Evaluation of Semiautomated IS6110-Based Restriction Fragment Length Polymorphism Typing for Mycobacterium tuberculosis in a High-Burden Setting

Abstract

The manual IS6110-based restriction fragment length polymorphism (RFLP) typing method is highly discriminatory; however, it is laborious and technically demanding, and data exchange remains a challenge. In an effort to improve IS6110-based RFLP to make it a faster format, DuPont Molecular Diagnostics recently introduced the IS6110-PvuII kit for semiautomated typing of Mycobacterium tuberculosis using the RiboPrinter microbial characterization system. This study aimed to evaluate the semiautomated RFLP typing against the standard manual method. A total of 112 isolates collected between 2013 and 2014 were included. All isolates were genotyped using manual and semiautomated RFLP typing methods. Clustering rates and discriminatory indexes were compared between methods. The overall performance of semiautomated RFLP compared to manual typing was excellent, with high discriminatory index (0.990 versus 0.995, respectively) and similar numbers of unique profiles (72 versus 74, respectively), numbers of clustered isolates (33 versus 31, respectively), cluster sizes (2 to 6 and 2 to 5 isolates, respectively), and clustering rates (21.9% and 17.1%, respectively). The semiautomated RFLP system is technically simple and significantly faster than the manual RFLP method (8 h versus 5 days). The analysis is fully automated and generates easily manageable databases of standardized fingerprints that can be easily exchanged between laboratories. Based on its high-throughput processing with minimal human effort, the semiautomated RFLP can be a very useful tool as a first-line method for routine typing of M. tuberculosis isolates, especially where Beijing strains are highly prevalent, followed by manual RFLP typing if resolution is not achieved, thereby saving time and labor.

INTRODUCTION

Genotyping of clinical isolates of Mycobacterium tuberculosis has improved our understanding of the epidemiology and patterns of transmission of tuberculosis (TB) (1, 2). Insertion sequence 6110 (IS6110)-based restriction fragment length polymorphism (RFLP) analysis is considered the current gold standard for M. tuberculosis genotyping, as it provides excellent discrimination (3, 4). IS6110-based RFLP typing has particular utility when applied to highly homogeneous strain groups, such as the Beijing family of M. tuberculosis. The Beijing family is apparently one of the most successful M. tuberculosis lineages, considered responsible for more than a quarter of the global TB epidemic (5). Many studies have reported that, unlike IS6110-RFLP, spoligotyping (6, 7) and MIRU-VNTR (mycobacterial interspersed repetitive-unit–variable-number tandem-repeat) typing (8, 9) have a tendency to overestimate clustering among clinical isolates of the Beijing family.

Despite its higher discriminatory power, several factors limit the usefulness of IS6110-RFLP genotyping for surveillance and outbreak management, especially in populations with high TB burden. IS6110-RFLP typing is time-consuming due to the requirement for culture and preparation of high-quality DNA, as well as being labor intensive and technically demanding (10, 11). Moreover, the resulting banding patterns (DNA fingerprints) are not standardized, as different laboratories use different reference standards to normalize IS6110 RFLP-generated gel images; thus, interlaboratory comparisons of the data remain a challenge (12, 13).

Although IS6110-RFLP typing is not amenable to full automation due to the requirement for M. tuberculosis culture, advances in instrumentation have enabled semiautomation of the RFLP process. The RiboPrinter system (DuPont Molecular Diagnostics, USA) recently introduced an IS6110-PvuII kit for M. tuberculosis typing. It is based on PvuII restriction endonuclease digestion of M. tuberculosis chromosomal DNA, followed by agarose gel separation and Southern blot hybridization, using probes for IS6110 and rRNA gene sequences. The semiautomated system uses sample preparations similar to those of manual RFLP; however, the typing process takes only 8 h. This approach addresses the limitations of IS6110-RFLP in terms of reduction in turnaround time and simplified technical performance, and the analysis is fully automated and generates easily manageable databases of standardized fingerprints that can be easily exchanged between laboratories.

The performance of assays based on RiboPrinter system has been validated for a number of pathogens, including Listeria (14), Salmonella (15), Staphylococcus spp., Escherichia coli, (16) and other bacteria (17). However, to date, no study has been undertaken to compare the assay with regard to its discriminative power, reproducibility, and turnaround time for IS6110-based typing of M. tuberculosis. In this study, we evaluated the performance of semiautomated IS6110-based RFLP (semiautomated RFLP) typing of M. tuberculosis clinical isolates against the standard manual IS6110-RFLP (manual RFLP) typing method.

