Fiber dissection in cadavers and DTI tractography studies in humans have investigated the structural anatomy of the SLF/AF complex [72, 73]. The different components of the perisylvian SLF were isolated and the tracts were followed until their cortical terminations. Three segments of the perisylvian SLF were detected: (1) anterior segment of the lateral SLF, that connects the supramarginal gyrus and superior temporal gyrus (in the region just posterior to the Heschl’s gyrus) with the ventral portion of the precentral gyrus (ventral premotor cortex), (2) posterior segment of the lateral SLF, that connects the posterior portion of the middle temporal gyrus with the angular gyrus, and (3) long segment of the AF, deeply located, stemming from the caudal part of the temporal lobe, mainly the inferior and middle temporal gyri, that arches around the insula and advances forward to end within the frontal lobe, essentially within the precentral gyrus and posterior portion of the inferior and middle frontal gyri. Based on these original data challenging the traditional view, it was suggested that the fibers from the posterior part of the superior temporal gyrus are part of the anterior portion of the perisylvian SLF and not of the AF [72].

In awake patients performing a picture-naming task, cortically, phonemic paraphasias have been generated by DEM of the inferior parietal lobule and inferior frontal gyrus in the dominant hemisphere [74]. Axonally speaking, these phonemic paraphasias were evoked when stimulating the AF [74, 75], possibly associated with repetition disorders [76]. This is in agreement with the theory by Geschwind [77], who postulated that damages of this bundle would produce conduction aphasia, including phonemic paraphasias and repetition disturbances, and this supports the role of the sub-part of the dorsal stream mediated by the AF in phonological processing. Interestingly, the posterior cortical origin of the AF within the posterior part of the inferior temporal gyrus corresponds to the visual object form area [72]. Indeed, this region represents a functional hub, involved both in semantic and phonological processing dedicated to visual material [78, 79]. Therefore, phonological processing sustained by the AF is performed in parallel to the semantic processes implemented by the ventral route (see below). In addition to this direct dorsal route, the indirect dorsal stream composed of the lateral SLF is involved in articulation and phonological working memory, as demonstrated by DEM. Cortical areas eliciting articulatory disorders are located in the ventral premotor cortex, supramarginal gyrus and posterior part of the superior temporal gyrus [80, 81]. Axonally, stimulation of the white matter under the fronto-parietal operculum and supramarginal gyrus, laterally and ventrally to the AF, elicited anarthria as well [74, 81]. This bundle corresponds to the part III of the SLF according to Makris et al. [82]. Indeed, this lateral operculo-opercular component of the SLF constitutes the articulatory (auditory-motor) loop, by connecting the supramarginal gyrus/posterior portion of the superior temporal gyrus (which receives feedback information from somatosensory and auditory areas) with the frontal operculum (which receives afferences bringing the phonological/phonetic information to be translated into articulatory motor programs and efferences toward the primary motor area) [72, 76, 81].

Using the same paradigm, DEM also supported that syntactic processing was mediated by delocalized cortical regions (including left inferior frontal gyrus and posterior middle temporal gyrus) connected by a sub-part of the left SLF. Interestingly, this sub-circuit is interacting but independent of the sub-network involved in naming, as demonstrated by a double dissociation between syntactic (especially grammatical gender) and naming processing during DEM. These findings support a parallel rather than serial theory, calling into question the principle of “lemma” [83].

The ventral stream connects the occipital, parietal and posterior temporal areas with the frontal lobe. This ventral route is referred by some authors as “extreme capsule fiber system” with reference to connectivity studies in the primate [84]. It seems nonetheless more adapted to talk about fascicles rather than “extreme capsule”, because the latter only considers a discrete anatomical structure while the former considers actual neural pathways with their cortical termination, in a connectomal view of brain processing. Indeed, if one takes account of the sole subcortical region without any considerations regarding the cortical epicenters connected by these white matter fibers, this does not allow the understanding of the whole eloquent network.

