Three different genetically engineered mouse strains were used for the present experiments: (1) a cross breed of PVcre/tdTomato mice (crossed B6;129P2-Pvalbtm1(cre)Arbr/J with B6.Cg-Gt(ROSA)26Sortm9(CAG-tdTomato)Hze/J mice) and GIN (Oliva et al., 2000) mice (PV-GIN; n = 6, male), which expressed the red fluorescent protein tdTomato in PV cells and green fluorescent protein (GFP) in a subset of SST cells, (2) SSTcre/tdTomato mice (crossed Ssttm2.1(cre)Zjh/J with B6.Cg-Gt(ROSA)26Sortm9(CAG-tdTomato)Hze/J mice) (n = 7, male) were used for visualizing all somatostatin cells, and (3) VIPcre/tdTomato mice (crossed VIPtm1(cre)Zjh with B6.Cg-Gt(ROSA)26Sortm9(CAG-tdTomato)Hze/J mice) (n = 9, male), were used to label all vasoactive intestinal polypeptide (VIP) cells. The age of animals was between 6 and 8 weeks. All used animals were obtained from the breeding facility of the University Medical Center Göttingen (Germany). Animal numbers and their suffering were restricted to the minimum. This study was carried out in accordance with the principles of the Basel Declaration and German laws on animal research (TierSchG and TierSchVersV, 2013). The protocol was approved by the LAVES (Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit).

The animals were anesthetized with an overdose of ketamin (Essex Tierarznei) and transcardially perfused with 0.9% NaCl solution to remove blood from vessels, followed by 4% paraformaldehyde dissolved in 0.1 M phosphate buffer (PB, pH 7.4). The brains were removed and the two hemispheres separated. The left hemispheres were flattened (Welker and Woolsey, 1974) whereas the right ones were kept unmodified, and all tissue was postfixed for 2 h in the same fixative. The right hemispheres were cut in the coronal plane, the left hemispheres tangentially on a vibratome (VT 1200S, Leica), both with a nominal section thickness of 50 μm.

The sections of PV-GIN animals were incubated with rabbit anti-GFP (Invitrogen) 1:2500 in TRIS-buffered saline with 0.3% Triton X-100 (TBST) for 2 days (in a cold room), followed by an anti-rabbit IgG coupled with Alexa-488 (Invitrogen) 1:500 in TBST for 2 h. The PV+ cells contained a sufficient amount of tdTomato, therefore they did not need further signal amplification. The sections of SSTcre/tdTomato and VIPcre/tdTomato animals were stained with guinea pig anti-vesicular glutamate transporter 2 (vGlut2; Millipore) 1:10,000 in TBST for 2 days, followed by an anti-guinea pig IgG coupled with Alexa 488 (Invitrogen) 1:500 in TBST for 2 h, in order to identify barrels. In a last step, sections of all strains were stained with DAPI (1:1000, Molecular Probes) to label cell nuclei. The sections were then embedded in AquaPolymount (Polysciences).

Microphotographs were taken with an epifluorescence microscope (AxioImager.M2, Zeiss, Jena, Germany; 5x A-Plan objective, NA 0.12 or 10x C-Plan Neofluar objective, NA 0.3) as “virtual tissue photomontages” controlled by Neurolucida software (MBF Bioscience, Colchester, VT, United States). Filter sets used were: for DAPI #49 (BP 365; BS 395, BP 445/50), for Alexa488/GFP #44 (BP 475/40; BS 500, BP 530/50), and for tdTomato #45 (Bp 560/40; BS 585, BP 630/75. To maintain illumination intensity comparable across sections, the dynamic range was set according to the saturation of a few (3–5) brightest pixels that belonged to somata.