MATERIALS AND METHODS

Study samples.

The validation of semiautomated RFLP typing was performed using 92 available M. tuberculosis strains collected as part of surveillance carried out from 2013 to 2014 at the Centre for Tuberculosis (CTB) in South Africa. In addition, 20 clinical isolates previously characterized by manual RFLP typing were included for verification. Genomic DNA was extracted from each of the M. tuberculosis isolates using the phenol-chloroform (CTAB) method as described previously (3). This study was approved by the Human Research Ethics Committee of the University of Pretoria.

Manual typing.

Manual typing was performed using the standardized IS6110-based RFLP technique as described previously (3). Briefly, genomic DNA was extracted using the CTAB method and was digested with PvuII, subjected to agarose gel electrophoresis, and transferred to a Hybond membrane, which was then hybridized with the IS6110 probe prepared using the enhanced chemiluminescence (ECL) assay kit (Amersham Biosciences, USA). Fingerprints were analyzed using BioNumerics (version 7.1; Applied Maths, Belgium) software.

Semiautomated typing.

Semiautomated RFLP typing was performed using the RiboPrinter microbial characterization system (DuPont Molecular Diagnostics, USA), with a customized IS6110 kit incorporating the PvuII restriction endonuclease. Genomic DNA was prepared from M. tuberculosis isolates using the CTAB method as described above. Samples were processed and analyzed according to the protocol described previously (18). Briefly, DNA was transferred in 30-μl aliquots to a sample carrier, and IS6110-PvuII consumables were loaded into the RiboPrinter system characterization unit. Within the instrument, bacterial DNA was digested with PvuII restriction enzyme and loaded into a 1% agarose gel containing 13 wells, five of which contained modified molecular size standard markers; fragments were separated by electrophoresis, transferred to nylon membrane, and hybridized with a prelabeled chemiluminescent probe. The probe used by the RiboPrinter system incorporates both IS6110 and M. tuberculosis rRNA gene sequences. As a consequence, additional PvuII endonuclease fragment(s) specific to the ribosomal genes will be present in the RiboPrinter system IS6110 banding patterns, providing additional information. For each DNA sample, the resulting pattern of positively hybridized bands was then converted to digital information by a charge-coupled-device camera and stored in the RiboPrinter database. Each piece of sample data was normalized to a common set of standard markers included in the adjacent well. Similarity coefficients were calculated based upon both band position and relative intensity. Strains were considered to have the same pattern if the similarity coefficient between their patterns was ≥0.93 and assigned to a specific ribogroup.

Reproducibility of semiautomated method.

Reproducibility of the semiautomated system was assessed by repeat testing of 22 strains with different banding patterns.

Data analyses.

The gel images produced by manual RFLP typing were imported into BioNumerics software (version 7.1; Applied Maths, Belgium), and the images were normalized by a standard reference (MTB14323) (3). Clusters were defined as two or more M. tuberculosis isolates with identical IS6110-RFLP patterns from different patients.

The semiautomated typing patterns (ribopattern) were analyzed using the on-board RiboPrinter software. When required, further examination of the automated ribogrouping was performed by visual evaluation of ribopatterns.

The level of similarity between DNA fingerprints of manual RFLP was estimated using the Dice coefficient, and cluster analysis of fingerprints was performed with the unweighted-pair group method using arithmetic averages (UPGMA). Both the optimization and position tolerance were set at 1% for all pairwise and cluster analyses. Simpson's index of diversity was used to evaluate the discriminatory ability of the two typing methods. The Simpson's index of diversity was calculated as described by Hunter and Gaston (19).

RESULTS

The study evaluated the discriminative power, reproducibility, and turnaround time of semiautomated RFLP against the standard manual RFLP typing. A total of 112 clinical M. tuberculosis strains were used. Comparative results of manual and semiautomated RFLP typing were available for 105 isolates, including 92 isolates from the validation and 13 of the 20 isolates used for verification. The remaining seven isolates did not yield sufficient growth for DNA extraction and were excluded.

