Resting state correlates of picture description informativeness in left vs. right hemisphere chronic stroke

No clue what this is or how the hell this is going to get survivors recovered. The mentors need to be fired for allowing this gobbledegook.

Resting state correlates of picture description informativeness in left vs. right hemisphere chronic stroke

Erin L. Meier^1* Shannon M. Sheppard¹ Rajani Sebastian² Shauna Berube¹ Emily B. Goldberg¹ Jennifer Shea¹ Colin M. Stein¹ Argye E. Hillis^1,2,3

• ¹Department of Neurology, Johns Hopkins University, Baltimore, MD, United States
• ²Department of Physical Medicine and Rehabilitation, Johns Hopkins University, Baltimore, MD, United States
• ³Department of Cognitive Science, Johns Hopkins University, Baltimore, MD, United States

Introduction: Despite a growing emphasis on discourse processing in clinical neuroscience, relatively little is known about the neurobiology of discourse production impairments. Individuals with a history of left or right hemisphere stroke can exhibit difficulty with communicating meaningful discourse content, which implies both cerebral hemispheres play a role in this skill. However, the extent to which successful production of discourse content relies on network connections within domain-specific vs. domain-general networks in either hemisphere is unknown.

Methods: In this study, 45 individuals with a history of either left or right hemisphere stroke completed resting state fMRI and the Cookie Theft picture description task.

Results: Participants did not differ in the total number of content units or the percentage of interpretative content units they produced. Stroke survivors with left hemisphere damage produced significantly fewer content units per second than individuals with right hemisphere stroke. Intrinsic connectivity of the left language network was significantly weaker in the left compared to the right hemisphere stroke group for specific connections. Greater efficiency of communication of picture scene content was associated with stronger left but weaker right frontotemporal connectivity of the language network in patients with a history of left hemisphere (but not right hemisphere) stroke. No significant relationships were found between picture description measures and connectivity of the dorsal attention, default mode, or salience networks or with connections between language and other network regions.

Discussion: These findings add to prior behavioral studies of picture description skills in stroke survivors and provide insight into the role of the language network vs. other intrinsic networks during discourse production.

1 Introduction

Discourse is the means through which people communicate their thoughts, feelings, and experiences during everyday conversation. Within formalist or structuralist frameworks, discourse is defined as the unit of language above the sentence level, whereas functionalist frameworks define discourse based on its capacity to meaningfully convey information within social contexts (1–4). Within the latter framework, a single word with significant meaning could be considered discourse (1), but more often, discourse samples contain multiple sentences that combine to form a coherent message. Discourse production likely relies on multiple cognitive systems, but the brain networks involved in different types of discourse genres (e.g., procedural, expository, narrative) remain underspecified.

Relevant to this topic is the fact that left and right hemisphere stroke survivors can exhibit discourse production impairments, albeit with different deficit profiles. Consistent with the notion that the left hemisphere (LH) is responsible for linguistic processing in most individuals, LH stroke survivors with aphasia often exhibit deficits in microstructural discourse properties, such as incorrect or omitted morphological and syntactic markers, reduced lexical diversity, and a decreased number of different word classes compared to normative samples [see reviews by Armstrong (1) and Linnik et al. (5)]. On the other hand, individuals with right hemisphere damage (RHD) most often demonstrate impaired macrostructural discourse skills such as poor discourse organization, tangentiality, and reduced local and global coherence, possibly due to cognitive-communication deficits in domains such as attention, executive functions, and pragmatics (6–8).

According to Armstrong (1), communication of meaningful information content (sometimes referred to as informativeness) falls in between micro- and macrostructural levels and may be deficient in individuals with a history of left or right hemisphere stroke. Discourse informativeness has been studied extensively via a variety of measures [e.g., content units, correct information units, lexical information units; see (5)] with mixed results. For example, some studies reported that people with aphasia (PWA) due to LH stroke produce significantly less meaningful discourse content, often with reduced efficiency, compared to neurologically healthy adults [e.g., (9–14)] while other studies reported no difference between PWA and controls or differences only for individuals with severe aphasia [e.g., (15–17)]. Similarly, lower informativeness scores in individuals with RHD compared to controls have been reported by some investigators [e.g., (9, 18–21)] but not others [e.g., (15, 16, 22)]. The type of information being conveyed may also matter, as there is some evidence that PWA demonstrate overall reduced lexical informativeness but not for main themes (10, 11), and content that requires inferential processing may be particularly susceptible in RHD (23, 24). Pertinent to this investigation is the fact that differences in discourse informativeness skills between individuals with LH damage (LHD) vs. RHD have not been established, partially due to the relative lack of studies [c.f. (15, 16, 19, 22)]. Given these inconsistent findings and the importance of informativeness in the success of discourse production, we focused on content unit (CU) measures derived from picture descriptions produced by our participants.

