Analysis of subcellular transcriptomes by RNA proximity labeling with Halo-seq

Abstract

Thousands of RNA species display nonuniform distribution within cells. However, quantification of the spatial patterns adopted by individual RNAs remains difficult, in part by a lack of quantitative tools for subcellular transcriptome analysis. In this study, we describe an RNA proximity labeling method that facilitates the quantification of subcellular RNA populations with high spatial specificity. This method, termed Halo-seq, pairs a light-activatable, radical generating small molecule with highly efficient Click chemistry to efficiently label and purify spatially defined RNA samples. We compared Halo-seq with previously reported similar methods and found that Halo-seq displayed a higher efficiency of RNA labeling, indicating that it is well suited to the investigation of small, precisely localized RNA populations. We then used Halo-seq to quantify nuclear, nucleolar and cytoplasmic transcriptomes, characterize their dynamic nature following perturbation, and identify RNA sequence features associated with their composition. Specifically, we found that RNAs containing AU-rich elements are relatively enriched in the nucleus. This enrichment becomes stronger upon treatment with the nuclear export inhibitor leptomycin B, both expanding the role of HuR in RNA export and generating a comprehensive set of transcripts whose export from the nucleus depends on HuR.

Figure 1.

Halo-seq facilitates the biotinylation of RNA transcripts in proximity to spatially restricted Halo-DBF molecules. (A) Overview of the Halo-seq procedure. A HaloTag protein domain is genetically fused to a protein which localizes to a subcellular location of interest. Because Halo ligands specifically bind HaloTags, this also spatially restricts a DBF Halo ligand. When irradiated with green light, DBF emits oxygen radicals that label nearby RNAs, resulting in their alkynylation. Alkynykated RNAs are substrates for in vitro biotinylation using ‘Click’ chemistry, which allows the localized RNAs to be separated from the bulk RNA sample with streptavidin pulldown, and quantified using high-throughput sequencing. (B) HaloTag protein domains fused to histone H2B and p65 are localized to chromatin, and the cytoplasm, respectively. HaloTag domains are visualized through the addition of a Halo ligand fluorophore. (C) RNA samples taken from cells expressing Halo fusions can be biotinylated in vitro, and this biotinylation is dependent upon the addition of a DBF Halo ligand to cells. (D) RNA samples taken from cells expressing Halo-p65 fusion. Cells were treated with DBF Halo ligand and then exposed to green light for 0, 1 or 5 min. RNA labeling, as assayed by the detection of biotinylated RNA following in vitro Click reactions, required both DBF and exposure to green light. (E) Alkynylated molecules can be visualized in situ by fusing them with fluorophores (e.g. Cy5-azide) using Click chemistry. Alkynylated molecules are restricted to the nucleus in cells containing H2B-Halo. They are only detectable in cells treated with both DBF and Cy5-azide, demonstrating the ability of HaloTag-restricted DBF to induce alkynylation of biomolecules.

Figure 2.

Halo-seq quantification of nuclear and cytoplasmic transcriptomes. (A) Differentially expressed genes in a comparison of pulldown and input RNA samples following Halo-seq RNA labeling using a H2B-Halo fusion. (B) As in A, but using a Halo-p65 fusion. (C) Halo-seq enrichments for selected RNA species known to be localized to the nucleus or cytoplasm. (D) Halo-seq enrichments for defined classes of RNAs. (E) Halo-seq enrichments for unspliced, intron-containing transcripts. The ratio of the abundance of unspliced transcripts to spliced transcripts was calculated for each gene. These ratios were then compared in input and pulldown samples. (F) Halo-seq enrichments for promoter-proximal upstream antisense transcripts. The ratio of the abundance of promoter-proximal upstream antisense to downstream transcripts was calculated for each gene. This ratio was then compared in input and pulldown samples. (G) Genes were binned by whether or not their 3′ UTR contained an AU-rich element (ARE). Gene enrichments in Halo-seq pulldown and input samples were then compared. (H) Genes were binned by the number of HuR binding sites (as defined by CLIP-seq) in their 3′ UTRs. Gene enrichments in Halo-seq pulldown and input samples were then compared. All significance tests were performed using a Wilcoxon rank-sum test. P value notation: * < 0.05, ** < 0.01, *** < 0.001, **** < 0.0001.

Figure 3.

