CRISPR Visualizer: rapid identification and visualization of CRISPR loci via an automated high-throughput processing pipeline

ABSTRACT

A CRISPR locus, defined by an array of repeat and spacer elements, constitutes a genetic record of the ceaseless battle between bacteria and viruses, showcasing the genomic integration of spacers acquired from invasive DNA. In particular, iterative spacer acquisitions represent unique evolutionary histories and are often useful for high-resolution bacterial genotyping, including comparative analysis of closely related organisms, clonal lineages, and clinical isolates. Current spacer visualization methods are typically tedious and can require manual data manipulation and curation, including spacer extraction at each CRISPR locus from genomes of interest. Here, we constructed a high-throughput extraction pipeline coupled with a local web-based visualization tool which enables CRISPR spacer and repeat extraction, rapid visualization, graphical comparison, and progressive multiple sequence alignment. We present the bioinformatic pipeline and investigate the loci of reference CRISPR-Cas systems and model organisms in 4 well-characterized subtypes. We illustrate how this analysis uncovers the evolutionary tracks and homology shared between various organisms through visual comparison of CRISPR spacers and repeats, driven through progressive alignments. Due to the ability to process unannotated genome files with minimal preparation and curation, this pipeline can be implemented promptly. Overall, this efficient high-throughput solution supports accelerated analysis of genomic data sets and enables and expedites genotyping efforts based on CRISPR loci.

1. Introduction

Due to the selective pressure of viral predation, bacteria have evolved multiple defence mechanisms against bacteriophage infection [1–3]. Innate immunity is often provided by restriction-modification and abortive-infection systems, while adaptive immunity is driven by a range of diversified CRISPR-Cas (clustered regularly interspaced short palindromic repeats-CRISPR associated proteins) systems [1,4]. Essential to CRISPR-Cas based adaptive immunity is a process called adaptation, in which invading foreign DNA is integrated into the CRISPR array to impede future attacks [5–13]. As a result, these repeat-spacer arrays represent a unique evolutionary record of attacks by bacteriophages and other invasive and mobile genetic elements. The industrial relevance of CRISPR-based genotyping has been demonstrated in several bacterial species widely found in the food supply chain, encompassing starter cultures and food spoilage organisms such as Streptococcus thermophilus [14], Bifidobacterium longum [15], and Lactobacillus buchneri [16]. CRISPR-based genotyping also exhibits significant medical relevance through its usage in identification of pathogens such as Mycobacterium tuberculosis [17,18], Salmonella enterica [19,20], Clostridium difficile [21], Escherichia coli and many others [22,23], including food-borne pathogens. Characterization of CRISPR-Cas systems has also revealed alternative functions such as roles in virulence regulation and DNA repair [24]. Understanding the origin and structure of CRISPR loci and their transcribed RNA (crRNA) is a shared thread running through many fields and functions: from the novice CRISPR biologist recently introduced to adaptive immunity, to the expert investigating a bacterial clade; from the genome assembler working to decode repetitive genomic regions, to the explorations of a phylogeneticist [17]. Though the genetic locus is the defining feature of CRISPR-Cas systems, their function hinges on CRISPR-derived RNA molecules that guide the activity and efficiency of these immune systems. Indeed, crRNA, and thus, CRISPR repeats and spacers, are the fundamental basis of CRISPR-Cas-based genome editing, with RNA-guides at its core. It is thus critical to discern and characterize the loci and sequences that define crRNAs. Yet, there is a surprising paucity of tools dedicated to understanding, manipulating and visualizing these loci. Pipeline analysis of the DNA sequences underlying active or ancient crRNA is a powerful tool in unsupervised genomic discovery, especially in an era where sequencing has become affordable, and high-throuput -omic technologies have yielded genomic and metagenomic sequencing data en masse. The resulting analytical bottleneck is hampered by the difficulty with which repeated regions are sequenced and assembled. Visualization and comparison of CRISPR loci bridges the fields of genome editing and CRISPR biology, providing a common denominator between disciplines. Although a handful of useful bioinformatic tools have been developed to facilitate CRISPR repeat and spacer identification, the space for large-scale discovery and comparison remains untapped [25–28]. These tools are largely centred around CRISPR-Cas detection and investigation of related features at the single-organism level, enabling deep exploration and discovery with narrow scope and are not optimized for direct comparison across organisms. Often, these jobs are limited to a single genome file at a time and can take several minutes to run per strain depending on the level of detail provided for each locus. Nevertheless, we used specific features of existing CRISPR tools (such as CRT, CRISPR finder, CRISPRCasFinder, CRISPRdb, CRISPRdetect, CRISPRone, CRISPRdisco and more) as a basis to develop a more efficient and scalable pipeline. Direct nucleotide comparison is time consuming and cumbersome in contrast to visual comparison, which is more intuitive and better suited for comparison across multiple organisms. Here, we present an automated pipeline that combines the power and speed of current bioinformatic tools with the user-interface of the modern web-browser for improved analysis of CRISPR arrays. We use this technology to explore reference systems and model strains across the two established CRISPR classes and 4 major types, with examples for reference subtypes: I-E, II-A, III-A, and V-A [29–31]. Interestingly, results illustrate the speed at which evolutionary relationships can be recognized and explored, and provide insights into their function and activity.

