International Collaborative Study To Compare Reverse Transcriptase PCR Assays for Detection and Genotyping of Noroviruses

Abstract

To allow more rapid and internationally standardized assessment of the spread of noroviruses (previously called Norwalk-like viruses [NLVs]) as important food-borne pathogens, harmonization of methods for their detection is needed. Diagnosis of NLVs in clinical diagnostic laboratories is usually performed by reverse transciptase PCR (RT-PCR) assays. In the present study, the performance of five different RT-PCR assays for the detection of NLVs was evaluated in an international collaborative study by five laboratories in five countries with a coded panel of 91 fecal specimens. The assays were tested for their sensitivity, detection limit, and ease of standardization. In total, NLVs could be detected by at least one RT-PCR assay in 69 (84%) of the samples that originally tested positive. Sensitivity ranged from 52 to 73% overall and from 54 to 100% and 58 to 85% for genogroup I and II viruses, respectively. In all, 64% of the false-negative results were obtained with a set of diluted stools (n = 20) that may have lost quality upon storage. Sensitivity was improved when these samples were excluded from analysis. No one single assay stood out as the best, although the p1 assay demonstrated the most satisfactory overall performance. To promote comparability of data, this assay will be recommended for newly starting groups in future collaborative studies.

Noroviruses (previously called Norwalk-like viruses [NLVs]) have emerged as the single most common cause of outbreaks as well as sporadic cases of acute gastroenteritis in children and adults (13, 20, 55). Epidemiological investigations of outbreaks have shown that the most important modes of transmission are person-to-person contacts and contaminated food and water (20, 31). Multistate and global outbreaks linked to the distribution of contaminated foods and water have been reported (4, 6, 14, 45). Moreover, the emergence of a common strain with global distribution has raised important questions on the likelihood of massive food-borne introduction of new NLV strains (43, 51).

Since human NLVs have not yet been cultivated in vitro, detection has traditionally relied on electron microscopy (EM) or immune-EM (11, 25). Although several reference centers still utilize EM for the routine investigation of outbreak samples, the successful cloning and sequencing of complete and partial NLV genomes has allowed the development of more sensitive reverse transciptase PCR (RT-PCR) methods. First-generation RT-PCR assays that used primers based on the prototype Norwalk virus performed poorly since the genetic diversity of NLV strains was much greater than anticipated initially and therefore the primers selected needed to be optimized (12, 26, 41, 44). Using sequence information of an increasing number of NLV strains, several research groups successfully developed RT-PCR assays based on improved primers targeting a conserved region of open reading frame 1 (ORF1) coding for the viral RNA polymerase (POL region) (1, 17, 21, 30, 34, 40, 50, 56). Subsequently, these assays have been used successfully in epidemiological studies for the diagnosis of NLVs in fecal specimens from both outbreaks and sporadic cases (1, 7, 13, 17, 37, 40, 46, 50, 51). In addition, primers directed to other regions of the NLV genome have been developed, including the 2C helicase, the 5′ end of the capsid region, and ORF3 (22, 38, 42, 52, 54, 56, 57). In general, however, assays based on these regions are less broadly reactive. With few exceptions (46), most assays for the detection of NLVs in stool samples are based on a single-round format, whereas for detection of low copy numbers in environmental samples (hemi)nested RT-PCR assays have successfully been developed and applied (23, 48).

Because NLVs cannot yet be cultured in vitro, antigenic typing has been limited to typing by solid-phase immune-EM and enzyme-linked immunosorbent assays based on recombinant capsid proteins (28, 35). Recently, the National Institute for Public Health and the Environment (RIVM; Bilthoven, The Netherlands) and the Centers for Disease Control proposed similar NLV genotyping schemes based on the diversity in the entire capsid gene (2, 33). Correlation between genotype and antigenic type was at least found for strains from nine genotypes (19, 52). The current consensus is that the NLVs commonly found in humans can genetically be divided into two genogroups (GI and GII), which further segregate into genotypes. At least 15 genotypes exist with an arbitrary cutoff of a 20% amino acid difference across capsid sequences (19, 53). In addition, two new NLV genogroups have been proposed: NLVs detected in fecal specimens of calves with Jena virus as the prototype virus (GIII [2, 36]) and genogroup GIV, which consists of human strains that cluster equidistantly with both GI and GII viruses, with Alphatron virus as the prototype virus (33, 53). For the majority of NLV strains tested in a panel of 31 strains, there was agreement between clustering of strains based on the capsid gene and a small region of the POL gene used for diagnosis (52). Consequently, the current used POL-based assays can provide indications of the level of diversity in epidemiological studies (1, 15, 37, 40, 46, 51). For comparability and exchange of data, it is important to agree on a specific region for detection and subsequent typing.

For interpretation of RT-PCR, methods other than gel electrophoresis are essential to prevent false-positive results. Hybridization and DNA sequencing of the RT-PCR products are commonly used methods for confirmation of RT-PCR products (1, 17, 21, 34, 50). Recently, NLV genotyping methods for high throughput have been described that target an overlapping region in the POL region used for diagnosis (5, 39, 40, 53). The lack of international harmonization of detection and genotyping methods for food- and water-borne viral infections, including NLV, has precluded full use of the molecular information for tracking of outbreaks across borders and for elucidation of the major transmission routes. Therefore, we initiated a collaborative research project funded by the European Union (EU) to allow more rapid and internationally harmonized assessment of the spread of food-borne viral pathogens (QLK1-1999-00594 [EU 5th Framework Program]). To understand how data from different regions can be aggregated, we compared five different RT-PCR assays for detection of NLVs with a panel of coded stool samples.

