RNAScope Fluorescence in situ hybridization (FISH), imaging, and analysis

Wild type mouse brains (P14) were flash frozen in 2-methylbutane. Brain tissue was then embedded in O.C.T. (Tissue Tek) and sectioned at 20 µm using a cryostat at 22˚C. Coronal or sagittal sections were mounted on charged glass slides (Coverfrost glass green, Fisher) and stored at −80˚C for no more than three months. Mounted sections were fixed in 4% PFA, pretreated for 30 min at room temperature, and probes (Advanced Cell Diagnostics (ACD), Newark, CA) were hybridized for 2 hrs at 40˚C. FISH was performed by following the protocol with the RNAScope® Multiplex Fluorescent Detection Reagent kit (ACD, 320851) (Wang et al., 2012a). A combination of the following probes was used: PlxnB1, PlxnB2, Sema4A, Sema4D, Aldh1l1, Gad1, Gad2, Slc17a7, and/or a negative control probe DapB (See Table 1). Slides were counterstained with DAPI and then mounted using flouromont-G. Sections were imaged using a Ni-E inverted microscope equipped with a Nikon C2 confocal head, and sixteen-bit images of four to five individual regions of CA1, CA3, and DG for each hippocampus (2 hippocampi/ animal) were acquired with a 60x objective. Within each experiment, images were acquired with identical detector settings for laser power, gain, and offset for each probe using the Nikon Elements AR software. Images were acquired as a z-stack (5–8 optical sections and 0.5 μm step size). Maximum intensity projections (MIPs) were created from each stack (FIJI). MIPs of channels were processed into a usable stack, using the StackPackBot (MATLAB; (Cicconet et al., 2017)), and channels were converted into eight-bit files for image analysis (FIJI).

RNAScope probe information

All analysis presented here was performed using automated MATLAB software assistant bots (Cicconet et al., 2017; Hrvatin et al., 2018). The automated MATLAB software identifies DAPI stained nuclei, and then counts the number of fluorescent probe puncta per nuclei. Given that the hippocampus has a high cell density, we first trained these bots in nuclear segmentation through use of a stacked Random Forest model (NucleiSegmentationBot), by using at least 10 images containing more than 3,000 nuclei. This Random Forest model trains on parameters such as background signal, nuclei contour, and the nuclei center. A watershed algorithm is then applied to these parameters to identify and split nuclei in close proximity to one another, creating individual nuclei masks, and assuring that fluorescent probe signals are not double-counted. DAPI nuclei masks are grown by 4 pixels (0.41µm/pixel) from the nucleus to account for RNAs that might be present in cell somas. Finally, the SpotsInNucleiBot was used in tandem with the nuclei segmentation training in order to generate the number of fluorescent puncta of up to three different probes per cell. These results were verified by an independent analysis performed on a subset of images using Halo (Indica labs; data not shown).

For the DapB, Aldh1l1, Plxnb1, Plxnb2, Sema4a, and Sema4d genes a cell was defined as expressing a specific RNA if it was found to contain ≥ 4 mRNA puncta; for the Gad1/Gad2 genes ≥ 6 mRNA puncta; and for the Slc17a7 gene ≥ 8 mRNA puncta. These thresholds were empirically determined based on the density of puncta observed for each probe set. Very rarely were fluorescent puncta observed above the established threshold with the DapB probe (0.1 +/− 0.09% DapB+ cells/all DAPI+ cells analyzed; Fig. S1A). Glutamatergic neurons are very densely packed in the pyramidal cell layer of the hippocampus, and Slc17a7 is well-expressed in these neurons. Thus, despite our higher thresholding of Slc17a7 positive puncta per cell, 9% of excitatory cells (Slc17a7+) co-labeled as inhibitory cells (Gad1/Gad2+), suggesting that these neurons are actually inhibitory and not excitatory. However we also cannot exclude the possibility that the co-expression of these markers (Slc17a7 and Gad1/Gad2) might simply be a biological feature. Nonetheless, this “double counting” does not affect our main conclusion as it only slightly increases the number of cells that are classified as excitatory neurons, and this misclassification is uniformly present in all tissue sections.

For visualization purposes, representative large field 10x images of the hippocampus (Figs. 1,​,5;5; Fig. S1,S2) were acquired with identical detector settings within an experiment and pairwise stitched together in FIJI (Preibisch et al., 2009). For single cell images (Fig. 5F), a Gaussian filter was used to decrease the DAPI+ nuclei noise for representative image visualization only. Statistical analysis was determined by a One-Way ANOVA (fixed factor was sub-region of the hippocampus) followed by a Tukey’s post hoc Test for significance (Fig. S1; SPSS).