2. Materials and methods

2.1. Extraction and conversion pipeline

The repeat and spacer sequence extraction is performed by MinCED (github.com/ctSkennerton/minced), which is derived from the CRISPR Recognition Tool (CRT) [26] . It is wrapped in a custom Bash script which enables parallel spacer/repeat extraction across multiple genomes using the shell’s native background job management system – drastically reducing execution time when processing large numbers of genomes. The custom script also facilitates the splitting of loci into individual rows and converts MinCED repeat output data to .fasta format and files for downstream analyses. The required user input is simple, and can consist of any genome, contig(s), or sequence file in .fasta format. If a specific file is not provided, all genome files in the current working directory will be analyzed. Extracted CRISPR spacer and repeat sequences are converted into an easily comparable colour and glyphicon (getbootstrap.com/docs/3.3/components/). Each spacer or repeat is represented by two components: a coloured square, representing the composition of a subset of the sequence; and a coloured symbol, representing the composition and length of the sequence. The conversion algorithm was written in Python – adapted from a VBA algorithm originally created by Philippe Horvath and colleagues [14]. This engine outputs the conversion data into a .json file available for uptake by a local server.

2.2. Reference CRISPR-Cas systems for select types and subtypes

Select genomes were sampled to encompass the previously defined reference strains across the two main CRISPR-Cas system classes, as well as the four major types (I, II, III, V) and select subtypes (I-E, II-A, III-A, V-A). Additional genomes within similar species carrying these CRISPR-Cas systems were randomly selected through a RefSeq search for comparative analysis with the appropriate reference loci [29,30,35] (Table 1).

Table 1.

Strains used in pipeline analysis.