Please consult Supplementary Figure 1 for a more intuitive illustration of the work flow. Consecutive tangential sections were aligned according to the pattern of vertically oriented cortical blood vessels. Neurolucida was used for registering cells and cortical domains. First, the barrels were identified and labeled individually, based on DAPI staining that was correlated with either intense PV neuropil or vGluT2 immunostaining, which is caused by thalamocortical projections into layer (L) IV. In coronal sections, the borders between LI, LII/LIII and LVa, LVb, and LVI were identified on the basis of DAPI staining. The pia mater and the LVI/white matter border were also delineated throughout the entire barrel field. Then, the location of all PV, SST, and VIP cells was registered independent of the laminar boundaries, thus the borders of the cortical domains did not bias the data collection. The resulting data files were analyzed in Neurolucida Explorer (MBF Biosciences). The following parameters were extracted: (i) cell numbers, (ii) areas of all delineated cortical domains (barrels, septa, and the related layers of the respective cortical columns), and (iii) the coordinates of all the GABAergic neurons with their distance from pia mater and white matter.

After identifying each barrel in coronal sections, the depth of the borders between cortical layers was measured for each, and the medio-lateral and rostro-caudal coordinates were also registered. These parameters were used to correct the measured volumina of the flattened tangential sections. After that procedure, the tangential sections were re-binned into 20 sections. In coronal sections we found the cortical thickness to be 1019,164 ± 38,550 μm. In case of 50 μm-thick sections it would contain a total number of ca. 20 tangential sections. By cutting the brain tangentially, we had diverse section numbers (15–19), depending on applied pressure during flattening. Therefore, we standardized the cortical thickness through re-binning the sections with the following methods. First we divided each section in 20 virtual subsections and divided the cell count of a given section equally between its virtual subsections. For example, in an animal with n = 17 sections (n = original section number), we got n’ = 340 subsections (n’ = number of a new subsection, n’ = n^∗ 20). Then, we created normalized tangential sections: one 20th of n’ of all subsections were binned together (in the example above, 17 subsections were binned in one new normalized section (Supplementary Figure 2).

The densities of cells of interest (PV+, SST+, VIP+) were plotted and the borders of cortical layers were defined, using the inflection points of these curves (Pohlkamp et al., 2014). Since the cortical and laminar thickness varies according to rostro-caudal and medio-lateral position, the extracted raw data had to be transformed to obtain statistically comparable quantities. The normalization of cortical thickness utilized landmarks, e.g., points of inflection on laminar borders as described previously for each of the markers examined (Pohlkamp et al., 2014). This normalization process transformed all the data into the same thickness range, therefore it made the comparison (or pooling) of coronally- versus tangentially-cut as well as more rostrally or medially versus more caudally or laterally located tissue possible.

To determine whether our data showed Gaussian distribution, we used the Shapiro–Wilk normality test. Since the Shapiro–Wilk test failed in all cases of studied cell types, the non-parametric Kruskal–Wallis test was used to determine statistically significant differences. Alpha levels were adjusted with the Bonferroni method. For comparison of numbers from tangential versus coronal sections, data were plotted and the correlation coefficient was calculated (Mystat, Systat Software, Inc., United States). We collected data from coronal sections to compare our results to previous literature. Tangential section were used to collect data about absolute numbers in the whole posteromedial barrelfield and because the identification of barrels and interbarrel septa was much more obvious. Coronal and tangential data series correlated highly with each other, therefore the density values of coronal and tangential section were pooled for the analysis.

These cells show a dominant location within the termination zones of the lemniscal thalamic projections (Chmielowska et al., 1989; Wimmer et al., 2010b), namely LIV and the LVb/VI border but also within the termination zone of paralemniscal thalamic projections (Wimmer et al., 2010b), namely layer Va (Figure 1A). Indeed, 26.17 ± 2.73% of all PV+ cells were located in LIV, 14.84 ± 1.94% in LVa and 27.15 ± 1.49% in LVb, whereas in layers II/III and VI together only 31.83 ± 1.99% were found. In LIV, the neuropil labeling led to the clear delineation of barrels, in coronal as well as tangential sections (Figure 1A,B).

We studied the PV+ cell distribution in coronal sections of 3 hemispheres and in tangential sections of another 3 hemispheres. The coronal and tangential sections correlated with each other in terms of cell number and distribution (n = 6; r² = 0.925) (Figure 3D), therefore they were pooled for further analysis (Table 1 and Figure 3A).

