Optimization of ribosome profiling using low-input brain tissue from fragile X syndrome model mice

Abstract

Dysregulated protein synthesis is a major underlying cause of many neurodevelopmental diseases including fragile X syndrome. In order to capture subtle but biologically significant differences in translation in these disorders, a robust technique is required. One powerful tool to study translational control is ribosome profiling, which is based on deep sequencing of mRNA fragments protected from ribonuclease (RNase) digestion by ribosomes. However, this approach has been mainly applied to rapidly dividing cells where translation is active and large amounts of starting material are readily available. The application of ribosome profiling to low-input brain tissue where translation is modest and gene expression changes between genotypes are expected to be small has not been carefully evaluated. Using hippocampal tissue from wide type and fragile X mental retardation 1 (Fmr1) knockout mice, we show that variable RNase digestion can lead to significant sample batch effects. We also establish GC content and ribosome footprint length as quality control metrics for RNase digestion. We performed RNase titration experiments for low-input samples to identify optimal conditions for this critical step that is often improperly conducted. Our data reveal that optimal RNase digestion is essential to ensure high quality and reproducibility of ribosome profiling for low-input brain tissue.

Figure 1.

Sample batch effects dramatically compromise the power of ribosome profiling. (A) Ribosome protected footprints (RPFs) from batch1–2 samples were mapped to Refseq mouse coding sequence (CDS) reference and quantified with RSEM. Counts were regularized log transformed, normalized with DESeq2, and used for principal component analysis (PCA) using the ‘prcomp’ function from the ‘stats’ package. Samples with the same genotypes are labeled with the same color and circled to show the genotype separation. Batches are labelled with different shapes. PCA analysis shows that the major (PC1) variance (var.) of ribosome profiling data from batch1–2 samples was derived from genotypes. (B) PCA analysis shows that the major (PC1) variance (var.) of ribosome profiling data from batch3–4 samples was derived from experimental batches. (C) Counts of RPFs mapped to CDS from batch1–2 were batch-corrected with the ‘Combat’ function from ‘sva’ package and used for the differential gene expression analysis with DESeq2. Volcano plot of batch1–2 shows the log₂ fold changes (KO/WT) of RPF abundance with –log₁₀ adjusted P value. Differentially expressed genes (DEGs) with adjusted P value less than 0.05 are colored red. (D) Volcano plot of batch3–4 shows the log2 fold changes (KO/WT) of RPF abundance with –log₁₀ adjusted P value. DEGs with adjusted P value less than 0.05 are colored red. (E) Overlap between the top FMRP CLIP genes (32) and DEGs identified with batch1–2 samples (red points in C, P = 2.3e–43, hypergeometric test) or batch3–4 samples (red points in D, P = 0.03, hypergeometric test). (F) RPF distributions on the serine peptidase inhibitor, clade A, member 3N (Serpina3n) gene. RPF number at each mRNA nucleotide position was calculated with the ‘plastid’ package, normalized to the library size, averaged across all replicates of batch1–2 (top panel) or batch3–4 (bottom panel), and plotted along the mRNA nucleotide positions with green and red triangles for annotated start and stop codons respectively. For visualization purposes, the curves were smoothed within a 30nt window. (G) RPF distributions on the gamma-aminobutyric acid (GABA) A receptor, subunit alpha 2 (Gabra2) gene.

Figure 2.

RPF GC content is correlated with sample batch effect. (A) Sequences of RPFs mapped to CDS were extracted from bam files with bedtools and samtools. Nucleotide composition at each position of RPFs for a representative sample from batch1 in Figure 1A was calculated and plotted. The mean GC percentage within the 10–20nt window was calculated and shown on the top. (B) Nucleotide composition at each position of RPFs mapped to CDS for a representative sample from batch3 in Figure 1B. (C) GC contents of RPFs from all the batch1-4 samples were calculated and plotted as a heatmap showing the correlation with the sample batches. Darker blue represents higher GC content. (D) The most abundant mRNA isoform in Batch1 WT1 sample estimated by RSEM was selected as the representative transcript for each gene across the Refseq transcriptome. GC contents of sequences for different mRNA regions in the curated and nonredundant transcriptome were calculated and plotted to visualize the medians. The lower and upper hinges correspond to the first and third quartiles. The whiskers extend from the hinges to the largest and smallest values no further than 1.5 fold of inter-quartile range. Outliers are not shown. Using full length of mRNAs as the reference, all pair-wise comparisons are statistically significant (***P < 0.001, Wilcoxon rank sum test). (E) Nucleotide composition at each position of RNA-seq reads mapped to CDS for the same sample in (D). The mean GC percentage within the 10–65nt window was calculated and shown on the top.

Figure 3.

RPF GC content is RNase-species independent. (A) 3.8 A₂₆₀ homogenate from hippocampi of one P35 male mouse was digested with 100ng RNase A (Sigma, # R4875) + 60U RNase T1 (Thermo Fisher Scientific, #EN0542)/A₂₆₀, at 25°C for 30min and applied to a 10–50% (w/v) sucrose gradient. (B) 3.8 A₂₆₀ homogenate from hippocampi of one P35 mouse was digested with 5U RNase I (Ambion, #AM2294)/A₂₆₀, at 25°C for 45min and applied to a 10–50% (w/v) sucrose gradient. (C) Nucleotide composition at each position of RPFs mapped to CDS from ribosomes in (A). (D) Nucleotide composition at each position of RPFs mapped to CDS from ribosomes in (B). (E) Nucleotide composition at each position of RPFs mapped to CDS from mouse embryonic stem cells (mESCs) (data from Ingolia et al.) (16). A 600 μl aliquot of lysate was treated with 15 μl RNase I at 100 U/μl for 45 min at 25°C. (F) Nucleotide composition at each position of RPFs mapped to CDS from human embryonic stem cell (hESC)-derived neurons (data from Grabole et al.) (42). 5 U TruSeq Ribo Profile Nuclease/A₂₆₀ at 25°C for 45 min.

