Experiments were performed on neonatal GAD67-GFP mice (P0, P4, P8, and P12; n = 36, consisting of at least three mice at each age for both single- and dual tracer studies). The GAD67-GFP knock-in line has previously been validated in our lab (Gay et al., 2018) and allows for easy visualization of GAD-positive LCIC modules (Lesicko et al., 2016, 2020; Gay et al., 2018; Lamb-Echegaray et al., 2019; Stinson et al., 2021; Brett et al., 2022). Specifics concerning the generation of the GAD67-GFP (Δneo) line are described elsewhere (Tamamaki et al., 2003; permission granted by Dr. Yuchio Yanagawa, Gunma University Graduate School of Medicine, Gunma, Japan). Heterozygous GAD-GFP males were crossed with C57BL/6J females and GFP-expressing progeny were identified before P4 using a Dark Reader Spot Lamp Goggle visualization system (Clare Chemical Research¹, Dolores, CO, Cat# SL10S). Equal numbers of males and females were used in the described experimentation and no sex-specific differences were noted. All procedures were performed in keeping with the US National Research Council’s Guide for the Care and Use of Laboratory Animals and received prior approval by the Institutional Animal Care and Use Committee (Protocol No. 20-1421).

Following an overdose of ketamine (200 mg/kg) and xylazine (20 mg/kg), mice at designated stages were perfused with chilled, oxygenated (95% O₂, 5% CO₂) artificial cerebrospinal fluid (aCSF; in mM; 126 NaCl, 3 KCl, 1.25 NaH₂PO₄, 10 dextrose, 20 NaHCO₃, 1.2 MgSO₄, 2.5 CaCl₂, pH 7.4). Brains were blocked in the coronal plane (just rostral to somatosensory cortex and just caudal to the midbrain) and varied in total thickness depending upon age. Under a dissecting microscope, gross placements of biocytin crystal (B1758, Sigma-Aldrich, St. Louis, MO) were made in somatosensory cortex at varying depths depending on age to maximize labeling of layer 5/6 corticocollicular projections. In double-labeling studies, the cerebellum was then removed to facilitate deposition of a 10,000 MW dextran (D22904, AlexaFluor 647 direct conjugate, ThermoFisherScientic, Waltham, MA) in the CNIC ipsilateral to the cortical biocytin placement. Tissue blocks were bubbled at room temperature in aCSF for 18–24 h to allow for complete filling of terminal endings in the LCIC. Prior to sectioning, tissue was postfixed and cryoprotected in a series of 4% paraformaldehyde solutions with increasing concentrations (10%–30%) of sucrose.

A rostrocaudal series of sections were cut at 50 μm on a sliding freezing microtome and collected in 0.1 M phosphate buffered saline (PBS, pH 7.4). Free-floating sections were rinsed three times for 5 min in PBS. In experiments combining tract-tracing with a matrix marker, a 2.5 h DyLight 549 streptavidin step (1:200, SA-5549, Vector Laboratories, Burlingame, CA, RRID:AB_2336408) for visualizing biocytin always preceded subsequent CR immunostaining. Following another round of PBS rinses, sections were blocked in 5% normal donkey serum (NDS) in PBS for 30 min. Tissue was then incubated in anti-CR primary made in rabbit (1:250, CR 7697, Swant, Burgdorf, Switzerland, RRID:AB_2619710) for 40 min at room temperature and then overnight at 4°C. An Alexa Fluor 350 donkey anti-rabbit IgG (1:25, A10039, Thermo Fisher Scientific, Waltham, MA, RRID:AB_2534015) was then applied prior to a final series of PBS rinses. Sections were mounted on charged slides, coverslipped with Pro Long Diamond (P36970, Thermo Fischer Scientific, Waltham, MA), and stored in the dark until imaging.

