Editor's evaluation: Hydrop enables droplet-based single-cell ATAC-seq and single-cell RNA-seq using dissolvable hydrogel beads
Year: 2022
Type: peer-review
Abstract: Article Figures and data Abstract Editor's evaluation eLife digest Introduction Results Discussion Materials and methods Data availability References Decision letter Author response Article and author information Metrics Abstract Single-cell RNA-seq and single-cell assay for transposase-accessible chromatin (ATAC-seq) technologies are used extensively to create cell type atlases for a wide range of organisms, tissues, and disease processes. To increase the scale of these atlases, lower the cost and pave the way for more specialized multiome assays, custom droplet microfluidics may provide solutions complementary to commercial setups. We developed HyDrop, a flexible and open-source droplet microfluidic platform encompassing three protocols. The first protocol involves creating dissolvable hydrogel beads with custom oligos that can be released in the droplets. In the second protocol, we demonstrate the use of these beads for HyDrop-ATAC, a low-cost noncommercial scATAC-seq protocol in droplets. After validating HyDrop-ATAC, we applied it to flash-frozen mouse cortex and generated 7996 high-quality single-cell chromatin accessibility profiles in a single run. In the third protocol, we adapt both the reaction chemistry and the capture sequence of the barcoded hydrogel bead to capture mRNA, and demonstrate a significant improvement in throughput and sensitivity compared to previous open-source droplet-based scRNA-seq assays (Drop-seq and inDrop). Similarly, we applied HyDrop-RNA to flash-frozen mouse cortex and generated 9508 single-cell transcriptomes closely matching reference single-cell gene expression data. Finally, we leveraged HyDrop-RNA’s high capture rate to analyze a small population of fluorescence-activated cell sorted neurons from the Drosophila brain, confirming the protocol’s applicability to low input samples and small cells. HyDrop is currently capable of generating single-cell data in high throughput and at a reduced cost compared to commercial methods, and we envision that HyDrop can be further developed to be compatible with novel (multi) omics protocols. Editor's evaluation De Rop et al. introduce a flexible microfluidics-based single-cell genomics technology that expands and improves previously existing custom droplet-based scRNA-seq protocols (inDrops and Drop-seq) in interesting directions: better data quality, simplified workflow, high-cell recovery, and flexibility towards other single-cell applications. https://doi.org/10.7554/eLife.73971.sa0 Decision letter Reviews on Sciety eLife's review process eLife digest Scientists are now able to determine the order of chemical blocks, or nucleic acids, that make up the genetic code. These sequencing tools can be used to identify which genes are active within a biological sample. They do this by extracting and analysing open chromatin (regions of DNA that are accessible to the cell’s machinery), or sequences of RNA (the molecular templates cells use to translate genes into working proteins). Initially, most sequencing tools could only provide an ‘averaged-out’ profile of the genes activated in bulk pieces of tissue which contain multiple types of cell. However, advances in technology have led to new methods that can extract and analyse open chromatin or RNA from individual cells. First, the cells are separated, via a technique called microfluidics, into tiny droplets of water along with a single bead that carries a unique barcode. The cell is then broken apart inside the droplet and the barcode within the bead gets released and attaches itself to the genetic material extracted from the cell. All the genetic material inside the droplets is then pooled together and sequenced. Researchers then use the barcode tags to identify which bits of RNA or DNA belong to each cell. Single-cell sequencing has many advantages, including being able to pinpoint precise genetic differences between healthy and abnormal cells, and to create cell atlases of whole organisms, tissues and microbial communities. But existing methods for extracting chromatin are very expensive, and there were no openly available tools for processing thousands of cells at speed. Furthermore, while several single-cell RNA sequencing tools are already freely available, they