Table 2 Overview of the main technological advances that allowed miniaturization and single-cell readout of genome-wide epigenetic profiling assays

WGBS Conventional WGBS profiling is compatible with a relatively low number of cells (Table 1). Recently, WGBS was adapted to enable single-cell profiling (scBS-Seq; single-cell bisulfite sequencing) [188]. Single cells were captured by fluorescence-activated cell sorting (FACS). To cope with the extensive DNA damage caused by the bisulfite treatment, Smallwood et al. [188] performed tagging of the DNA fragments with sequencing adaptors after bisulfite conversion as developed by Miura et al. [189], instead of before conversion as performed in traditional WGBS. scBS-Seq allows to get coverage of up to 48.4% over all CpGs. A subsequent study by Farlik et al. [190] used a similar approach for scBS-Seq but adapted it such that the whole process of library preparation following bisulfite treatment and cleanup is performed in a single tube, minimizing DNA loss and reducing contamination risk [190].

ChIP-Seq. Traditionally, ChIP-Seq requires a large number of cells (at least several hundred thousands). However, improvements in the sample preparation procedure to prepare the ChIPped DNA for sequencing allowed to perform ChIP-Seq profiling on 1 × 10⁴ cells for H3K4me3 and H3K27me3 [191–194] and recently even 200 cells for H3K4me3 [195]. Also, the use of MNase for chromatin digestion has been shown to facilitate low-input ChIP-Seq [196, 197]. An alternative approach for downscaling the number of cells for ChIP-Seq is to use carrier material, such as inert proteins and/or mRNA, which do not interfere with the ChIP-Seq procedure but increase efficiency and sensitivity [198]. This strategy allowed to perform ChIP-Seq on the TF Estrogen Receptor (ER) on 1 × 10⁴ cells. Similarly, bacterial DNA has been used as carrier, although this comes at the cost of increased sequencing depth as the bacterial DNA remains included in the sequencing procedure [199]. In more recent studies aiming to obtain ChIP-Seq information from low numbers of cells, barcodes or adaptors for sequencing are ligated or transposed before or during the ChIP procedure instead of after the ChIP. ChIPmentation, the use of transposase to add adaptors to DNA during the ChIP, was shown to be highly efficient and compatible with as low as 1 × 10³ cells [200]. An alternative recent strategy for low-input ChIP-Seq relies on the addition of histone octamers during ChIP to outcompete unspecific binding [201]. Ligation of adaptors before the ChIP (indexing-first ChIP (iChIP)) allows to pool multiple samples during the ChIP-Seq procedure, after which the sequence tags can be mapped back to the original sample [202, 203]. Bernstein and his coworkers developed this further using direct adaptor ligation on MNAse treated chromatin in an automated droplet-based microfluidic device to obtain single-cell resolution for H3K4me3 and H3K4me2 ChIP-Seq [152]. Efficient immunoprecipitations were performed by pooling 100 single cells with the addition of carrier material that is not amplified during preparation of the ChIPped DNA for sequencing. This workflow enables the profiling of thousands of individual cells in parallel, mainly due to the continuous flow of droplets that is being generated to capture the individual cells (Fig. 4). Inherent to single-cell enrichment techniques, the coverage per single cell is sparse (~1000 unique reads per cell) and does not allow comprehensive analysis of protein binding sites in individual cells. However, the single-cell ChIP-Seq was shown to be very powerful in identifying functionally-relevant subpopulations within embryonic stem cells [152].

ATAC-Seq/DNAseI-Seq. ATAC-Seq has been downscaled to less than 200 cells [135, 175]. Next to this, Buenrostro et al. [151] reported ATAC-Seq to be compatible with single-cell profiling by performing transposition on single cells captured on a commercial microfluidics platform (Fluidigm C1; Fig. 4). This allows capturing of 96 single cells in parallel and subsequent processing steps toward a full library ready for sequencing. Together, this automated epigenetic platform represents the first of its kind in which a single-cell suspension is loaded on a platform that subsequently generates a full library for sequencing without any further manual intervention. An alternative approach for single-cell ATAC-Seq has been developed by Cusanovich et al. [204]. They performed the transposase reaction in intact nuclei on small pools, while simultaneously performing indexing of the tagged sides. Pooling followed by redistribution of the small cell numbers combined with the introduction of a second barcode for each cell allowed to map back the tags obtained after sequencing to individual cells. The advantage of this strategy is that it allows for a higher throughput as shown by the 15,000 individual cells profiled by Cusanovich et al. [204]. Recently, also DNaseI-Seq has been further developed to facilitate low-input profiling (between 1 × 10² and 1 × 10⁴ cells) as well as single-cell profiling [115]. Critically, after FACS sorting of single cells followed by lysis and DNaseI digestion, large amounts of circular plasmid DNA were added during further sample preparation for sequencing. The genomic coverage of both DNaseI-Seq and ATAC-Seq in single cells is inherently low due to the fact that each cell only contains two copies of the genome. The average number of sequence reads per cell was about 317,000 reads for DNaseI-Seq [115] and 73,000 [151] or 35,000 [204] reads for ATAC-Seq after deep sequencing of the libraries. Clearly, these numbers of sequencing reads do not allow to investigate individual genomic loci within single cells. Rather, the computational analysis in both studies makes use of DNaseI hypersensitive sites (DHSs) determined in pools of cells in order to call DHSs in single cells. Despite this limitation, the single-cell chromatin accessibility assays were shown to be powerful in identifying cell-type specific transcription factors, and their variation on genomic binding within individual cells on a global scale [115, 151].

4C-Seq and HiC-Seq. 4C-Seq and HiC-Seq are relatively new techniques [182–184], for which optimization to low cell numbers have not been extensively reported yet. However, it has been shown that HiC-Seq is compatible with single-cell profiling by performing in-nuclei DNA digestion and ligation and subsequent manual picking of individual nuclei. Using single-cell HiC-Seq it was shown that the large megabase-sized TADs that have been identified in large populations of cells are also present in single cells [205, 206]. Furthermore, single-cell HiC-Seq was shown to be very powerful to reconstruct chromosome folding. Although providing information at single-cell resolution, the single-cell HiC-Seq protocol requires 1 × 10⁷ cells as starting material to facilitate the early steps of the protocol. Inherent to the HiC-Seq protocol, the resolution obtained in individual cells is low. Currently, between 10,000 and 30,000 ligation events are profiled per cell [205].