Currently, evidence regarding the neural substrates of discourse production impairments in left vs. right hemisphere stroke survivors is scant, and only a couple investigations (9, 16) have included measures that reflect the production of meaningful content. Using region-based lesion symptom mapping, Agis et al. (9) investigated which regions in the core LH language network and their RH homologs were implicated in reduced word-level content in acute left and right hemisphere stroke survivors, respectively. In the LH stroke group, they found that the total number of CUs produced when describing the Cookie Theft picture (25) was predicted by a model that included damage involving the left supramarginal gyrus (SMG), angular gyrus (AG), superior temporal gyrus (STG), middle temporal gyrus (MTG), and inferior temporal gyrus (ITG), as well as total lesion volume, but only damage to left ITG and lesion volume were significant independent predictors. While no single variable was significantly predictive in patients with acute RHD, a model that included infarct in the right inferior frontal gyrus, pars opercularis (IFGop), SMG, AG, STG, the superior longitudinal fasciculus, the sagittal striatum, and lesion volume was significant. Using whole-brain voxel-based morphometry, Schneider et al. (16) found that greater lexical informativeness during sequential scene descriptions was associated with greater gray matter density in the left primary sensory cortex and left insula—but not RH regions—across a sample of 10 patients with chronic LHD, 10 individuals with chronic RHD, and 10 neurologically healthy controls. In other lesion mapping studies that included LH stroke survivors with aphasia (but not individuals with RHD) (26–29), impaired production of meaningful discourse content was linked to damage to cortical regions spanning left frontal, temporal, and parietal lobes as well as several underlying LH white matter association tracts.

Collectively, lesion symptom mapping studies of informativeness deficits during discourse have primarily implicated LH structures which are traditionally associated with semantic or articulatory processes rather than RH regions or LH areas outside the canonical language network [cf. e.g., RH stroke findings in Agis et al. (9) and domain-general regions reported in Alyahya et al. (26)]. However, a major caveat to this conclusion is that most studies to date have been in individuals with post-stroke aphasia, and thus, the majority of analyses have been restricted to parts of the brain typically lesioned in PWA, i.e., language network regions within the left middle cerebral artery territory. Task-based functional imaging circumvents this issue but comes with added methodological limitations, such as stimulus-correlated motion artifacts in functional imaging time series data induced by overt speaking. As an alternative, resting state analysis allows for the delineation of intrinsic network markers that can then be correlated with participant performance on offline production tasks. Using such an approach, Duncan and Small (30) found that an increase in resting state network modularity from before to after an imitation-based therapy in 19 PWA was significantly associated with an increase in correct information units produced during retelling of the Cinderella story. In a follow-up investigation of dynamic functional connectivity, Duncan and Small (31) reported that treatment-related increases in correct information units were also significantly associated with increased dwell time in a state in which there was segregation of the default mode, dorsal attention, executive control, language, frontoparietal, sensorimotor, and visual networks. In other words, findings from these two studies indicate that treatment-induced improvements in discourse informativeness were associated with an improved balance between modularity and segregation of intrinsic domain-general and language-specific networks in PWA due to LH stroke.

The current study is motivated by the need to better understand the neurobiology of spoken discourse and the neural substrates that underlie impaired production of meaningful discourse content in individuals with LHD or RHD due to stroke. Our study had two aims. First, we evaluated differences between individuals with early chronic LH vs. RH stroke in three picture description measures derived from the Cookie Theft picture description task (25): (1) total number of CUs, primarily reflecting lexical-semantic skills; (2) the percentage of interpretative CUs, reflecting a combination of lexical-semantic and inferencing skills; and (3) the number of CUs produced per second, reflecting communication efficiency of meaningful scene content. Consistent with prior studies with overlapping measures [e.g., (9, 19, 24)], we hypothesized that individuals with LHD would produce fewer total CUs than individuals with RHD, but that several individuals within the RHD group would still demonstrate impaired CU production. Given the likely cause of discourse impairments following LH and RH stroke and the prior literature [e.g., (9, 12, 14, 23, 24)], we predicted that there would be no statically significant differences between groups in the percentage of interpretive CUs or communication efficiency of salient content.

Our second aim was to examine how picture description measures relate to resting state functional connectivity (rs-FC) of four intrinsic bilateral networks—the language network [LN; (32–36)], the core default mode network [DMN; (37–39)], the dorsal attention network [DAN; (40)], and the cingulo-opercular salience network [SN; (41)]—within LHD and RHD groups (Aim 2a) and between groups (Aim 2b). We interrogated these four networks to test the overarching hypothesis that the production of CUs within connected speech requires not only core linguistic processes mediated by the LN, but also domain-general cognitive skills processed within other intrinsic networks. More specifically, we hypothesized that all three CU measures would be positively associated with LN rs-FC in both groups, given that the production of CUs relies heavily on core stages of word retrieval (i.e., conceptual processing and semantic retrieval, lexical access, phonological retrieval and encoding, and articulation) and in some cases, morphosyntactic processes involved in the production of phrase-level utterances (42–47). We also predicted that all three measures would be positively associated with the within-network connectivity of other task-positive networks (i.e., DAN and SN) given that the DAN plays a role in directing attention to target stimuli to accomplish task goals (40, 48) and the SN is important for maintaining sustained attention to task (49–51). The DMN is considered a task-negative network which engages during rest but disengages during goal-directed activity (37, 39, 52). In other clinical populations (e.g., autism, dementia, schizophrenia), the integrity of the DMN is considered a biomarker for overall brain health (53–55). As such, we predicted that higher within-network connectivity of the DMN would likely also be related to better picture description ability in individuals with LHD or RHD [although cf. e.g., McCarthy et al. (56) and Weisssman et al. (57) for findings demonstrating worse behavioral outcomes coinciding with hyperconnectivity]. Given the importance of balanced network modularity and segregation in general and for discourse informativeness in prior studies in aphasia specifically (30, 31, 58), we expected that greater connectivity of between-network connections would be associated with worse CU scores. Finally, across networks, we predicted that relationships between picture description measures and left intra-hemispheric connectivity would be stronger in the RHD group than the LHD group whereas the opposite would be true of right intra-hemispheric connectivity.