When extrapolated in number to 1 mm³, PV+ cells preferred LIV and LV (a and b) in approximately equal numbers, were much sparser in LII/III and LVI and avoided LI (Table 1 and Figure 1A,B, 3; significance levels are listed in Supplementary Table 1, Kruskal–Wallis test was used to determine statistically significant differences). Thus, PV+ cell numbers showed a differential distribution between layers, which was very similar in the two analyzed columnar compartments (see also below).

In layers II throughout VI, the distribution of PV+ cells was found to be similar for barrel- and septum-related compartments. However, this more or less even distribution of cell bodies was masked by the differential distribution of PV+ neuropil, which strongly preferred barrels (Figure 1B and Table 1). There were slightly more cells in barrels versus septa (Figure 2A,D,G,J, 3A, LIVc versus LIVs), but this difference was not statistically significant (p = 0.885; significance levels are listed in Supplementary Table 2, Kruskal–Wallis test was used to determine statistically significant differences). Considering soma and neuropil labeling in PV+ neurons, in LIV this resulted in a complementary distribution profile, contrasting with VIP+ and SST+ GABAergic neurons (see below).

The overall distribution of SST+ cells was unique in several aspects (Figure 1C,D). Whereas PV+ cells showed an increased density in layers IV and Vb (as noted above), SST+ cells strongly preferred infragranular layers V-VI. In fact, 73.69 ± 5.37% of the cells were located in the infragranular layers, whereas only 26.31 ± 5.37% were found in layers I-IV. This suggests that projections like the one from M1 with heavy terminal labeling in infragranular layers might be a preferential input to these neurons (Kinnischtzke et al., 2014) but also that paralemniscal inputs have ample opportunity in targeting this class of cells in LI and LVa, whereas lemniscal inputs should preferentially innervate cells in LVb/LVI (Wimmer et al., 2010b; Audette et al., 2017).

We studied the SST+ cell distribution in coronal sections of 3 hemispheres and in tangential sections of another three hemispheres, too. The coronal and tangential sections (total n = 6) correlated highly with each other (r² = 0.94) (Figure 3D), therefore the neurons were pooled for further analysis (Table 2 and Figure 3B).

The SST+ neurons were similar in number in layers Va, Vb, and VI. Each of these layers possessed significantly more cells than LIV, LII/III, and LI, respectively. LII/III SST+ cells did not differ in number from LIV cells. The lowest number of SST+ cells was found in LI, which was significantly different from all other layers (significance levels are listed in Supplementary Table 3; Kruskal–Wallis test was used to determine statistically significant differences).

When comparing the barrel with the septal compartment, we found that SST+ cells were distributed more or less homogenously between these compartments in all layers, except LIV (Table 2). As shown in Figure 2B,E,H,K, 3B, in LIV, the septal compartment contained significantly more cells than the barrel compartment (p < 0.0001) (significance levels are listed in Supplementary Table 4; Kruskal–Wallis test was used to determine statistically significant differences).

Interestingly, VIP+ cells show an inverse relationship to SST+ cells, as they display a strong preference for the supragranular layers (Figure 1E). Indeed, 70.24% of all cells were located in layers I-III, whereas only 29.76% were found in layers IV-VI. Since the subpopulation located in LI and LII/III shows extensive dendritic arborization in layer I (Prönneke et al., 2015), projections like those from motor cortex or basal forebrain can strongly target these cells (Mechawar et al., 2000; Lee et al., 2013), as does the lemniscal thalamus in deep LII/III (Staiger et al., 1996b; Wall et al., 2016).

We studied the VIP+ cell distribution in coronal sections of three hemispheres and in tangential sections of another three hemispheres. The cell numbers in these sections (total n = 6) were again correlated highly with each other (r² = 0.973) (Figure 3D), thus, the values were pooled for further analysis (Table 3 and Figure 3C).

VIP+ cells in LII/III outnumbered those in all other layers significantly whereas LI had significantly lower numbers than all other layers (except LVI). Also the gradient from LIV to LVI reached statistical significance, with each deeper layer housing less neurons than the one located pialward (significance levels are listed in Supplementary Table 5; Kruskal–Wallis test was used to determine statistically significant differences).