are not very sensitive or practical to use. Here, De Rop et al. have developed a new open-source platform called HyDrop that overcomes these barriers. The method entails a new type of barcoded bead and optimised elements of existing microfluidics protocols using open-source reagents. These changes created a more user-friendly workflow and increased sensitivity of sequencing at no additional cost. De Rop et al. used their new platform to screen the RNA and open chromatin of thousands of individuals cells from the brains of mice and flies. HyDrop outperformed other open-source methods when working in RNA-sequencing mode. It also provides the first open-source tool for sequencing open chromatin in single cells. Further improvements are expected as researchers tweak the platform, which for now provides an affordable alternative to existing methods. Introduction In the past 5 years, droplet microfluidics have been applied extensively to partition single cells and sequence their nucleic acid content. Two methods that pioneered the field, inDrop (Klein et al., 2015) and Drop-seq (Macosko et al., 2015), both rely on the same working principle: individual cells are rapidly encapsulated into a nanoliter droplet together with a barcoded bead. Barcoding primers carried by the beads inside the emulsion are then used to index each individual cell’s mRNA. This process occurs either inside the droplet, where the cell’s mRNA is reverse transcribed using barcoded primers released by the barcoded bead (inDrop) or after emulsion breaking, where the cellular mRNA is anchored onto barcodes carried by resin beads (Drop-seq). The thousands of single-cell transcriptomes can then be processed for next-generation sequencing in a pooled manner. The high throughput and low per-cell reagent consumption associated with this approach have allowed researchers to profile gene expression of tens of thousands of single cells (Kalish et al., 2020; Yap et al., 2021; Karaiskos et al., 2017) at an affordable cost. However, the labor requirements of current open-source droplet microfluidic protocols combined with their limited sensitivity have hindered widespread adoption. Furthermore, academic development of droplet-microfluidic single-cell sequencing technology after inDrop and Drop-seq has been limited. To our knowledge, only one noncommercial droplet-based single-cell assay for transposase accessible chromatin (scATAC-seq) protocol has been published so far (Chen et al., 2019). Despite its elegant conceptual solution to droplet-based combined scATAC and scRNA-seq, the SNARE-seq protocol is still labor intensive, and the use of resin beads leads to reduced cell capture rates. These resin beads are loaded at dilute concentrations to prevent microfluidic obstruction, and as a result, many droplets are empty, leading to a low cell capture rate (~2%), and reagent waste (Zhang et al., 2019). Instead, deformable hydrogel beads can be stacked and loaded into droplets at a fixed rate without the risk of microfluidic failure, thereby increasing cell capture rate to >50% (Zhang et al., 2019; Abate et al., 2009). Several commercial solutions have emerged since inDrop and Drop-seq (Zheng et al., 2017; Sarkar et al., 2018), and their application has been used to generate hundreds of thousands (Davie et al., 2018; Svensson et al., 2020) - and recently millions (Datlinger et al., 2021; Ren et al., 2021) - of single-cell transcriptomes at a high sensitivity. However, the high cost of these commercial protocols remains prohibitive for many research applications, and their fixed nature limits custom protocol development. In order to increase the sensitivity and user-friendliness of open-source microfluidic protocols, and to provide a more flexible and open platform, we developed HyDrop, a new hydrogel-based droplet microfluidic method for high-throughput scRNA-seq or scATAC-seq. We adapted inDrop’s original isothermal extension bead barcoding protocol (Zilionis et al., 2017) to a linear amplification workflow to generate more uniformly barcoded beads. Next, we applied a custom hydrogel bead production process similar to a recently published protocol (Wang et al., 2020) to generate dissolvable beads, improving barcoded primer release and diffusion. We also optimized Drop-seq’s pooled template-switching reverse transcription strategy for application inside the cell/bead emulsion, and optimized the