Introduction
The Inferior Frontal Occipital Fasciculus (IFOF) is a major anterior-to-posterior white matter pathway in the ventral human brain. It connects occipital, parietal and posterior temporal regions, implicated in word and object recognition, to prefrontal areas associated with cognitive control, semantic retrieval and speech production (Caverzasi et al.
2014; Duffau et al.
2013; Martino et al.
2010). In line with this functional anatomy, IFOF has long been associated with language and semantic cognition (Almairac et al.
2015; Binder et al.
2009; Han et al.
2013; Sierpowska et al.
2019). Stimulation of IFOF during neurosurgery elicits semantic errors in picture naming tasks (Duffau et al.
2005,
2008), while damage to IFOF is associated with semantic impairment (Han et al.
2013; Souter et al.
2022; Surbeck et al.
2020). Similarly, variation in the integrity of IFOF in healthy participants is associated with performance on lexical-orthographic tasks (Vandermosten et al.
2012) and bilingual aspects of language (Mohades et al.
2012). Nevertheless, there are many unresolved questions about the contribution of IFOF to cognition, including whether this tract supports both conceptual and visually-guided (non-conceptual) aspects of cognition, whether there are differences between semantic tasks employing pictures and words, and whether there are functional distinctions across subdivisions of IFOF.
Contemporary theoretical accounts emphasize that semantic cognition and language emerge from interacting component processes, which have dissociable neurocognitive bases (Lambon Ralph et al.
2017) and which might draw on different white matter tracts (Duffau et al.
2009,
2013; Moritz-Gasser et al.
2013; Rauschecker
2012). For example, a dorsal language pathway is thought to support acoustic − motor mappings and articulatory sequences, while a ventral language pathway supports semantic processing (Saur et al.
2008): by this account, conceptual retrieval draws on the interaction of heteromodal conceptual representations in an anterior temporal ‘hub’ with unimodal ‘spoke’ representations in visual and auditory cortex (Patterson et al.
2007). The inferior longitudinal fasciculus (ILF) links occipital cortex to anterior temporal regions and is consequently well-placed to support rapid and automatic access to heteromodal conceptual knowledge from visual inputs (Duffau et al.
2013; Herbet et al.
2018; Saur et al.
2008; Turken and Dronkers
2011). However, this pathway is thought to be insufficient for semantic cognition since we need to retrieve conceptual information in a flexible fashion to generate adaptive thoughts and behaviour and this is thought to involve the interaction of semantic knowledge with control processes. Recent studies have suggested that a distributed network including posterior middle temporal gyrus and anterior inferior frontal cortex is important for our capacity to focus on semantic information that is relevant to the evolving context or our current goals (Davey et al.
2016; Gao et al.
2021; Jackson
2021). This ‘semantic control network’ shows the strongest recruitment in fMRI studies when non-dominant aspects of knowledge are needed, or when there is ambiguity or conflict between concepts (Badre et al.
2005; Jackson
2021; Noonan et al.
2013; Thompson-Schill et al.
1997). Because IFOF connects posterior temporal to prefrontal regions, it is likely to play a critical role in the controlled application of knowledge during semantic cognition (Duffau et al.
2013; Giampiccolo and Duffau
2022; Nugiel et al.
2016). Patients with poor control of semantic cognition in the context of stroke aphasia have highly consistent damage to IFOF and ILF despite highly variable lesions affecting left inferior prefrontal or posterior temporal areas (Souter et al.
2022). Moreover, for phonological and semantic verbal fluency tasks with high executive demands, IFOF integrity is correlated with performance (Almairac et al.
2015; Nugiel et al.
2016).
IFOF also shows broader connectivity beyond semantic control regions, connecting occipital and parietal regions to broad swathes of lateral, anterior and ventral frontal regions (Catani and Thiebaut de Schotten
2008; Duffau et al.
2013) linked to cognitive control, decision making and speech production. Consequently, IFOF is likely to be important for controlled visually-guided cognition
beyond the semantic domain; for example, Walsh et al. (
2011) found that reduced microstructural integrity of the IFOF is associated with poor object working memory performance, and other studies have established a link between IFOF dysfunction and unilateral neglect (Herbet et al.
2017b; Karnath et al.
2011; Urbanski et al.
2008,
2011). Given this connectivity with visual cortex, it is also important to establish if the functional role of IFOF varies across semantic tasks involving pictures and words. If IFOF is critical for linking visual input regions to prefrontal cortex, this tract might play a greater role in semantic tasks that utilize picture inputs, such as picture naming and picture association. Alternatively, IFOF may support connections critical to the putative heteromodal semantic control network, linking posterior middle temporal gyrus to anterior portions of inferior frontal cortex: both of these regions respond to control demands across word and picture semantic tasks (Krieger-Redwood et al.
