this study specifically evaluated if (at least under conditions of LH damage) the RH can make a unique contribution to language processing (i.e. a contribution that is independent of that of the language-dominant LH), as well as if the RH contribution is specific to language processing, as opposed to domain-general processing. The findings we report allow us to better characterize the role of the RH in language processing in the context of LH damage, and also have potential clinical value for the PPA population, as future neuromodulation studies might choose to target the RH on the basis of its role in language processing.

Diffusion tensor imaging and tractography

Analysis of WM data has greatly increased in the past couple of decades, with much of the research involving Diffusion Tensor Imaging (DTI). DTI, by evaluating how water diffuses in the brain, enables the localization and the reconstruction of WM fiber tracts, a technique known as tractography, and the specific characteristics of the water diffusion properties of these fibers provide the basis for inferences regarding their structural features and integrity.

A variety of WM metrics have been reported in the literature, including fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), radial diffusivity (RD), and volume (e.g. number of streamlines/fibers). In the context of neurological diseases, it is generally assumed that lower MD values and higher FA values indicate greater tract integrity (e.g. Galantucci et al. 2011; Agosta et al. 2013; Ivanova et al. 2016). Greater tract integrity is typically associated with more intact or efficient processing (de Zubicaray et al. 2011; Rollans et al. 2017; but see De Erausquin and Alba-Ferrara 2013; Meekes et al. 2019). Despite the fact that FA and MD values are expected to be complementary, previous work indicates that MD values might be a more sensitive measurement of WM changes in neurodegenerative diseases. For example, several investigations have reported MD changes accompanied by fewer or no FA changes in various neurodegenerative disease populations when compared to healthy controls (HC) (see Cherubini et al. 2010; Zheng et al. 2014; Li et al. 2015; Elahi et al. 2017), likely indicating that FA and MD capture different types of microstructural changes that may vary across different neurological diseases.

Several WM tracts have been associated with language processing. Perhaps the most widely discussed are the dorsal tracts: the arcuate fasciculus which is typically divided into three segments (the long, anterior and posterior segments), and the superior longitudinal fasciculus, which has been characterized as consisting of up to five components (I, II, III, arcuate/IV, and temporoparietal)¹. The different components of these tracts have been previously implicated in various language processes, such as semantic processing (Whitwell et al. 2010; Mandelli et al. 2014), syntax den Ouden et al. 2019; Galantucci et al. 2011; Wilson et al. 2011), spelling (Banfi et al. 2019) and phonological processing (Lebel and Beaulieu 2009; Yeatman et al. 2011). More ventrally located tracts often examined in language research are the uncinate fasciculus, the inferior longitudinal fasciculus, and the inferior fronto-occipital fasciculus. Prior studies have shown relationships between these tracts with, for example, semantic processing (Bello et al. 2008; Agosta et al. 2010; Whitwell et al. 2010; Galantucci et al. 2011; Catani et al. 2013; Basilakos et al. 2014; Almairac et al. 2015), syntax (Wilson et al. 2011; Mahoney et al. 2013; Ivanova et al. 2016), and spelling (Gebauer et al. 2012; Vandermosten et al. 2012; Banfi et al. 2019; Cheema et al. 2022). In addition, WM tracts that are sometimes, albeit less frequently, examined in language studies include the extreme capsule (Saur et al. 2008; Kümmerer et al. 2013; Kourtidou et al. 2021), the frontal aslant tract (Catani et al. 2013), and the left middle longitudinal fasciculus (Luo et al. 2020).

WM integrity and language processing in PPA

Previous studies comparing WM structural integrity in PPA and HC have reported several differences between the two groups for the various tracts discussed above, primarily in the LH (de Oliveira et al. 2011; Magnin et al. 2012; Agosta et al. 2013; Catani et al. 2013; Mandelli et al. 2014), but sometimes also for the homologous tracts in the RH (Agosta et al. 2010; Whitwell et al. 2010; Galantucci et al. 2011; Grossman et al. 2013; Mahoney et al. 2013; Schwindt et al. 2013; Iaccarino et al. 2015; D’Anna et al. 2016; Marcotte et al. 2017; Tetzloff et al. 2018; Bouchard et al. 2019). These studies often compared specific PPA variants to healthy individuals, and across studies, different methodologies were used to identify the different tracts. Despite these methodological differences, across all studies, the PPA groups exhibited lower FA/tract volume values and/or higher MD/AD/RD values compared to HC individuals in both hemispheres.

Investigations of the relationship between the WM integrity of the language tracts with behavioral measures of language functions in PPA are even further limited in number and have almost exclusively considered the WM tracts of the LH. A small set of studies also included the RH but found no significant associations with language processing (Wilson et al. 2010; Grossman et al. 2013; Luo et al. 2020). In the LH, naming/verbal fluency performance has been shown to be associated with one or more measures of WM integrity in the left uncinate fasciculus and the left frontal aslant tract (Catani et al. 2013), the left middle longitudinal fasciculus (Luo et al. 2020), and the left superior longitudinal fasciculus II and III (Mandelli et al. 2014). Single-word comprehension performance has been shown to be related to the integrity of the left inferior longitudinal fasciculus (Agosta et al. 2010; Mandelli et al. 2014), the left long arcuate fasciculus (Agosta et al. 2010), the left uncinate fasciculus (Catani et al. 2013), and the left middle longitudinal fasciculus (Luo et al. 2020). Finally, syntactic comprehension and production have been shown to have strong associations with the integrity of the left long arcuate fasciculus, as well as the left superior longitudinal fasciculus II and III (Wilson et al. 2010; Mandelli et al. 2014).

In summary, further investigations are needed to better understand if and how RH WM tract properties are associated with language processing in the PPA population. Note that, as we discuss in the next section, in other populations, such as HC and people with stroke aphasia, there has been evidence of a relationship between the RH and language processing.