Class	Type/Subtype	Strain	Accession
Class 1	I-E	Escherichia coli O111:H11 str. CVM9455	NZ_AKAX00000000.1
		Escherichia coli O26:H11	NZ_BEVX00000000.1
		Escherichia coli K-12 strain ER3476	NZ_CP010440.1
		Escherichia coli strain K-12 NEB 5-alpha	NZ_CP017100.1
		Escherichia coli strain STEC 690	NZ_LOFJ00000000.1
		Escherichia coli strain 90	NZ_NSKO00000000.1
	III-A	Staphylococcus epidermidis strain 987 SEPI	NZ_JUOW00000000.1
		Staphylococcus epidermidis 125 SEPI	NZ_JVXA00000000.1
		Staphylococcus epidermidis strain ABKWR	NZ_LYLX00000000.1
		Staphylococcus epidermidis strain A1KF08	NZ_LYPX00000000.1
		Staphylococcus epidermidis strain Bt1p3	NZ_LYPY00000000.1
		Staphylococcus epidermidis AOKB02	NZ_LYPZ00000000.1
		Staphylococcus epidermidis KED12	NZ_LYQA00000000.1
		Staphylococcus epidermidis XE_C06	NZ_LYQB00000000.1
		Staphylococcus epidermidis ENVL_370	NZ_LYVZ00000000.1
		Staphylococcus epidermidis Mt1p16	NZ_MAJJ00000000.1
Class 2	II-A	Streptococcus thermophilus ND03	NC_017563.1
		Streptococcus thermophilus strain APC151	NZ_CP019935.1
		Streptococcus thermophilus KLDS3.1012	NZ_LHSK00000000.1
		Streptococcus thermophilus St-10	NZ_PHHG00000000.1
		Streptococcus thermophilus St-9	NZ_PHHH00000000.1
	V-A	Francisella novicida Fx1	NC_017450.1
		Francisella tularensis subsp. novicida GA99-3549	NZ_AAYF00000000.1
		Francisella tularensis subsp. novicida F6168	NZ_CP009353.1
		Francisella tularensis subsp. novicida strain AL97-2214	NZ_CP009653.1
		Francisella tularensis subsp. novicida strain DPG 3A-IS	NZ_CP010103.1
		Francisella tularensis subsp. novicida strain FAI	NZ_JOOT00000000.1

2.3. Web-based visualization tool

The web-based visualization tool is written in HTML5, CSS3, and jQuery and is locally served through port 4444 by SimpleHTTPServer in Python 2 or http.server in Python 3. User-generated changes to the data set can be saved via HTML5 local storage or exported to disc. Our custom multiple sequence alignment algorithm is implemented in JavaScript and is comprised of three distinct stages of execution. The primary stage uses the Needleman-Wunsch algorithm [32] to evaluate an exhaustive combination of global pairwise alignments, generating raw similarity scores for each pair of sequences from empirically selected match, mismatch, and gap penalty values: 4, −1, −2, respectively. These values are hard-coded into multipleSeqAlignment.js and can be modified as needed. The second stage uses UPGMA [33] to recursively generate a guide tree based on the similarity matrix. Branching order is determined by highest pairwise similarity score. If multiple maximum scores exist, one is chosen at random. Progressive alignment is implemented in stage three, based on the guide tree’s branching order and uses a sum-of-pairs scoring approach [34]. End gaps are not penalized and are removed from the user-interface to yield a more concise output – reduction in unnecessary gaps towards end of some sequences. While typical progressive alignments are concerned with comparison and gap insertion at the nucleotide level, we are concerned with comparison and gap insertion at the whole spacer level, enabling alignments of entire spacers as opposed to alignments of individual nucleotides. To this end, logic at the nucleotide level has been abstracted away by converting the nucleotide sequence of each spacer to an integer for speed of computational comparison, then by comparing integers as nucleotides are typically compared – resulting in a match, mismatch, or gap score. Nucleotide-to-gap (integer-to-gap) is scored as a typical gap (−2). The gap-to-gap score of −3 was determined empirically after observing the tendency for gap-to-gap alignments to be favoured and inserted in long stretches when its score was set to anything greater than or equal to the singular gap score of −2.

Additional visual manipulation is possible in the user-interface, including row and spacer sorting, reverse-complementation, and spacer/repeat view toggles. This software can be installed from the command-line and is also available as a Docker image. Command-line installation requires Bash, Java, Python, pip, Biopython, and Numpy. To download the pipeline and for additional detail on installation and use, please see the repository at (github.com/CRISPRlab/CRISPRviz).

2.4. Performance evaluation

Executions of 1, 10, and 100 genomes were normalized by using copies of a single genome: NC_015428. This genome was chosen based on its average size (~ 2.5 Mb) and familiarity. Each sample run was executed 5 times (n = 5). The machine used for execution is a 2016 MacBook Pro laptop with 8GB RAM and a 2.7 GHz Intel Core i5 processor. Subsequent runs on higher power processors yielded significantly shorter execution times, however, the conservative execution times were chosen for display as they are more likely to better represent the technology used by the majority of typical end users.

