Sections were deparaffinized, hydrated, and boiled in 10 mM citrate buffer (pH 6) for 20 min for antigen retrieval, rinsed in Tris-buffered saline (TBS, pH 7.6, 0.05 M), and incubated in the primary antibodies overnight in a humid chamber. After rinsing, they were incubated in the corresponding biotinylated secondary antibodies (rabbit anti-mouse IgG or goat anti-rabbit IgG; Dako, Glostrup, Denmark), diluted at 1:200 in TBS, followed by incubation with avidin-biotin complex (ABC, DAKO) in TBS. Bound peroxidase was revealed using 0.04% 3,3-diaminobenzidine (Sigma, United States), 0.05% ammonium nickel (II) sulfate, and 0.03% hydrogen peroxide in TBS, pH 7.6. Sections were dehydrated, cleared, and coverslipped using Eukitt (O. Kindler, Freiburg, Germany). Negative controls omitted the primary antibodies.

The following primary antibodies were used: Mouse monoclonal anti-reelin antibody 142 [IgG1, 1:500, (gift of A. Goffinet), 1/500; Rabbit polyclonal anti-Calretinin, Swant, 7699/4, 1/3000; Mouse monoclonal antibody anti-PCNA, Thermo Scientific, Ab-1 (clone PC10) 1/1000; Rabbit polyclonal anti-Tbr1, Abcam, ab31940, 1/300; synthetic peptide within human Vimentin aa 400 to the C-terminus (acetyl), 1/200, Abcam]; Rabbit polyclonal anti-nestin, Abcam, ab 93666, 1/100; Rabbit polyclonal anti-MAP2, Sigma, HPA 012828, 1/100; Rabbit polyclonal anti-Eomes (Tbr2) Sigma, HPA028896, 1/100.

Antigens were immunolabeled sequentially by using primary antibodies (Tbr1 and CR; CR and PCNA) generated in rabbit. The first antibody was developed using DAB/nickel as chromogen. Thereafter, sections were rinsed in TBS and incubated overnight with the second antibody. After incubation with the biotinylated secondary antibodies and ABC as described above, sections were developed by using DAB alone as chromogen. Sections were dehydrated, cleared in xylene, and cover-slipped with Eukitt (Freiburg, Germany). Photographs were taken with a Zeiss Axio microscope equipped with an AxioCam MRc5 digital camera and AxioVision LE 4.6 software. Images were processed using Adobe Photoshop CS2 for adjustment of brightness and contrast.

The insula of perinatal (32–40 GW) brains displayed an adult-like gyration pattern (Figures 1B,C), although the case in Figures 1B,C, 40 GW) presented only two short anterior gyri with one accessory anterior gyrus, and two long posterior gyri on each side. Another 40 GW brain and the 32 GW case showed the more common configuration of three short anterior and two long posterior gyri. As in the adult (Mesulam and Mufson, 1985), three main modalities of insular cytoarchitecture were recognizable at term, following a gradient from rostro-ventral to caudo-dorsal, independently of the sulcal pattern: The antero-basal sector of the insula near the limen (Field IC of Von Economo and Koskinas, 1925; Figure 1A) was agranular, showing a prominent layer V but absence of the inner granular layer IV (Figure 1D). At levels rostral to the limen, the insular cortex was continuous with the POC via a small transition area where the neurons lacked any recognizable lamination, with superficial medium-sized pyramidal cells and deeper smaller pyramidal and non-pyramidal cells distributed apparently at random. The anterior gyri had a variable prominence of layer IV and were thus considered dysgranular (Figure 1E), whereas in the posterior and dorsal insula layer IV was wider, more cell-dense, and radially organized (Figure 1F), features characteristic of a granular isocortex. Nonetheless, the width of layer IV was variable, even along the same gyrus. The local heterogeneities of layer IV may reflect the immaturity of the perinatal brain, but may also be due to laminar distortions when a gyrus changes orientation or undergoes additional folding.