The right hemisphere’s role in language processing

The idea that the RH contributes to language processing is not new (for a review, see Lindell 2006), yet the nature of the contribution is still a matter of debate. Findings of good language recovery after severe LH damage early in life, such as in prenatal and perinatal strokes and hemispherectomies, provide robust evidence that, under at least specific circumstances, the RH can instantiate a fully developed language system (Krynauw 1950; McFie 1961; White 1961; Lenneberg 1967; Guzzetta et al. 2008; Newport et al. 2017). However, despite evidence of RH engagement/activation during language tasks in healthy adults (e.g. Chee et al. 1999; Booth et al. 2002; Hartwigsen et al. 2010a, 2010b; Yablonski et al. 2019), RH damage does not seem to affect core language processes (Ross and Mesulam 1979; Ross 1981; Pell 1999). This suggests that, in the healthy adult brain, RH language processes are not normally necessary for language processing and, therefore, that they are, at best, redundant. The fact that LH damage actually results in language deficits also indicates that RH language processes are also normally not sufficient for successful language processing. This implies that any redundant RH capacities in the healthy adult brain are, at best, not fully developed. This still leaves open the possibility that, in the context of LH damage to the adult brain, the RH may make nonredundant contributions or expand its capacity for language processing. Thus, the focus of the current study is specifically on furthering our understanding of the RH’s language capacities in the face of damage to the language-dominant LH.

Evidence relevant to understanding the RH’s contribution to language processing in the context of LH damage comes from neuroimaging studies of language reorganization in the brain after LH brain injury showing activation of RH brain regions in language processing that either is not found in healthy individuals or is greater in brain-injured compared to healthy individuals (Karbe et al. 1998; Skipper-Kallal et al. 2017; Kiran et al. 2019). For example, Tao and Rapp (2019, 2020) reported changes in functional connectivity in both the LH and the RH that were associated with behavioral improvements from before to after language therapy. Brain activation studies also have provided evidence for the dynamic contributions of the RH over time during recovery from LH damage. In a seminal longitudinal study, Saur et al. (2006) showed that, after stroke, RH activation was reduced in the acute phase, it then increased in the early subacute phase with peak activation in the right frontal cortex and then, in the chronic phase, normalized (compared to HC individuals), with a re-shift of peak activation back to the LH language regions (see also, Hartwigsen and Saur 2019). There is also relevant evidence of RH contributions to language processing from neurostimulation studies showing, for example, that Transcranial Magnetic Stimulation (TMS) administered to RH language areas can disrupt language processing in the stroke-damaged LH but not in the healthy brain (Thiel et al. 2006).

Furthermore, there is evidence of structural changes in the RH that have been shown to be related to changes in language behavior. With regard to gray matter, Xing et al. (2016) showed that LH stroke survivors had increased RH gray matter volume relative to healthy individuals and that this increase was correlated with spontaneous speech, naming, and repetition performance. Along similar lines, Hope et al. (2017), in a longitudinal study, showed a correlation between the rate of behavioral change in naming and the rate of gray matter structural change in the RH of individuals with post-stroke aphasia. There is also a small amount of data from studies examining RH WM properties, following stroke. For example, Forkel et al. (2014) reported that, in post-stroke aphasia, the volume of the right long arcuate fasciculus was positively correlated with longitudinal language recovery. Along similar lines, there have also been studies reporting that FA values of the right long arcuate fasciculus were associated with long-term effects of Melodic Intonation Therapy² (Schlaug et al. 2009, 2010; Wan et al. 2014).

Finally, some work has more directly considered the relationship between LH damage and RH recruitment for language. For example, there is evidence showing that the strength of the RH and language relationship depends on the extent of LH damage (for discussion see Anglade et al. 2014 and Hamilton et al. 2011). Karbe et al. (1998) showed activation of RH regions in individuals with large LH infarcts and severe aphasia that were not activated in individuals with smaller LH infarcts and milder aphasia symptoms. More recently, a study of WM properties in individuals with post-stroke aphasia reported a correlation between the degree of integrity of RH WM tracts and language performance that differed between subgroups with different degrees of LH damage, pointing to a compensatory role of the RH WM tracts (Kourtidou et al. 2021).

Despite these findings, the nature of the RH functionality is not well-understood. Importantly, we do not understand the extent to which these RH processes function independently of remaining LH functions. To determine this, it would be important to establish that the adult RH, at least under certain conditions, makes unique contributions (above and beyond LH contributions) to explaining variance in language behaviors. Another important limitation of current reports on contributions of the RH to language processing is that they do not specifically examine if the observed relationships between RH properties and language behaviors are specific to language. This leaves open the possibility that these RH capacities could be domain-general, facilitating performance not only for language tasks, but for other cognitive functions as well. We attempt to address both of these issues in the current investigation. In doing so, this work provides a better understanding of the nature of the contributions of the RH to language processing that is relevant both for our basic scientific understanding of the neural underpinnings of language processing as well as for its clinical applications.

Current study

By examining a population with primarily LH damage, the study aims to further our understanding of the nature of limited language processes that the RH might support in the context of LH damage, with a novel focus on WM integrity. Specifically, in a group of individuals with PPA, we studied: (i) the WM integrity of the LH language tracts and their RH homologs and (ii) the overall relationship between language skills and WM integrity in the two hemispheres. Also, in order to better characterize RH language processing, we more specifically examined (iii) the degree of the domain-specificity of the RH contributions (i.e., whether they specifically supported language processes or, instead, provided support to cognitive processes more generally), and (iv) whether the RH uniquely contributed to language processing, independently of LH contributions.

Materials and methods

Participants

The current study included data from 33 individuals (16 female) with clinically diagnosed PPA with ages ranging from 51 to 80 (mean = 66; standard deviation = 6.8). These individuals were enrolled in a treatment investigation at the Johns Hopkins Medical Institutions (JHMI) (see Clinical Trial Identifier: NCT02606422 on clinicaltrials.gov), for which they were evaluated on a set of speech and language tests and magnetic resonance imaging (MRI) protocols. PPA diagnosis was determined based on the Gorno-Tempini et al. (2011) consensus guidelines. All participants (or their caregivers) gave written informed consent, and the study was approved by the JHMI institutional review board. A control group consisted of 20 age-matched neurologically HC (15 female) ranging in age from 46 to 81 years (mean = 69.3). The HC individuals were enrolled in the BIOCARD study (Marilyn Albert, PI), an ongoing longitudinal investigation, supported by the National Institute of Health (grant number: U19 AG033655).