3. Results

3.1. Bioinformatic pipeline overview

CRISPR Visualizer is comprised of two main components – the extraction pipeline and conversion engine, and the user-facing web-based frontend. The extraction pipeline and conversion engine consist of a string searching algorithm that identifies CRISPR spacers and repeats through MinCED, as well as a conversion engine that converts sequence composition and length into RGB values and a corresponding symbol, respectively (Fig. 1A). The web-interface is comprised of a ‘File actions’ menu, a display area where rows of spacers and repeats are located, individual row action buttons, as well as a bank of action buttons that apply to the entire display area (Fig. 1B). The ‘File actions’ menu (Fig. 1C) includes saving and loading from local browser storage, import and export to disc in .json format, as well as a management feature which lists all saved browser-based entries (Fig. 1D). The row display area exhibits the spacers and repeats of a single genome per row or, can alternatively show an individual CRISPR locus per row when the pipeline is run with the split (-x) option. Each row and individual spacer are draggable – to enable fine tuning of the alignment through manual row rearrangement and gap manipulation. Adjacent to each row is a group of action buttons that apply specifically to the repeats and spacers of that row. This button collection allows the manual addition and removal of gaps (represented by an ‘x’ icon), row reversal, reverse-complementation, and row deletion (Fig. 1E). Beneath the row display area is a similar bank of buttons whose actions apply to all rows collectively. The first row of these buttons includes two sort features: by ascending and descending length, a direction reversal toggle, and a reverse-complement toggle. The second row includes a progressive multiple sequence alignment function for spacer alignment, and toggle buttons to show spacers, repeats, or spacers and repeats simultaneously (Fig. 2).

Figure 1.

Overview of bioinformatic pipeline. A) The pipeline extracts spacer and repeat sequences from .fasta files, which are converted into unique color/shape combinations representing the sequence composition and length of each unit. B) The web-interface contains a File actions menu, a row display area, and a bank of action buttons that control the current view of the display area. C) The File actions menu allows saving and loading to local browser storage, a local storage management feature, and import and export to disk. D) The ‘Manage’ menu, accessed from the File actions menu, displays items saved locally in browser storage. E) Each row contains actions buttons that can perform a specific task on that row: add or remove gaps, reverse the row direction, show each item’s reverse complement, and delete row.

Figure 2.

Action buttons that control the overall display area. The bottom of the web-interface contains a bank of action buttons that allow manipulation of the rows in the display area. The first row of buttons allows sorting rows by length, direction reversal, and displaying the reverse complement. The second row of buttons allows multiple sequence alignment and toggles to show spacers, repeats, or repeats and spacers together.

3.2. Reference type and subtype analysis

Reference and model strains and genomes, representing both Class 1 and Class 2 across the four major CRISPR-Cas system types and 4 model subtypes, were analyzed using the pipeline – encompassing 4 species and 27 strains in total: Escherichia coli as the Type I-E reference, Streptococcus thermophilus as the Type II-A reference, Staphylococcus epidermidis as the Type III-A reference, and Francisella novicida as the Type V-A model [31]. Importantly, these specific systems were selected because they have been defined as the canonical references for these subtypes [31], and some have been extensively used as models to unravel the molecular mechanisms of action of CRISPR-Cas systems (i.e. E. coli for Type I-E; S. thermophilus for Type II-A, S. epidermidis for Type III-A, and F. novicida for Type V-A).

Type I-E – Across 6 strains of Escherichia coli, including reference strain K12, only a single Type I-E system was found. 4 conserved spacer arrays were identified, as well as 2 unique spacer arrays – both belonging to NZ_NSKO00000000 (Fig. 3A). Repeats from NZ_BEVX00000000.1, NZ_LOFJ00000000.1, and NZ_NSKO00000000.1 display a modified terminal repeat shown in yellow, which is common and instrumental in determining locus orientation. Repeats directly adjacent to Type I-E cas genes are highly similar, while repeats that are located ~ 25 kb from the Type I-E cas genes show slight variation. These repeats likely evolved from a single ancestral array and were physically split over time.