Regarding structural anatomy, the ventral route is composed of direct and indirect pathways. The direct pathway is represented by the IFOF, that has never been described in animals, explaining the controversy about its role [46]. In humans, the IFOF is a ventral long-distance association bundle that connects the occipital lobe, parietal lobe and the postero-temporal cortex with the frontal lobe. Recent anatomic dissections combined with DTI have investigated the main course of the IFOF [85, 86]. From the posterior cortex, it runs within the sagittal stratum in the superior and lateral part of the atrium; it reaches the roof of the sphenoidal horn in the temporal lobe; it joins the ventral part of the external/extreme capsule and it runs under the insula at the posterior two-thirds of the temporal stem; then it joins the frontal lobe [85]. Two layers of the IFOF have been described [86]. The superficial and dorsal layer connects the posterior portion of the superior and middle occipital gyri, the superior parietal lobule and the posterior part of the superior temporal gyrus to the inferior frontal gyrus (pars triangularis and opercularis). The deep and ventral subcomponent connects the posterior portion of the inferior occipital gyrus, the posterior temporal-basal area including the Fusa (fusiform area at the occipito-temporal junction) and the posterior part of the middle temporal gyrus to the frontal lobe – orbito-frontal cortex, middle frontal gyrus and dorsolateral prefrontal cortex [85, 86].

In parallel, the ventral stream is underpinned by an indirect pathway, constituted by the anterior part of the inferior longitudinal fascicle (ILF) (running below the IFOF), that links the posterior occipitotemporal region (Fusa) and the temporal pole (TP), then relayed by the uncinate fasciculus (UF), that connects the TP to the basifrontal areas by running within the anterior third of the temporal stem (in front of the IFOF) [85]. Of note, the posterior part of the ILF links the occipital lobe to the posterior occipitotemporal junction (visual object form area) [70, 79]. This means that this indirect route connects the occipital/Fusa to the orbito-frontal cortex – thus partially overlapped with the IFOF. Finally, while previously described in monkey, another pathway has recently been observed in humans, the middle longitudinal fascicle (MdLF), that connects the angular gyrus with the superior temporal gyrus up to the TP and courses under the superior temporal sulcus, lateral and superior to the IFOF [89].

In the awake patient, during picture naming, DEM of the IFOF, at least in the dominant hemisphere, elicited semantic paraphasias either associative (e.g. /key/ for /padlock/) or coordinate (e.g. /tiger/ for /lion/) in more than 85% of cases [90]. It did not matter what portion of the IFOF was stimulated (parieto-occipital junction, temporal, subinsular or frontal part) [43, 56]. These language disorders were mainly induced by stimulating the superficial layer of the IFOF. Interestingly, semantic paraphasias were never observed during stimulation of the dorsal route (SLF) [74]. Moreover, IFOF stimulation may also generate verbal perseveration [91], raising the question of its role in semantic control.

The functional role of the indirect ventral pathway is still debated. On one hand, this indirect route connects areas involved in verbal semantic processing such as Fusa and lateral frontal cortex [92]. Moreover, the major cortical relay between the ILF and UF is the TP, which is a hub, i.e. a functional epicenter enabling a plurimodal integration of the multiple data coming from the unimodal systems (subserved by ILF, UF and MdLF), explaining its role in semantics and its implication in semantic dementia when bilaterally damaged [93]. On the other hand, except for the posterior part of the ILF which is involved in visual recognition and reading [78, 79, 87, 88] and for which injury generates alexia [78, 79], the indirect pathway can be functionally compensated when unilaterally damaged [94–96]. This was also confirmed by language recovery following anterior temporal lobectomy in tumor and in epilepsy surgery [94, 97]. Even if very mild and selective deficit may persist, as concerning proper name retrieval after resection of the UF [98], or a more difficult lexical access after resection of the anterior part of the ILF combined with resection of the posterior part of the inferior temporal cortex [99], this is a good illustration of the concept of “subcortical plasticity”, in which a sub-network (IFOF, direct pathway) is able to bypass another sub-network (indirect pathway) and to functionally compensate it [14, 96]. Similarly, DEM of MdLF and resection of its anterior part failed to induce any functional disorders [100], demonstrating that this fascicle converging to the TP can also be compensated.