Behavioral data

For the PPA group only, spelling, naming, and syntactic comprehension were assessed using a spelling to dictation task, two picture naming tasks, and a sentence comprehension task, respectively. The spelling to dictation task included both real words and pseudowords, ranging from 73 to 134 and 19 to 34 items, respectively (Goodman and Caramazza 1985). For each item, spelling accuracy was evaluated at the letter level, providing percentage accuracy scores per item that were averaged across items to get an overall spelling accuracy score³. The two spoken picture naming tasks were the Boston Naming Test (Kaplan et al. 2001) and the Hopkins Assessment for Naming Actions (Breining et al. 2015), which evaluate the naming of nouns and verbs, respectively. A percentage accuracy score was calculated for each task and the two scores were then averaged together to obtain an overall naming score for each participant. Syntactic comprehension was evaluated with the SOAP test (Love and Oster 2002), which assesses the comprehension of subject-relative, object-relative, active and passive voice sentences. A percentage accuracy score was calculated by averaging together the correct responses across all four sub-categories of the test. In addition to the language tasks, Spatial-Span (SpSpan) Forward and Backward tasks (i.e. the Corsi block-tapping task; Corsi 1972) were used to assess non-language cognitive functioning. See Supplementary Material Section 1 (Table S1) for the data from these tasks for all PPA participants.

MRI data acquisition

Magnetization-prepared rapid acquisition with gradient echo (MPRAGE) structural scans and DTI scans were acquired on a 3T Philips Achieva MRI scanner with a 32-channel head coil at the FM Kirby Research Center for Functional Brain Imaging at the Johns Hopkins Kennedy Krieger Institute. For MPRAGE, we used a 3D inversion recovery sequence with the following parameters: repetition time (TR)/echo time (TE)/inversion time (TI) = 8.1/3.7/842 ms, resolution = 1 × 1 × 1 mm³, field of view (FOV) = 224 × 224 mm, acquired in the axial plane, SENSE factor = 2, flip angle = 8°, acquisition time = 4.5 min. For DTI, we used a multislice, single-shot spin-echo echo-planar imaging sequence providing whole-brain coverage with the following parameters: TR/TE = 7,324/75 ms, SENSE factor = 2.5, FOV = 212 × 212, matrix size = 96 × 96 reconstructed to 256 × 256, and acquisition time = 4.27 min. Each DTI scan included 32 diffusion-weighted volumes, each with different noncollinear diffusion directions with b = 700 s/mm², and one volume with no diffusion weighting (b = 0 s/mm²). Two identical DTI datasets were consecutively acquired for each subject to increase the signal-to-noise ratio and were concatenated for the analysis.

DTI data analysis

The diffusion data were analyzed and visualized using ExploreDTI (Leemans et al. 2009). The preprocessing consisted of (i) signal drift correction (Vos et al. 2017), (ii) Gibbs ringing correction (Perrone et al. 2015), (iii) diffusion tensor estimation using a linear least squares method, and (iv) eddy current distortion and subject motion correction (Leemans and Jones 2009) that was carried out by co-registering the DTI data on the individual subject, unnormalized, and high-resolution structural image (i.e. a T1-weighted image). The unnormalized T1-weighted image was selected for co-registration over normalized images because brain malformations caused by the disease could cause distortions in the normalized brain that could, in turn, affect tract reconstruction and segmentation (Brett et al. 2001). As a result, all DTI data analyses were performed in native space. Once the data were preprocessed, whole-brain tractography was performed with a deterministic tracking algorithm, estimating the principal diffusion orientation at each seed point. Streamlines were propagated with a step-size of 1 mm along this principal direction. In this way, a pathway was traced through the DTI data until either the FA values fell below 0.2 or the direction of the pathway changed through an angle >30°.

For this study, we considered eight bilateral tracts: the three segments of the arcuate fasciculus (long, anterior, posterior), anterior and posterior inferior fronto-occipital fasciculus, inferior longitudinal fasciculus, uncinate fasciculus, and the spinothalamic tract—see Fig. 1. The inferior fronto-occipital fasciculus, inferior longitudinal fasciculus, uncinate fasciculus, and the long, anterior, and posterior segments of the arcuate fasciculus were selected as the language tracts of interest, based on the abundance of evidence implicating these tracts in the LH in language processing across populations. The spino-thalamic tract was included as a non-language “control” tract to investigate if WM disruption was limited to language tracts affected by PPA, or if instead, the observed WM disruption might be due to general factors (such as age) that would affect WM integrity more generally throughout the hemisphere. The spino-thalamic tract has not been implicated in language processing in the past and is instead associated with somatosensory function, especially with pain perception (Hodge and Apkarian 1990)⁴.

Fig. 1.

The tracts investigated in the current study. IFOF = inferior Fronto-occipital fasciculus; ILF = inferior longitudinal fasciculus; UF = uncinate fasciculus; AF = arcuate fasciculus; STT = spino-thalamic tract. The dotted red line across the IFOF represents the division of the tract into anterior and posterior components.

The fiber tracts of interest were isolated by utilizing the hand-drawn region-of-interest (ROI) selection method (Bucci et al. 2013; Kaur et al. 2014). Representative ROIs are shown in Supplementary Material Section 2 (Fig. S1). For extracting the inferior fronto-occipital fasciculus, the inferior longitudinal fasciculus, and the uncinate fasciculus we followed the guidelines of Wakana et al. (2007), and for delineating the three segments of the arcuate fasciculus, we followed Catani et al. (2005). For the spino-thalamic tract extraction, we followed Hong et al. (2010).

MD values were selected as the primary measure of interest because, as indicated in the Introduction, several investigations have shown MD changes in the context of fewer or no FA changes in various neurodegenerative populations when compared to HC. Although the precise causes of these differences are not easy to identify, it has been argued that increase in MD values may reflect enlargement in extracellular space due to, for example, the myelin loss associated with neurodegeneration (see Cherubini et al. 2010; Zheng et al. 2014; Li et al. 2015; Elahi et al. 2017).

For each participant’s tract, a mean MD value was calculated by averaging the MD values of all the voxels in the tract⁵. These mean MD values were used for all the statistical analyses discussed below. See Supplementary Material Section 8 for the results of statistical analyses that were also carried out using FA/AD/RD and volume values (Tables S7 and S8).