Type II-A – The 5 selected strains of Streptococcus thermophilus yielded repeats and spacers from 3 distinct CRISPR-Cas systems, 2 distinct  Type II-A systems and a Type III-A system.

3 clusters of conserved spacer arrays were found, as well as 3 unique spacer arrays belonging to NZ_PHHG00000000.1, NZ_PHHH00000000.3, and NZ_PHHH00000000.2 (Fig. 3B), illustrating their distinct genotypic characteristics that may be reflective of their individual environments. There are 3 distinct groups of repeats, each in close proximity to a different CRISPR-Cas system. There is significant variation among these 3 repeat groups, indicating an independent evolutionary association with the most proximal CRISPR-Cas system. Repeats belonging to the Type II-A Locus 2 system and the Type III-A system each have a modified terminal repeat. The wide range of diversity in both spacer and array length could warrant additional inquiry into the level of CRISPR activity and the organism’s biological niche.

Figure 3.

Visualization and alignment of select model systems. A) Alignment of Type I-E model strain Escherichia coli K12 spacers reveals 4 conserved families of loci and 2 unique spacer arrays. Repeat alignment shows distinct repeat groups that share repeat sequence and length, and are split by position relative to cas genes. B) Type II-A model species Streptococcus thermophilus spacer analysis reveals a diverse range of length and sequence composition. Repeat alignment displays 3 distinct spacer and repeat groups. C) Alignment of Type III-A model species Staphylococcus epidermidis spacers displays 2 conserved families of loci and one unique array. Repeat analysis shows 2 distinct repeat groups. D) Alignment of Type V-A model strain Francisella cf. novicida Fx1 spacers yield 3 conserved locus families and 2 unique loci. Two individual repeat groups can be seen after alignment.

Type III-A – Analysis of 10 of Staphylococcus epidermidis genomes revealed a single Type III-A system. 2 spacer arrays were highly conserved while a single unique spacer array was identified in NZ_JUOW00000000.1 (Fig. 3C). Repeats are clustered into 2 highly conserved groups: the first of which is proximally associated with Type III-A cas genes, and a second group which is lacking associated cas genes. There is some variation in the repeat sequences between these two groups as well as physical distance, indicating initial evolution from a single ancestral array followed by point mutations over time. Almost all repeat arrays contain a single modified terminal repeat – providing an evident indication of locus orientation.

Type V-A – We analyzed 6 strains of Francisella novicida, revealing 2 distinct CRISPR-Cas systems: a Type V-A system and a Type II-B system. 3 highly conserved spacer arrays were identified, followed by 2 unique spacer arrays which were both found in the reference strain NC_017450 (Fig. 3D). 2 distinct repeat groups emerge when arranged by sequence variation, each belonging to a respective CRISPR-Cas system. All Type V-A repeat arrays contain a modified terminal repeat, while only 1 Type II-B repeat array contains a modified terminal repeat.

3.3. Performance evaluation

To measure execution time, the pipeline was run against 1, 10, and 100 genomes and elapsed seconds necessary to complete the extraction and conversion process were recorded (Fig. 4). Each batch of genomes (1, 10, or 100) was run through the pipeline a total of 5 times to assess replicative consistency. The average execution time for a single genome was 2.93 seconds, while the average time until completion for 10 genomes was 10.72 seconds. To test a large-scale batch execution, 100 genomes were processed in parallel – resulting in an average of 114.38 seconds before total completion.

Figure 4.

Execution time. To demonstrate the pipeline’s ability to scale, batches of 1, 10, and 100 genomes were analyzed and the seconds until completion were recorded.

4. Discussion and conclusion

Successful execution of the pipeline presented here is solely based on output generated from MinCED extraction. At times, repeated nucleotides in spacer sequences or poorly assembled genomes can cause the spacer/repeat boundary to be shifted, misrepresenting the spacer and repeat lengths in the web-interface. Spacers and repeats experiencing this issue can be manually corrected in the underlying extraction files, however, the issue can be avoided primarily by using high quality genomes.