In summary, DEM has enabled a re-examination of the classical Broca–Wernicke localizationist model of language. The new findings from axonal stimulation provide a distributed framework for future studies of language networks and for management of patients with aphasia. This new model is based on multiple direct and indirect corticosubcortical interacting subnetworks involved in syntactic, semantic, phonological and articulatory processes. It offers several advantages in comparison with previous models, especially by explaining double dissociations during damage of the ventral versus dorsal stream (semantic and phonemic disorders, respectively). Also, it takes into account the cortical and subcortical anatomical constraints [43].

Surgical resection of DLGG within the pars opercularis and/or triangularis of the left inferior frontal gyrus can be performed without generating permanent language deficits [108–111]. Language compensation may be underlain by the recruitment of adjacent regions, primarily the ventral premotor cortex (vPMC), the pars orbitaris of the inferior frontal gyrus, the dorsolateral prefrontal cortex and the insula [108]. Indeed, it has recently been demonstrated that the left vPMC (located behind the pars opercularis) cannot be (totally) removed because it was in fact the final common pathway for speech – while Broca’s area can be compensated after resection [81, 112]. Indeed, extensive neuropsychological examination following resection of Broca’s area confirmed a complete functional recovery [109]. A transopercular surgical approach through Broca’s area, even though not invaded by the tumor, has even been reported in insular DLGG in order to avoid splitting the Sylvian fissure and therefore reducing the risk of vascular injuries: no persistent language deficits have been observed [113, 115]. Interestingly, a recent probabilistic map for crucial cortical epicenters of human brain functions, based on over 700 DEM data obtained in 165 consecutive patients who underwent awake mapping for DLGG resection, challenged the classical theories of brain organization by demonstrating that Broca’s area is not the speech output area [115, 116].

The language compensation following resection of DLGG involving the posterior part of the left “dominant” superior temporal gyrus (and its junction with the inferior parietal lobule) could be explained by the fact that this complex function is organized in multiple parallel networks. As a result, in addition to the recruitment of areas immediately adjacent to the surgical cavity (e.f. the supramarginal gyrus), the long term reshaping could also involve progressive recruitment of remote regions within the left dominant hemisphere—such as the pars triangularis of inferior frontal gyrus or other left frontolateral regions—as well as controlateral sites in the right hemisphere because of transcallosal disinhibition [117]. Indeed, a recent study combining functional MRI and DEM evidenced the existence of a wide bilateral cortico-subcortical network able to compensate surgical resection of left Wernicke’s area invaded by glioma [118].

Despite a possible hemiparesis after right insular DLGG removal, likely because this region is a non-primary motor area, and possible transient speech disturbances following left dominant insular DLGG resection, all patients recovered in a personal experience—except in rare cases (2%) of deep stroke due to a damage of the lenticulo-striate arteries [114, 119–121]. QoL was even improved in about a third of patients who had epilepsy which was refractory to preoperative medical treatment, and in whom removal of the insular tissue (and even when necessary additional removal of the temporomesial structures), controlled the seizures. Moreover, it has been possible to remove the claustrum with no cognitive disorders (despite its role suggested in consciousness) in right non-dominant fronto-temporo-insular DLGG involving the deep grey nuclei [122]. The right striatum was also resected when invaded by DLGG without causing motor deficit or movement disorders, even after a long-term follow-up over 10 years: this compensation can likely be explained by a recruitment of parallel subcortical circuits such as pallido-luyso-pallidal, strio-nigro-striate, cortico-strio-nigro-thalamo-cortical and cortico-luysal networks [123].

Despite transient cerebral facial paralysis, the bilateral hemispheric cortical representation of this function explains why all patients have recovered [124]. However, if the insula is also invaded, a transitory Foix-Chavany-Marie syndrome may be produced with transient bilateral transient orofacial pharyngeal laryngeal paralysis [125].