At 32 GW, the insula showed the same basic folding and lamination pattern as at 40 GW, even though neurons appeared slightly less mature, with a higher cell density than at term. Transitional (Figure 1G), dysgranular (Figure 1H) and granular (Figure 1I) regions were clearly established. At both 32 and 40 GW, the outer granular layer II (Figures 1E,F) was more cell-dense than in the adult, due to the inside-out migration gradient of the cortex, according to which layer II is the last layer to develop (Angevine and Sidman, 1961; Rakic, 1974).

We conclude that the insula acquires an adult-like gyration and architectonic pattern during the last trimester of gestation. We did not detect the VEN (Allman et al., 2011), possibly because our material did not include the fronto-insular transition area where they are more numerous, and because they mature at later stages (Allman et al., 2005).

We tried to identify the possible proliferative sources and their migratory routes into the insula through analysis of radial glia architecture. At 11–12 GW, basal or outer radial glia (oRG) appeared in the SVZ at the PSB, which now occupied a PCNA+ and Tbr2+ wedge-shaped area at the intersection of IC and EC. This SVZ was wider than the SVZ in the adjacent cortex (Figures 6A,B). Radial glia processes originating in the SVZ at the PSB were positive for vimentin and nestin, and assembled in a distinct RGF coursing along the prospective EC. Following the curvature of the PSB, the RGF originating from its SVZ formed a continuous band of radial glia fibers. According to their origin, four subdivisions of the RGF could be established, which remained constant during the first half of gestation: PF, fronto-parietal (FP), PT, and T, temporal radial glia fibers, which took different directions to reach their destination in the insula. Descending RGF component FP merged with ascending RGF component T (Figures 5B, 6D,E), whereas component PF descended without an ascending counterpart (Figure 5A). Component PT at the caudal end of the BG (putamen islands interspersed between the posterior limb of the IC) (Figure 5C) coursed straight to the caudal insula, where also an intermediate zone was recognizable. Along its course, the RGF fanned out and entered the insula at different dorso-ventral levels, maintaining an initially parallel orientation up to the point where descending and ascending components met (Figure 5B).

At 13/14 GW, migrating neurons, still positive for Tbr1, followed the direction of the RGF into the insula, populating its subplate and CP, clearly delimiting cortical and subcortical territories (Figures 6C,G). After this age, migrating neurons were Tbr1-negative. The wedge-shaped SVZ, marked with vimentin (Figures 6D–F), nestin, PCNA (Figures 6A,D), and Tbr2 (Figures 6B,F), progressively increased in width, partially entering the IC, and expanded in parallel with the proliferation of the outer SVZ (oSVZ) in the adjacent opercula. We reconstructed the course and orientation of the RGF components into the insula in a horizontal section at 21 GW (Figure 7). As expected, the RGF arose from both the dorsal and ventral oSVZ at the PSB, although the dorsal RGF predominated. The orientation of the fibers leaving the fascicle and entering the insular subplate is shown for different levels (Figure 7, from 1–4). At levels 1 and 2, fibers emerging from the RGF component FP took a descending course, while at level 3 (component T) they ascended. Level 4 shows radial glia fibers in the intermediate zone of the temporal operculum, which seemed to bend and course toward subplate and CP of the superior TC rather than into the insula. The RGF was compressed in the EC and contained also vimentin+ cells. However, PCNA+ mitoses were rare, and we did not observe Tbr2+ progenitor cells in this location; this indicates that the territory of the RGF was not an extension of the proliferative oSVZ.

Importantly, the PF PSB (RGF component PF) extended farther rostrally than the temporal one (RGF component T) (Figure 5A); in consequence, the most anterior region of the insula received radial glia fibers only from the PF PSB. The middle and posterior insular regions, in turn, received radial fiber-mediated migrations from the dorsal (FP) and ventral (temporal) PSB (Figure 5B). Even more caudally, the posterior end of the insula was close to the PT junction, and was also populated by large numbers of oRG cells and processes. At this level, an intermediate zone was recognizable as a fiber-rich layer continuous with the intermediate zone of the adjacent cortices (Figure 5C).

