Culture-Independent Detection and Identification of Leptospira Serovars

For most species, the optimized 11108-assay had a detection limit of 10 GEs with expected T_m regardless of diluent (i.e., PBS or NHS). However, to achieve a similar level of detection with borgpetersenii, and santarosai-specific primer pairs, extended runs (>40 to 45 cycles) were required. Under optimized assay conditions, the detection of the borgpetersenii, and santarosai-specific primers was 10² to 10³ GEs.

As summarized in Table 1, CFX96-generated T_m’s ranged from 80°C to 85°C. Negative controls and NTCs were invariably negative, confirming specific detection of group I pathogenic Leptospira. Intra- and inter-run T_m’s were remarkably consistent (σ = 0.0°C) among replicates, with T_m’s defined for each allele (Table 1). At the manufacturer-recommended resolution of 0.2°C, nine interrogans/kirschneri sequence variants or alleles produced only four T_m’s, ranging from 80.2 to 80.8°C, though all could be differentiated at 0.1°C-resolution (not shown). Nonetheless, several important serovars could be distinguished based on 11108-amplicon derived T_m’s, including svCopenhageni/svIcterohaemorrhagiae (preferred hosts = peri-domiciliary rats), svLai (field mice), svHardjo-prajitno (cattle), and svCanicola/svPomona (dogs and pigs, respectively). T_m’s were species-dependent, as expected based upon species-variable GC%. With interrogans, kirschneri, and noguchii alleles producing T_m’s ranging from 8°C to 81°C; borgpetersenii, 81.°C to 82.5°C; weilii, 83°C; and santarosai, >84°C (Table 1; Fig. S2); joint inference of species and serotype could be made. Therefore, interrogans svHardjo subtype hardjoprajitno and borgpetersenii svHardjo bovis could be readily distinguished.

We also analyzed reference strains obtained from Peru and Sri Lanka, where we have conducted prospective field studies. All strains originating from Sri Lanka (svGeyaweera/svWeersinghe, svRatnapura, svCeylonica, and svAlice) could be reliably differentiated from each other (80.6°C, 80.4°C, 81.6°C, 85°C), and from svCopenhageni/svIcterohaemorrhagiae (80.2°C). In contrast, Peruvian reference strains originating from outside Iquitos, specifically those belonging to L. santarosai, produced inconsistent melt peaks, precluding confident T_m assignment, which is likely due to reduced primer annealing and, thus, amplification efficiency, from these (as-yet) unknown sequence variants, suggesting that alternative forward and/or reverse primers might be needed (as was the case for some interrogans [e.g., svManilae] and noguchii strains).

Owing to a different design, MIC-derived T_m’s were on average ~1°C higher than those produced by the CFX96 but were also reliable (σ ≤ 0.05). Statistical comparison of the means showed that the superior precision of the MIC improved allele resolution (up to 100 potential distinct T_m’s at 0.05°C resolution, and 200 at 0.025°C resolution). Accordingly, previously unresolved (but distinct) alleles in svBataviae, svCopenhageni/svIcterohaemorrhagiae, svCynopteri, svHardjo-prajitno/svWolffi, svMankarso, svAustralis/svBratislava/svDjasiman/svGrippotyphosa/svPyrogenes, svAutumnalis, svCanicola/svPomona, and svManilae were assigned distinctive T_m’s (Table 1). Overall, all nine interrogans/kirschneri sequence variants could be reliably distinguished based upon mean T_m’s (P < 0.05).

During the 5-year-period of December 2002 to June 2007, 144 Leptospira strains were isolated from humans (50), peri-domiciliary rats (34), livestock (32), and wild animals (28) in Iquitos and surrounding areas. The majority (65.3%) were either interrogans, kirschneri, or noguchii. Fifty (34.7%) were identified as santarosai, with 27 (or 54%) derived from livestock (i.e., buffaloes, cattle, and pigs) representing 84.4% of all isolates obtained from these sources. In contrast, only 14 human isolates were classified as santarosai (28%), as were six from wild animals (21.4%), two from rats (5.9%), and none from bats or dogs. To test the reliability and accuracy of our optimized 11108-based NAAT, we analyzed a sample of these isolates, inclusive of all species and Serogroups found in the region. Twenty-three were taxonomically classified as interrogans, 13 as santarosai, seven as L. noguchii, and two as L. kirschneri. Table 2 summarizes the 11108 qPCR-assay results of all human and rat isolates analyzed, each producing a reliable T_m.