It is important to note that there is considerable controversy in the literature with respect to the identification and naming of some of these tracts. As discussed in the Introduction, some tracts are referred to in the literature interchangeably, as arcuate fasciculus segments or superior longitudinal fasciculus components, depending on the tract segmentation protocol and other researcher preferences. While there is no study explicitly investigating the overlap between arcuate fasciculus and superior longitudinal fasciculus, reviews of the literature suggest that: the long arcuate fasciculus corresponds to the superior longitudinal fasciculus-arcuate, the anterior arcuate fasciculus to the superior longitudinal fasciculus II and/or III, and the posterior arcuate fasciculus to the superior longitudinal fasciculus-temporoparietal (Becker et al. 2022; Janelle et al. 2022). In our study, we use the arcuate fasciculus terminology because: (i) separating the superior longitudinal fasciculus II from the superior longitudinal fasciculus III was not possible for all the participants, therefore we extracted only one bundle of fibers, the anterior arcuate fasciculus, that included the fibers of both the superior longitudinal fasciculus II and III; (ii) we did not include the superior longitudinal fasciculus I which, based on the available literature, is the superior longitudinal fasciculus component least implicated in language processing.

Controversy also surrounds the distinction between inferior fronto-occipital fasciculus and the extreme capsule. The inferior fronto-occipital fasciculus connects the occipital lobe to the orbitofrontal cortex, while the extreme capsule is usually identified as a bundle of fibers that begins from the middle part of the superior temporal gyrus, lies dorsal to the uncinate fasciculus and terminates in the pars triangularis (BA45) (Petrides and Pandya 1988; Frey et al. 2008). Given the proximity of the extreme capsule to the anterior parts of the inferior fronto-occipital fasciculus it has been extremely difficult to distinguish the two sets of fibers (Friederici 2009)⁶. In the current study, in an attempt to capture distinct functions of the anterior portions of the inferior fronto-occipital fasciculus that might correspond to the extreme capsule, we divided the inferior fronto-occipital fasciculus into anterior and posterior segments (See Fig. 1).

Statistical analyses

Analysis 1 examined the effects of PPA on WM integrity in the two hemispheres, while Analyses 2–4 examined the relationship between WM integrity in the two hemispheres and behavioral measures. Analysis 2 examined the association between the WM integrity of the language tracts and language performance separately for each of the two hemispheres. Analysis 3 evaluated the domain-specificity of the WM language networks in the two hemispheres by examining the relationship between WM integrity and performance on a spatial working memory task. Finally, Analysis 4 specifically examined the unique contribution of each hemisphere’s WM tract integrity in explaining variability in language performance. We use the term “unique contribution” to refer to the variability in language behavior that is associated only with the variability in the WM integrity of one hemisphere, independently of any variability that could be associated with the homologous hemisphere. Note that Analyses 2, 3, and 4 evaluated data from the PPA group only⁷.

All analyses were performed in R (R Core Team 2013), using the “stats” package for the linear regression models and the “lme4” package for the linear mixed-effects models (LMEM) (v1.1-21; Bates et al. 2014). The variance inflation factor (VIF) was calculated for all models using the package “car” (v3.0-3; Fox 2015; Fox and Monette 1992). All VIFs were below 3, while a problematic amount of collinearity is usually indicated when VIF > 5 (James et al. 2013). R² values for linear regression models reflect the adjusted R², extracted using the package “stats” (v3.5.2; R Core Team 2013) for Analyses 2 and 4 and the package “relaimpo” (v2.2-6; Groemping 2007) for Analysis 3. R² values for LMEMs reflect the conditional R² extracted using the package “MuMIn” (v1.42.1; Barton 2018).

Analysis 1: the effects of PPA on WM tract integrity

This analysis considered the LH language tracts and their RH homologs, as well as the spino-thalamic tract in both hemispheres. Two LMEMs (Baayen et al. 2008), one per hemisphere, were evaluated to determine if the MD values of the WM tracts differed between the individuals in the PPA and the HC groups. As mentioned earlier, MD values were selected as the primary measure of interest but analyses for FA/AD/RD values and volume were also carried out and are reported in Supplementary Material Section 8 (Table S7). For each model, the dependent variable consisted of all individual participant MD values for each of the seven language tracts. Therefore, each participant contributed seven data points in each model (i.e. one data point per tract, per hemisphere). Fixed effects were Group (PPA vs. HC), Tract Type (language vs. non-language), the interaction of the two, as well as Age and Gender (to account for the differences in gender breakdown between the two groups). Random effects included random intercepts for Tract and Subject, as well as random slopes of Group by-Tract. See Supplementary Material Section 4 for the specification of the full model structure.

In this analysis, we were interested in the fixed effect of Group, as well as in the interaction of Group *  Tract Type. In other words, we investigated whether the two groups (PPA and HC) were on average different from one another with respect to the MD values of the language and the non-language tracts, as well as whether the differences between groups were different for the language and non-language tracts. Investigating whether there were statistically significant differences between the two groups with respect to specific tracts was beyond the scope of this analysis.

Analysis 2: association between WM tract integrity and language performance for each hemisphere

For Analysis 2 we compared pairs of regression models with and without language variables (henceforth, the language and null models, respectively) to test for a relationship between language performance and WM tract integrity by looking at the change in variability explained (i.e. change in R²) when language performance variables were included (i.e. the variance in WM in each hemisphere that is specifically associated with language performance). For the language tracts, we evaluated a total of four LMEMs (i.e. one null and one language model per hemisphere). For both the null and the language models, the dependent variable consisted of the MD values of the PPA individuals for each language tract in each hemisphere. Therefore, each participant contributed seven data points to each model (i.e. one data point per tract, per hemisphere). The null models only included Age⁸ as a fixed effect, while the language models included Age as well as variables corresponding to the behavioral scores for the three language domains: Naming, Syntax, and Spelling. For both the null and the language models the random effects were by-Tract and by-Subject intercepts, while for the language models, random effects also included by-Tract slopes for each of the three language variables. For the non-language tract, we compared null and language simple linear regression models in each hemisphere that had the same structure as the LMEMs described above but did not include the random effects. See Supplementary Material Section 4 for the specification of the full model structure. R² values for linear regression models reflect the adjusted R², while R² values for LMEMs reflect the conditional R².