When comparing the columnar and septal compartments, there was no obvious difference in the supragranular and infragranular layers. However, in LIV we found significantly more cells in septa than in barrels (p < 0.0001; Kruskal–Wallis test was used to determine statistically significant differences), which can be seen in Figure 1F, 2C,F,I,L, 3C. This columnar segregation was therefore restricted to the granular layer where the cell density in septa was approximately two times higher than in barrels (significance levels are listed in Supplementary Table 6; Kruskal–Wallis test was used to determine statistically significant differences).

In summary, we could reproduce the general pattern of a dominance of PV+ cells in lemniscal thalamorecipient layers (LIV and Vb) but show a frequent occurrence on the paralemniscal thalamorecipient LVa, too. SST+ showed a strong abundance in LVa/b and LVI whereas VIP+ were preferentially distributed in LII/III and LIV (Supplementary Figure 3). In addition, when comparing barrel-related columns with septum-associated compartments, only in LIV we could detect a statistically significant preference of SST+ and VIP+ cells for the latter (Supplementary Figure 3).

We compared and finally pooled results obtained from coronal and tangential sections of the barrel cortex (Woolsey and van der Loos, 1970). The advantage of frontal sectioning is that laminar borders can be identified with high certainty, whereas in tangential sections the barrel versus septal regions can be delineated unequivocally. Since tangential sectioning requires a certain compression of the cortex, and the cortex is not evenly thick along its medio-lateral axis, the distance of the cells from the pia mater had to be adjusted, in order to match laminar location in tangential with coronal sections. After adjustment, the cell density read-out from coronal and tangential sections was virtually identical, allowing us to pool the data. Our laminar density results of VIP+ cells are in very good agreement with that previously described for the barrel cortex (Prönneke et al., 2015) and with PV+ and SST+ cell counts from other ongoing projects (own unpublished results).

Here, we used staining for vGlut2 and DAPI as pre- and postsynaptic markers, respectively, to determine barrel and septum borders, building on our previous experience (Wagener et al., 2010). Because of putative plasticity of thalamocortical fibers (Wimmer et al., 2010a), the vGlut2-defined barrel borders could vary, thus the additional use of DAPI staining is important to define the actual barrel size and further helps to distinguish barrel and septal compartments (Erzurumlu and Kind, 2001).

It should also be noted that we have counted tdTomato-expressing cell bodies distributed across the various layers and columns, which does not necessarily mean that they are specifically and comprehensively reflecting the entire population of interest, as represented by the different cre-driver lines. However, previous studies already showed a reasonable overlap of tdTomato-fluorescence with the respective markers (GAD1, PV, SST, and VIP (Taniguchi et al., 2011; Pfeffer et al., 2013); and our own published and still ongoing efforts to characterize these mouse lines revealed a very high specificity and an excellent sensitivity (Prönneke et al., 2015).

To the best of our knowledge, the numbers of GABAergic neurons in mouse barrel cortex had not been established before and the partitioning of GABAergic neurons between barrel- and septal column was completely lacking. It has to be acknowledged that there is a comprehensive quantification with an immunohistochemical approach available, the area-densities reported there being in general agreement with our results (Xu et al., 2010). Here, we started to overcome this lack of detailed knowledge by counting soma numbers of neurons expressing tdTomato-fluorescence under the control of three marker genes that are currently viewed as the best candidates to separate the population of GABAergic neurons into three different classes (Taniguchi et al., 2011; Tremblay et al., 2016).

In PVcre/tdTomato mice, we could detect a distribution across layers that was both, qualitatively and quantitatively similar to the rat somatosensory cortex (van Brederode et al., 1991; Ren et al., 1992). We showed numbers peaking in layers IV, Va, and Vb, which would mean that major layers receiving lemniscal or paralemniscal input (Wimmer et al., 2010b) house a feedforward inhibition motif in their circuitry (Cruikshank et al., 2010; Naka and Adesnik, 2016), which does strongly influence their in vivo physiology (Bruno and Sakmann, 2006; de Kock et al., 2007; Yu et al., 2016).