Based on CAAT (and/or PFGE-fingerprinting), 23 interrogans strains belonged to either serogroup Icterohaemorrhagiae (with all but two definitively typed as svCopenhageni) or Canicola (svCanicola, and two closely related but seemingly distinct strains, BEL039 and CBC1203R). All produced consistent T_m’s which were verified by BLAST. Thus far, only two strains isolated in Iquitos, HAI1536 and MOR069, have been shown to have similar 11108 amplicons but different CAAT or PFGE fingerprints among the 19 analyzed that were classified as interrogans/kirschneri/noguchii. Similarly, isolate MMD4803—presumed to be distinct based on PFGE fingerprinting—could not be resolved from either of two HAI1378-derived isolates, nor CBC621. Overall, of the 33 isolates analyzed, only two produced 11108-inferred IDs that disagreed with CAAT- or PFGE-based assignments, indicating correct inferences approximately 94% of the time, regardless of isolation date or source (human, clinical versus mammalian reservoir). Species inferences were reliable, and the assay correctly differentiated alleles/sequence variants 100% of the time. As is the case for most “exotic” serovars, sequences were not publicly available for Bataviae strains, VAR132 (80.4°C), MOR069 (80.6°C), nor Shermani strains, HAI1378(U) and CBC621 (84.8°C), precluding definitive serovar assignment. However, these were all confirmed via BLAST best-hits to be distinct and derived from 11108-amplicon variants, and based upon T_m-defined species-ranges, grouped with other serotypes from the expected species, i.e., noguchii and santarosai, respectively. Taken together these findings indicate that the 11108-assay is highly reliable and capable of producing consistent results with strains isolated several years apart and originating from various hosts. Therefore, although T_m values might vary based upon hardware/software options utilized, our approach consistently distinguished sequence variants.

Pilot experiments using archived culture-positive patient serum-samples from Sri Lanka (12 isolates, sample age range = 3 to 5 years old) and Iquitos (25, sample age range 11 to 15 years old) produced mostly consistent results. Ten of 12 recently archived serum-samples were 11108 qPCR positive (~83% sensitivity). Eight produced consistent amplification (i.e., were positive ≥67% of the time) across replicates and repeat runs (minimum of nine replicates), and two amplified less consistently, being positive <67% of the time. That said, all produced expected T_m’s with respect to their cognate isolate. Fig. 1 shows melt-curves from six of these archived samples. Amplification of older samples (i.e., 11 to 15 years old) was less reliable, although all were positive with amplification evident in at least one replicate in all runs (i.e., ~33%). And regrettably, most produced indefinite or quasi-stable T_m’s (±0.1°C), though some, e.g., serum-sample PAD451 (81.0°C), produced reliable T_m’s agreeing with CAAT-based classification of its cognate isolate, indicating that further refinement of our assay could allow retrospective analysis of bio banked serum-samples.

Forty-four dog urine samples were initially classified as culture negative, leading to the presumptive identification of 11 false-positive samples (culture negatives that were positive by at least one real-time qPCR assay). Because culture is inefficient (<10% success rate, in our hands), these presumed false positives were verified to contain Leptospira DNA by shotgun metagenomics sequencing (SMS). All were positive by 11108-based real-time qPCR, of which four were also positive by lipL32 qPCR (not shown). Partial Leptospira genomes were reconstructed from all but two, which produced mostly low-quality reads. Of the initial 84 samples, complete data were produced for 80: 47 verified positives and 33 negatives. Assay results were used to construct a contingency table to assess sensitivity and specificity (Table 3). The sensitivity (true positives detected) was clearly superior to the lipL32-based assay used for comparison while maintaining 100% specificity.

