Correspondence: DNA shape is insufficient to explain binding

Proteins bind DNA through combinations of DNA base and shape recognition1. DNA base recognition refers to a unique arrangement of protein interactions with functional groups on the four DNA bases. Shape recognition refers to protein interactions with specific twists and turns of short stretches of DNA that may deviate from the average three-dimensional shape of B-form DNA. A recent study by Zentner et al.2 characterized the genome-wide binding of S. cerevisiae DNA binding proteins Abf1, Rap1 and Reb1, reporting many thousands of novel, low-scoring binding sites that lacked a consensus motif sequence. The sites were deemed significant because they reportedly possessed a DNA shape that was highly similar to that of the protein’s cognate sites and very different from random sites. We show here that when random sites are processed in precisely the same manner as highand low-scoring sites, including using a 50 bp search window (which by error was not done in Zentner et al.2), the low-scoring sites were no different than random, thereby invalidating the applicable conclusions. Since other analyses on slow sites were interpreted based on these invalid conclusions, we find an overall lack of evidence supporting the conclusion that Abf1, Rap1 and Reb1 predominately read DNA shape to recognize thousands of novel ‘low-scoring’ sites. In Figure 7 of Zentner et al.,2 it was reported that the DNA shape3 of low-scoring sites was on average highly similar to the DNA shapes associated with high-scoring sites and significantly different from random sites. From this result, it was concluded that the favourable DNA shape recognition at low-scoring sites captures transient scanning interactions. The analysis used a 50 bp search space centred over highand low-scoring ChEC-seq (chromatin endogenous cleavage with high-throughput sequencing) peak midpoints, so as to find the best match to a previously published consensus motif4. The P value threshold was set such that up to three mismatches to the consensus motif were allowed. From this, a DNA shape analysis was performed and compared across sites. We repeated the analysis and obtained precisely the same results for highand low-scoring sites (Fig. 1a, red versus blue traces). Most critically, Zentner et al.2 further reported that the same number of random sites, as a negative control, had on average no particular shape property. However, when we repeated this control3, we obtained a shape pattern that was essentially no different from the putative low-scoring sites (Fig. 1a, green versus blue traces). If we excluded the 50 bp search (that is, performed a 1 bp search), then we obtained the pattern reported by Zentner et al.2 (Fig. 1a, black traces). Our reanalysis shows that average DNA shape at low-scoring sites is indistinguishable from random if the best motif is sought equivalently in both data sets. Therefore, this DNA shape analysis3 provides no evidence of shape specificity at putative low-scoring sites, and no evidence that ChEC-seq peaks at low-scoring sites are a product of specific DNA shape recognition. To understand the source of ChEC-seq peaks at putative low-scoring sites, we next investigated the spatial relationship between high-scoring and low-scoring sites. We found that low-scoring sites were physically closer to high-scoring sites in the genome compared to random sites (Fig. 1b). Of the low-scoring sites, B50% were within 250 bp of a high-scoring site (compared to B8% of random sites), indicating that these low-scoring sites generally exist within the same nucleosomedepleted region (NDRs) as high-scoring sites. Consistent with this, Zentner et al.2 reported that AT-rich sequences, which are a well-known property of NDRs, are enriched at low-scoring sites. Since micrococcal nuclease (MNase) preferentially cleaves DNA in NDRs5, we surmise that the ChEC-seq peaks associated with putative low-scoring sites arise mostly from non-specific MNase cleavage events that are near high-scoring sites in accessible chromatin, perhaps after cleavage release from high-scoring sites. This is in accord with their temporally slow appearance (Fig. 3 in Zentner et al.2). Since the high-scoring sites for Abf1, Rap1 and Reb1 do not appreciably overlap, it follows that the positionally linked low-scoring sites would also not overlap, nor overlap with an untargeted MNase-only ‘negative’ control as reported in Zentner et al.2 Figure 4 in Zentner et al.2 reports the distribution of Abf1 X-ChIP-seq (chromatin immunoprecipitation with highthroughput sequencing) peaks around low-scoring sites/motifs. We note a local minimum directly at these motifs, and local maxima B50–100 bp away. These observations are consistent with high-scoring sites, which X-ChIP-seq is measuring, being physically close to but not coincident with putative low-scoring sites. Figure 5 in Zentner et al.2 reports MNase-derived DOI: 10.1038/ncomms15643 OPEN