Halo-seq quantification of the nucleolar transcriptome. (A) Visualization of the subcellular location of histone H2B and fibrillarin Halo domain fusions. Halo fusion proteins were visualized using a fluorescent Halo ligand. (B) RNA dot blot of RNA collected from cells expressing a Halo-fibrillarin fusion protein. (C) Differentially expressed genes in a comparison of pulldown and input RNA samples following Halo-seq RNA labeling using a Halo-fibrillarin fusion. (D) Halo-seq enrichments for selected RNA species known to be localized to the nucleus, nucleolus, or cytoplasm. (E) Halo-seq enrichments for defined classes of RNAs. (F) Halo-seq enrichments for unspliced, intron-containing transcripts. As in Figure 2E, the ratio of the abundance of unspliced transcripts to spliced transcripts was calculated for each gene. This ratio was then compared in input and pulldown samples. (G) Enriched gene ontology terms derived from RNAs identified as localized to the nucleolus. All significance tests were performed using a Wilcoxon rank-sum test. P value notation: * < 0.05, ** < 0.01, *** < 0.001, **** < 0.0001.

Figure 4.

Comparison of Halo-seq with similar methods for quantifying RNA localization transcriptome-wide. (A) RNA biotinylation, as assayed by RNA dot blot, in cells expressing a HaloTag domain and treated with Halo ligand forms of dibromofluorescein (DBF) or fluorescein. (B) Comparison of RNA biotinylation efficiency of Halo-seq and CAP-seq, a similar RNA proximity labeling method that uses miniSOG2, an enzymatic singlet oxygen generator. For the Halo-expressing cells, the experimental and control conditions correspond to the addition or omission of DBF, respectively. For the miniSOG2-expressing cells, the experimental and control conditions correspond to the treatment or omission of blue light irradiation. (C) Comparison of RNA biotinylation efficiency of Halo-seq and APEX-seq. For the Halo-expressing cells, the experimental and control conditions correspond to the addition or omission of DBF, respectively. For the APEX2-expressing cells, the experimental and control conditions correspond to the inclusion or omission of hydrogen peroxide. (D) Comparison of nuclear enrichments for selected gene classes using data produced by Halo-seq and CeFra-seq. The Halo-seq data corresponds to enrichments from the H2B-Halo-mediated labeling experiment described previously. The CeFra-seq data corresponds to enrichments in a biochemically defined nuclear fraction compared to total RNA from an unfractionated sample. (E) As in D, comparison of cytosolic enrichments for selected gene classes. The Halo-seq data corresponds to enrichments from the Halo-p65-mediated labeling experiment described earlier. The CeFra-seq data corresponds to enrichments in a biochemically defined cytosolic fraction compared to total RNA from an unfractionated sample. (F) Enrichments for unspliced, intron-containing transcripts. The ratio of the abundance of unspliced transcripts to spliced transcripts was calculated for each gene. For Halo-seq samples, this ratio was then compared in input and pulldown samples. For CeFra-seq samples, this ratio was then compared in fractionated (either nuclear or cytosolic) and total, unfractionated samples. All significance tests were performed using a Wilcoxon rank-sum test. P value notation: * < 0.05, ** < 0.01, *** < 0.001, **** < 0.0001.

Figure 5.

Halo-seq identifies transcriptome-wide changes in RNA localization following treatment with leptomycin B. (A) Visualization of the subcellular localization of Halo fusion proteins using a fluorescent Halo ligand. Note that while the p65 fusion is normally cytoplasmic, upon treatment with LMB, it becomes nuclear, indicating inhibition of CRM1 mediated export upon LMB treatment. (B) Volcano plot depicting changes in Halo-seq H2B pulldown enrichment (i.e. nuclear localization) following LMB treatment. (C) Changes in H2B pulldown enrichment following LMB treatment for selected gene classes. (D). Kmers of length 6 enriched in the 3′ UTRs of genes whose H2B pulldown enrichment was sensitive to LMB compared to the 3′ UTRs of genes whose H2B pulldown enrichment was not sensitive to LMB. (E) Genes were binned by the number of HuR binding sites (as defined by CLIP-seq) in their 3′ UTRs. Enrichments in H2B pulldown samples from LMB treated and untreated samples were then compared. (F) Abundance changes in total RNA (left) and nuclear RNA (right) following LMB treatment as a function of the number of HuR binding sites in the 3′ UTRs of transcripts. Total RNA and nuclear RNA samples correspond to the input and pulldown samples from H2B-targeted Halo-seq experiments. All significance tests were performed using a Wilcoxon rank-sum test. P value notation: * < 0.05, ** < 0.01, *** < 0.001, **** < 0.0001.

Figure 6.

Model for observed differences in nuclear and cytoplasmic transcript abundances. (A) ARE-containing RNAs were relatively less abundant in the cytoplasm than the nucleus in our Halo-seq experiments. Cytoplasmic localization of many proteins that drive ARE-mediated RNA degradation, including TTP, could be a major contributing factor to this observation. (B) Upon LMB treatment, ARE-containing RNAs become even more relatively enriched in the nucleus. AREs are often bound by HuR, which both participates in RNA export and stabilizes cytoplasmic ARE-containing transcripts. HuR export from the nucleus is blocked by LMB, resulting in decreased export of ARE-containing transcripts and/or their increased vulnerability to degradation in the cytoplasm.