2015).
Research has also revealed dorsal and ventral subdivisions within IFOF, which may have different functional associations (Martino et al.
2010; Rollans and Cummine
2018; Roux et al.
2021; Sarubbo et al.
2013). Martino et al. (
2010) found the superficial dorsal subcomponent of IFOF connects superior occipital regions associated with visually-guided action, plus posterior superior temporal and parietal regions, to posterior parts of inferior frontal gyrus: this pathway is potentially suited to the control of action and visual − spatial processing. In contrast, a deep ventral pathway was found to connect inferior occipital and ventral temporal regions implicated in object recognition, plus heteromodal posterior middle temporal gyrus, to diverse frontal lobe regions, including anterior and ventral regions within the default mode network and dorsolateral prefrontal regions associated with cognitive control: this pathway is therefore suited to supporting visually guided decision making and heteromodal semantic control. Sarubbo et al. (
2013) corroborated this division of the IFOF into dorsal and ventral components, using a combination of the Klingler method and single-subject DTI, showing that the dorsal aspect terminates in the posterior inferior frontal gyrus, while the ventral component can be further divided in three minor bundles terminating in middle frontal gyrus and dorsolateral prefrontal cortex, lateral orbitofrontal cortex, and frontal pole. In line with this multi-bundle—as opposed to dichotomous—IFOF architecture, other studies have found a five-layer organization when streamline tractographies are initiated from frontal areas (e.g. Wu et al.
2016). Almairac et al. (
2015) proposed that the ventral pathway might be important for semantic access, while the dorsal pathway supports phonology; the dorsal route is also implicated in reading and writing (Motomura et al.
2014). However, there are alternative functional interpretations. Rollans and Cummine (
2018) found that Fractional Anisotropy (FA) in dorsal and ventral IFOF was linked to Go/No-Go and naming tasks respectively, suggesting the dorsal pathway may be crucial for controlled action driven by visual input beyond the domain of language or semantic cognition, while the ventral pathway supports more abstract conceptual behaviour.
Finally, given that language and semantic control are thought to be highly left-lateralised (Frost et al.
1999; Gonzalez Alam et al.
2019; Josse and Tzourio-Mazoyer Nathalie
2004), while visually-guided non-semantic cognition and domain-general executive processes are more bilateral (Bartolomeo and Seidel Malkinson
2019; Duncan
2010; Hellige and Michimata
1989; Hugdahl et al.
2015), there may be important functional dissociations between left and right IFOF. Using tract-based spatial statistics (TBSS), Rollans et al. (
2017) found that IFOF was associated with gross picture naming differences in the left hemisphere, but more subtle differences in naming performance in the right hemisphere. Herbet et al. (
2017a,
b) showed that direct electrical stimulation of the IFOF in the right hemisphere is associated with poorer performance in non-verbal semantic processing. Moreover, in a meta-analysis performed by Vigneau et al. (
2011), the cognitive load of semantic tasks determined the contribution of right IFOF: when the task involved working memory to manipulate verbal content, or the capacity to switch between categories in a verbal fluency test, the association with right IFOF was stronger.
In the current study, we investigated these diverse hypotheses about the functional relevance of subdivisions of left and right IFOF using a semantic Go/No-Go task. In semantic conditions, decisions to produce or withhold a prepotent button press response were made on the basis of conceptual content (whether a visually-presented stimulus was an animal or a manmade object), while in non-semantic conditions, these decisions were based on the shape of the box enclosing scrambled images, such that semantic access was not necessary for the task. The semantic conditions also compared written words and pictures, to allow us to consider the extent to which effects related to language, transmodal aspects of semantic cognition or visual semantic processes necessary only for pictorial tasks. Semantic efficiency should show an association with tracts that link visual inputs to conceptual regions needed to select an appropriate response based on meaning, particularly for verbal trials, since written word inputs do not contain any superficial cues to the behaviourally relevant categorical distinction, unlike pictures. In contrast, the inhibition efficiency for both semantic and non-semantic Go/No-Go tasks is expected to relate to tracts that connect visual regions to dorsal parts of prefrontal cortex supporting action selection and inhibition. In summary, given the distinct anatomy of dorsal and ventral IFOF, we expected dorsal IFOF to be relevant to visually-guided Go/No-Go behaviour across semantic and non-semantic domains, while ventral IFOF was expected to be associated with controlled semantic behaviour.
