A missense mutation in Kcnc3 causes hippocampal learning deficits in mice

Significance Using an unbiased genetic screen, we uncovered a novel missense mutation in the voltage-gated potassium channel, subfamily C member 3 gene (Kcnc3) that decreases the activity of hippocampal neurons and causes defects in learning and memory in a fear-conditioning task. These findings provide evidence that Kcnc3 is important for hippocampal encoding of memories and contribute significantly to our understanding of potassium currents in the hippocampus and their role in learning.

USA) in a sound-attenuating chamber (ETS-Lindgren Acoustic Systems, Cedar Park, TX, USA), maintaining body temperature at 37.5°C throughout testing. Subdermal neurology needle electrodes (Neuroline Subdermal, Ambu Inc., Columbia, MD, USA) were inserted at the vertex (+), ventrolateral to the left ear (-), and the left thigh (ground) to record responses. ABRs were elicited with tone bursts at 8, 16, and 32 kHz (0.5 ms rise/fall Blackman ramp, 1 ms duration, alternating phase) presented to the left ear of the animal at the rate of 21/s. ABRs were band-pass filtered below 100 Hz and above 3000 Hz, amplified, then computer averaged and displayed. The stimulus intensity was reduced in 10-20 dB steps and finally 5 dB steps to identify the lowest intensity at which a repeatable ABR waveform was detectable for each animal. ABR waveforms were generated from averages of 1052 stimuli for each stimulus condition, and data were stored digitally for later offline measurements and analyses.
Olfactory discrimination test. Mice were presented 5 different odors (water, lemon oil, vanilla oil, male odor and female odor in sequence) with 3 trials for a length of 2 mins per odor and the amount of sniffing time was recorded. The inter-trial interval was 1 min (4).

Open field test.
The open field used was a 55.8 cm x 55.8 cm x 35.6 cm (length x width x height) square arena with a white Plexiglas floor and white laminate walls (Phenome Technologies, Lincolnshire, IL). The behavior of 10 to 20-weekold males was observed and position in the arena over time was recorded using a video camera-based computer tracking system (Limelight 4, Actimetrics). Mice were then removed and returned to their home cages. The arena was cleaned with diluted Quatricide Detergent/Disinfectant (Pharmacal Research Labs, Waterbury, CT) between trials to avoid olfactory cues. The arena image was divided equally into 25 virtual zones (5 x 5). The 15 zones along the edge were defined as the peripheral region and the 10 zones in the middle were defined as the center region (5). The mice were allowed to explore the arena for 1 hr which also served as habituation for the novel position test on the next day. The behavior from the first 5 mins were analyzed to evaluate the anxiety level and the behavior from the first 30 mins were analyzed to assess the locomotor activity.
Novel position test. Two days after the open field test, mice were released into the same arena with visual cues on the arena walls. Two identical silicone vials were placed in the area as shown in Fig. 2f. Mice were allowed to interact with vials for 30 mins and returned to the home cage. One hr after training, one of the silicone vials was moved to a new location and mice were placed back to the arena for 10 mins. The mouse's position was continuously recorded by Limelight 4. The location of the novel position was counterbalanced during tests. The objects were thoroughly cleaned with Quatricide between tests to minimize olfactory cues. During offline analysis after the experiment, a two-cm virtual band around the silicone vial was drawn in Limelight 4 and the nose position within the band was tracked to reflect the exploration. Data are presented as a discrimination index, which is the difference of the amount of interaction time (novel position -familiar position) divided by the total interaction time (novel position + familiar position) (6).
Barnes maze test. The circular maze table was 92 cm in diameter, 105 cm in height with 20 equally spaced holes (5 cm diameter) around the perimeter (7,8). The maze was positioned at the center of a square area (1.7 x 1.7 m 2 ) surrounded by curtains and each side of the curtain was decorated with visual cues. Two bright lights were turned on during testing (~1500 lux at the maze level). During training, 19 of the holes were blocked and one hole led to an escape box (target hole). Each mouse was released in the middle of the maze by removing a cylindrical black start tube and allowed to explore the maze for 3 mins. If the mouse entered the target hole, it was returned back to its home cage after 1 min. If the mouse failed to find the target hole, it was gently guided to the target hole and allowed to stay in the escape box for 1 min. Fifteen mins later, the mouse was trained again, and such training was repeated for a total of 4 times each day for 4 consecutive days. The target hole for each mouse was randomized. The total latency to the target hole was recorded when all four paws of the mouse were inside the escape box. On the fifth and twelfth day, probe trials were run where all the holes were blocked and the amount of time that mouse spent in the virtual target hole area within 2 mins was measured. The maze was cleaned with diluted Quatricide and dried between each trial to remove olfactory cues. Mouse behavior was recorded with Limelight 4.

