Table 1.
Orientation concepts overview.
We show all orientation concepts in use for CRISPR arrays that are referenced within this work. We give them labels to simplify referencing the different orientation concepts. We also provide references for the concepts and tools that rely on the respective orientation concept.
Fig 1.
We show the steps taken by CRISPR-evOr to arrive at a prediction for a given group of spacer alignments. The left and right column show the same steps of reconstructing the ancestral history of a spacer alignment with SpacerPlacer. They differ only by their orientation, i.e. the spacer alignments are turned around compared to each other. Note, that SpacerPlacer is heavily reliant on the partial spacer insertion order (PSIO) and thus, this difference can have major impact on the reconstruction. Furthermore, in this case here, no tree was provided to SpacerPlacer and thus, SpacerPlacer estimates a different tree based on the spacer arrays for each orientation. After both reconstructions are performed, SpacerPlacer obtains maximum likelihoods of the respective reconstructions based on the reconstructed events. Then by comparing the ratio of both likelihoods, CRISPR-evOr decides for one of the orientations or, if the likelihoods are too close, determines that no confident prediction can be made.
Fig 2.
Example of a group of arrays with high confidence orientation prediction.
A and B show reconstructions of the same alignment of spacer arrays. B has the orientation provided by CRISPRCasdb, while A is simply the reconstruction of the same arrays, but arranged in reverse order. Note, that in B all spacers are reconstructed to be inserted (green) at the root and thus the exceptional amount of deletion events (red) along the tree (also shown in red in the spacer alignment). The insertions are produced through the first spacer 1 (light yellow) which is present at the root, as it is present in all arrays, and “pulls up” all other spacer insertions through the spacer insertion order. In contrast, in A the insertions are distributed more evenly along the tree and thus the reconstruction avoids a lot of deletions. Comparing both reconstructions clearly indicates that the orientation of A is more likely under the assumption of polarized insertions. To make the figures more concise, acquisition and deletion events in the tree are consolidated, e.g. 9 - 1 in green/red indicates that spacers 9 through 1 were acquired/deleted at this branch. At the leafs “+ (x) +” indicates that x spacers were acquired. They are the leftmost spacers in the alignment for the respective leafs (which are not found in any other array). The shown example is a group of Klebsiella pneumoniae (Cas type I-E) from CRISPRCasdb with small adjustments for readability.
Fig 3.
Comparison between acquisition- and PAM-orientation.
We compare the predictions between the PAM-orientation and CRISPR-evOr broken down according to Cas type. Only subtypes with at least 5 groups with more than 50 arrays are shown. While there are many arrays were both methods can not make confident decisions, in cases where they are confident, they almost always agree, but type II-C deviates from this behavior.
Fig 4.
Comparison with repeat based orientation methods.
We show the agreement of orientation predictions between CRISPRDirection, CRISPRstrand and CRISPR-evOr where all methods have high confidence broken down according to Cas type. Cas types were filtered as in Fig 3. At the bottom, we show Venn diagrams of the same data for three interesting subtypes with distinct behavior. Clearly, all methods agree quite often, especially for well-researched Cas types like I-E and I-F. However, there are substantial differences for predictions for type I-B, II-A and II-C. This indicates differences in the characteristics used for orientation prediction such as their repeats, leader and acquisition behavior. CRISPRDirection and CRISPRstrand were developed/trained using the more prevalent and well-studied Cas types, they may struggle for specific less common Cas types. Venn diagramms for the remaining types can be found in Fig G in S1 Material.