The number of bands for both manual and semiautomated RFLP typing per strain ranged from 1 to 19. Of the 105 M. tuberculosis isolates, a total of 13 clusters (31 isolates, 29.5%) were identified by manual typing, the largest containing 5 isolates, and 74 isolates (70.5%) were unique, giving a strain-clustering rate of 17.1% (Table 1). In comparison, 10 clusters, comprising 33 (31.5%) isolates, and 72 (68.5%) unique patterns were identified by semiautomated RFLP typing, giving a clustering rate of 21.9% (Table 1). The discriminatory powers of semiautomated RFLP and manual RFLP, as shown by Simpson index, were close (0.990 and 0.995, respectively) (Table 1). A dendrogram for the manual RFLP method is shown in Fig. 1.

TABLE 1.

Discriminatory power of IS6110-RFLP typing compared to RiboPrinter system

RFLP method	n	No. of:				Range of cluster size	Simpson's diversity index (D)	Clustering rate (%)
		Types	Unique isolates	Clustered isolates	Clusters
Manual	105	84	73	31	13	2–5	0.995	17.1
Semiautomated	105	82	72	33	10	2–6	0.990	21.9

FIG 1.

Manual RFLP dendrogram of 105 isolates. The dendrogram was constructed using the Dice similarity coefficient and the UPGMA algorithm with the software BioNumerics v 7.1. Isolates in boxes are isolates with identical RFLP patterns.

Manual RFLP clusters were designated C1 to C13 (Fig. 1), while semiautomated RFLP clusters were designated R1 to R10. The summary of the ribogroups is shown in Table 2. The banding patterns of isolates clustered by manual RFLP and semiautomated RFLP were compared. Three clusters identified by manual RFLP (C2, C8, and C13) shared fully identical patterns by semiautomated RFLP typing (R1, R4, and R7). Isolates in semiautomated cluster R6 were subdivided into two subclusters (C11 and C12) with identical manual RFLP patterns. Similarly, cluster R5 was subdivided into two subclusters (C9 and C10) and included one more isolate that was split off by manual RFLP typing. In clusters C3 and C7, the semiautomated RFLP system included isolates (R2 and R3) whose fingerprints differed from the fingerprint of the clustered isolates by the presence of one or two additional bands (Fig. 1 and Table 2).

TABLE 2.

Isolates grouped into ribogroups (clusters) by the semiautomated system

Cluster	Isolate
R1	10132
	64993
R2	TT179
	38616
	62903
	57671
	51998
R3	1277
	47763
	49656
	41018
	65920
R4	36775
	22753
R5	M223
	23642
	83820
	86755
	98605
	27406
R6	37003
	53267
	44786
	51187
R7	31160
	R987
R8	21980
	40091
R9	39162
	65048
R10	R182
	R910
	R550

Four clusters defined by manual RFLP (C1, C4, C5, and C6), comprising 2 isolates each, were not clustered by the semiautomated system due to the low-intensity bands. The semiautomated software considers intensity of the bands, in addition to molecular weight, when isolate comparisons are performed. Additional visual comparison of the gel images of manual RFLP and semiautomated RFLP typing confirmed identical patterns of the isolates in each of the four clusters. Conversely, three semiautomated RFLP clusters (R8, R9, and R10) showing differences by one or two bands were not clustered in manual RFLP typing.

Additional visual comparison of the gel images of the two methods was performed. The two methods provided identical fingerprinting patterns for 84/105 (80%) isolates. In 21/105 (20%) isolates, closely located bands were not well resolved by the semiautomated system, in which two fragments often appeared as one.

In the semiautomated system, gel separation and Southern transfer occur simultaneously, which reduces the turnaround time significantly. After sample preparation, the turnaround times for manual and semiautomated RFLP typing were approximately 4 days and 8 h, respectively. However, gel analysis is required for interpretation of the manual RFLP results, which can lengthen turnaround for final results by 1 day. The turnaround time of the semiautomated compared to the manual method is summarized in a flow diagram (Fig. 2).

FIG 2.

Workflow diagram of manual and semiautomated RFLP methods. Mtb, M. tuberculosis.

Reproducibility of semiautomated system.

All duplicate samples included were consistently reproducible, with only one pair of duplicate samples differing by the absence of a faint band. Figure 3 shows the banding patterns of the duplicate samples obtained by the semiautomated method.

FIG 3.

Reproducibility of the semiautomated system. Reproducibility was performed by repeat testing of 22 strains with different banding patterns.