The distinct subregions of the insular lobe were thus connected more or less abundantly via the RGF with the proliferative SVZ along the PSB. The differential availability of progenitor cells may be the basis for the differential layering of the subareas of the insular lobe. It is remarkable that the insular lobe, despite its rather distant relationship with the proliferating oSVZ, and the limited progenitor pool at the PSB, is able to fold once migration is finished (Figure 1B).

Vimentin, a marker of radial glia, does not define the PSB. To determine the origin of the RGF at the PSB at midgestation (21GW), we compared adjacent sections stained for vimentin and Tbr2 (Figure 8). When reconstructing the photomosaics for both markers from high magnification microphotographs, we noticed that the periventricular layers, inner and outer SVZ, were quite different in the various lobes. Similarly, the fiber tracts also differed, and widely varied in thickness. Particularly in the frontal and parietal lobes, the intermediate zone (future white matter) was at this time point much wider than in the temporal lobe. In parallel, Tbr2+ and PCNA+ cells extended much farther into the IZ in FP areas than in the TC. In the TC, the anterior commissure seemed to represent an obstacle for the RGF, since it was traversed neither by the RGF nor did it contain Tbr2+ cells. We limited our study to the PSB and the adjacent regions, where the Tbr2+ progenitor cells became less and less numerous toward the IC (Figures 8B,D), and were basically absent at the level of the insula. The RGF was more difficult to discern than in earlier stages, but still very thick, prominent radial glia fibers emerged at the PSB and crossed the inner fibrous layer and adjacent IC (Figures 8A,C).

While at early fetal stages the PSB extended medially beyond the cortico-striatal sulcus, at midgestation it had shifted laterally in the temporal lobe, but not in the frontal lobe. Future studies will show how the PSB in different lobes behaves toward the end of cortical neurogenesis.

During the first half of gestation, the insular cortex had a uniform structure, with a CP formed by densely aggregated immature neurons. On the whole, the insular CP was narrower than that of the adjacent opercular areas. The first clear appearance of layering was at midgestation (20/21 GW), when MAP2 (Figure 9D) and Tbr1 (Figures 9C,F) immunostaining (in maturing human cortical neurons, Tbr1 is cytoplasmic) indicated the presence of a distinct layer V, or inner pyramidal layer, which was particular prominent in the anterior agranular insula (Figure 9C). However, compared with the Betz cells in the adjacent primary motor cortex (Figure 9E), the dimensions and proportions of layer V pyramids in the anterior insula were rather reduced, as well as their positivity for MAP2. The deep layer VI and subplate were also Tbr1+, although in this case staining was nuclear (Figure 9F). The subplate was directly continuous with the subplate of the adjacent opercular cortices, but considerably reduced in width. Remarkably, the rostral and intermediate insula lacked an intermediate zone, characterized by horizontal fibers traversed by clusters of migrating neurons (Bystron et al., 2008; Figures 6E,G, 8C). In the posterior insula, the intermediate zone was present, traversed by calretinin+ fibers from the posterior limb of the IC (not shown).

Also at midgestation, we observed a difference in the distribution of the Reelin+ axonal plexus of the Cajal–Retzius cells in the lower marginal zone (Meyer and González-Gómez, 2017): The plexus, characteristic of isocortex and important for laminar arrangement of neurons, was absent in the periallocortical transition area (Figure 9B), but appeared in the more dorsal, isocortical insula. Cajal–Retzius cells were, however, abundant in the upper marginal zone all over the insula.

In the ages examined after midgestation, 24 and 25 GW, the overall immaturity of the insular CP persisted, and there was no evidence for a future differentiation into an agranular, dysgranular and granular cortex. Since at 32 GW the insula presented an adult-like morphology, cytoarchitectonic maturation would take place during the interval between 25 and 32 GW. Similarly, in our coronal sections at 24 and 25 GW, we were unable to distinguish incipient insular sulcation and gyration.

