Search

Contains fulltext : 195277.pdf (Publisher’s version ) (Open Access)

The TFIIH subunit XPB is involved in combined Xeroderma Pigmentosum and Cockayne syndrome (XP-B/CS). Our analyses reveal that XPB interacts functionally with KAT2A, a histone acetyltransferase (HAT) that belongs to the hSAGA and hATAC complexes. XPB interacts with KAT2A-containing complexes on chromatin and an XP-B/CS mutation specifically elicits KAT2A-mediated large-scale chromatin decondensation. In XP-B/CS cells, the abnormal recruitment of TFIIH and KAT2A to chromatin causes inappropriate acetylation of histone H3K9, leading to aberrant formation of transcription initiation complexes on the promoters of several hundred genes and their subsequent overexpression. Significantly, this cascade of events is similarly sensitive to KAT2A HAT inhibition or to the rescue with wildtype XPB. In agreement, the XP-B/CS mutation increases KAT2A HAT activity in vitro. Our results unveil a tight connection between TFIIH and KAT2A that controls higher-order chromatin structure and gene expression and provide new insights into transcriptional misregulation in a cancer-prone DNA repair-deficient disorder.

MIT specificity scores calculated by the CRISPOR website for 28 guide sequences calculated based on predicted off-targets with up to four, five, and six mismatches. Only one label is shown for identical guide sequences from two different studies. A change from four to five allowed mismatches used in the scoring results in a change of the specificity scores but only in minor changes of the ranking of the guide sequences by specificity score. (PDF 23 kb)

Extended version of Fig. 4 . This figure includes the datasets not shown in Fig. 4. Shown are: Wang 2015 data, both human and mouse data from Doench 2014, Doench 2016, both cell lines tested by Chari et al., Housden score and Housden dataset, Liu dataset and the score-like efficiency heuristics from Ren et al. and Farboud et al. Labeling is similar to Fig. 4: datasets are indicated along the y-axis and scores for predicting guide actvity along the x-axis. Data points where the training data of the algorithm has been processed with the algorithm, so likely affected by over-fitting, are shown in grey. (PDF 22 kb)

Similar to Fig. 3 but using off-target predictions by the CRISPR Design website ( http://crispr.mit.edu/ ). a For the 28 tested guide sequences, MIT guide specificity scores as calculated by the MIT website (x-axis), number of off-targets (y-axis), and sum of off-target modification frequencies (circle size). b The specificity of the 28 tested guide sequences (blue) versus the specificity scores of all unique 20mers in human coding regions (green). The specificity score histogram was calculated by running 1000 randomly selected guide sequences from human coding regions through the CRISPR Design website ( http://crispr.mit.edu/ ). (PDF 40 kb)

Precision/recall curves for the large-scale datasets. The precision/recall plot for the Wong score on the Chari dataset looks different to that in the Wong et al. article as our study is analyzing only full datasets and Wong et al. used cross-fold data. The datasets by Hart, Doench 2016, Koike-Yusa, or Wang/Xu give the best impression of the Wong score on independent data that was not used for training by Wong et al. (PDF 87 kb)

Ratio of off-target to on-target cleavage for validated guide sequences. The two guides RAG1B and RAG1A are not shown on this plot as their on-target cleavage was not determined in the study by Frock et al. [28]. Studies in the legend are referenced by the first author’s name; in the case of Kim et al., the cell type is also indicated. For each guide, the sum of all off-target modification frequencies was divided by the on-target modification frequency, e.g., a ratio of 2 indicates that cleavage is twice as frequent on all off-targets taken together than on the target. To better show the two outliers, a portion of the x-axis and y-axis was cut out. The CRISPOR website and Genome Browser tracks show a warning message for guide sequences with a GC content >75 %. (PDF 66 kb)

MIT specificity scores calculated by the CRISPOR website for 28 guide sequences calculated based on predicted off-targets with up to four, five, and six mismatches. Only one label is shown for identical guide sequences from two different studies. A change from four to five allowed mismatches used in the scoring results in a change of the specificity scores but only in minor changes of the ranking of the guide sequences by specificity score. (PDF 23 kb)