MATERIALS AND METHODS

European consortium.

The consortium consists of 12 partner laboratories in nine countries (p1 [RIVM], p2 [University Central Hospital, Helsinki, Finland], p3 [Staten Serum Institute, Copenhagen, Denmark], p4 [Swedish Institute for Infectious Disease Control, Solna, Sweden], p5 [Central Public Health Service, London, United Kingdom], p6 [Robert Koch Institut, Berlin, Germany], p7 [Instituto de Salud Carlos III, Majadahonda, Spain], p8 [IFREMER, Nantes, France], p9 [CHU du Bocage, Dijon, France], p10 [Istituto Superiore di Sanità, Rome, Italy], p11 [University of Barcelona, Barcelona, Spain], and p12 [University of Valencia, Valencia, Spain]) across Europe. Among these, 10 laboratories are involved in the diagnosis of NLVs in humans on a routine basis or as a national or regional reference laboratory (Table 1). Coordination of the project was contracted out to RIVM (M. P. G. Koopmans [EU contract QLK1-1999-00594]). Overall, the aim of this consortium was to allow more rapid and internationally standardized assessment of the spread of food-borne viruses, including elucidation of the mechanisms of emergence of novel variants, by harmonization of methods, the use of a common database, and epidemiological follow-up of international food-borne viral infections. There are four work packages, namely, (molecular) virology (WP1), database (WP2), epidemiology (WP3), and food microbiology (WP4). The work described here was performed within WP1. The objectives of this work package were to compare the performance of existing methods for detection and (geno)typing of NLV and hepatitis A virus by using a coded panel of stool specimens, resulting in the selection of consensus assays for use in outbreak investigations across Europe.

TABLE 1.

NLV RT-PCR assays used by the different partner laboratories

Laboratorya	Viral RNA extraction method (Supplier)	Primers or primer pairs used for NLV diagnosisb	Reference
p1*	Boom method	JV12-JV13	49
p2*	Tripure (Roche)	Ni, NV69, NVp110	39
p3	QIAamp Viral RNA Mini Kit (Qiagen)	JV12-JV13	49
p4	Boom method	JV12-JV13	49
p5*	Boom method	Ni, Ando, E3	18
p6*	QIAamp Viral RNA Mini Kit (Qiagen)	32, 33, 35, 36 (32a, 33a, 35a)	45
p7	RNaid (Bio 101)	JV12-JV13	49
p8	High Pure Viral RNA (Roche)	NV36, Ni, Ando, NVp110	33
p9	QIAamp Viral RNA Mini Kit (Qiagen)	NV36, Ni, NVp110	33
p10	Boom method	JV12-JV13	49
p13*	Nuclisens (Organon Teknika)	Mon 431, 432, 433, and 434	16

^aRT-PCR assays evaluated in this study are indicated with an asterisk.

^bAndo, SR48/50/52 primer combination (1). Primer pairs are indicated with a hyphen.

Inventory of methods.

One of the objectives of the present study was to compare methods and, if possible, to select a consensus assay for the molecular detection of NLVs to be used throughout Europe, with a special focus on multinational outbreaks. Information on the methods used by the different partner laboratories for RNA extraction, RT-PCR detection, and genotyping of NLVs was obtained by using a questionnaire.

Assembling and coding of the stool panel.

A coded stool panel of 82 samples was compiled from stool specimen collections from five different partner laboratories (p1, p2, p4, p5, and p6). The specimens had been collected over a 4-year period of time (1997 to 2000), were derived from both outbreaks and sporadic cases of gastroenteritis, and had previously been tested for NLVs by RT-PCR (p1, p2, p4, p5, and p6) and EM (p2 and p4). The specimens had been stored either as original stool samples at 4°C (p1, p2, p4, and p5) or as 10% suspensions in phosphate-buffered saline at −20°C (p6). All samples were shipped frozen to the p1 laboratory (RIVM) and were coded by an epidemiologist who had no prior knowledge of contents or origin of the samples. In addition, nine stool samples representing nine different NLV genotypes were included as positive controls. The panel samples were frozen at −70°C and shipped on dry ice to four different participating laboratories. All laboratories were asked to assess the presence of NLVs in the specimens by using their routine diagnostic workup and, in addition, to determine the detection limit of the samples containing viruses from nine genotypes. Instead of a stool panel, viral-RNA extracts of a complete panel were shipped to the central reference laboratory for NLV detection in the United States (Viral Gastroenteritis Section, Centers for Disease Control), designated the p13 laboratory.

RNA extraction and RT-PCR.

