Optimize single-nucleus ACE-Seq (snACE-Seq) to provide cell type classification,

4.1.1 Introduction

DNA from a single cell (~5-10 pg). This application is especially important as the field of epigenetics, by definition, defines differences between cells that contain the same genetic code; therefore, analyzing bulk samples that average epigenetic data from many cells can only provide ensemble information and likely obscures important potential findings.

4.1.2 Preliminary results

One limitation in our strategy for direct application in single cells lies in the arena of library preparation. At present, our current library preparation protocol would not be efficient enough to generate complex libraries from such a small amount of input DNA. Therefore, we are currently piloting a new library generation protocol developed as a collaboration between Joe Ecker and Swift Biosciences for use in single-cell bisulfite sequencing (Luo et al., 2017). The major difference is the initial use of a random priming step before their patented Adaptase reaction that should help to increase the complexity of the library. Initial attempts at using this library preparation protocol have proven successful for both phage and mammalian DNA with 0.5 ng gDNA input (~50-100 cells).

Another important limitation for optimizing our traditional workflow for snACE-Seq is our ability to shear genomic DNA from single cells. In our traditional workflow, we have been able to purify genomic DNA from a bulk cellular population, shear using sonication, and then input this sheared DNA into our ACE-Seq reactions. However, such limited DNA input from single cells prevents purification and mechanical shearing without significant loss of DNA. Our first attempt to generate fragmented DNA from single cells actually relies on bisulfite treatment, which, as discussed previously, is known to fragment DNA. This treatment can be directly applied to lysed single cells using the Zymo EZ-96 DNA Methylation Direct Kit. Initial attempts at using bisulfite treatment followed by ACE-Seq to localize hmC has been somewhat successful, in that DNA does appear to be fragmented and we are able to achieve both robust bisulfite conversion of Cs and A3A conversion of mCs. Importantly, one unknown factor going into this

approach was whether the CMS adduct generated by the bisulfite reaction with hmC would be sufficient to prevent deamination of hmC by A3A in lieu of glucosylation. Indeed, we were able to achieve protection of hmC at levels similarly high as our traditional ACE-Seq workflow utilizing glucosylation via βGT (Figure 4-1).

Currently, we are utilizing the approaches outlined above to provide a framework for optimizing ACE-Seq conditions for snACE-Seq. For example, we are currently testing whether

generated by each protein preparation. Results to this end have been complicated by the use of many different preps, though it does seem possible that we could decrease the amount of A3A in each reaction from 5 uM to at least 2 uM while maintaining similarly high levels of conversion (Fig 4-2). This is a promising first step, as it will allow us to, in effect, more than double the amount of reactions we can perform with a single preparation of enzyme.

4.1.3 Next steps

This project is still very much in its preliminary stages, so there are many different approaches we are currently trying out for the snACE-Seq adaptation. For example, we are not planning on utilizing bisulfite as our fragmentation step in the long term, as it comes with many of the caveats discussed previously (Tanaka and Okamoto, 2007), the most notable of which is its destructive nature which limits the complexity of any resulting library for subsequent Illumina

Figure 4-2. Conversion efficiencies of ACE-Seq on lambda (C/mC) and T4-hmC spike-in controls with 2 uM enzyme

sequencing. One potential option we will explore is the use of an enzymatic “fragmentase,” which is now commercially available from New England Biolabs. Use of this enzyme will continue to improve our nondestructive method (as the enzyme should fragment the genome but not produce abasic sites or other destructive effects), and the size of average fragments can be controlled through time of incubation with the enzyme. While this requires much initial optimization both to achieve the right fragment sizes and to determine optimal conditions so as to move from

fragmentation to BGT/A3A treatment without comprising the efficiency of the latter steps. We are also pursuing another approach previously applied to single-cell Hi-C, where barcodes and Tn5 transposase are directly applied to permeabilized nuclei to simultaneously barcode and fragment the gDNA (Ramani et al., 2017). As discussed below in Section 4.3, we are also interested in characterizing and optimizing a hyperactive chimeric APOBEC variant for use in both traditional ACE-Seq and possibly in snACE-Seq, which could even further lessen the amount of enzyme required for complete conversion in each step.

Once optimal conditions are achieved for fragmentation, hmC protection, and C/mC conversion, we will perform a pilot experiment on 384 individual sorted nuclei from wildtype murine cortical neurons from different stages of embryonic development using a nuclear isolation technique developed in the Wu and Zhou labs (P. Hu et al., 2017). Initially, we will use snACE- Seq in conjunction with single-nucleus Drop-Seq (snDrop-Seq), a method for massively parallel RNA sequencing from sorted nuclei developed in Hao Wu’s lab. Using both of these methods together, we aim to address the biological question of how intragenic hmCG levels correlate to

correlation between single cell hmC levels and localization, either in certain genomic elements or in specific genes of interest, and provide functional evidence for certain phenomena we observe in single cells.

4.2 Develop mC-ACE-Seq for nondestructive, direct readout of genomic mC utilizing a