Analysis of strains containing CRISPR arrays from reference subtypes I-E, II-A, III-A, and V-A resulted in readily identifiable conserved and distinct loci across subtype genomes. Recognition of closely related strains through CRISPR-based genotyping was facilitated through the application of multiple sequence alignment. This custom implementation of a progressive alignment algorithm is optimized to support genotypic identification through spacer and repeat analysis and is best suited for use against somewhat similar to highly similar spacer sequences. The current alignment model uses a single pass for tree building and pairwise alignment scoring, and could benefit from iterative refinement, as well as a weighted sum-of-pairs approach for improved tree building. Bacterial strains that share highly conserved spacer arrays can be located quickly and assessed for putative CRISPR activity in silico based on the presence of additional spacers, and thus acquisition events; compared to genetically similar strains. Orientation of the CRISPR locus can also be determined in cases where the terminal repeat is modified (it often times carries a mutation at its 3ʹ end), as acquisition is strictly directional and occurs only at one end of the locus. However, additional inspection of the relative locations of cas1 and cas2 genes may also be used for further verification [5,36]. Of course, when transcriptional data is available, orientation and boundaries of RNA transcripts are also critical and valuable in these analyses. The development of a robust orientation detection function could be extremely advantageous for future discovery, and RNA studies underway will play a critical role in shaping the next iteration of this pipeline. CRISPR repeat and spacer lengths typically cluster around a canonical standard based on their respective subtype. Visualizing this signature is particularly helpful when mining the burgeoning quantities of uncharacterized metagenomic samples for CRISPR loci, supporting analysis of currently known systems and promoting discovery of novel systems.

The brief execution time of a substantial number of genomes in parallel supports high-throughput analysis, however, visually evaluating such a considerable number of arrays in cases of distantly related organisms may not be especially advantageous – primarily when performing progressive alignment. In these cases, the spacer and repeat files generated by the pipeline serve as a practical starting point for deeper analysis of the individual locus or organism, typically represented in a text-based table format or SQL-based report generated from spreadsheet tools like Excel or created from standalone local databases like postgres, mysql, etc.

5. Conclusion

The availability of genomic sequencing data has increased at a staggering rate and the need for automated bioinformatic tools for high-power analysis has never been greater, especially with the rise of CRISPR. Current methods for identification, comparison, and visualization of CRISPR loci across various organisms are time consuming and typically lack support and capacity for high-throughput analysis. This pipeline facilitates swift implementation and provides the necessary framework for the rapid generation of large amounts of spacer and repeat data across many organisms in parallel, as well as subsequent comparison through visualization and alignment. It is our hope that this tool will provide a convenient and effective platform for rapid CRISPR analysis across academia and industry alike; across the nascent field of CRISPR biology and throughout the accomplished arena of genome engineering. The pipeline could serve as an epidemiological tool, facilitating tracking of microevolution in diverging pathogenic strains, as well as a genomic tool, providing ancestral clues to bolster genomic and phylogenetic reconstruction from metagenomic samples. As the importance and necessity of CRISPR-Cas-based genome engineering continues to grow, a proficient understanding of CRISPR loci and associated crRNA in particular is paramount, given the importance of their sequences, structures, and transcriptional profiles, as they guide CRISPR effector nucleases to provide immunity in their native host, and enable a variety of applications when engineered.

Funding Statement

The authors acknowledge support from NC State University and the NC Ag Foundation;

Acknowledgments

We would like to thank Phillippe Horvath and colleagues for the initial development and implementation of the Excel-based CRISPR Spacer Macro, Claudio Hidalgo Cantabrana and Katelyn Brandt for their time, testing, and input throughout the development of the pipeline, as well as Cassandra Cañez for her graphic design expertise. This research was funded by NCSU and the NC Ag Foundation.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed