A middle ground where executive control meets semantics: the neural substrates of semantic control are topographically sandwiched between the multiple-demand and default-mode systems

Abstract Semantic control is the capability to operate on meaningful representations, selectively focusing on certain aspects of meaning while purposefully ignoring other aspects based on one’s behavioral aim. This ability is especially vital for comprehending figurative/ambiguous language. It remains unclear why and how regions involved in semantic control seem reliably juxtaposed alongside other functionally specialized regions in the association cortex, prompting speculation about the relationship between topography and function. We investigated this issue by characterizing how semantic control regions topographically relate to the default-mode network (associated with memory and abstract cognition) and multiple-demand network (associated with executive control). Topographically, we established that semantic control areas were sandwiched by the default-mode and multi-demand networks, forming an orderly arrangement observed both at the individual and group level. Functionally, semantic control regions exhibited “hybrid” responses, fusing generic preferences for cognitively demanding operation (multiple-demand) and for meaningful representations (default-mode) into a domain-specific preference for difficult operation on meaningful representations. When projected onto the principal gradient of human connectome, the neural activity of semantic control showed a robustly dissociable trajectory from visuospatial control, implying different roles in the functional transition from sensation to cognition. We discuss why the hybrid functional profile of semantic control regions might result from their intermediate topographical positions on the cortex.


Supplemental Stimuli
The Semantic task included two conditions (Easy and Hard). A quadruplet of words were shown in each trial, containing an oddball (target) that was semantically inconsisent with the remaining three (Foil 1 -3). Note, the target's location vaired randomly from trial to trial, equally likley to be in any of the four positions and be reacted to by any of the four designated buttons in the actual experiment. (Supplemental Figure S1-1)

Semantic Easy Semantic Hard
Within the left IFG, the cluster of visuospatial difficulty was situated in the posterior subpart (pars opercularis), and visuospatial easiness was situated in the anterior subpart (pars orbitalis). Semantic control involved nearly the entire chunk of the left IFG and partially overlapped with the clusters of visuospatial difficulty and easiness. The middle section of the left IFG (pars triangularis) was almost exclusive to semantic control. Results were thresholded at p < 0.001 (voxelwise) and p < 0.05 (whole-brain familywise correction for clusters). Visualisation here was confined to only showing activation in the IFG mask.
(Supplemental Figure S1-2) In the bilateral occipital and temporal lobes, visuospatial difficulty involved mostly the inferior temporal gyri and occipital cortex, while visuospatial easiness involved the superior temporal gyri and temporal poles.  Fedorenko et al. (2013); in their group-level data, various frontoparietal areas, collectively forming the MD network, showed robustly greater activation for hard conditions than easier conditions. The MD network was then segregated using anatomical atlas, producing 10 cortical parcels ( Figure S2 Simple effects showed that the semantic difficulty effect was greater in the left hemisphere (consistent with the left-lateralisation of language functions); the semantic effect was also greater in prefrontal areas (mid-frontal gyrus, insula, anterior cingulate cortex) than in those regions typically implicated in saccade and visuomotor tasks (frontal eye fields, superior parietal lobules). We interpret this pattern (more prefrontal activation, less visuomotor regions) as reflecting the Semantic task's emphasis on semantics over visuomotor processes. Simple effects also showed that the visuospatial difficulty effect was statistically significant in every subregion of the MD system and in both hemispheres. It is worth noting that comparing the sizes of semantic difficulty versus visuospatial difficulty is not appropriate, because of differential baselines in behavioural data (namely, the reaction times of Semantic-Easy were slower than those of Visuospatial-Easy).
Next, we situated the activation of semantic control (i.e., Sem.-Hard > Sem.-Easy) in the results of meta-analysis of 'semantic' (Figure S2-C) and 'executive control' (Figure S2-D). We used the term-based search function of NeuroSynth, surveying for 'semantic' (based on 4,031 fMRI studies and 40,030 activation points); in another analysis, we used 'executive' (based on 786 fMRI studies and 28,937 activation points). The meta-analysis results were thresholded at q < 0.01 (corrected for voxelwise false discovery rate). As shown in Figure S2-C and S2-D, the clusters of semantic control of the present study (red clusters) are situated in the meta-analysis outcomes of semantic processes (yellow clusters in S2-C) and executive control (blue clusters in S2-D).
In Figure S2-C, the semantic control clusters partially overlap with the expansive set of regions robustly appearing in the semantic fMRI literature. Note that the semantic control clusters overlap with the meta-analysis mostly in the left prefrontal cortex and posterior mid-temporal gyrus (pMTG), but no overlap is found in the anterior temporal cortex (ATL). This pattern is consistent with previous fMRI literature that semantic control relies on the left prefrontal cortex and pMTG, while semantic representation depends on the ATL. Next, in Figure S2-D, while there is overlap between semantic control and the meta-analysis of executive process in many frontoparietal areas, the right posterior parietal lobule (well-stablished in the literature of spatial attention and saccade) seems minimally engaged by semantic control.
Taken together, we found that activation was heightened in most subregions of the MD system when tasks became difficult, observed both for the Semantic or Visuospatial task. It is, however, important to note that the MD network's role in a language task has been demonstrated to reflect maintaining attentional focus on the task, rather than semantic processes per se (Diachek, Blank, Siegelman, Affourtit, & Fedorenko, 2020). Moreover, semantic control was relativley lateralised to MD areas in the left hemisphere, and relied more on prefrontal areas (relative to parietal areas), whereas visuospatial control was bilateral and relied on both prefrontal and parietal lobes. In sum, the data showed that MD regions were more engaged by visuospatial than semantic processes.