For real-time qPCR-based serovar identification (ID), genes comprising the core Leptospira genome were first defined using the Pan-Genome Ortholog Clustering Tool, PanOCT (31), and those predicted to encode proteins participating in LPS assembly or export, e.g., O-antigen polymerases (rfbY), O-antigen/polysaccharide flippases, rfbX and LIC_11108, and the Leptospira lipopolysaccharide export system (Lpt), including msbA, lptA, and lptD, were used to construct exhaustive amino-acid-guided multiple sequence alignments (MSAs) of coding sequences (CDS) retrieved from GenBank—anticipating that these would be serogroup- and (potentially) serovar-dependent, due to their stereospecificity (32). LIC_11108 was chosen for initial evaluation based on several criteria (the most notable summarized in the preceding paragraph). From these MSAs, hypervariable regions were identified and targeted for amplification by primers complementary to conserved flanking sequence (Fig. S1).

As none of these flanking regions were 100% conserved, degenerate primers 11108f: 5′–TTR AAN GAR AGT ATA AAA CTT CC–3′, and 11108r: 5′–AAC GGR CTY TTT CAA TAY GC–3′ (Table 4)—specific for group I pathogenic Leptospira—were synthesized. Pilot experiments were conducted using 11108f and 11108r to confirm amplification of globally distributed Leptospira species: L. interrogans, L. kirschneri, L. noguchii, L. borgpetersenii, L. weilii, and L. santarosai. Amplification of L. mayottensis, which has a more restricted distribution, was assessed via in silico PCR of available genome sequences. Baseline assay conditions were established using degenerate primers, following which optimal conditions were determined using deconvoluted species-specific primers (Table 4). Cognizant of our intended longer-term development goals, we opted for an affordable SYB green-based NAAT that could be easily customized for probe-based assays (should the need arise) and evaluated its performance on affordable hardware/software combinations, including a four-channel magnetic induction thermal cycler (MIC) manufactured by Bio Molecular Systems (~40% cost of CFX96).

Genomic DNA (gDNA) for qPCR (and sequencing) was prepared from interrogans svLai, santarosai svShermani, and borgpetersenii svTarassovi using the manufacturer-recommended extraction protocol for Gram-negative bacteria (GeneJET DNA purification kit, Thermo Fisher Scientific) and then amplified on a CFX96 thermal cycler (Bio-Rad). Amplicons were cloned into pCR-2.1 TOPO to create species-specific reference plasmids, which were transformed into E. coli following manufacturer’s directions (TA Cloning Kit, Invitrogen). After routine minipreps (GeneJET Plasmid DNA Miniprep Kit, Thermo Fisher Scientific), purified (reference) plasmids were quantified (individually) by fluorometry (Qubit, Thermo Fisher Scientific), and their copy number estimated. This was followed by serial dilution in either PBS buffer or heat-inactivated normal human serum (NHS); re-extraction was carried out using the QIAamp DNA Blood minikit (Qiagen) per manufacturer-recommended protocol for serum, and used as template for qPCR.

As 11108 genes occur once per Leptospira genome, copy numbers could be expressed as genome equivalents (GEs). The detection limit was defined as minimum number of GEs producing the expected T_m’s among all technical replicates. Reference plasmids were also used as template in amplifications evaluating the propensity of deconvoluted primer pairs for cross-species amplification.

Reaction mixes consisted of the following: 10 μL of Precision Melt Supermix, 2 μL of template gDNA, and 4 μL of both forward and reverse primers, yielding final (empirically optimized) primer concentrations of 0.4 μM per 20 μL reaction. The finalized thermal cycling profile consisted of a 3-min hold at 95°C (denaturation), followed by 44 cycles of 95°C for 10 s, 62°C for 30 s (empirically optimized annealing temperature) and 72°C for 1 min (extension), and a final hold at 72°C for 7 min to ensure full extension of PCR product. Amplification was followed by melt-curve peak identification (via Precision Melt Analysis Software) with fluorescence measured at 0.2°C increments (default). All samples were run in triplicate, and runs were repeated at least three times to produce a minimum of nine data points/per sample. Cq values of ≤ 40 cycles were considered positive, but postcutoff amplification (i.e., Cq > 40) with appropriate melt peaks were recorded, and later verified by sequencing. Milli Q water was run in parallel, in triplicate, as a no template control (NTC). gDNA from L. licerasiae Varillal svVarillal (archetype of group II, comprising low virulence Leptospira species), the genetically related L. wolffii svKhorat, and premixed gDNA from Gram-negative and Gram-positive bacteria, Citrobacter rodentium (ATCC DBS 100), Escherichia coli (EPEC E2348/69), Clostridium difficile (ATCC BAA-9689), and Staphylococcus aureus (MARSA, USA 300), were used as negative controls (to assess nonspecific “off-target” amplification). For each serovar, a mean T_m was calculated after first identifying and then removing outlier data points via robust regression and outlier removal (ROUT) (33) analysis (Q = 10%, Prism v7.0a). Sequence-variant specific T_m’s were defined via comparison of sample-means by two-tailed, nonparametric t test (P < 0.05, Prism v7.0a).