Figure 4.

RPF GC content and length depend on the RNase digestion protocol. (A) Lysates from human iPSC neuron samples spanning a wide range of amounts were digested with 100 ng RNase A + 60U RNase T1/A₂₆₀ at 25°C for 30 min. Monosomal RNA was extracted from monosomal fractions of sucrose gradients and quantified with Nanodrop. GC contents were calculated as in Figure 2A and the peaks of length distributions of RPFs mapped to CDS were also determined. Scatter plots with Pearson correlation coefficients show the negative correlation between 80S monosomal RNA amounts (log₂ scale) and the GC contents (black) or RPF lengths (red). (B) Lysates from human iPSC samples were digested with 20 ng RNase A + 12 U RNase T1/A₂₆₀ at 25°C for 30 min. Scatter plots with Pearson correlation coefficients show the negative correlation between 80S monosomal RNA amounts (log₂ scale) and the GC contents (black) or RPF lengths (red). (C) Nucleotide composition at each position of RPFs mapped to CDS from mESC-derived neurons with an alternative protocol of RNase digestion (data from Zappulo et al.) (43). 70 U RNase I at 25°C for 40 min.

Figure 5.

The GC-content correlated batch effects are caused by incomplete RNase digestion. (A) Hippocampi from one P35 WT mouse were homogenized and the homogenate was aliquoted for the titration experiment. 0.5 unit A₂₆₀ homogenate containing 2 μg RNA (measured with Qubit HS RNA kit) in 0.3 ml volume was used for digestion at each RNase concentration. Digested homogenates were separated on 10–50% (w/v) sucrose gradients. Profile of hippocampal ribosomes after the digestion at the lowest concentration1 [Conc.1, 4.8ng RNase A (Ambion, #AM2270) + 0.6 U RNase T1 (Thermo Fisher Scientific, #EN0542)/μg RNA × 2 μg RNA in 0.3 ml at 25°C for 30 min] and sucrose gradient fractionation. (B) Profile of hippocampal ribosomes after the digestion at the concentration2 (Conc.2, 24 ng RNase A + 3U RNase T1/μg RNA × 2 μg RNA in 0.3 ml at 25°C for 30 min) and sucrose gradient fractionation. (C) Profile of hippocampal ribosomes after the digestion at the concentration3 (Conc.3, 120 ng RNase A + 15 U RNase T1/μg RNA × 2 μg RNA in 0.3 ml at 25°C for 30 min) and sucrose gradient fractionation. (D) Profile of hippocampal ribosomes after the digestion at the concentration4 (Conc.4, 600 ng RNase A + 75 U RNase T1/μg RNA × 2 μg RNA in 0.3 ml at 25°C for 30 min) and sucrose gradient fractionation. (E) Profile of hippocampal ribosomes after the digestion at the highest concentration5 (Conc.5, 3000 ng RNase A + 375 U RNase T1/μg RNA × 2 μg RNA RNA in 0.3 ml at 25°C for 30 min) and sucrose gradient fractionation. (F) Scatter plots with Pearson correlation coefficients show the negative correlation between RNase concentrations (log₅ scale) and the GC contents (black) or RPF lengths (red).

Figure 6.

Optimized RNase digestion generates ribosome profiling data with higher quality and reproducibility. (A) rRNA and tRNA contaminates were filtered out from RPF reads with Bowtie2 and unmapped reads were next mapped to mm10 with Tophat2. ‘Unmapped’ and ‘Multimapped’ reads were defined based on the Tophat2 outputs. PCR-derived ‘Duplicates’ were identified based on the unique molecule identifier (UMI). The uniquely mapped reads after all the upstream filtering were classified as ‘Unique’ and used for downstream analyses. Stacked bar plots show the percentage of RPFs uniquely mapped to the transcriptome under different RNase concentrations. (B) The number of ‘Unique’ reads in (A) mapped to various gene regions were calculated by intersecting bam files with USCS bed annotations by bedtools and samtools. Stacked bar plots show the percentage of ‘Unique’ RPFs in (A) mapped to CDS under different RNase concentrations. (C) The P-site offsets and frame preferences of ‘Unique’ RPFs in (A) mapped to CDS were calculated with ‘plastid’ package. The best frame resolution across all the RPF lengths was selected to represent each RNase concentration. Stacked bar plots show the percentage of ‘Unique’ RPFs in (A) mapped to the same frame as the annotated CDS (Frame1) under different RNase concentrations. (D) ‘Unique’ reads in (A) were mapped to Refseq mouse CDS reference and quantified with RSEM. Counts of mapped RPFs were regularized log transformed and normalized with DESeq2. Scatter plots and Pearson correlation coefficients show the reproducibility of RPF abundance quantification among various RNase concentrations. (E) Principal component analysis (PCA) shows the similarity between Conc.4 and Conc.5 samples. (F) Distinct RPF distributions under different RNase concentrations on actin beta (Actb) mRNA. RPF number at each mRNA nucleotide position was calculated with the ‘plastid’ package, normalized to the mean read density on CDS, and plotted along the mRNA nucleotide positions with green and red triangles for annotated start and stop codons respectively. (G) Distinct RPF distributions under different RNase concentrations on the calcium/calmodulin-dependent protein kinase II alpha (Camk2a) mRNA.