assay’s sensitivity by testing several different cDNA library preparation strategies. The combination of these adaptations resulted in a significantly increased sensitivity for HyDrop-RNA compared to inDrop and Drop-seq at no additional cost, and in a more user-friendly workflow. Additionally, the dissolution of the hydrogel beads stabilizes the cell/bead emulsion during linear amplification thermocycling. This change allowed us to implement to our knowledge the first open-source single-cell ATAC-seq in droplets using hydrogel beads. We applied both technologies to mouse cerebral cortex and generated single-cell ATAC-seq and single-cell RNA-seq data that is highly concordant with reference data. Results Generation of dissolvable hydrogel beads with barcodes for scATAC-seq and scRNA-seq We generated barcoded hydrogel beads that can dissolve and release their embedded barcoded oligonucleotide. Polyacrylamide beads incorporating disulfide cross-linkers and short oligonucleotide polymerase chain reaction (PCR) priming sites were generated by droplet microfluidics similar to a recently published method (Wang et al., 2020). A custom droplet microfluidic chip (Figure 3—figure supplement 1) was employed to produce beads of approximately 50 μm diameter. These hydrogel beads were then barcoded using a modified three-round split-pool linear amplification synthesis method (Klein et al., 2015; Zilionis et al., 2017), resulting in 96 × 96 × 96 (884,736) barcode possibilities (see Supplementary file 1 – ‘Molecular sequence description of HyDrop bead barcoding’ for an in-depth visualization of the nucleic acid sequences in every step). The terminal sequence used in the final round of barcoding can be varied depending on the assay the beads will be used for (see Methods, Figure 1a). A sequence complementary to the Tn5 transposase adapter was used to capture tagmented chromatin fragments in scATAC-seq and a unique molecular identifier (UMI) (Kivioja et al., 2011)+ poly (dT) sequence was used to capture and count poly (A)+ mRNA in scRNA-seq (protocols described further). The barcoded beads were stored in a glycerol-based freezing buffer at –80°C in order to prevent loss of primers over time (Figure 1—figure supplement 1). Figure 1 with 6 supplements see all Download asset Open asset Technical overview of HyDrop barcoded bead production and quality control. (a) Split-pool process for barcoding of dissolvable hydrogel beads for single-cell RNA-sequencing (scRNA-seq) and single-cell assay for transposase-accessible chromatin (scATAC-seq). Unbarcoded hydrogel beads are sequentially distributed over 96 wells, sub-barcoded using linear amplification polymerase chain reaction (PCR), repooled, and re-distributed two more times to generate 96 × 96 × 96 (884,736) possible barcode combinations. Different 3-prime terminal capture sequences are possible depending on the oligonucleotide sequence appended in the last step. (b) Semiquantitative assessment of bead primer incorporation by fluorescence in-situ hybridisation (FISH) using a fluorescein amidite (FAM) probe after every sub-barcoding step shows that bead fluorescence uniformity is retained throughout the barcoding process. (c) FISH with FAM probes complementary to only one of 96 sub-barcode possibilities shows that approximately 1/96 beads exhibit fluorescence for a selected sub-barcode probe. Fluorescence signal is overlaid with a brightfield image at 50% transparency to indicate positions of nonfluorescent beads (see Figure 1—figure supplements 2–6 for additional quality control experiments and full images). We validated the extension of the hydrogel bead primers using fluorescent probes complementary to the beads 3-prime terminal sequence (Zilionis et al., 2017; Figure 1b and Figure 1—figure supplement 2) and the sub-barcode purity using fluorescent probes complementary to 1 of the 96 sub-barcode possibilities (Figure 1c, Figure 1—figure supplements 2 and 3, Tables S12). These experiments showed that there was no significant loss of primers or mixing of barcodes throughout the barcoding process, and that the beads are uniform in size and primer content. Additional testing revealed that our modified linear amplification barcoding method produced more uniformly barcoded beads compared to the isothermal amplification protocol described in inDrop (Figure 1—figure supplement 2). Furthermore, the disulfide moieties incorporated in