Analysis 3: domain-specificity of the WM language networks in the two hemispheres

This analysis evaluated the domain-specificity of the WM language networks in the two hemispheres to ascertain if any relationships between language processing and WM integrity identified in Analysis 2 reflected a relationship with language specifically, or whether it could reflect a relation with non-language cognitive functioning more generally. Non-language, executive cognitive functioning was indexed by performance on a SpSpan Forward and a SpSpan Backward task. There is a consensus that executive processing (particularly in the visuospatial domain; Jonides et al. 1993; Thiebaut de Schotten et al. 2011) is primarily supported by the RH. Therefore, if a relationship between RH and language processing is found over and above a relationship with non-language cognitive functioning (as captured by the spatial processing tasks), it would constitute strong evidence for the contribution of the RH specifically to language processing.

To that end, similar to Analysis 2, we compared LMEMs with and without language performance variables to evaluate change in variability explained (i.e. change in R²) when language performance variables were included. Specifically, we compared models with Age and the two SpSpan scores (Forward and Backward) (henceforth SpSpan models) versus LMEMs with Age, the two SpSpan scores, and the Language scores of the three language tasks as fixed effects (henceforth, SpSpan +Language models), separately for each hemisphere. For both SpSpan and SpSpan +Language models, the dependent variable was the MD values of the PPA individuals for each language tract in each hemisphere. As before, each participant contributed seven data points in each model (i.e. one data point per tract, per hemisphere). Random effects were by-Tract and by-Subject intercepts for all models, as well as by-Tract slopes for the SpSpan scores for the SpSpan models, and by-Tract slopes for SpSpan scores and language variables for the SpSpan +Language models. See Supplementary Section 4 for the specification of the full model structure.

A significant difference between SpSpan and SpSpan +Language models would indicate a relationship between language scores and tract integrity over and above the relationship between SpSpan scores and tract integrity. R² values for LMEMs reflect the conditional R².

Analysis 4: unique variance in language performance explained by each hemisphere’s language tracts

The goal of Analysis 2 was to capture the overall relationship of language processing with the WM integrity of language tracts for each hemisphere. The WM integrity of tracts is often highly correlated across the two hemispheres, with interhemispheric correlation values, in this study, ranging from 0.41 to 0.86 (see Supplementary Material Section 3—Table S2). This raises the concern that any relationship we might find in Analysis 2 between RH WM tract integrity and language processing reflects the LH-language relationship and that a RH effect does not actually represent an independent RH contribution to language processing. To address this concern, Analysis 4 specifically evaluated the unique contribution of each hemisphere’s tract integrity to explaining the observed variability in language behaviors.

To that end, we examined linear regression models that simultaneously included the MD values of both the LH language tracts and their RH homologs as independent variables, with language domain scores as the dependent variable, and Age as a covariate. Specifically, we examined a separate model for each of the three language domains and each of the seven tracts, for a total of 21 models. Each model provided an estimate of the extent to which the tract-specific variability in the MD values of the LH and RH tracts explained the variability in language scores (i.e. unique-R²). This was achieved by comparing the models with both LH and RH values to reduced models that had only RH or LH values (but not both). In other words, for each of the 21 models that had both LH and RH values two comparisons were performed: one comparison with a model that only had the LH values (i.e. providing the unique-R² of the RH) and one comparison with a model that only had the RH values (i.e. providing the unique-R² of the LH). These unique-R² values estimate the variability that is uniquely explained by a tract in each hemisphere, independently of any variability explained jointly with its homologous tract. For each pair of homologous tracts, we then calculated the hemispheric difference in unique contribution by subtracting the unique-R² of each left tract from the unique-R² of its right homolog. A negative unique-R² difference value indicates that a LH tract explained more unique variance in language performance than its RH homolog, and a positive unique-R² difference indicates the reverse. These difference values were then evaluated to determine: (i) if the magnitude of the differences significantly favored the LH or the RH for certain tracts and (ii) if the distribution (i.e. the number) of unique-R² difference values across tracts significantly favored the LH or the RH (regardless of the magnitude). R² values for linear regression models reflect the adjusted R².

In sum, in the “magnitude” analysis, the outcome of interest is the amount of unique-R² difference for each of the 21 models, while in the “distribution” analysis, the outcome of interest is the number of tracts that favor one hemisphere over the other. In other words, for the “magnitude” analysis, we aimed to evaluate if the amount of unique-R² difference we obtained in each of the 21 models was greater than what could have arisen by chance, while for the “distribution” analysis we aimed to evaluate if the number of tracts we found that explained more variance in one hemisphere than the other was a number larger than what could have arisen by chance. The statistical significance of these effects was evaluated via Monte Carlo permutation testing, with 10,000 permutations. For each permutation, the MD values were randomly shuffled across all participants, tracts, and hemispheres, the same 21 models were constructed, and the unique-R² difference values were extracted from the shuffled data. For both sets of analyses, the permutation approach allowed the observed magnitude and distribution values to be compared to distributions of 10,000 chance values, accounting for the multiple comparisons.

Results

Analysis 1: the effects of PPA on WM tract integrity

For the LH, the results showed a statistically significant main effect of Group (P = 0.004), with higher MD values for the PPA compared to the HC group, as well as a statistically significant interaction effect of Group * Tract Type (P = 0.014), showing that while the PPA group had higher MD values compared to the HC group across all tracts, this difference was greater for the language tracts compared to the non-language tract (Fig. 2A). For the RH, neither the main effect of Group, nor the interaction of Group * Tract Type was statistically significant (Fig. 2B). Beta coefficients for the effect of Group by-tract are graphically depicted in Fig. 3. Statistical significance for the effect of Group separately for each tract was determined with a post hoc analysis comprising a set of simple linear regression models reported in Supplementary Material Section 5 (Table S4). Similar effects were also found for FA/AD/RD values as well as for volume (see Supplementary Material Section 8—Table S8).

Fig. 2.