qPCR results were verified by conventional Sanger sequencing of cloned amplification products. Twenty-five randomly chosen colonies were subcultured into LB broth, and plasmids were isolated using a QIAprep Spin Miniprep Kit (Qiagen) and then sent for sequencing. Consensus sequences were aligned using the MAFFT multiple sequence alignment (MSA) algorithm with default parameter settings. Individual consensus sequences were submitted to the Basic Local Alignment Search Tool (BLAST) server to confirm specificity (34). For serovars consistently yielding inexact or a quasi-stable T_m’s, we sequenced 10 randomly selected clones producing six reads per clone (for a total of 60), and then compared clone-specific consensus sequences to identify variants.

Reaction mixes and thermal cycling profile were as described above, however, postamplification melt-curve analysis absorbance readings were taken at increments of 0.025°C, exploiting the superior precision of MIC. Negative controls and NTCs were run in triplicate, in parallel, as described above. And as before, outliers were removed by ROUT analysis before comparison of serovar-associated T_m by two-tailed, nonparametric t test. Data were visualized via a box-and-whisker plot.

To assess assay reliability, amplifications were done using different hardware/software combinations: (i) CFX96 with melt-curve analysis via Precision Melt Software; (ii) CFX96 with default melt-curve analysis (i.e., without Precision Melt Analysis Software); and (iii) MIC with bundled melt-curve analysis software.

To test accuracy (and thus, its potential epidemiological utility), we analyzed 45 Leptospira isolates obtained over a 5-year period from febrile patients, domesticated animals, and rats in Iquitos, Peru (Table S2), and 25 isolates from hospitalized patients in Sri Lanka, assessing agreement among T_m’s, PFGE- and CAAT-derived serovar assignments. Outlier removal, statistical comparisons, and data visualization were as described above. Discrepancies among T_m’s, PFGE and CAAT-based serotyping were resolved by Sanger-sequencing (of amplified products).

Assay sensitivity and specificity were determined using 84 canine urine samples, which comprised 40 culture positives and 44 randomly selected culture negatives and were sourced from a study of canine leptospirosis in Sokoto, Nigeria (35), as well as 12 culture-positive acute-phase serum-samples from a prospective study of human leptospirosis in Sri Lanka (36). Because Leptospira species and serovars in Sokoto and study sites in Sri Lanka are not well-characterized, all samples were amplified using premixed, species-appropriate primers. Shotgun metagenomics sequencing (SMS) data were generated from qPCR-positive, culture-negative samples to identify potential false-positives.

For comparison, dog urine samples were also assayed for the presence of leptospiral DNA by lipL32 qPCR using the following forward and reverse primers: lipL32_178f (5′–TCT GTG ATC AAC TAT TAC GGA TAC–3′) and lipL32_419r (5′–ATC CAA GTA TCA AAC CAA TGT GG–3′), specific for virulent group I pathogenic species. The (previously) empirically optimized lipL32-based thermal cycling profile consisted of the following: 3 min denaturation at 95°C, 45 amplification cycles (95°C for 10 s/62°C for 30 s/72°C for 60 s), 7 min extension at 72°C followed by postamplification melt-peak identification. For qPCR, gDNA from svManilae was used as a positive control and as before, premixed gDNA from four bacterial species was used as a negative control.