Supplemental Results 3
We contrasted the task-induced activation of Sem.-Hard vs. Vis.-Hard, because the two conditions had statistically matched performance levels. Results of whole-brain analysis was thresholded at p < 0.001 (voxelwise intensity), p < 0.05 (FWE-corrected for cluster-level multiple comparisons). As shown in Figure S2, the contrast 'Sem.-Hard > Vis.-Hard' identified a distributed set of regions the left inferior frontal gyrus (IFG), left superior temporal gyrus, bilateral temporal ATLs (temporal poles), bilateral supramarginal gyri, medial prefrontal cortex, medial temporal lobes (including the hippocampus). Collectively, these regions exhibited the typical topography of the semantic/language network (e.g., Diachek et al., 2020); their topography also resembled the pattern of transmodal areas that are situated at the 'abstract-cognitive' end of principal cortical gradient (Margulies et al., 2016). The reverse comparison (Vis.-Hard > Sem.-Hard) revealed many regions well-documented in the literature of spatial attention, saccade, and various externally-directed visuomotor processesthe bilateral superior parietal lobules, intraparietal sulci, frontal eye fields. The topography of these regions resembled the multiple-demand network (Fedorenko, Duncan, & Kanwisher, 2013), as well as the dorsal-attention network and visual network (Yeo et al., 2011).
It is worth emphasising the functional subdivision (semantic control vs. semantic representation) within the semantic system (for review, see Lambon Ralph, Jefferies, Patterson, & Rogers, 2017). To detect such subdivision, different contrasts had to be used: First, the contrast of 'Sem.-Hard > Vis.-Hard' revealed the typical configuration of entire semantic network, which included both semantic representation regions (e.g., the ATLs) and semantic control regions (the IFG and pMTG). These two types of semantic regions both preferred semantics to visuospatial processing, hence they were both identified by this contrast. Second, to identify the subregions specifically sensitive to semantic control, a contrast differentiating the harder semantic operations from the easier ones was necessary (Sem.-Hard > Sem.-Easy); this contrast identified the subsystem of semantic control (the IFG and pMTG) while leaving out the subsystem of semantic representation (the ATL).