Depiction of the interaction effect between Group (PPA vs. HC) and Tract Type (language vs. non-language) on MD values, separately for the left (A) and the right (B) hemisphere.

Fig. 3.

Beta coefficients for the effect of Group by-Tract, a random effect in the LMEMs evaluating the effect of Group on MD values. The higher the beta coefficient value (i.e. blue color), the higher the MD value of the PPA group compared to the HC group (see Supplementary Material Section 5—Table S3 for specific beta coefficients). Higher MD values are typically interpreted to indicate lower tract integrity. The spino-thalamic tract, the non-language tract, is not depicted as no significant differences were found between the two groups for this tract in either of the two hemispheres.

Analysis 2: association between WM tract integrity and language performance for each hemisphere

The changes in R² values between the null and the language models (i.e. variability in MD values uniquely associated with language performance) of each hemisphere are presented in Table 1, for the language tracts⁹. The R² attributable to language was 10.3% and 5% for the LH and RH, respectively. These values were significant for both hemispheres (LH: P < 0.001; RH: P < 0.001), indicating a significant association between language scores and tract integrity in each hemisphere. For the non-language tract there were no significant differences between the null and language models in either of the two hemispheres (LH: P = 0.692; RH: P = 0.851).

Table 1.

R² values for the null and language models for each hemisphere and their differences.

	Left	Right
Null model	63.9%	73.9%
Language model	74.2%	78.9%
Difference	10.3%	5%

The same analyses were also carried out using FA/AD/RD values and volume, instead of MD values (see Supplementary Material Section 8) and, as expected given previous findings in neurodegenerative populations (Cherubini et al. 2010; Zheng et al. 2014; Li et al. 2015; Elahi et al. 2017), these other measures were less sensitive than MD.

Analysis 3: domain-specificity of the WM language networks in the two hemispheres

The change in R² values between the SpSpan and the SpSpan +Language models of each hemisphere are presented in Table 2. The change in R² between the SpSpan and the SpSpan +Language models were 10.7% and 4.2% for the LH and RH, respectively. These values were significant for both hemispheres (LH: P < 0.001; RH: P = 0.04), indicating a significant relationship between language scores and tract integrity in both hemispheres that was beyond any contribution from domain-general cognitive function/s that SpSpan¹⁰ and language processing might share.

Table 2.

R² values for the SpSpan and the SpSpan +language models for each hemisphere and their difference.

	Left	Right
SpSpan model	64.7%	75.2%
SpSpan +Language model	75.4%	79.4%
Difference	10.7%	4.2%

Analysis 4: unique variance in language performance explained by each hemisphere’s language tracts

Table 3 reports, for each tract in each hemisphere, the unique-R² representing the amount of variability in language performance associated with the MD values of the RH and LH homolog pairs (also graphically depicted in Fig. 4), as well as the unique-R² difference value. Two findings in particular stand out from visual inspection of the table. First, there are two tracts, the uncinate fasciculus and the anterior inferior fronto-occipital fasciculus, for which the magnitude of the LH unique R² values were much higher than the RH values (i.e. the unique-R² difference was highly negative). Second, the majority of tracts (67%) had greater unique-R² in the RH than the LH. We evaluated each of these findings statistically.

Table 3.

Unique-R² of each language domain explained by the MD values of the LH and RH language tracts, and the difference in the unique-R² between the homologous tracts. Positive difference values indicate higher unique-R² for the RH, while negative difference values indicate higher unique-R² for the LH. Asterisks (*) indicate statistically significant effects (P < 0.05).

		Unique LH	Unique RH	Difference (RH–LH)
SYNTAX	ILF	2.33%	4.43%	2.10%
	UF	5.99%	5.85%	−0.15%
	AF long	5.56%	6.99%	1.42%
	AF ant.	2.92%	7.62%	4.70%
	AF post.	3.75%	5.60%	1.86%
	IFOF ant.	1.05%	5.00%	3.95%
	IFOF post.	4.10%	1.94%	−2.16%
NAMING	ILF	2.95%	10.30%	7.35%
	UF	12.00%	0.03%	−11.98%*
	AF long	1.42%	2.39%	0.97%
	AF ant.	0.73%	1.39%	0.66%
	AF post.	2.18%	0.47%	−1.72%
	IFOF ant.	14.67%	2.14%	−12.53%*
	IFOF post.	2.85%	3.31%	0.46%
SPELLING	ILF	0.28%	6.53%	6.25%
	UF	0.30%	0.20%	−0.10%
	AF long	2.16%	2.98%	0.82%
	AF ant.	0.79%	6.69%	5.90%
	AF post.	0.49%	8.90%	8.41%
	IFOF ant.	1.91%	1.74%	−0.16%
	IFOF post.	0.35%	6.18%	5.83%

MD, mean diffusivity; LH, left hemisphere; RH, right hemisphere; ILF, inferior longitudinal fasciculus; UF, uncinate fasciculus; AF, arcuate fasciculus; IFOF, inferior fronto-occipital fasciculus.

Fig. 4.

Unique-R² of language performance explained by the MD values of the homologous tracts. The uncinate fasciculus and the anterior inferior fronto-occipital fasciculus in Naming are two tracts in which the difference in magnitude of the unique R² between the two hemispheres is statistically significant, with greater values in the left hemisphere compared to the right. The finding that for 67% (14/21) of the tracts, there is more unique R² for the RH than for the left, represents a statistically significant distribution of unique-R² values (both results were evaluated by permutation test).

First, with respect to the magnitude of R², of the 21 models tested using a Monte-Carlo correction for multiple comparisons, the unique-R² difference values were statistically significant for two tracts predicting Naming scores: (i) uncinate fasciculus (P = 0.026), and (ii) anterior inferior fronto-occipital fasciculus (P = 0.021). Second, the analysis of the distribution of unique-R² difference values found that 67% of the RH tracts explained more unique variance compared to their LH counterparts and that the probability of observing this RH-lateralized distribution by chance, as evaluated by the Monte-Carlo permutation test described above, had a statistical significance of P = 0.049.