Discussion
This study examined the associations between distinct IFOF subdivisions in the left and right hemispheres and behavioural inhibition guided by semantic versus perceptual properties of the stimuli. Inhibition efficiency showed a significant interaction with tract integrity for both the dorsal and ventral IFOF in the left hemisphere but there were no significant associations in the right hemisphere. In left dorsal IFOF, tract integrity interacted with the efficiency scores of the inhibition task across both semantic and non-semantic domains: there was an association with performance on both the Word and Hard Perceptual conditions; behaviourally, these were also the two hardest conditions. The integrity of the dorsal IFOF tract also predicted the magnitude of the effect of difficulty. Left ventral IFOF showed a somewhat different pattern: the integrity of this tract predicted performance on the Word task, with no association for non-semantic inhibitory control. This association with tract integrity was greater for the verbal than the picture-based semantic decisions—and this semantic modality effect could not be readily explained in terms of an influence of general task difficulty since left ventral IFOF did not show a difficulty effect for the non-semantic trials. Lastly, we investigated hemispheric differences in these tract-inhibition associations. All the task conditions, except Picture decisions, showed better performance in participants who had stronger left than right hemisphere IFOF tracts, although this pattern only reached statistical significance in the dorsal IFOF.
These findings are highly consistent with the proposal of functional differences between the dorsal and ventral subdivisions of the IFOF in the left hemisphere, with dorsal IFOF showing greater involvement in the regulation of difficult perceptual Go/No-Go trials, while ventral IFOF is engaged exclusively when inhibition is guided by abstract meaning (Rollans
2016). The semantic effect in the ventral IFOF was stronger for the verbal task, perhaps because decision-making in these trials required abstract categorical information to be more fully accessed. Unlike written words, the picture semantic trials provided visual feature cues about the category of each stimulus: for example, animals have shared visual features, such as eyes and tails, and the presence of these features even in the absence of conceptual identification could have been sufficient to drive an appropriate response to pictures. Tract integrity in dorsal IFOF, in contrast, was linked to the more difficult semantic and non-semantic decisions; like ventral IFOF, the dorsal subdivision showed stronger behavioural associations with word than picture performance, but there are alternative interpretations of this difference. First, the effect of modality in the semantic inhibition task in left dorsal IFOF may have reflected the greater need for orthographic to phonological conversion processes and/or visual − spatial processes needed to read words, since previous studies have previously implicated dorsal IFOF in reading (Motomura et al.
2014) and phonology (Almairac et al.
2015). However, this interpretation cannot explain the distinction between easy and harder non-semantic performance. An alternative interpretation is that left dorsal IFOF supports more difficult visually guided decision-making, as for both non-semantic decisions (are the lines of a box off vertical?) and semantic decisions (is the object an animal or manmade?), behavioural associations with left dorsal IFOF were stronger when the information needed for decision-making was less salient. Overall, left dorsal IFOF was associated with more demanding task conditions across domains (both word and hard perceptual trials), indicating that its function is not specific to language or semantic cognition; this pattern of results suggests that dorsal IFOF contributes to the ‘multiple demand network’ that supports domain-general executive processing (Duncan
2010,
2001). The functional distinction between dorsal and ventral IFOF is also consistent with the proposal that there are dissociable yet spatially proximal mechanisms supporting different aspects of cognitive control in the prefrontal cortex, including the possibility that the multiple demand network can be distinguished from semantic control mechanisms (e.g., Chiou et al.
2023; Gao et al.
2021; Whitney et al.
2011), and the observation that parcellations of resting-state connectivity patterns identify distinct networks linked to control (Dixon et al.
2018; Schaefer et al.
2018). The differences between dorsal and ventral IFOF reported here also fit well with the findings of Rollans (
2016), who found a dissociation between visually guided inhibition and picture naming: however, the current study utilizes cleaner contrasts between similar task conditions to show this dissociation.
We also found evidence that our Go/No-go task was more reliant on the left hemisphere overall, for both semantic and non-semantic decisions. This is surprising since a functional neuroimaging study of the same task showed responses that were strongly right-lateralized in dorsolateral prefrontal cortex (Gonzalez Alam et al.
2018), alongside bilateral occipital − temporal activation. Only the contrast of semantic over non-semantic decisions revealed clusters in left prefrontal regions including in left inferior frontal gyrus; consequently, only semantic conditions of this task might be expected to be more reliant on white matter tracts projecting from occipital − temporal to prefrontal cortex in the left hemisphere. The unexpected hemispheric effects in the current study could reflect differences between tractography and BOLD fMRI, since left IFOF might be more important than right IFOF for some aspects of the task and yet connections between left and right prefrontal cortex could still result in a stronger response in the right hemisphere if, for example, right pre-supplementary motor and dorsolateral prefrontal regions play a stronger role in motor control (Aron et al.