Surprisingly, in tangential sections we did not find a significant difference between barrel and septal compartments, although PV has been used as a marker for barrels (Sukhov et al., 2016). This means that barrels are highlighted in PV staining by virtue of the neuropil labeling (mainly being axonal boutons) and not by somatodendritic profiles. This could be explained by anisotropic axonal arbors of basket cells displaying a bias toward barrel centers (Munoz et al., 2014; Koelbl et al., 2015). Further evidence that there is a profound difference between the organization of PV+ cells’ dendrites versus axons in barrels versus septa comes from a recent study that showed their gap junction coupling to occur in a cell type- and location-specific manner (Shigematsu et al., 2018).

In SSTcre/tdTomato mice, we were able to show that all infragranular layers express high numbers of these cells, where they recently have been shown also to be most diverse in structure and function (Nigro et al., 2018). In contrast, the fewer cells located in LII/III all seem to be Martinotti cells (Munoz et al., 2017; Nigro et al., 2018), which was also consistently reported in the GIN mouse line (Oliva et al., 2000; Ma et al., 2006; Walker et al., 2016). The minority population in LI remains to be better characterized (Ma et al., 2014).

The analysis in tangential sections, revealing a preferential location in septa when compared to barrels, is without precedent. One has to know that septa in mice are very small compartments, as compared to rats, thus it is very difficult to specifically record from them, in order to assess their function (Welker and Woolsey, 1974). We can only speculate that the increased number of SST+ neurons (but also VIP+ cells, see below) might serve to flexibly adjust the receptive field size of septal neurons, which is usually larger than that of barrel-related column neurons (Brecht and Sakmann, 2002).

In VIPcre/tdTomato mice, we obtained cell numbers across the different layers that were in good agreement with previous reports (Xu et al., 2010; Prönneke et al., 2015), pinpointing a preferential role of these neurons to relay information originating from diverse sources via axonal projections into layers I-III. This information is then relayed to all layers of the respective barrel column- or septum-associated compartment by virtue of their “column-filling” axonal arbors, probably impinging on excitatory and inhibitory neurons in parallel (Garcia-Junco-Clemente et al., 2017; Kuchibhotla et al., 2017; Zhou et al., 2017).

As noted already above, a novel finding was the differential distribution of VIP+ somata in barrels versus septa, with approximately twice as many cells in the septa than in the barrels. Interestingly, a previous study has reported a differential distribution of VIP+ boutons, which were more numerous in the side region than within the hollow (Zilles et al., 1993). Since a “side” is the barrel wall and the septum pooled (Welker and Woolsey, 1974) and since VIP neurons in layer II/III and IV have a radially restricted axonal arbor (Prönneke et al., 2015), these findings are complementary to ours.

As noted above, the soma location alone is a very limited predictor of the input space of a neuron. However, since the dendritic arbors of GABAergic neurons are usually more compact than those of pyramidal cells, this could serve as a reasonable first approximation to consider the layer- and column-associated location of the soma as the main input space of the respective neuron (De Felipe et al., 2013).

Thus, we suggest that PV+ neurons are more or less uniformly distributed to fulfill the need of all cortical circuits to operate in a tight excitation-inhibition-balance (Isaacson and Scanziani, 2011), no matter whether feedforward pathways from the thalamus or local pathways activate them. SST+ cells, with their strong bias to infragranular layers and to LIV septa, are rather heterogeneous in their input and output domains, making a more specific functional interpretation than that they are providing widespread dendritic inhibition difficult (Urban-Ciecko and Barth, 2016; Yavorska and Wehr, 2016). VIP+ neurons, with their strong bias to supragranular layers and also LIV septa, can directly participate in sensory processing, potentially by strong disinhibitory mechanisms (Karnani et al., 2016a,b; Walker et al., 2016; Feldmeyer et al., 2018) but they also relay more global salience- or reward-related signals, originating in (pre-)motor cortex or in subcortical neuromodulatory centers to the column (Lee et al., 2013; Kepecs and Fishell, 2014).

The enrichment of SST+ and VIP+ neurons in LIV LIV septa might help to task-dependently switch cortical activity flow (and thus sensory processing) between the nested, partly distinct, partly overlapping, circuits that originate in or are wired through barrel- versus septum-associated compartments of the barrel cortex (Alloway, 2008; Diamond et al., 2008; Feldmeyer et al., 2013). The next task will be to perform optogenetic stimulation of these putative input pathways and record from single identified GABAergic neurons, in order to test our predictions.