The human insular lobe is shaped by the curvature of the lateral ventricle, and lies embedded between PF, frontal, and parietal areas dorsally, and temporal areas ventrally, from which it is separated by the Sylvian fissure and the circular sulcus. Its position external to putamen and IC prevents direct contacts with the periventricular proliferative zones, and radial migrations to the insula have to take indirect, unusual routes to reach and populate the distant lobe. Using radial glia architectonics and the pallial markers Tbr1 and Tbr2 (Hevner et al., 2003; Englund et al., 2005), we identified the PSB as the origin of the neurons destined to form the insular cortex. During early fetal development, the PSB follows the curvature of the ventricle, and thus extends from PF to temporal levels. We suggest that local differences of the PSB along its extent through the various lobes contribute to the multifaceted lamination pattern of the insular cortex. The different timing of the entrance of the IC into the cortex traversing the PSB (Krsnik et al., 2017), with rostral parts preceding more caudal parts, is an additional influencing factor, which will require further studies.

The finding that the architecture of the insula is contingent on size and orientation of the lateral ventricle and the curved shape of the PSB, explains the differences in insular structure reported in a variety of mammals. Even though comparative studies tend to emphasize common principles of brain structure in order to establish homologies (Cat: Clascá et al., 1997; monkey: Gallay et al., 2012; Evrard et al., 2014), anatomical studies in a wide variety of mammalian species usually not studied in the laboratory, including the dolphin (Jacobs et al., 1984; Casanova et al., 2010), demonstrated an extremely variable shape, general organization, lamination and cellular specialization of the insula, to the point that there is no recognizable common model of organization of the mammalian insular cortex (Butti and Hof, 2010). The relationship between insula and claustrum is similarly controversial. We neglected the claustrum, because we did not detect migratory mechanisms similar to those of the insula (Mufson and Mesulam, 1982; Nieuwenhuys, 2012). Developmental gene expression studies postulated that insula and insular claustrum are formed from the lateral pallium (Watson and Puelles, 2017). It may be argued that the mouse is not the best model for the human cortex, taking into account recent studies suggesting that the ancestor of mammals was probably a gyrencephalic animal (O’Leary et al., 2013; Lewitus et al., 2014). The magnitude of species differences suggests that the development of the insula is best understood when examined together with the anatomical landmarks that define this lobule in a given species.

In any case, the human insula should be considered in the context of its complex cognitive, social and emotional functions, including empathy, altruistic behavior, self-awareness, interoception, mindfulness, and consciousness (Craig, 2009; Fan et al., 2011; Fischer et al., 2016; Tusche et al., 2016; Laneri et al., 2017), or, as expressed by Craig (2010), the “sentient-self.” In keeping with these human-specific functions, a volumetric comparison of the insula of human and non-human primates revealed that in terms of absolute volumes the left and right agranular insula are among the most enlarged cortical areas relative to the chimpanzee (Bauernfeind et al., 2013). The degree of “granularity” of a cytoarchitectonic area is thus unrelated to its involvement in networks engaged in human cognitive functions such as those attributed to the anterior insula.

The PSB has been extensively studied in terms of comparative anatomy, establishing homologies with the anterior dorsal ventricular ridge (ADVR) of sauropsids, and the fate of the cells in the LCS in sauropsids and mammals (Molnár and Butler, 2002), as well as in terms of developmental gene expression. Gene expression studies in the PSB of mice showed that the PSB is the main source of POC, claustrum, olfactory bulb, olfactory tubercle, and amygdala, and that the cells destined to these centers migrate ventrally via the LCS (Medina et al., 2004; Carney et al., 2006; Cocas et al., 2011). The insula as a PSB derivative has received less attention, probably because it is so unconspicuous in the rodent. The PSB is certainly an important landmark in the rodent brain, where it represents the boundary between pallium and subpallium, and is implicated in dorsoventral patterning of the telencephalon (Stoykova et al., 2000; Yun et al., 2001), but to what extent is it relevant for the human brain? Molnár and Butler (2002) recognized the evolutionary potential of the PSB, and also its function as an initial barrier zone for crossing cortico-thalamic and thalamo-cortical axons in the IC. However, the rodent studies did not take into account the enormous regression of olfactory structures in the microsmatic human, which rests importance from this aspect of the LCS. Instead, we show that the early descending migrations from the human PSB are mostly destined to the insula, although they also contribute Tbr1+ and calretinin+ cells to the POC. Furthermore, the rostro-caudal extension of the PSB is hugely increased due to the size increase of both, cortical progenitor zones and GE. Another important factor is the IC, which crosses the PSB (Molnár et al., 2012), and thus represents an additional obstacle for radial migration from the PSB to the insula. We suggest that the prominence of the PSB in human is an important factor for the development of the insular lobe. Furthermore, in the human brain, the insula does not represent the most lateral part of the cortex as in the rodent, but rather emerges as the core cortical region that makes possible the mirror arrangement of dorsal (FP) and ventral (temporal) cortical lobes. The central anatomical position of the insula in the human telencephalon is also paralleled by a similar central functional relevance, since especially the dorsal anterior insula can be considered a critical hub in connectivity networks of the human brain (Uddin et al., 2014).

