Background

Spontaneous DNA damage, which is usually caused by exposure to environmental chemicals or the free radicals generated from metabolic processes, frequently occurs in our genome. If the damage is not repaired correctly, it could lead to permanent change in the genomic sequence and sabotage gene functions (Ames et al., 1993; De Bont and van Larebeke, 2004; Tudek et al., 2010; Dizdaroglu et al., 2015). Besides mutagenesis, DNA damage could also perturb the local epigenetic stability and, as a result, alter gene expression patterns (Dabin et al., 2016). The non-uniform distribution of DNA damage on our genome could result from various levels of vulnerability to DNA damage in different genes. With high-damage genes being the potential Achilles’ heels of our genome, the alteration in expression of these genes could directly contribute to the development of human multifactorial diseases. Therefore, the accurate examination of the distribution of DNA damage and identification of high-damage genes are greatly desirable.

So far, high-performance liquid chromatography (HPLC)-based methods, gas chromatography/mass spectrometry (GC/MS)-based methods, and chromatin immunoprecipitation (ChIP)-based methods have been the primary approaches in measuring DNA damage. HPLC and GC/MS based methods are mainly used to measure the global levels of various DNA modifications (De Bont and van Larebeke, 2004; Wagner et al., 1992; Galashevskaya et al., 2013). However, since HPLC and GC/MS-based approaches require the DNA samples to be hydrolyzed into mononucleotides, the sequence context information cannot be recovered. Furthermore, the extensive treatment at the DNA hydrolysis step could introduce a significant amount of artificial DNA damage, which leads to large technical variations. On the other hand, ChIP-based methods could provide a sequence-based measurement of DNA damage by using antibodies to pull down the DNA fragments with damaged DNA (Yoshihara et al., 2014; Garcia-Nieto et al., 2017; Ding et al., 2017; Poetsch et al., 2018; Wu et al., 2018; Amente et al., 2019). However, such methods largely depend on the availability of antibodies that could recognize specific types of DNA damage, and do not have single-base resolution. The artificial damage induced during the DNA shearing step could lead to the overestimation of DNA damage levels.

Recently, we have reported a new single-cell whole-genome amplification (WGA) method, called linear copying and splitting based whole genome amplification (LCS-WGA), which can allow us to achieve genome-wide profiling of DNA damage (Zhu et al., 2021), since DNA damage can induce base misincorporation during amplification (Purmal et al., 1994, 1998; Gehrke et al., 2013). These misincorporated bases can then be directly identified in the sequencing data. We refer to damage-associated single-nucleotide variants as damSNVs. The main advantage of the single-cell WGA-based approach is that, without the requirement of either DNA hydrolysis or shearing, the artificial DNA damage is significantly reduced. Meanwhile, the genomic context of DNA damage can be effectively detected.

The only major technical challenge to single-cell WGA is the large number of amplification errors, which could limit the accuracy in single-cell WGA based approaches (Zong et al., 2012; Chen et al., 2017; Luquette et al., 2019; Bohrson et al., 2019; Xing et al., 2021). In LCS-WGA, we have overcome this technical hurdle by employing linear copying and then a split-amplification scheme. Briefly, we first perform three cycles of low-temperature annealing and extension with temperature ramping to generate multiple copies of the genomic DNA of the single cell as the preamplification step (Figure 1A). In this process, the DNA copies directly generated from the original genomic DNA are termed as semiamplicons, and the DNA products that are copied from the semiamplicons are termed full amplicons. It is worth pointing out that the semiamplicons are linearly produced original genomic DNA of the single cell. The full amplicons are nonlinearly copied from semiamplicons. To quench the nonlinearly produced full amplicons from amplification in the downstream multiple displacement amplification (MDA) reaction, we perform a double-stranded conversion, to convert all the full amplicons into double-stranded DNA, while the semiamplicons are kept as single-stranded DNA. Next, we split the preamplification products into three tubes, to perform independent MDA reactions. As MDA only amplifies single-stranded DNA, the double-stranded full amplicons cannot be amplified and, as a result, only linearly copied DNA products and original genomic DNA are amplified in these MDA reactions. Importantly, since the amplification errors in semiamplicons are independently produced during the preamplification step, we can cross-compare the sequencing result of different splits, to effectively filter out amplification errors. Here, we employ a two-split detection criterion, by which we only keep the variants detected in at least two split samples in the variant calling step (Figure 1B). Based on amplification errors in single split reaction, we estimate that the false positive rate due to amplification errors is 8.3 × 10^-10 per base when we apply two-split detection criterion (Zhu et al., 2021).

With this new method, we have successfully examined the DNA damage in human neurons and identified the association between the non-uniform distribution of DNA damage and 3D chromatin structure, as well as the connection between high-damage genes and the altered gene expressions in Alzheimer's disease and autism. This method can be readily applied to different biological contexts, to quantify damage levels in cells and, more importantly, identify high-damage genes. We believe that the characterization of the damagenome in different cell types from different tissues could shed new light on complex human diseases.

Figure 1. Scheme of LCS-WGA. (A) The scheme of pre-amplification and split-MDA. (B) The scheme to distinguish true de novo variants from amplification errors.