DISCUSSION

A variety of efforts are being made to adapt molecular typing of M. tuberculosis to a faster format. This study compared the semiautomated strain typing of M. tuberculosis based on the IS6110 sequences against the standard manual RFLP typing. The overall performance of semiautomated compared to manual RFLP typing was excellent, with a high discriminatory index (0.990 versus 0.995), similar numbers of unique profiles (72 versus 74, respectively), numbers of clustered isolates (33 versus 31, respectively), and cluster sizes (2 to 6 and 2 to 5, respectively), and similar clustering rates (21.9% versus 17.1%, respectively).

Semiautomated RFLP has shown excellent capability for determining patterns, especially for isolates with fewer bands. A small proportion (20%) of closely located fragments, however, were poorly resolved by the semiautomated RFLP typing. This was due to the size of the gel, duration of the run, and electrophoresis conditions which reduce the quality of the band separation and discrimination. Unfortunately, these parameters cannot be changed in the automated system. Patterns with very large numbers of DNA fragments are better resolved by the manual RFLP method rather than the semiautomated system, as the electrophoresis conditions are optimized with larger gels and a longer duration of electrophoresis (overnight). It should be noted that the use of the semiautomated system to compare strains would not result in any clusters being missed. However, for strains with large numbers of bands, overclustering might occur, which can be resolved by visual examination or further testing when required.

The semiautomated software uses similarity cutoffs of 93% to determine clusters, compared to 98 to 100% similarity for manual RFLP typing. As the M. tuberculosis genome is highly conserved compared to other bacteria, the 93% cutoff might not provide sufficient discrimination for epidemiologic investigations. Some of the isolates in R9 and R10 with 1 to 2 band differences were grouped in the same cluster. It should be noted that we used identical patterns to define clusters for the manual RFLP method in this study, although we took into account the variation in exact band position between separate manual RFLP runs. If 98% similarity had been used (±1 band difference), the 2 semiautomated clusters (R9 and R10) would have clustered by the manual method. Therefore, the software should categorize clusters with a similarity index of 98% and higher into the same cluster. This modification would enhance the performance of this instrument when typing M. tuberculosis for epidemiologic purposes.

Due to the low intensity of some bands, four clusters defined by manual RFLP were not clustered by the semiautomated system. The semiautomated system analyzes the band pattern based on position and intensity, whereas for manual RFLP, the way to analyze the bands is chosen by the operator. Genetic similarity is based on band position and the distance between bands and not on band intensity. The parameters of analysis used by the semiautomated software can lead to some misclassification of isolates. In such cases, visual evaluation of the gel images can be made to verify the results.

The semiautomated RFLP system is technically simple and significantly faster for genotyping than the manual method, reducing the turnaround time by 4 days while requiring far less hands-on manipulation. It can automatically process up to eight isolates at one time, with results available about 8 h from sample input. The instrument can accept new sample batches every 2 h, and 32 samples can be completed in 8 h. With manual typing, 5 days are needed to complete the analysis of 24 strains. In addition, there is no gel analysis required by the semiautomated RFLP system; acquisition and normalization of ribopatterns are performed automatically using the on-board software of the instrument. Identical sample patterns are automatically clustered into ribogroups, allowing the user to instantly view previous occurrences of any particular pattern. It can be a very valuable tool as a first-line choice for typing in epidemiological surveillance, followed by manual RFLP typing as the occasion demands to save labor and time.

Comparison of agarose gel results between separate runs is not always straightforward. Differences between fragments and exact base pair size variations can be difficult to detect due to variations in fragment migration and gel resolution. However, the semiautomated system has a standardized methodology and data analysis, and it was found to be highly reproducible in this study and gave excellent comparisons between runs.

The disadvantages of the semiautomated system include the high cost of initial instrument setup and the ongoing cost of maintenance. Placing the instrument in a centralized facility may be cost-effective, as the RiboPrinter system can be used for diverse applications. The price of consumables for the semiautomated RFLP system and manual RFLP typing is $45 and $30, respectively. Nevertheless, taking into account the reduced labor time using the semiautomated system would offset the additional consumable costs.

Conclusions.

In this study, the semiautomated RFLP has shown good performance and offers many advantages with respect to standardization, speed, and technical simplicity. With improvement to the instrument software, the assay can be a simple and robust first-line method for routine typing of M. tuberculosis isolates, especially where Beijing strains are highly prevalent. If more discrimination is needed, manual IS6110 RFLP typing can be used. Further studies are needed in different geographical settings.

Funding Statement

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sector.