Radial glia has important roles in cortex development: It serves as a guidance substrate for radially migrating neurons (Rakic, 1971), and is also the principal progenitor cell type of the developing telencephalon (Malatesta et al., 2000; Miyata et al., 2001; Noctor et al., 2001; Tamamaki et al., 2001). In early embryonic stages, radial glia somata are confined to the VZ as apical radial glia, while in later stages, basal or oRG forms the proliferating cell population in the SVZ (Fietz et al., 2010; Hansen et al., 2010; Reillo et al., 2011). An oSVZ (Smart et al., 2002) is more prominent in gyrencephalic brains than in the lissencephalic rodent, and has been suggested to play key roles in the folding of the neocortex in gyrencephalic species because of its abundance in oRG that promote cortical expansion (Fietz et al., 2010; Hansen et al., 2010; Lewitus et al., 2013; Martínez-Martínez et al., 2016). On the other hand, comparisons of lissencephalic primates, gyrencephalic rodents, and carnivores suggested that the cytoarchitectonic subdivisions of the SVZ are an evolutionary trend and not a primate-specific feature, and that a substantial population of oRG exists unrelated to the degree of cortical folding (García-Moreno et al., 2012; Kelava et al., 2012; Martínez-Cerdeño et al., 2012, 2016). The expression of transcription factors Pax6 and Tbr2 varies between the various types of progenitor cells and species (Hansen et al., 2010; Kelava et al., 2012; Betizeau et al., 2013; Cunningham et al., 2013), although the expression of Tbr2 in neural precursor cells can be used for defining the boundaries of the SVZ both developmentally and evolutionarily (Martínez-Cerdeño et al., 2016). It is not clear whether also the intermediate progenitor cells, which lack a polarized process, express the pan-radial glia marker vimentin. In any case, the vimentin-expressing radial glia cells in the SVZ of our human material displayed the various morphotypes described in the monkey (Betizeau et al., 2013), where they indicate differential mitotic potentials and cell-cycle parameters. Radial glia-guided locomotion is the migration mode of excitatory cortical neurons on their route through intermediate zone, subplate and CP, until they reach the marginal zone, detach from the radial glia fiber, and change to a somatic translocation mode (Tissir and Goffinet, 2003; Sekine et al., 2011). The critical question is the degree of horizontal dispersion along the radial glia route. Radially migrating neurons may change from one radial glia fiber to an adjacent one, so that the strictly radial orientation of a radial glia fiber does not necessarily imply a similar radial course of the migrating neurons (Reillo et al., 2011; Gertz and Kriegstein, 2015). It is thus possible that the origin of neurons in the insula might be more extensive than described here, and that the SVZ of the adjacent opercula contribute neuroblasts dispersing tangentially into the insula. We suggest, however, that it is precisely the absence of tangential dispersion between adjacent lobes that leads to the enormous growth of the opercula, versus the restricted expansion of the insula. The distinct RGF into the insula is another argument for an origin from a specific sector of the oSVZ at the PSB. The RGF represents the main radial migration substrate from the PSB to the insula, and is set apart from the non-fasciculated oRG fibers originating from the oSVZ of the opercula. It is possible though that tangential dispersion within the spatially compressed RGF contributes to the inhomogeneities of layer IV. An important feature of the human RGF is that it is composed of descending and ascending fibers and thus connects the derivatives of the FP and temporal SVZ, which act in concert in the formation of the insula. It appeared, however, in our material that the dominant source of insular neurons is the dorsal PSB, whereas the temporal PSB contribution is less substantial. Future studies using region-specific markers might solve this question.