both the bead’s polymer matrix and oligonucleotide linker are cleaved when exposed to reducing conditions, such as dithiothreitol (DTT). This chemical method of release is more user-friendly compared to the UV-mediated (Klein et al., 2015; Zilionis et al., 2017) primer release as the beads do not have to be shielded from light. In addition to improved primer release compared to nondissolvable beads (Figure 1—figure supplement 4, Table S3), dissolvable beads did not disrupt the emulsion during the thermocycling (Figure 1—figure supplement 5). We hypothesize that this effect is due to the lower physical stress generated by the dissolved hydrogel as opposed to the solid hydrogel. This added benefit allowed us to implement scATAC-seq (see further) on HyDrop. Finally, by varying the concentration of the acrydite primer during bead synthesis, lower or higher amounts of cleavable barcoded primers could be generated. We found that when the concentration of acrydite primer incorporated in the hydrogel matrix was high (50 μM, similar to InDrop), excess unreacted barcodes could not be sufficiently filtered out in further downstream steps. These primers were then carried over to subsequent reactions, leading to random barcoding of free fragments after droplet merging, and subsequently to cell-mixed expression or chromatin accessibility profiles. The bead primer concentration with an optimal balance between sensitivity and library purity was found to be 12 µM for both scATAC-seq and scRNA-seq (Figure 1—figure supplement 6). Implementation and accuracy assessment of HyDrop-ATAC We implemented a new open-source protocol for single-cell ATAC-seq using HyDrop’s dissolvable barcoded hydrogel beads. Nuclei were Tn5 tagmented in bulk and coencapsulated with HyDrop-ATAC beads. Pitstop, a selective small molecule clathrin inhibitor, was supplemented during nuclei extraction and tagmentation to increase nucleus permeability to Tn5 (Mulqueen et al., 2019). Inside the droplet, the hydrogel beads dissolve and release their uniquely barcoded primers inside the droplet due to the presence of DTT carried by the nuclei/PCR mix. Subsequent thermocycling of the emulsion denatures the Tn5 protein complex and releases accessible chromatin fragments within the droplet. These fragments were then linearly amplified and cell indexed by the bead’s barcoded primers after which the emulsion was broken and the indexed ATAC fragments were pooled, PCR amplified, and sequenced (Figure 2, see – Supplementary file 2 ‘Molecular sequence description of HyDrop-ATAC’ for an in-depth visualization of the nucleic acid sequences in every step). Figure 2 Download asset Open asset Schematic overview of HyDrop-ATAC. Nuclear membrane is visualized in salmon, water droplet is visualized in blue. Nuclei were Tn5 tagmented in bulk and co-encapsulated with HyDrop-ATAC beads, where the hydrogel beads dissolve and release their uniquely barcoded primers. Thermocycling of the emulsion releases accessible chromatin fragments which are then linearly amplified and cell indexed by the bead’s barcoded primers within the droplet. The emulsion is then broken and the indexed ATAC fragments are pooled, PCR amplified, and sequenced. In order to coencapsulate beads and nuclei with a high capture rate and minimal microfluidic complexity, we developed a custom microfluidic chip (Figure 3a and b). The chip design features one inlet for beads, one inlet for cells or nuclei, and one inlet for the emulsion carrier oil. This configuration is slightly more convenient to operate compared to inDrop’s four channel setup. Several layers of passive filters near the inlet ports mitigate dust and debris buildup during droplet generation to prevent obstruction of the channels. Beads and nuclei were loaded via a tip reservoir to reduce nonlinear flow behavior and the potential accumulation of cells/nuclei and hydrogel beads associated with narrow tubes (Sinha et al., 2019; Hur et al., 2011; Figure 3b). Due to the stability of all flows and the deformable nature of the hydrogel beads, >90% occupancy of hydrogel beads in droplets could be achieved (Abate et al., 2009; Figure 3c), resulting in a final cell recovery of ~65%. Figure 3 with 1 supplement see all Download asset Open asset Microfluidics setup employed in HyDrop emulsion generation. (a) 3d rendering of HyDrop chip design. The design has three inlets: one each for carrier oil, barcoded hydrogel beads and cell/reaction mix (see Figure 3—figure supplement 1 for full design). Passive filters at each inlet prevent dust and debris from entering the droplet generating junction. (b) Diagram and snapshot of cell/bead encapsulation into microdroplets. (c) Microfluidic chip setup on the Onyx integrated microfluidic instrument. Cells and beads are loaded into pipette tips and plugged into a HyDrop chip. Flow of oil and aqueous phases is achieved by Onyx displacement syringe pumps. To assess the purity of scATAC-seq libraries generated by HyDrop-ATAC, we performed two mixed species experiments. First, we generated single-nuclei ATAC-seq libraries from a 50:50 mixture of human breast cancer (MCF-7) and a mouse melanoma cell line generated previously (Dankort et al., 2009). We developed a custom preprocessing and mapping pipeline for HyDrop-ATAC data (Dankort et al., 2009). After filtering the cell barcodes for a minimum transcription start sites (TSS) enrichment score of 7 and unique fragment count of 1000, we recovered 1353 cells from an input of 2000, and a median of 2705 unique fragments per cell. About 88% of all barcode reads were paired with the whitelist when one mismatch was allowed along the entire length of the 30 bp barcode. An additional 1.16% and 1.13% of HyDrop barcodes could be assigned to the whitelist when two and three mismatches were allowed, respectively. The remaining 9% of barcode sequences could not be identified and are most likely a result of frameshift due to incorporation errors in the synthetic oligonucleotide production process. Notably, only 39.7% of barcode reads could be matched with the whitelist in a public inDrop dataset (Kalish et al., 2020), when one mismatch was allowed. When prefiltered cell barcode sequences were used as a whitelist for a public mouse retina Drop-seq dataset (Macosko et al., 2015), 49.5% of reads could be matched. We identified 98.4% of cells as either human or mouse at a minimum purity of 95% fragments mapping to either species (Figure 4a). Next, we generated libraries from a mixture of MCF-7/PC-3/Mouse cortex (45:45:10) to evaluate whether two human cell types can be distinguished. A spike-in of 10%mouse cells was used as an internal control. We recovered 2602 human cells, 466 mouse cells, and 93 species doublets after filtering for 95% species purity. Clustering human cells (together with the MCF-7 cells from the first species mixing experiment to evaluate batch effects) recovered two distinct populations, each exhibiting specific ATAC-seq peaks near MCF-7 or PC-3 marker genes (Figure 4—figure supplement 1a). Aggregated reads per cluster showed typical ATAC-seq profiles concordant with public bulk ATAC-seq data (Consortium, 2012; Figure 4b, Figure 4—figure supplement 1). Figure 4 with 1 supplement see all Download asset Open asset Validation of HyDrop-ATAC on mixed species cell lines. (a) Scatterplot of the number of unique fragments detected in a 50:50 mixture of human Michigan Cancer Foundation-7 (MCF-7) and mouse melanoma cells colored by local density estimation. (b) reads per genome coverage (RPGC)-normalized aggregate genome tracks comparing HyDrop-ATAC and bulk ATAC-seq profiles of human MCF-7 and prostate cancer-3 (PC-3) cell lines around the estrogen receptor 1 (ESR1) locus, scaled to of all four enrichment profile of reads around transcription start see Figure 4—figure supplement 1 for of these cells and count with public data. Supplementary data available for of HyDrop-ATAC on flash-frozen mouse cortex cerebral cellular and specific accessibility profiles To evaluate the of HyDrop-ATAC on we generated single-cell libraries from mouse were sequenced to approximately After filtering for a minimum of unique a enrichment score of and cells detected as doublets by et al., we recovered a of 7996 single Cells the filters a median of fragments per a median enrichment score of and a median of of fragments in high-quality cells and low of signal (Figure the number of unique fragments per cell is lower that of commercial methods per cell for see HyDrop-ATAC in of enrichment and of reads in peaks We used et al., to reduce the of the dataset and the et al., to cluster cells (Figure using the ATAC signal around several et al., 2021; et al., recovered distinct cell similar to previously published scATAC-seq mouse cortex data et al., 2018; et al., 2020; Figure supplement 1). we identified