Aside from documentation of somatosensory cortex tracer placements, data collection focused exclusively on the ipsilateral IC, as corticofugal projections to the contralateral LCIC are sparse. Verification of accurate cortical and CNIC tracer deposits were made using age-matched sections in The Atlas of the Developing Mouse Brain (Paxinos et al., 2007). Sections throughout the rostrocaudal extent of the IC were imaged using epiflourescent microscopy (Nikon Eclipse Ti-2 microscope equipped with a monochrome, Hamamatsu ORCA-Flash 4.0 V3 sCMOS camera, Japan; PlanApo objectives 10× : NA = 0.30, 20× : NA = 0.75, and 40× : NA = 1.30). Filter sets (Chroma Technology, Bellows Falls, VT) were designed with careful attention to the spectra of the various fluorophores to ensure no cross-channel bleed through (Alexa Fluor 350 filter set: excitation 381–403 nm emission 417–477; GFP filter set: excitation 446–486 emission 500–550; biocytin-streptavidin DyLight 549 filter set: excitation 542–566 emission 582–636; dextran AlexaFluor 647 filter set: excitation 593–643 emission 663–733. An extended depth of field (EDF) algorithm was used to generate two-dimensional images from acquired Z-stacks (Elements Software; Nikon) using only focused portions of each optical slice. Monochrome channel acquisitions were pseudocolored accordingly (blue: CR, green: GAD, biocytin: red, dextran: cyan) and saved as lossless JPEG2000 files.

Quantification of somatosensory afferent pattern alignment with emerging LCIC compartments, as well as assessments of multisensory pattern segregation, focused on mid-rostrocaudal regions of the LCIC where the modular-matrix framework was most readily apparent. Data from a minimum of three mice were analyzed at each age and multiple sections (a minimum of two) were sampled to account for any potential regional LCIC variability. LCIC layer 2 sampling and generation of brightness plot profiles has been previously described in detail (Wallace et al., 2016; Dillingham et al., 2017; Gay et al., 2018; Lamb-Echegaray et al., 2019; Stinson et al., 2021; Brett et al., 2022). In brief, separate image channels were converted to grayscale and exported as uncompressed TIFFS prior to importing into ImageJ² (NIH, Bethesda, MD, RRID:SCR_003070). A freehand tool (line thickness 100 μm) was used to sample LCIC layer 2 from ventral-to-dorsal bisecting GAD-positive modules. Sampling contours were approximated through presumptive LCIC layer 2 in P0 mice, as GAD-defined modules are not easily discerned at this earliest postnatal time point. An ROI function was employed to apply the same sampling contour to each subsequent channel sampled of the same image. Generated signals (single tracer studies: somatosensory vs. GAD/CR, dual tracer studies: somatosensory vs. auditory) were compiled as multi-channel brightness profiles to visualize relative signal overlap/non-overlap. Raw data values for each waveform were exported for cross-correlation and kurtosis analyses described below. The latter analysis used F_S and F_A to refer to sampled somatosensory and auditory fluorescence values, respectively.

Cross-correlation analyses (Microsoft Excel, Redmond, WA) were performed to numerically quantify the similarity of two signal patterns relative to each other (e.g., somatosensory and GAD, or somatosensory and CR). Cross-correlation function y-intercepts (i.e., no spatial shift or relative displacement) serve as an objective measure for assessing signal matching/mismatch. Values range from +1.0 to −1.0, with higher values indicating strong signal overlap, whereas lower values indicate waveforms that are non-overlapping or out-of-phase. Single factor ANOVA assessed similarity among developmental stages. Based on ANOVA high confidence dissimilarity (p < 0.05), subsequent comparisons between age pairs were made using independent, two-tailed Student’s t-tests assuming unequal variance (Type 3) with statistical significance (p < 0.05). Similar t-tests (albeit one-tailed) were performed comparing y-intercepts of somatosensory vs. GAD and somatosensory vs. CR signal pairings.

To quantify the extent of segregation of developing multimodal LCIC inputs, we build upon established unbiased methods applied to emerging retinogeniculate patterns (Torborg and Feller, 2004; Jaubert-Miazza et al., 2005; Torborg et al., 2005). These previous methods processed afferent data in four main stages: (1) a “rolling ball” algorithm correcting for uneven background; (2) logarithmic ratios (R values) of ipsilateral and contralateral fluorescence intensities, F_I and F_C respectively; R = log₁₀(F_I/F_C); (3) R value histograms revealing the extent of segregation in their shape; and finally (4) variances of R-distributions quantifying the extent of segregation in a single metric. Typically, overlapping patterns produce unimodal R distributions, while decreased projection overlap (or segregation) produce bimodal R distributions. Testing for overlap/segregation therefore requires quantifying the shape of R distributions.