Viral RNA was extracted from the panel samples according to published standard procedures used in each partner laboratory (18, 34, 40, 46, 50). The essentials of the procedures are listed in Table 1. Laboratories p1, p5 and p6 used a silica-based method (8) and laboratory p2 used Tripure reagent (Roche) as described previously (40). For RT-PCR, each partner laboratory tested the RNA extracts by using published assay formats and primer pairs (Table 2). Laboratory p1 used a separate step for the RT reaction, followed by a PCR assay with the primer pair JV12-JV13 (50); laboratory p2 used a separate RT with NVp110, followed by PCR with the primers NVp110, Ni, and NVp69 (40); laboratory p5 used two RT-PCR assays with the E3-Ni and E3-Ando primer pairs, respectively (18); and laboratory p6 used a recently published nested RT-PCR assay format (46), with some modifications. Briefly, for the first PCR an equimolar mixture of primers 32 and 32a and antisense primer 36 were used. In the second PCR a mixture of primers 33, 33a, 35, and 35a was used (Table 2). Laboratory p13 tested RNA extracted by p1 by using a single-tube RT-PCR assay targeting the 3′ end of ORF1 (region B) (16). Detection of NLV RT-PCR products was confirmed by using Southern hybridization and reverse line blot hybridization (RLB) (50, 53) (p1), microplate hybridization (40) (p2), and sequencing (p1, p5, p6, and p13).

TABLE 2.

DNA sequence and location of oligonucleotide primers used in RT-PCR assays in this study^a

Primer	Orientation	DNA sequence (5′-3′)b	nt position (range)c	Reference
NV36	+	ATAAAAGTTGGCATGAACA	4487-4505	53
32	+	ATGAATATGAATGAAGATGG	4499-4518	45
32a	+	ATGAACACAATAGATGATGG	4499-4518	45
JV12	+	ATACCACTATGATGCAGATTA	4552-4572	49
33	+	TACCACTATGATGCAGATTA	4553-4572	45
33a	+	TATCACTATGATGCTGACTA	4553-4572	45
p290	+	GATTACTCCAAGTGGGACTCCAC	4568-4590	30
NV69d	+	GGCCTGCCATCTGGATTGCC	4733-4752	53
SR48/50/52d	+	GTGAACAGYATAAAYCAYTGG	4766-4786	1
Nid	+	GAATTCCATCGCCCACTGGCT	4495-4515*	17
JV13	−	TCATCATCACCATAGAAAGAG	4858-4878	49
E3	−	ATCTCATCATCACCATA	4865-4881	17
NVp110	−	ACDATYTCATCATCACCATA	4865-4883	33
SR33	−	TGTCACGATCTCATCATCACC	4868-4888	1
35	−	GTTGACACAATCTCATCATC	4871-4890	45
35a	−	ACAATYTCATCATCICCAT	4866-4884	45
p289	−	TGACAATGTAATCATCACCATA	4865-4886	30
NV35d	−	CTTGTTGGTTTGAGGCCATAT	4936-4956	53
36d	−	ATTGGTCCTTCTGTTTTGTC	4961-4980	45
Mon431	+	TGGACIAGRGGICCYAAYCA	5093-5112	16
Mon432	+	TGGACICGYGGICCYAAYCA	5093-5112	16
Mon433	−	GAAYCTCATCCAYCTGAACAT	5285-5305	16
Mon434	−	GAASCGCATCCARCGGAACAT	5285-5305	16

^aFor reference, several other well-known detection primers are included (NV35, p290, p289, and SR33). Primer sequences grouped between lines of space overlap and have been aligned, except as noted in footnote d.

^bN = A + T + G + T; D = A + T + G; Y = C + T, S = C + G; R = A + G; I = inosine.

^cNumbers refer to the nucleotide positions of Hu/NLV/Norwalk/68/US (M87661) or, where indicated by an asterisk, of Hu/NLV/Lordsdale/93/UK (X86557).

^dThese primers are not aligned with the other primers in this table.

Detection limits.

To determine the detection limit of each RT-PCR assay, participants were asked to prepare and test 10-fold serial dilutions of purified RNA prepared from undiluted fecal samples containing nine different NLV genotypes as determined by sequence analysis (50). The PCR titer was calculated per milliliter of original stool suspension and was based on the highest dilution giving a clear positive signal after RT-PCR.

NLV genotyping by RLB.

For high-throughput screening, NLVs were genotyped at the p1 laboratory by RLB, which was performed as described previously (53). Briefly, all panel sample extracts were amplified by using the standard RT-PCR, including a biotin-labeled JV13 primer. The amplified biotinylated products were hybridized to a set of 18 immobilized oligonucleotide probes, each corresponding to a genogroup (I or II) or genotype (53). Hybridization was performed in a miniblotter system (MN45; Immunetics, Cambridge, Mass.) that can analyze 40 samples simultaneously. After hybridization, bound PCR products were detected by chemiluminescence (53). Phylogenetic analysis of a 145-nucleotide (nt) region of the POL gene was used as the “gold standard” for genotype identification (53).

DNA sequencing and phylogenetic analysis.

All extracts that resulted in NLV RT-PCR products of the appropriate size (327 bp for p1, 113 bp for p5, 338 bp for p6, and 213 bp for p13) were sequenced in both directions by using dye terminator chemistry. PCR products of the p2 laboratory were not sequenced. The NLV genogroups and genotypes were assigned according to a recentlyproposed classification system (2, 20, 33). Briefly, genotypes within genogroup I or II (GI or GII) are represented by a cryptogram of the prototype strain, followed by the genogroup and cluster number in parentheses as follows: Hu/NLV/GI/Norwalk/1968/US (GI.1), Hu/NLV/GI/Southampton/1991/UK (GI.2), Hu/NLV/GI/DesertShield395/1990/SA (GI.3), Hu/NLV/GI/Chiba407/1987/JP (GI.4), Hu/NLV/GI/Musgrove/1989/UK (GI.5), Hu/NLV/GI/Hesse3/1997/DE (GI.6), Hu/NLV/GI/Winchester/1994/UK (GI.7), Hu/NLV/GII/Hawaii/1971/US (GII.1), Hu/NLV/GII/Melksham/1994/UK (GII.2), Hu/NLV/GII/Toronto24/1991/CA (GII.3), Hu/NLV/GII/Bristol/1993/UK (GII.4), Hu/NLV/GII/Hillingdon/1990/UK (GII.5), Hu/NLVGII/Seacroft/1990/UK (GII.6), Hu/NLV/GII/Leeds/1990/UK (GII.7), and Hu/NLV/GII/Amsterdam/1998/NL (GII.8). In addition, we included sequences of strains belonging to a distinct branch within a genotype (Hu/NLV/GI/Stavanger/1995/NO (GI.3b), Hu/NLV/GII/Wortley/1990/UK (GII.1b), a sequence that tentatively is assigned to a fourth genogroup (Hu/NLV/GIV/Alphatron/1998/NL (GIV.1), and viruses of which the grouping in ORF1 is different than in ORF2 in the phylogenetic analyses.

Genetic classification of the strains was performed by using nucleotide sequences of a 145-nt consensus POL region for strains from p1 and p6. A multiple alignment was constructed by using CLUSTAL W (v1.4) and imported into the Treecon software package (49). To compile a phylogenetic tree, distances of 100 bootstrapped data sets were calculated by using the Jukes and Cantor correction for evolutionary rate (49).

Combining data and selection criteria for the assays.

After each laboratory returned its results for detection (p1, p2, p5, p6, and p13), genotyping (p1), and sequencing (p1, p5, p6, and p13) to the coordinator (p1), the code of the panel was broken. Assays were ranked based on an arbitrary scoring system for sensitivity and detection limit based on the actual data from the panel evaluation from lowest (score 1) to highest (score 5). In addition, having a single-round format was assigned two points, whereas having a nested format was assigned the score 0, since the latter was considered less desirable to use in a clinical diagnostic setting.

RESULTS

Inventory of methods.

All partner laboratories used an assay targeting the POL gene in the NLV genome. The overlap of the amplified region between the assays, however, was 63 nt only. Based on the inventory (Table 1), we selected four assays (p1, p2, p5, and p6) for comparative evaluation (18, 40, 46, 50). The assays reflect the diversity of RT-PCR detection methods for NLVs used by different partner institutes. In addition, at the p13 laboratory, a recently developed RT-PCR assay was used that targets a different region within ORF1 with no sequence overlap at all (16).

Sensitivity of five different RT-PCR assays for the detection of NLV.

After we broke the code of the panel, we found that 82 of the 91 samples of the stool panel had previously tested positive for NLV by either RT-PCR (75 samples) or by EM (7 samples). The remaining nine samples had been included as negative controls and originally tested negative by RT-PCR. Therefore, all calculations of the sensitivity rate were performed by using the 82 NLV-positive samples as a 100% score (Table 3). In total, 69 (84%) of the samples tested positive in at least one of the assays. The overall sensitivity rate of the assays (considering any positive test result) ranged from 52 to 73% (73% for p6, 67% for p1, 60% for p13, 59% for p2, and 52% for p5). When we compared the different assays, we found that the p1 assay had the highest percentage of positive samples when three (91%) or four (100%) of the five assays yielded positive results (Table 3).

TABLE 3.

Summary of results of comparative evaluation of RT-PCR assays for the detection of NLVs

Analysis	n	RT-PCR result
		p1	p2	p5	p6	p13
Sensitivity for all samples
% Positive overall	82	67	59	52	73	60
% Positive samples in four of five assays	44	100	84	86	91	91
% Positive samples in three of five assays	56	91	79	77	91	82
% Positive samples in two of five assays	64	86	75	67	88	77
% Positive samples in one of five assays	69	80	70	62	87	71
Sensitivity according to origin
% Positive of p6 samples	20	35	40	20	65	15
% Positive of p2 samples	20	70	65	55	70	70
% Positive of p4 samples	19	74	53	74	74	84
% Positive of p5 samples	14	79	79	57	86	71
% Positive of p1 samples	9	100	67	67	89	67
Total	82
Sensitivity according to genotype (no. of sensitive strains)
GI strain (cluster)
Chiba (GI.4)	4	4	1	4	4	4
Stavanger (GI.3b)	3	3	2	2	2	2
Southampton (GI.2)	2	2	2	2	2	2
Hesse (GI.6)	3	3	2	2	3	2
Winchester (GI.7)	1	1	0	1	1	1
Total	13	13	7	11	12	11
% Positive of original (n = 13)		100	54	85	92	85
GII strain (cluster)
Bristol (GII.4)	18	16	18	12	16	12
Hawaii (GII.1)	13	7	9	5	10	8
Leeds (GII.7)	9	5	4	4	9	7
Toronto (GII.3)	2	2	2	2	2	2
Rotterdam (GII.3*)a	2	2	1	2	2	1
Amsterdam (GII.8)	4	2	1	2	3	3
Hillingdon (GII.5)	3	2	2	1	1	1
Wortley (GII.1b)	2	2	2	1	1	2
Melksham (GII.2)	1	1	1	1	1	1
GGII Finland (NA)	1	1	1	1	1	1
Alphatron (GIV.1)	1	1	0	1	1	0
Total	55	41	41	32	47	38
% Positive of original (n = 55)		75	75	58	85	69
Sensitivity excluding p6 samples
% Positive of total	62	81	66	65	81	77

^aRotterdam forms a distinct group in the POL region and groups with Toronto (GII.3) in the complete capsid gene.

Sensitivity related to origin of samples.

To identify whether there was a relationship between performance of an assay and the geographic origin of the samples or different sample storage conditions, the results were summarized based on the origin of the panel samples (Table 3). From the 20 p6 samples that had previously tested positive the p6 assay scored 65%, whereas in the other assays a range of from 15 to 40% of the samples scored as positive. For the panel samples of other geographical origin, the sensitivity rates ranged from 55 to 70% for the p2 samples, 53 to 84% for the p4 samples, 57 to 86% for the p5 samples and 67 to 100% for the p1 samples. Of the 11 samples that had been submitted to the panel as positive samples but that tested negative upon retesting, seven (64%) were of p6 origin and had been stored as 10% fecal suspensions at −20°C. When p6 samples were excluded from analysis, the overall sensitivity rate of the assays increased significantly (65% for p5, 66% for p2, 77% for p13, 81% for p6, and 81% for p1 [Table 3]).

Sensitivity related to genotype.

The nucleotide sequences of all RT-PCR products of the expected sizes for the p1, p5, p6, and p13 samples were determined. A stretch of 145 nt that overlapped sequences of both p1 and p6 and sequences from representative strains of 15 different genotypes (52) was used to create a multiple alignment from which a phylogenetic tree was constructed, and most panel strains could be assigned to one of the existing genotypes (20, 52) (Table 3). NLV sequences were obtained from 66 panel strains (13 GI, 52 GII, and 1 GIV strains; Fig. 1) in the POL region (p1, p5, and p6 together). For three strains, sequences of less than 145 nt were obtained and therefore were not included in the phylogenetic analysis. One of these strains was detected in a sample in which a completely different NLV sequence was detected by a different partner laboratory (sample 3, Table 4). Overall, the RT-PCR assays detected between 54 to 100% for the GI strains and between 58 to 85% for the GII and GIV strains (Table 3). The sensitivity of detection by genotype was lower for the p2 assay for GI strains (54%) and for the p5 assay for GII and GIV strains (58%). The performance results for each RT-PCR assay, including the nine samples that originally tested negative, are shown in Table 4.

FIG. 1.

Phylogenetic relationships based on a 145-bp region of the RNA POL gene showing the relationships among 66 panel strains from which a sequence was obtained by p1 or p6, as well as NLV strains representing the individual genotypes or clusters. Full capsid numbers are shown in parentheses as follows: Bristol (GII.4), Wortley (GII.1b), Hawaii (GII.1), GII Finland, Rotterdam, Melksham (GII.2), Hillingdon (GII.5), Toronto (GII.3), Amsterdam (GII.8), Leeds (GII.7), Stavanger (GI.3b), Winchester (GI.7), Chiba (GI.4), Southampton (GI.2), Sindlesham (GI.6), and Alphatron (GIV.1). Several other NLV strains (Lordsdale, Snow Mountain virus, Girlington, Desert Shield virus, Hesse, Queens Arms, and Musgrove) were included as a reference. Bootstrap values of >50 of the internal nodes are indicated.

TABLE 4.

Overall RT-PCR results for 91 Stool Samples

Stool panel		RT-PCR result					Cluster	RLB type		Stool panel		RT-PCR result					Cluster	RLB type
No.	Origin	p1	p2	p5	p6	p13			No.	Origin	p1	p2	p5	p6	p13
1	p6	+	+	−	+	−	GII.4
2	p6	−	−	−	−	−
3	p6	+	−	−	+	−	GI.2 and GII.1a
4	p2	+	+	+	+	+	Rotterdamc	Rotterdamc
5	p4	−	−	−	−	−
6	p2	+	+	−	+	−	GI.6	GI
7e	p1	−	−	−	−	−
8	p6	−	−	−	−	−
9	p6	−	−	+	+	+	GII.8
10	p6	−	+	−	+	−	GII.4
11	p2	−	+	−	+	+	GII.7
12	p5	−	+	−	+	+	GII.1
13e	p1	−	−	−	−	−
14	p5	−	+	−	+	+	GII.7
15	p4	+	−	+	+	+	GI.4	Chiba
16	p6	−	−	−	+	−	GII.8
17	p6	−	−	−	−	−
18f	p2	+	−	−	−	+	GII.8	Leeds
19	p5	+	+	+	−	+	GII.1	Hawaii
20	p6	−	−	−	−	−
21	p2	+	+	+	+	+	GI.2	Southampton
22f	p2	−	−	−	−	−
23	p6	+	+	+	+	+	GII.8
24e	p1	−	−	−	−	−
25	p5	+	−	+	+	−	Rotterdamc	GII
26f	p4	−	−	−	−	−
27e	p5	+	+	+	+	+	GII.4	Bristol
28	p6	+	+	−	+	−	GII.4
29	p5	+	−	−	+	+	GII.7	Leeds
30	p5	+	+	−	+	+	GII.1	Hawaii
31	p4	+	−	+	+	+	GI.3b	GI
32	p2	+	+	+	+	+	GGIIFinland	GII
33	p6	−	+	−	+	−	GII.4
34	p2	+	+	+	+	+	GII.5	Hillingdon
35	p5	+	−	+	+	+	GI.6	GI
36e	p1	−	−	−	+	−	GII.1
37	p6	+	+	+	+	−	GII.4	Bristol
38	p4	+	+	+	+	+	GII.4	GII
39	p6	−	−	−	−	−
40	p4	+	+	+	+	+	GII.4	Bristol
41f	p4	−	−	−	+	+	GII.7
42	p5	+	+	+	+	+	GI.4	Chiba
43	p6	−	−	−	+	−	GII.1
44	p2	+	+	+	+	+	GI.6	Sindleshamc
45	p2	+	+	−	+	+	GII.1	Hawaii

^aTwo different NLV sequences were detected by two different laboratories; only from GI.2 was a 145-nt sequence obtained.

^bNo sequence was obtained; genotyping was performed by RLB only.

^cStrains cluster differently in the POL and capsid regions: Sindlesham groups with Hesse (GI.6) in the complete capsid region but forms a distinct group in the POL region. Similarly, Rotterdam forms a distinct group in the POL region and groups with Toronto (GII.3) in the complete capsid gene.

^dSamples 83 to 91 were included as NLV-positive controls.

^eSamples that originally tested negative by RT-PCR.

^fSamples that originally tested positive by EM.

Detection limit.

To calculate the detection limit of each RT-PCR assay, each partner laboratory determined the RT-PCR endpoint titers of nine different NLV strains. The detection limit differed from 10³-fold for the Southampton strain up to 10⁷-fold for the Toronto strain (Fig. 2).

FIG. 2.

Comparison of detection limit of the five different NLV detection assays with RNA from nine different NLV genotypes. For more information about the partner laboratories (p), see Table 1 and Materials and Methods.

Ranking of the assays.

After we compared the overall sensitivity, the detection limit, and the PCR format (single-round or nested) of each of the assays evaluated in the present study, it was clear that no single assay stands out as best based on all of these criteria. However, the p1 assay scored higher for sensitivity and for not having a nested format.

Genotyping. (i) Sequencing.

In order to classify the NLV strains in the present study, we combined and extended three previously published NLV classification schemes (2, 20, 33) based on NLV strains for which complete ORF2 sequences were available. To assign genotypes by using POL sequences, we used our previously defined working criteria: >85% similarity based on the nucleotide sequence of the POL fragment for GI strains and 90% for GII strains (52). Phylogenetic analysis of 66 POL sequences obtained in the present study revealed that they could be grouped into 17 clusters; 5 within GI, 11 within GII, and 1 (Alphatron; GIV) clustering almost equidistant from GI and GII (Fig. 1). Of these, 16 correlate with known genotypes (Southampton [GI.2], Chiba [GI.4], Hesse [GI.6], Winchester [GI.7], Hawaii [GII.1], Melksham [GII.2], Toronto [GII.3], Bristol [GII.4], Hillingdon [GII.5], Seacroft [GII.6], Leeds [GII.7], Amsterdam [GII.8], and Alphatron [GIV.1]) or subclusters (e.g., Stavanger [GI.3b], Wortley [GII.Ib], and Rotterdam). One potentially new genotype could be recognized, strain “GII Finland,” with an ∼15% nucleotide difference from the currently known GII clusters (Fig. 1).

(ii) RLB.

Genotyping of NLVs by RLB was performed with a membrane to which a set of 18 probes was bound as described previously (Fig. 3) (53). In all, 50 of the 55 (91%) samples tested positive by RLB and, of these, 37 (74%) could be genotyped directly when compared with the sequence of the POL region. Of the 13 untypeable strains, 11 reacted only with a genogroup probe (GI or GII) and were negative with a genotype-specific probe (four Bristol strains, two Sindlesham strains, two Leeds strains, one Rotterdam strain, one Desert Shield strain, and one untypeable GII strain). Two strains that were determined to belong to the Wortley subcluster (GII1b) by sequencing were incorrectly genotyped by RLB as Bristol and Melksham, respectively.

FIG. 3.

Detection and simultaneous typing of NLVs from 30 panel strains (strains 31 to 60). Probes specific for the detection of genogroup I (GIa and b) and genogroup II (GII) and for 15 ORF1 clusters are indicated (53). The name of each probe is shown with the representative full capsid number (e.g., GII.4) in parentheses. ✽, The Rotterdam probe detects strains that are GII.3 in the capsid region, but that group separately in the POL region; p, positive Bristol (GII.4) control; n, negative control (water). Note that probes specific for strains of the clusters Seacroft (GII.6), Amsterdam (GII.8), and Alphatron (GIV.1) were not included on the membrane.

DISCUSSION

To establish an international laboratory surveillance network for the detection of multinational outbreaks of NLV gastroenteritis across Europe, a harmonization of methods for NLV diagnosis is needed. Molecular detection of NLVs has found widespread use since the development of RT-PCR assays that allow successful amplification of viruses that are often shed in low numbers and are genetically extremely heterogeneous (1, 20). By using these assays as the diagnostic tool, it has been established that NLVs are the most common cause of outbreaks of gastroenteritis worldwide (15, 33, 37, 43, 51). Harmonized detection and genetic comparison of NLV strains ideally requires an assay that has high sensitivity and specificity, that is easy to standardize, that preferably (though not necessarily) uses a single-round format, and that results in a product that is sufficiently long for reliable phylogenetic typing.

In the present study, we selected five different RT-PCR assays and tested their performance with a panel of 91 stool specimens. To have the best representation of circulating NLV strains, this panel included specimens from five different countries, specimens that were EM positive but had previously tested as RT-PCR negative, and specimens representing at least nine different NLV genotypes. Overall, the sensitivity was acceptable (>84%) for all assays with the samples that were unambiguously positive (four of five or more tested positive). The sensitivity went down to as low as 62% (p5) when all positive results were included. This suggests differences in the detection limit for the assays with decreasing sensitivity, which was observed most prominently for the p2, p5, and p13 assays. Indeed, for the p5 test, detection limits were rather low, suggesting that this is the explanation for the lower sensitivity. The p2 assay detected viruses in highly diluted samples but appears to be more selective, since only 54% of the GI viruses were detected with this test. This may be explained by the limited number of GI probes that are used in the microplate hybridization assay which, given the high sequence diversity between NLV strains, should be significantly increased to detect all currently known genotypes. This finding implies that the assays should be used with caution: while they may work well with outbreak specimens (multiple samples), problems may arise when the assays are used in community surveillance or environmental studies. A nested format was considered to be less favorable when used as a consensus assay in a laboratory surveillance network because of the higher risk of cross-contamination in clinical diagnostic laboratories. Copurification of RT-PCR inhibitors during RNA extraction, lack of primer specificity, or inadequate interpretation of EM data are possible explanations for samples that previously tested positive for NLVs by EM (p2 and p4) but could not be confirmed by RT-PCR.

Four of the five RT-PCR assays evaluated in the present study include primers that target the POL region involving GLPSG and YGDD amino acid motifs and are among the most frequently used in the field (1, 17, 34, 40) or have been used for screening NLVs in large-scale epidemiological studies (13, 51). To extend our comparative analysis, we included a recently developed RT-PCR assay targeting a different region at the 3′ end of ORF1 (16). Other regions of the genome have been used as targets as well, including 2C helicase (54, 57), the 5′ end of the capsid region (21, 42), and ORF3 (52, 56). However, since these regions have a greater genetic diversity than the POL region, they are less frequently used. It remains to be seen whether data obtained for genotype classification by using the region B fragment (16) correlate with clustering based on the capsid gene, as has been done for the p1 assay, since there was no clear advantage of using this region over the existing assays. Furthermore, a strong disadvantage would be the loss of the current POL sequence database. Therefore, we favor the assays that target the overlapping region within ORF1. To promote further harmonization, we recommend for newly starting groups the p1 assay as the method of choice for the EU network. It is clear, however, that the overall sensitivity is less than optimal for all assays. This could be due to the loss of RNA upon storage and shipping. Of the panel samples that had previously tested positive for NLV, only 84% were found to be positive in at least one of the five RT-PCR assays, raising questions regarding the stability of the viruses in these false-negative samples. The preferred storage condition for diagnostic work is fresh stool specimens at 4°C, but samples were divided into aliquots and frozen for shipment. Since the majority of the false-negative samples originated from one laboratory from which 10% fecal suspensions in phosphate-buffered saline were included in the panel, instead of the original undiluted fecal sample, it is conceivable that the viruses in these samples may have deteriorated during storage, shipment, or freeze-thawing, as has been described previously (9). Therefore, we recommend the use of raw stool samples stored at −70°C or lower for future primer evaluation studies.