and by accessibility and respectively. we were able to further with specific ATAC-seq signal near either or and with accessibility near either or Finally, HyDrop-ATAC data revealed distinct specific accessible (Figure and b). Figure 5 with 1 supplement see all Download asset Open asset of HyDrop-ATAC on flash-frozen mouse cortex cellular enrichment of ATAC fragments near transcription start sites (TSS) enrichment per barcode of reads in peaks per (c) and rate per barcode in mouse cortex HyDrop-ATAC data. A minimum enrichment of 5 and a unique number of fragments of are used as to cells from Cells are colored by local density estimation. of 7996 mouse cortex nuclei with cell type by accessibility near marker cells cortex neurons and neurons layers 2–6 near neurons and neurons Figure supplement 1 for cluster marker gene Supplementary data available for and Figure 6 Download asset Open asset accessible between cell types recovered by HyDrop-ATAC on mouse cerebral (a) accessibility of per (b) aggregate genome the one for each Implementation of HyDrop-RNA as a method between inDrop and We implemented a new scRNA-seq assay using barcoded bead primers a 3-prime poly (dT) cells or nuclei were in a reverse mix and coencapsulated into with 3-prime poly (dT) HyDrop beads. The same microfluidic chip design was used for both HyDrop-RNA and HyDrop-ATAC. Cells are inside the droplets with the buffer in which the barcoded beads were barcoded primers were released from the hydrogel bead after to DTT in the reverse mix. transcription inside the emulsion thousands of barcoded single-cell cDNA libraries in The emulsion was then broken and the single-cell libraries were processed further in a pooled (Figure see Supplementary file Figure 7 Download asset Open asset Schematic overview of HyDrop-ATAC. cells or nuclei are in a reverse mix and coencapsulated into with 3-prime poly (dT) HyDrop beads. Cells are inside the droplets and barcoded primers are released from the hydrogel bead. transcription inside the emulsion using a template-switching thousands of barcoded single-cell cDNA libraries in which are processed further in a pooled after the ‘Molecular sequence description of for an in-depth visualization of the nucleic acid sequences in every to the protocol (Zilionis et al., both assays are on hydrogel beads, HyDrop-RNA significantly from HyDrop-RNA a reverse transcription technique to an in priming workflow. This change and up the protocol significantly with no in sensitivity. To the assay’s we the of several to the protocol’s reaction chemistry by testing on a 50:50 mouse We first the use of pooled after reverse transcription to barcode primers. We that these barcoded primers could during the subsequent bulk polymerase chain reaction leading to a loss of purity of associated with a barcode. from the increase in cells we found that improved assay purity (Figure supplement 1). We then the of a nucleic acid in the as it has been to increase assay sensitivity due to increased stability of the complex et al., We also the addition of and to the reverse transcription the addition of as a molecular and as a of the complex has been to assay sensitivity in et al., 2020). Finally, we whether a second synthesis library approach could the In order to we compared both and to the RNA from the first and the of both DNA polymerase and et al., 2020; et al., 2021) fragment for second We found that the and protocol supplemented with performed a median of and genes per cell with a species purity of (Figure and Figure supplement 1). the method was used in all further HyDrop-RNA experiments. this protocol to a 50:50 mouse mixture recovered human and mouse cells with species doublets at a of 95% species with a median of and genes per cell (Figure Figure with 2 supplements see all Download asset Open asset Validation and of HyDrop-RNA on species mixed cell line samples and mouse cerebral (a) of unique molecular identifier (UMI) and gene count of HyDrop-RNA with and without merging, with the use of a nucleic acid and with and fragment library lines and Figure supplement 1 for species purity of these experiments. (b) Scatterplot of human and mouse detected in a 50:50 mixture of human MCF-7 and mouse melanoma
Source: eLife
Author: Naama Barkai
Institution: Weizmann Institute of Science
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: gold
Estimated APC paid: $2,500
OpenAlex ID: W4226098844