Sequencing and Gene Expression Analysis
Stranded mRNA-seq. Raw RNA-seq data (fastq files) were trimmed based on quality score and N calling. For quality score, reads showing poor quality scores (< 20) from both 5'-end 10bp and 3'-end 38bp windows were trimmed. For N calling, reads containing Ns at 3'-end 15bp window were trimmed. Short trimmed reads (< 30 bp) and reads with poor average quality scores (< 21) or low read accuracy values (< -1) were also trimmed. From the read accuracy value cut off, trimmed reads could be filtered if they had < 50% probability to be accurate. Read accuracy values were calculated using the following formula, where k is the length of a trimmed read: After the quality score and N calling trimming, reads were further trimmed to remove contamination of adapters and polyA signals using cutadapt (version 1.14) (9) using the following parameters: Adapter trimming: cutadapt -g AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTC CGATCT -a AGATCGGAAGAGCACACGTCTGAACTCCAGTCAC -m 30 PolyA signal trimming: cutadapt -g T{100} -m 30 The sequence qualities of fastq files were tested before and after the trimming process using both FastQC (version 0.11.8) (Babraham Bioinformatics -FastQC A Quality Control tool for High Throughput Sequence Data: https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and FastQ Screen (version 0.13.0) (10).

featureCounts -s 2 -O -t exon -g gene_id
From read counts aligned to the forward direction of transcripts, RPKM values were calculated after adding one more count to avoid missing value errors during log2 transformation (13).
Lowly expressed genes in the hippocampus were filtered out using uniquely mapped read counts and mean read ratio (read ratio = read count /(transcript length(bp) / 3,000)). Genes were filtered if both measures were < 3. To overcome variation in mapped read numbers across samples, genes were further filtered if the read counts and read ratios of sample with minimum mapped reads were < 2. Therefore,16,009 expressed genes (35.464%) out of 45,141 genes (transcript length ≧ 200bp) remained after filtering. The distribution of log2(RPKM) was close to a normal distribution without showing noise-level expressed genes (Fig.  S10).
Mapped read data were normalized to decrease variance introduced by uncontrolled influences. A normalized RPKM was calculated, considering both gene length and GC content, from read count using CQN (conditional quantile normalization) R package (14). MA plots showed that the normalized RPKM values generate the expected MA plots and are unlikely to be the previously calculated RPKM values (Fig. S10). Though the MA plot of the previously calculated RPKM values showed mostly positive log(fold change) (M > 0) in high expressed genes (e.g. A > 10), the pattern disappeared at the MA plot of the normalized RPKM values.
Outlier samples were removed if their Z score of overall connectivities across all genes was > 2. Connectivity is a measure of how correlated a gene is with all other genes in the dataset. Outlier samples also tend to have a large clustering coefficient. The clustering coefficient show how gene clusters are intraconnected, and high clustering coefficient values imply that all genes are connected with each other. Therefore, high clustering coefficient and low connectivity implies little replication of a true biological system in the samples. Outlier samples with low connectivities were assessed for high clustering coefficients (Fig. S7), leading to the filtering of two additional outlier samples (C05_2_MT_Ctl_DDG and C13_3_MT_FCT_DDG; see Dataset S4 for detailed information).
From principal component analysis of the metadata, allowed for identification first and second principal components containing sequence statistics contributed by the metadata components that influence gene expression. From the principal components and metadata components, one-way anova tests were performed across all samples to find confounding factors. Metadata components were considered as independent compounding factors if they showed significant pvalues from the anova as well as poor correlations with the first and second principal compoents. Using limma R package, the influences of independent confounding factors were regressed out from gene expression data using the linear model below: Wherein log2(normalized RPKM) is the gene expression data, and RIN is RNA integrity numbers. PF_HQ_ERROR_RATE, INTERGENIC_BASES, and AT_DROPOUT are from the summary statistics from Picard (http://broadinstitute.github.io/picard/). PC1 and PC2 are the first and second principal components. From the gene expression data, 32 differential gene expression analyses were performed for control vs FCT, wild type vs Clueless, dorsal vs ventral, and DG vs CA (Dataset S5). DEGs were identified if their pvalues were below 0.01.

Figure S1: Breeding scheme for QTL mapping and heritability analysis. a)
Wild type C57BL/10J female mice were mated with wild type C57BL/6J male mice to produce F1 mice (WTB10B6F1). WTB10B6F1 mice were intercrossed to produce ~ 250 F2 mice (WTB10B6F2) for mapping potential QTLs that contributed to differences in contextual fear conditioning phenotype between C57BL/10J and C57BL/6J. Similarly, to map the causative locus in Clueless mutants, wild type C57BL/10J female mice were mated with Clueless male mice (B6J background) to produce F1. F1 mice with low contextual freezing scores were intercrossed to produce ~ 250 F2 mice for mapping. All mice were tested concurrently. b) B6J, B10J and WTB10B6F1 are three isogenic lines. We took the mean of the 3 variances to represent the environmental variance. The variance from ClueB10B6F2 was caused by both genetic and environmental factors. Based on the equation and the SD from Fig. 1e, the heritability was         and CA (bottom), respectively. c) Gene set enrichment analysis comparing immediate early genes (IEGs) and Npas4 target genes (Npas4) in response to fear conditioning in WT and Clueless mice. WT mice had more IEGs and NPAS4 target genes with differential expression after fear conditioning than Clueless mice. DDG = dorsal dentate gyrus, VDG = ventral dentate gyrus, DCA = dorsal cornu ammonis, and VCA = ventral cornu ammonis. Darker red boxes indicate more genes with changes in expression.        Background_gene sheet shows the background gene list that used for gene set enrichments, which was a list of all (16,009 genes) expressed in the mouse hippocampus.