The “rigid” somatotopic organization of the homunculus needs to be replaced by a more dynamic vision of the functioning of sensorimotor areas, based on the fact that there are redundancies, both within the primary motor cortex and between the pre-and retrocentral regions (see above). Stimulation of the precentral gyrus can regularly produce somatosensory responses, whereas stimulation of the retrocentral gyrus may lead to involuntary muscle contractions or even to stop movement—indicating the existence of a complex central network and not a succession of discrete independent sites [53]. Unmasking of the latent subnetworks can therefore be demonstrated in real time in the operating theater, allowing sensorimotor maps to be reorganized within an hour and permitting more complete resection without causing deficit: these are short term plasticity mechanisms [54, 55]. Moreover, long-term redistribution effects (such as years later after the first operation) may also occur, optimizing tumor resection during a reoperation compared to the initial surgery, including surgery in the”knob of the hand” [117, 125, 126] (see below).

Due to a dynamic organization of the whole central region referred to above, results using pre- and post-operative functional neuroimaging have suggested the possible recruitment of “redundant” eloquent sites around the cavity, within the postcentral gyrus, after removal of a DLGG involving the primary somatosentory cortex. This is in accordance with the intraoperative DEM data, showing unmasking of redundant somatosensory sites during resection, likely explained by the decrease of the cortico-cortical inhibition. In addition, recruitment of secondary somatosensory areas and/or posterior parietal cortex, primary motor area (due to strong anatomo-functional connections between the pre- and retro-central gyri), and controlateral primary somatosensory area may contribute to functional compensation to various degrees [117, 127].

This resection generally causes a typical syndrome, with transient symptoms of akinesia (which may be complete) and mutism (particularly after surgery to the left supplementary motor area), which improves rapidly over around 10 days and then resolves after a few weeks (usually after intensive rehabilitation) [128, 129]. Pre- and postoperative task-based and resting-state funtional MRI has shown recruitment of the supplementary motor area and contralateral premotor cortex, as well as a modulation of the intra-hemispheric and inter-hemispheric connectivity, contributing to recovery [130, 131]. A more detailed cognitive examination however may reveal persistent subtle but objective long-term problems, particularly with complex movements and bimanual coordination [130]. For this reason, preservation of networks subserving movement control may be considered in some patients in order to preserve an excellent quality of life (in a pianist for example) [38, 65–67].

Interestingly, while neuroplasticity is constrained by subcortical connectivity (see above), some parts of white matter pathways can be resected with no severe permanent functional deficit. Indeed, this overall picture is more complex in some respects, as witnessed by the results of tract-based cluster analyses reported in a recent probabilistic atlas of functional plasticity [6]. In this study, the vast majority of individual tracts were generally divided into a subsection with a low and a subsection with a higher functional compensation index—suggesting that groups of fibres within the same tract differ in their neuroplastic potential. This confirms observations made in clinical practice. A telling example is the ILF, which connects the temporal pole to the occipital cortex and the occipitotemporal junction. Although the posterior part of the ILF is always functional for a set of cognitive processes (including reading aloud [79] and visual recognition [78]), its anterior part appears to abandon its functional role once infiltrated by a DLGG. These gradients of plasticity in the basal inferotemporal system can be analysed with regard to the connectivity patterns in the occipitotemporal area. In addition to projections from the ILF, this region receives widespread neural connections from the posterior segment of the SLF, the IFOF and the AF or even the vertical occipital fasciculus. Accordingly, one can speculate that the information broadcast by the ILF towards the temporal pole under normal circumstances could be redistributed via other connectivities if functional compensation is prompted by glioma growth [94, 96]. In the same vein, a part of the corpus callosum invaded by DLGG may also be removed with no morbidity [132].

However, as previously mentioned, in spite of these rare exceptions, the axonal connectivity should be preserved in the vast majority of cases, to allow the occurrence of neuroplasticity mechanisms within a wide bilateral cortico-subcortical circuit—also involving the deep grey nuclei and cerebellum, as recently evidenced by means of resting-state functional MRI before and after surgery for DLGG [133].