Discussion

This study investigated the integrity of the white matter (WM tracts in a group of individuals with PPA compared to a group of healthy control (HC) individuals, as well as the relationship between WM tract integrity and language processing in PPA in the two hemispheres. The key findings were as follows: (i) In terms of WM tract integrity, we found that, in this sample of PPA individuals, the disease significantly affected the integrity of the language tracts only in the LH (Analysis 1); (ii) With regard to the relationship between WM tract integrity and language processing, we found that they were significantly associated in both hemispheres (Analysis 2); (iii) Furthermore, we found evidence of domain-specificity for the relationship between language processing and WM tract integrity in both hemispheres, indicating that the role of these WM tracts in language cannot be reduced to general (domain-independent) support for cognitive processing (Analysis 3); (iv) We found that the integrity of each hemisphere’s WM language tracts explained unique variance in language performance. Specifically, the LH’s unique contribution was more focal (involving the left uncinate fasciculus and anterior inferior fronto-occipital fasciculus), while the RH’s unique contribution involved smaller effects that were more broadly distributed across tracts (Analysis 4). This set of results provides additional support for the already well-established strong relationship between language and WM tract integrity in the LH and, more significantly, they constitute novel evidence of the contribution of the RH to language processing, at least in the context of LH damage.

White matter integrity in the two hemispheres in PPA

Analysis 1 found evidence that PPA affected the WM integrity of the language tracts only in the LH. As discussed in the Introduction, some of the previous studies that investigated differences in WM integrity between PPA and HC individuals have reported effects in both hemispheres (e.g. Galantucci et al. 2011; Agosta et al. 2012; Grossman et al. 2013), which we did not find. However, methodological differences might explain this discrepancy. First, since the data for this study were collected, more updated scanning parameters have been developed and these might be more sensitive. Second, the majority of previous studies have looked at the three PPA variants separately. Given that the three PPA variants are associated with distinct patterns of atrophy in the LH (Gorno-Tempini et al. 2004, 2011), it is possible that they also suffer from distinct patterns of damage in the RH. However, in the current study, the samples for each variant were too small to be assessed separately, and, therefore, in this study the PPA participants were treated as a single group. This may have reduced the power to detect WM changes that were differentially distributed across RH tracts. Thus, while we cannot be confident that the RH WM was not affected in this group, it is evident that the LH is more severely affected. As shown in Supplementary Material Section 5—Table S4, except for the posterior inferior fronto-occipital fasciculus, all LH language tracts showed a statistically significant or marginally significant difference between the two groups, with the PPA group having greater MD values (i.e. greater damage) compared to the HC group. The long component of the arcuate fasciculus, the inferior longitudinal fasciculus, and the uncinate fasciculus showed the greatest effects.

White matter integrity and language processing in the two hemispheres

Previous studies examining the relationship between WM integrity and language processing in both hemispheres are scarce; in this work we examined this relationship by carrying out three different analyses. In Analysis 2, rather than focusing on individual tracts and their relationship with specific language functions as in previous studies (for example, PPA: Catani et al. 2013; Grossman et al. 2013; stroke aphasia: Ivanova et al. 2016; healthy individuals: Nugiel et al. 2016; Yablonski et al. 2019), we statistically evaluated the overall relationship of language processing with WM integrity for each hemisphere. By comparing regression models (separately for the LH and RH) that included or did not include language performance, we showed that language performance was significantly associated with the integrity of both the LH and the RH WM tracts. However, given that the WM integrity of the two hemispheres is often correlated (see Supplementary Material Section 3—Table S2), one might argue that the RH effect merely reflects the LH WM-language relationship. If that were the case, then the RH effect identified in Analysis 2 would not represent an independent RH contribution to language processing. To address this possibility, Analysis 4 specifically examined the difference in the unique contributions of the homologous tracts of the two hemispheres in explaining variance in language performance. The results showed unique contributions of WM integrity to language processing for both hemispheres, with the RH’s unique contribution being broadly distributed across tracts, while the LH’s unique contribution was focal and localized to specific tracts and language domains.

The analyses considered language performance across three language domains—spoken naming, syntactic processing, and spelling. The statistically unique contribution of LH WM integrity to language processing was specifically related to spoken naming performance. The two tracts that showed this highly specific association with spoken naming were the left uncinate fasciculus and anterior inferior fronto-occipital fasciculus, both of which have been previously implicated in spoken naming (McDonald et al. 2008; Papagno et al. 2011; Catani et al. 2013). If these two LH tracts had been relatively intact in this group, this result could be interpreted as indicating that these tracts were able to sustain their “original” role and that there had not been pressure for reorganization or RH recruitment to support naming. However, the results from Analysis 1 showed that both tracts were severely damaged in this group of PPA individuals (see Fig. 3 and Supplementary Material Section 5). One alternative explanation for the continued relevance of these LH tracts, in the absence of evidence of clear involvement of their RH homologs, could be that naming is a highly left-lateralized process that is not easily transferred to the RH, even in the context of severe LH damage. This account, however, would seem to be at odds with studies showing a relationship between RH WM integrity and naming (Stamatakis et al. 2011; Kourtidou et al. 2021). Clearly, there is more to be investigated to clarify the conditions under which RH involvement comes online, including the specific behavioral tasks, measures of WM tract integrity, and patterns of LH damage that may be relevant.

The role of the RH in language processing in the context of LH damage

With respect to the contribution of the RH to language processing, we have provided, for the first time, evidence of unique contribution of the RH to language processing that differs from the unique contribution of the LH. Specifically, we found that the RH’s contribution to language processing was more modest compared to that of the LH, since, unlike in the LH, there were no specific tract-language relationships that were individually statistically significant. However, the RH’s contribution to language processing was more widespread across WM tracts compared to that of the LH. A possible interpretation is that the RH WM tracts are not as highly specialized as those in the LH with respect to the various language processes they support. Whether the lack of specialization is an intrinsic limitation of the RH's role in language processing or is specifically associated with neurodegenerative disease is a question to be investigated in future studies. These might, for example, examine LH stroke where it would be important to ascertain if similar analysis approaches would also reveal nonspecific RH contributions.

