Although cells derived from the same clone, tissue or organism have identical genomes, their gene expression varies, as single-cell RNA-sequencing data have shown. “We need to understand the determinants of that heterogeneity in expression,” says Keji Zhao at the US National Institutes of Health, “and we need to understand single-cell chromatin states, chromatin accessibility.”

Comparing cellular expression profiles yields information on the genes actively expressed in each cell, but identifying regions of open chromatin provides additional information about where regulatory regions such as enhancers and promoters lie. It also reveals a gene's potential. “At the moment of the analysis, a gene may not be expressed, but it may be expressed under the right conditions,” says Zhao.

The traditional method for probing open chromatin (termed DNase-seq) has been to digest the DNA with DNase I nuclease and then isolate and sequence short DNA fragments (fewer than 200 base pairs (bp)) that comprise regions not protected by histones and thus are open to cleavage. Large consortia such as ENCODE have profiled such DNase-hypersensitive sites (DHSs) in populations of cells and drawn what Zhao considers a “pretty comprehensive” map of cells' DHS landscapes. ENCODE profiled around 90,000 DHSs in one cell type, but a critical limitation is that they needed 10 million cells to do so. “You want a more sensitive method,” explains Zhao, “because practically for cells from primary tissue the numbers are very limited.”

To increase the sensitivity of DNase-seq, Zhao's team added large excesses of circular carrier DNA to minimize loss of the tiny amounts of nuclease-digested fragments and to specifically amplify and enrich only those short DHS fragments. PCR adaptors do not ligate to circular DNA, so it cannot be amplified, and nonspecific DNase fragments amplify inefficiently because they tend to be much larger than 200 bp. With this single-cell (sc) DNase-seq strategy, the researchers were able to profile DHSs using very low input—from 10,000 cells down to a single cell—and their results agree with those from ENCODE data. In a single cell they identified more than 30,000 DHSs, from 100 cells they detected 50,000 DHSs, and with 1,000 cells they detected 60,000 DHSs, which comes close to ENCODE's 90,000.

Of course, it is difficult to validate a DHS unique to a single cell. Unlike transcriptional profiling, which can be validated with imaging techniques such as fluorescence in situ hybridization, there is no orthogonal technique for DNase-seq. Zhao's team used DHSs derived from bulk genomic DNA or ENCODE data to confirm their results and saw that DHS patterns even from single cells were highly reproducible and correlated with modifications on histones typical for active chromatin.

Changes in gene expression and chromatin accessibility are also a hallmark of cancer cells, and the researchers tried scDNase-seq on formalin-fixed paraffin-embedded (FFPE) tumor samples. Such samples present a particular challenge to molecular analysis, because the DNA is damaged during the harsh fixation procedure. “We did not expect that it would work,” recalls Zhao, but they were pleasantly surprised to see that chromatin features such as DHSs were retained in the samples.

They found a mutation in a patient with thyroid cancer that affected the binding of a tumor suppressor. Opening the huge repertoire of FFPE samples to DHS analysis is likely to yield a wealth of important information on how the regulation of gene expression goes awry in disease.