The human neocortex displays a large diversity of size and density of neurons, which are arranged in six horizontal layers of variable width. This diversity is the foundation of the cytoarchitectonic areas described by Brodmann (1909) and Von Economo and Koskinas (1925).

A major criterion in these classifications is the differentiation of the granular layers II and IV, which define the degree of granularity of a given area. Primary sensory areas have a particularly prominent layer IV, which is the main target of thalamo-cortical fibers, and populated by its principal neurons: spiny stellate cells, star pyramids, and small to very small “dwarf pyramids” (Von Economo and Koskinas, 1925). Spiny stellate cells are glutamatergic excitatory neurons (Conti et al., 1989; Feldmeyer et al., 1999), which establish asymmetric synapses mainly with dendritic spines (Saint Marie and Peters, 1985), give rise to interlaminar projections (Qi and Feldmeyer, 2016), and even connect adjacent areas (Meyer and Albus, 1981). In human auditory cortex, layer IV is populated by transitional forms between spiny stellate and small pyramidal cells (Meyer et al., 1989). However, layer IV is also prominent in many other cortical areas, such as the posterior insula, where anatomical and functional studies stressed its afferent sensory (gustatory, auditory, and vestibular) input (Augustine, 1996; Kurth et al., 2010).

There is a general agreement in the basic classification of the insula into agranular, dysgranular and granular parts (Mesulam and Mufson, 1985), even though the number of subareas varies between authors (reviewed by Nieuwenhuys, 2012). In our perinatal material we tried, following Von Economo and Koskinas (1925), to section the insula in a plane perpendicular to the main axis of most insular gyri. However, the insular gyri undergo subtle changes in orientation, have multiple dimples and small subsulci, which altogether distort the naturally vertical columnar arrangement of all layers including layer IV, and may give the impression of a distinct cytoarchitectonic subdivision.

As described here, the insula derives from the PSB, which in turn is continuous with the VZ and SVZ of the adjacent lobes. In this sense, the agranular character of the anterior insula reflects the trend of the agranular PF and frontal areas, whereas the granular caudal insula resembles the hypergranular character of the adjacent parietal and temporal sensory areas. The dysgranular character of the intermediate areas of the insula may be attributed to the contribution and possible intermixture of RGF compartments FP and T, as well as to the possibility of tangential dispersion within the RGF.

In the literature on the oSVZ, it is often emphasized that the subgranular layers V and VI derive from the VZ and inner SVZ, whereas the supragranular layers III and II originate from the oSVZ (e.g., Smart et al., 2002; Lukaszewicz et al., 2005; Nowakowski et al., 2016), leaving open the question of the origin of layer IV. Species differences may account for this apparent neglect. As shown by Martínez-Cerdeño et al. (2012), the peak in number of Tbr2+ progenitors and mitotic divisions in the oSVZ of the macaque somatosensory cortex occurs during the generation of layer IV, whereas in rat and ferret this peak is at the end of cortical neurogenesis, when the supragranular layers are born. It is thus tempting to propose that layer IV of the human insula derives predominantly from the oSVZ. The degree of granularity of a given cytoarchitectonic area may thus depend on the availability of oSVZ progenitors characterized by high-output cycling parameters during a precise time window (Betizeau et al., 2013), along with the presence of an adequate migration substrate. The generation of a cell-rich layer IV is an important issue particularly in the primate and human brain, where the classification into anatomical and cytoarchitectonic areas is largely based on the granularity of the inner granular layer. Interestingly, the “highest” cognitive functions within the insular lobe are attributed to the agranular anterior insula, which demonstrates that the agranular character of a cortical area is not an indication of less complex functions. It would be desirable that future ontogenetic and phylogenetic studies of the cerebral cortex become more focused on the peculiarities of the human brain (Clowry et al., 2010).