Despite the lack of clear association of specific RH tracts with specific language domains, we were, nonetheless, able to show that the RH’s contribution was specific to language processing. Analysis 4 provided evidence of a significant relationship between language scores and WM tract integrity in both hemispheres that was beyond any contribution from domain-general cognitive functions such as those that might be shared by SpSpan and language processing. This result is not surprising for the LH, which has been shown to support language-specific processes, but to the best of our knowledge, this is the first evidence that distinguishes RH support for language processes from support for general cognitive processing.

One question that has been repeatedly raised is whether the RH plays a compensatory or maladaptive role in the aftermath of LH damage. This determination depends on whether the RH changes are found to be associated with improved (compensatory) or impaired (maladaptive) performance. Evidence for maladaptive performance comes from studies showing that increased RH activation in individuals with aphasia is not always associated with improved language performance (Belin et al. 1996; Cao et al. 1999; Winhuisen et al. 2005; Postman-Caucheteux et al. 2010; Allendorfer et al. 2012). One hypothesis positing a maladaptive role is that LH damage disrupts the balance of normal inhibition between the hemispheres, increasing the relative inhibition by the RH of LH language areas and, in this way, increasing the difficulty of recovery (for discussion see: Hamilton et al. 2011; Turkeltaub 2015).

Evidence of the possible compensatory contributions of the RH comes largely from intervention studies. Many have shown that improved performance is associated with structural or functional changes in GM. For example, there has been evidence of increased segregation of functional networks in the RH as a result of treatment (Tao and Rapp 2019, 2020). A very small number of studies have examined the question of compensatory or maladaptive consequences related to the WM integrity of the RH. For example, there is evidence of a positive relationship between language therapy effects and the integrity of WM tracts in the RH (e.g., the superior longitudinal fasciculus) in PPA (Zhao et al. 2021).

Stimulation of the RH could turn out to be beneficial whether the RH has a compensatory or maladaptive role in language recovery, depending on the underlying neurostimulation mechanisms. If the RH provides compensatory support, then targeting the RH could enhance language processing, but it may also be the case that if RH contributions are determined to be maladaptive, stimulation could serve to diminish its maladaptive contributions (Turkeltaub 2015). For example, given the Zhao et al. (2021) findings of apositive relationship between language therapy effects and the integrity of white matter tracts in the right superior longitudinal fasciculus, the right hemisphere cortical and subcortical regions connected by relevant WM tracts (e.g. the right inferior frontal gyrus) could be considered as stimulation targets. Previous research has already explored this possibility, with several studies showing that stimulation over the right frontal lobe, especially over the right pars triangularis is associated with improved naming performance (Naeser et al. 2002, 2005; Martin et al. 2004, 2009; Winhuisen et al. 2005; Barwood et al. 2011; Kang et al. 2011; Weiduschat et al. 2011; Chieffo et al. 2014) as well as improvements in spontaneous elicited speech (Hamilton et al. 2010) in post-stroke aphasia.

Finally, the finding that individual variability in WM tract robustness in the RH is associated with individual variability in language performance in the PPA group raises a key question: Does the variability in RH WM that is associated with language performance predate the LH damage or is it a consequence of it? In other words, are those individuals with more robust RH WM tracts able to achieve better language performance in the face of LH damage, or are the RH WM changes we observed triggered by the LH damage itself? While this is still an open question, there is evidence suggesting that, at least in post-stroke aphasia, RH recruitment is more prominent in situations of large-scale LH damage (Karbe et al. 1998; Gold and Kertesz 2000; Ansaldo et al. 2002; Kourtidou et al. 2021).

Study limitations

The present study provides important new evidence for the role of the two hemispheres in language processing. However, certain limitations need to be considered. First, the lack of behavioral language data for the HC did not allow us to assess the relationship between language processing and WM tract integrity in the healthy brain and how this is specifically affected by the neural changes caused by PPA. It is also important to acknowledge that our investigation of the domain-specificity of the WM language networks (Analysis 3) only made use of a single non-language domain, that of visuospatial working memory. While visuospatial working memory falls under the broad category of executive functions, this task only captures specific aspects of executive functioning. Therefore, the conclusions we reach here are limited to the cognitive functions implicated in this task and cannot be generalized across executive functions more generally. Another limitation is inherent to the tractography method and involves the difficulty of isolating the WM fibers of a given tract from nearby fibers. As discussed, there is considerable controversy in the literature with respect to the identification and the naming of some of the tracts discussed in this paper, hence caution is required when drawing conclusions about the contribution of specific tracts to different language functions. It is also important to note that the DTI acquisition parameters used in this study (32 diffusion-weighted volumes at b = 700 s/mm²) are rather conservative compared to the DTI acquisition parameters more commonly used in the field (i.e. greater number of volumes, at greater b-values) and this might impose additional limitations with respect to tract extraction and the metrics of WM integrity. Another limitation is that the group of PPA individuals used in this study included small and unequal size samples per variant that did not allow us to look at each variant separately. It is possible that the effects we report in this study are not homogeneous across the three variants of PPA, hence future work should investigate if and how the effects reported in this study differ across PPA variants. Lastly, this is a cross-sectional study, with PPA individuals at different stages of disease progression. However, it is possible that at different stages of the disease, the relationship between language and WM integrity in the two hemispheres may differ. A longitudinal study could examine how the progression of the disease affects the relationship between language and WM integrity of each of the two hemispheres.

Conclusions

The current investigation allows us to confirm the effects of PPA on the WM integrity of the language tracts in the LH and provides strong evidence of the relationship between language processing and WM integrity in both hemispheres. To the best of our knowledge, this is the first study to show evidence of unique contributions of both the LH and the RH in language processing, as well as evidence that the contribution of both hemispheres is specific to language processing and does not simply support domain-general processing. These findings allow us to better characterize the role of the RH in language processing in the context of LH damage.

Funding

This work was supported by grants from the National Institutes of Health (NIH)/National Institute of Deafness and Communication Disorder (NICDC) through award R01 DC014475 to K.T. and award DC006740 to B.R.

Conflict of interest statement: None declared.

Author contributions

K.N., K.T. and B.R. designed the study; K.N. and C.L. performed the Diffusion MRI data preprocessing and white matter tract extraction; K.N. and R.W. performed the statistical analysis of the data. All authors contributed to discussing and interpreting the results as well as writing and editing the article.

Data availability

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