Accordingly, we also perform the same first three stages, but in the last stage employ a different R distribution statistic (kurtosis rather than variance) to quantify R distribution shape to assess relative input separation. Variance is the second central moment (i.e., average squared deviations from the mean) and a measure of distribution spread. Kurtosis is the fourth central moment (i.e., average fourth power deviations from the mean; normalized by variance) and a direct measure of distribution shape. Unimodal distributions yield more positive kurtosis values, while bimodal distributions yield more negative kurtosis values. In sum, we also process our dual tract-tracing data in four stages: (1) applying a “rolling ball” filter in ImageJ (diameter = 20 pixels) correcting for uneven background fluorescence (i.e., regions of the LCIC exhibiting no axonal labeling); (2) logarithmic ratios (R values) combining fluorescence intensities of somatosensory and auditory inputs, F_S and F_A respectively; R = log₁₀(F_S/F_A); (3) histograms of R values showing extent of segregation in their shape; and (4) kurtoses of R-distributions to quantify extent of segregation in single statistics.

To determine whether somatosensory inputs to the LCIC share the same critical period defined for auditory inputs (Lamb-Echegaray et al., 2019) and its emerging compartments, unilateral biocytin crystal placements were made in somatosensory cortex (Figure 1) in a developmental series of GAD67-GFP living preparations. Given the sparse nature of the contralateral corticocollicular pathway, analyses focused exclusively on the LCIC ipsilateral to the tracer placement. Terminals arising from somatosensory cortex were present in the LCIC at birth and diffusely distributed (Figure 2A), lacking any regional specificity. GAD-positive modules at this age are not readily apparent (Figure 2B), in keeping with previous reports (Dillingham et al., 2017; Gay et al., 2018; Lamb-Echegaray et al., 2019; Stinson et al., 2021; Brett et al., 2022). As layer 2 modular fields emerge at P4 and continue to develop through P12, a preference of somatosensory endings for these defined zones becomes increasingly apparent (Figures 2C–H). By P12, projection specificity appears adult-like, with elaborate arborizations within modular confines and little evidence of significant labeling in the surrounding matrix.

Qualitative observations of increased matching of somatosensory and GAD patterns with age (i.e., in-phase signal waveform oscillations, arrows in Figures 2D,F,H) were confirmed with cross-correlation analyses. The positively sloped linear regression of y-intercept values plotted by developmental stage supports the notion that initially dispersed somatosensory inputs become increasingly specific for LCIC modules (Figure 3A). Single factor ANOVA showed significant mean dissimilarity among age groups (p < 0.001). Thus, t-tests comparing age pairs were performed that showed cross-correlation y-intercept values statistically different for P0 vs. P4 (p = 0.019), P0 vs. P8 (p = 0.009), and P0 vs. P12 (p = 0.005), but not other age pairs (p > 0.05). Off-origin cross-correlation function maxima and minima occur at certain relative spatial shifts between the two series, signifying aligned and mis-aligned signals, respectively. A representative P12 cross-correlation function (Figure 3B) shows a distinctly positive y-intercept (i.e., strong signal matching with no spatial shift) and pronounced off-origin peaks and troughs that reflect the periodic nature of the LCIC modular-matrix framework at this age. Taken together, these data suggest a similar period of projection shaping for top-down somatosensory inputs that gradually sharpen and overlap emerging LCIC modular domains.

To further confirm the establishment of highly refined somatosensory inputs for LCIC modules by its critical period closure (P12), analogous labeling studies were performed coupled with immunostaining for the established matrix marker, calretinin. As anticipated, concentrated somatosensory terminal fields overlapping layer 2 GAD cell clusters filled voids in the encompassing CR-positive matrix (Figure 4, arrowheads). This complementary patterning was reliably observed in additional P12 mice (Figure 5A). Higher magnification reveals dense somatosensory terminal fields sculpted such that they nearly exclusively reside within modular confines (Figures 5B,C, dashed contours). Three-channel LCIC layer 2 sampling and corresponding brightness plot profiles illustrate oscillatory waveforms, with matching somatosensory and GAD signals that are mismatched with CR fluctuations (Figures 5D,E, arrows).