2014; Cai et al.
2012; Hannah and Aron
2021; Rae et al.
2014). One relevant aspect of the task could be inhibition efficiency: whilst the right-lateralised activation observed in fMRI studies might be related to motor inhibition, our study used efficiency scores of an inhibition task as a metric. In support of this, Hirose et al. (
2012) found a separation of inhibitory processes using an efficiency index: while the right hemisphere showed the usual neural substrates associated with response inhibition, inhibition efficiency specifically was associated with a set of structures in the left hemisphere, including temporal and frontal regions which might be subserved by the IFOF.
There are of course some limitations of our approach. First, the functional differentiation of the IFOF may extend beyond the left and right dorsal and ventral subdivisions examined here. In a study with diffusion data that resolved “kissing fibers” and with “high-angular resolution”, Wu et al. (
2016) proposed five divisions of the IFOF, involving orbito-frontal, dorsolateral frontal, angular gyrus and marginal gyrus portions. Other researchers have proposed graded and continuous variation of the location of tract terminations across IFOF; by this view, while there are functional dissociations within IFOF in line with our data, this functional variation does not necessarily reflect distinct tract subdivisions (Weiller et al.
2021). Secondly, given that IFOF has been previously associated with the integrity of the semantic control network (Nugiel et al.
2016), which links heteromodal posterior middle temporal and inferior frontal regions implicated in the retrieval of non-dominant conceptual information, we cannot rule out the possibility of different tract integrity findings in studies that directly manipulate the need to control conceptual retrieval. Our semantic conditions required participants to use word and picture meaning to decide whether to press a button or withhold this response—but did not require conceptual retrieval itself to be controlled. Future studies could clarify whether distinct aspects of IFOF support (i) the control of meaning retrieval (using classic semantic control manipulations such as contrasts of weak and strong associations); (ii) the control of behaviour based on meaning (as in the semantic conditions of the current task) and/or (iii) the control of behaviour based on non-meaningful aspects of visual inputs (as in the Easy/Hard Perceptual conditions used here). This type of study might provide new information about the partial separation of the semantic control network from the domain-general multiple-demand network (Davey et al.
2016; Gao et al.
2021; Gonzalez Alam et al.
2018; Wang et al.
2020). Thirdly, the current investigation was restricted to examining subdivisions of the IFOF (since tract tracing for each individual participant is time-consuming and there are statistical limits on the number of tracts that can be investigated given our sample size). However, IFOF does not underpin semantic cognition alone and in future work, it will be important to consider how interactions between tract strengths for IFOF, uncinate fasciculus and ILF underpin distinct aspects of semantic and non-semantic cognition. A related issue is that the study did not include a control tract beyond the semantic and cognitive control domains. As a consequence, we cannot fully confirm the specificity of our findings. However, since semantic cognition and cognitive control both draw on highly distributed networks (e.g., Duncan
2010; Jackson
2021), it is challenging to confidently identify tracts that would not be expected to make
any contribution to these functions. Finally, there are some potential methodological weaknesses in our analysis. We used an angle threshold of 35° to perform streamline tractography, in line with some earlier investigations that have used angles under 45° (Forkel et al.
2014; Wakana et al.
2007) and with strategies and recommendations given our tract of interest and resolution (Mori et al.
2002; Mori and Van Zijl
2002), yet other studies have used 60° (Caverzasi et al.
2014; Wu et al.
2016), or 45° (Hau et al.
2016). Using a smaller angle of threshold has some advantages (as well as potential disadvantages): the IFOF has a complex course that crosses other tracts, and a smaller angle is thought to make it easier to exclude fibres that are not part of IFOF (Thomas et al.
2014). Also, since IFOF is a relatively small tract, using a smaller angle threshold might allow for more detailed tract reconstruction (Mori and Van Zijl
2002). However, fibres targeting the superior parietal cortex and middle/superior frontal gyrus may be difficult to track with this parameter value, given their geometrical orientation.
In conclusion, using an individual differences approach in healthy participants, we show broad involvement of IFOF in both semantic cognition and visually guided decision-making. These findings are potentially consistent with recent neuroanatomical accounts which suggest that IFOF connects prefrontal cortex to both posterior heteromodal semantic regions in posterior temporal cortex (although the presence of these connections is variable across subjects) and visual regions in occipital and ventral temporal cortex (Catani and Thiebaut de Schotten
2008; Duffau et al.
2013; Giampiccolo and Duffau
2022; Martino et al.
2010; Nugiel et al.
2016).
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.