Similar to Fig. 3 but using off-target predictions by the CRISPR Design website ( http://crispr.mit.edu/ ). a For the 28 tested guide sequences, MIT guide specificity scores as calculated by the MIT website (x-axis), number of off-targets (y-axis), and sum of off-target modification frequencies (circle size). b The specificity of the 28 tested guide sequences (blue) versus the specificity scores of all unique 20mers in human coding regions (green). The specificity score histogram was calculated by running 1000 randomly selected guide sequences from human coding regions through the CRISPR Design website ( http://crispr.mit.edu/ ). (PDF 40 kb)

Overlap of off-target detection for the EMX1 and VEGFA guides tested by different assays. Off-targets are only shown if they were detected by at least a single study and with a frequency of 0.1 %. See Additional file 1: Tables S1 and Additional file 4: Table S2 for the modification frequencies and additional details on the off-targets for the guides EMX1 and VEGFA, respectively. Additional file 4: Table S2 also includes the data by Hsu et al. [7], who quantified cleavage at putative off-target loci predicted by the CRISPR Design website ( http://crispr.mit.edu/ ) with targeted deep sequencing, Tsai et al. [3], who isolated double-strand breaks with modified oligonucleotides followed by sequencing, Frock et al. [28], who detected translocations, and Kim et al. [33] and Kim et al. [27], who performed whole-genome sequencing to find CRISPR-induced modifications. For details on the different studies, see Additional file 1: Table S1. (PDF 17 kb)

Overlap of off-target detection for the EMX1 and VEGFA guides tested by different assays. Off-targets are only shown if they were detected by at least a single study and with a frequency of 0.1 %. See Additional file 1: Tables S1 and Additional file 4: Table S2 for the modification frequencies and additional details on the off-targets for the guides EMX1 and VEGFA, respectively. Additional file 4: Table S2 also includes the data by Hsu et al. [7], who quantified cleavage at putative off-target loci predicted by the CRISPR Design website ( http://crispr.mit.edu/ ) with targeted deep sequencing, Tsai et al. [3], who isolated double-strand breaks with modified oligonucleotides followed by sequencing, Frock et al. [28], who detected translocations, and Kim et al. [33] and Kim et al. [27], who performed whole-genome sequencing to find CRISPR-induced modifications. For details on the different studies, see Additional file 1: Table S1. (PDF 17 kb)

Precision/recall for the top quartile against the top quartile of the predictions. Positives are the top 25 % of each assay. Precision is defined as the ratio True positives/(True positives + False positives) or intuitively the ability of a score not to label as positive a sample that is negative. The recall is the ratio True positives/(True positives + False negatives) or intuitively the ability of a score to find all the positive samples. When the cost of the assay is high and there are many candidate guide sequences, e.g., for a long exon in rats, the priority may be on precision or, conversely, it may be on recall for a short exon in Drosophila. Data are separated into three parts: (1) U6-base data, (2) T7 in vitro data, (3) data from this study. The rules by Ren and Farboud are already binary; all other scores were considered a positive if the rank-percent of the score exceeded 75 %. The 75 rank-percent cutoffs were: Housden, 6.8; Wang, 78; Chari, 53; Doench, 32; Moreno-Mateos, 60; Xu, 42; Fusi, 62; Wong, 0. The 75 % cutoff for the Wong score is indeed 0 as most values of this score are 0. The Wong score has a good precision on the large U6-based datasets but a relatively low recall. Among the heuristics, the -GG rule from Farboud et al. has high precision for T7 in vitro transcription datasets, except for the Farboud dataset where some guides have been designed to fulfill the rule, so it is not unbiased. In U6-based datasets, most scores show a similar precision, but the Fusi et al. score has generally higher recall. (PDF 147 kb)

Ratio of off-target to on-target cleavage for validated guide sequences. The two guides RAG1B and RAG1A are not shown on this plot as their on-target cleavage was not determined in the study by Frock et al. [28]. Studies in the legend are referenced by the first author’s name; in the case of Kim et al., the cell type is also indicated. For each guide, the sum of all off-target modification frequencies was divided by the on-target modification frequency, e.g., a ratio of 2 indicates that cleavage is twice as frequent on all off-targets taken together than on the target. To better show the two outliers, a portion of the x-axis and y-axis was cut out. The CRISPOR website and Genome Browser tracks show a warning message for guide sequences with a GC content >75 %. (PDF 66 kb)