Comparisons of cross-correlation y-intercept values for somatosensory/GAD vs. somatosensory/CR signal combinations further demonstrate discrete afferent patterns at P12 that target LCIC modules (Figure 6). Tightly clustered positive y-intercepts for somatosensory/GAD labeling indicate reliable signal matching or registry. In contrast, more negative cross-correlation y-intercept values for the same axonal distributions with respect to CR matrix labeling confirm non-overlapping patterns or signal mismatch (Figure 6A). Such quantitative findings support our qualitative observations (Figures 4, 5) that somatosensory terminals spatially align with modular domains and are offset from the encompassing matrix. The y-intercept median and standard deviation for biocytin and GAD signals were + 0.45 and 0.12, respectively, whereas that for biocytin and CR signals were—0.07 and 0.22. Not surprisingly given these dissimilar medians and overall distributions, these two data sets proved to be statistically different from each other (p < 0.001). A representative P12 cross-correlation function for biocytin and GAD shows a positive y-intercept (i.e., strong overlap at zero shift) with an off-origin trough and peak indicative of strong signal mismatching and matching respectively at increasing relative shifts (Figure 6B). A contrasting P12 cross-correlation function for biocytin and CR exhibits a negative y-intercept (i.e., non-overlap at zero shift) with an off-origin peak and trough indicative of subsequent strong signal matching and mismatching with increasing relative shifts (Figure 6C).

These findings, taken together with previous work from our lab characterizing developing auditory patterns (Lamb-Echegaray et al., 2019), suggest multimodal LCIC inputs share a similar critical period of shaping that yields complementary projection arrangements. Furthermore, they point towards emergence of discrete maps from initially intermingling distributions that undergo a process of segregation. To directly test such notions, we performed dual tracer studies simultaneously labeling somatosensory (corticocollicular) and auditory (arising from the CNIC) inputs in GAD67-GFP mice (Figure 7A) over the defined critical period (P0–P12). In each case, coronal sections containing placement centers for both dyes were documented and matched with corresponding atlas plates to verify tracer location (Figures 7B,B’,C,C’; plates modified for simplicity from Paxinos et al., 2007). As in the single-labeling experiments, placements centered in somatosensory cortex were made sufficiently deep to include corticocollicular cells of origin located in layers 5/6 (Figures 7B,B’, see Slater et al., 2013; Stebbings et al., 2014). CNIC placements were appropriately large to fill this subdivision, while not bleeding into the neighboring LCIC or DCIC (Figures 7C,C’).

At birth, both somatosensory and auditory terminals are present in the LCIC, albeit sparse and unorganized (Figures 8A–D). Results at P4 were somewhat variable, with some cases still exhibiting intermingling patterns and inconclusive modules (Figures 8E–H), while others show clear indications of an emerging modularity and initial projection separation. These findings suggest P4 as a pivotal timepoint for the developing LCIC compartmental framework and its interfacing multimodal afferents. By P8, GAD-positive modules are easily discerned and projection segregation is in progress, as somatosensory inputs preferentially terminate within modular zones and auditory inputs the surrounding matrix (Figures 8I–L). By P12, the process of segregation is largely complete, with adult-like projection patterns that terminate almost exclusively within their respective compartments (Figures 8M–P, Supplementary Figure 1). Sampling of tracer channels at P12 consistently generate periodic waveforms with out-of-phase signal fluctuations, signifying fully segregated multimodal patterns (Figure 9).

The kurtosis statistic was used to quantify the shape of R distributions which were calculated from somatosensory/auditory fluorescence ratios. Generally higher kurtosis values at earlier ages, contrasting with negative kurtosis values at later ages, indicate gradual changes in distribution shape from unimodal to bimodal with increasing age (Figure 10). Since overlapping projections produce unimodal R distributions and segregated projections produce bimodal R distributions, this kurtosis trend demonstrates increasing input separation with age. Linear regression depicts this developmental trend (Figure 10A). Representative R value histograms at each age reveal the transition from unimodal (overlapping) to bimodal (non-overlapping) from P0 to P12 (Figure 10B).