Several factors can affect the sensitivity and specificity of RT-PCR assays, including the viral RNA extraction method, the primers used in amplification, and the methods used for confirmation of test results. For extraction of viral RNA, all laboratories participating in the present study used a guanidine thiocyanate-based extraction procedure to release viral RNA assay and subsequent purification by either binding onto silica beads (8) or precipitation by ethanol (10). Although direct heat release of NLVs from fecal extracts has been described (47), the guanidine isothiocyanate-silica method has been reported to be the most successful approach to removing inhibitors from fecal specimens (24). Differing efficiencies of extraction and RT among the different assays might also explain discrepancies that could be addressed with viral cDNA as the starting material. Visual interpretation of gels after gel electrophoresis yielded bands in the right range for several samples in the p1, p5, and p13 assays, demonstrating the necessity to confirm the specificity of the amplicons by a second method such as DNA sequencing of the RT-PCR products, Southern hybridization, microplate hybridization (40), or RLB (53). Nonspecifically amplified DNA can easily be generated because low annealing temperatures (i.e., 37°C) are used during PCR to tolerate some mismatches between primers and NLV templates (3, 51, 54). This is considered essential for the successful generic detection of such a genetically diverse group of viruses.

All five RT-PCR assays detected the majority of the genetic clusters present in the panel. However, there were considerable differences among the performance of the assays because only 34% of the samples tested positive in all assays. Overall, the p2 assay performed relatively poorly in detecting GI strains, and the p5 assay performed poorly for the GII strains. The reason for the latter result is unclear since both assays use primer Ni, which detects primarily GII strains (17). Primer NV69 (p2) was developed based on the Norwalk virus sequence in the early 1990s when few other NLV sequences were available (54).

Although it is the most conserved region of the genome, the POL region, which is the target of most of the primers used in the present study, still has a high level of sequence diversity (1, 52). This template variability has a direct effect on the primer homology and thus for the detection limit for different primer pairs for different genetic clusters. Thus far, only for the p1 assay has the detection limit been estimated to be 3 to 30 RNA viral particles after a Hawaian strain was mixed with known concentrations of 80-nm latex beads (33). In the present study, we compared the detection limit of the different assays by using RNAs from nine different NLV genotypes, and the results showed that there is significant variation in detectable quantities of different NLV strains with no consistent pattern.

Recently, two similar schemes for the genetic classification of NLVs have been proposed (2, 33) and were summarized by Green et al. (20). The current consensus is that the entire capsid gene sequence is needed to define genotypes. To date, at least 15 different genotypes have been recognized based on >80% amino acid similarity in the complete capsid gene (19, 53). With few exceptions, NLV strains cluster similarly when different regions of the genome are analyzed (52). In our study, genotyping by RLB showed a high level of agreement with POL typing after sequencing and phylogenetic analyses. However, several genotype-specific probes missed corresponding strains and, therefore, need to be refined. Other high-throughput NLV genotyping systems have recently been described (5, 39) and require further evaluation. The elegance of RLB is its straightforwardness of standardization between laboratories, since probe-labeled membranes can be prepared at a central facility and used by others. In addition, new probes directed toward newly emerging strains can easily be added. The limitations of this typing method include the minimum size of the amplicon (160 bp) necessary to target the probes on the membrane (53) and the limited sequence variation in the POL region between some clusters (e.g., Hawaii and Wortley), which makes the design of cluster-specific probes challenging. The amplicon size limitation precludes the use of assays p2 and p5 in combination with this method. In addition, RLB clearly cannot be used to compare closely related strains for which sequence analysis will remain the method of choice. Genotyping by microplate hybridization is another approach that could be used for high-throughput typing (40). However, the temperature (50°C) of hybridization, as well as subsequent washing steps, prevent an easy transfer of probes used in RLB to an enzyme-linked immunosorbent assay format. Ideally, genotyping should be performed with capsid sequences as a template since these most reliably reflect antigenicity and possibly serotype epitopes. Thus far, the majority of strains show good agreement between polymerase and capsid clustering (52), but as more and more sequences become available and recombinants are likely to circulate (29, 52; H. Vennema, unpublished data), it remains to be seen if the POL region alone is the appropriate region for the genotyping of NLVs.

In conclusion, this is the first study that evaluates five different RT-PCR assays for the detection of NLVs. Although no single assay stands out as the best by all criteria, evaluation of sensitivity, detection limit, assay format, and successful implementation in several other laboratories prompts us to recommend the p1 assay to laboratories that want to newly initiate NLV diagnostics. However, it must be stressed that no single assay at present detects all variants of NLV and that followup testing of negative stool samples with additional primer sets should be considered for all outbreaks matching Kaplan's criteria (32). Due to the complementarities of some of the assays the sequence data generated in the present study, together with the increasing database of NLV strains, make it possible to design assays that may result in higher overall sensitivity and broad reactivity. Finally, although the addition of probes could further improve its performance, we feel that the RLB method may help to standardize methods for genotyping of NLV strains across Europe.

13Originp1p2p5p6p13GGIIFinlandSouthampton 46p2−−−−−47p6−−−+−GII.148p4+−+++GI.4Chiba 49p4+++−+GII.4GII 50p4+++++GII.1Hawaii 51p6−−−−−52p4+++++GII.4GII 53p2+++−+GII.1bBristol 54p4−−−−−55p5++++−GII.1Hawaii 56p5+++++GII.4Bristol 57p4+++++GII.4Bristol 58p4−−−+−GII.159p5−−−−−60p5−−−−−61p5+++++GII.4Bristol 62p4+++++GII.1Hawaii 63p4+++++GII.4Bristol 64p5+−−++GII.865p5−++−+GII.166p2−−−++GII.767p2+++++GII.4Bristol 68p2+++++GII.2Melksham 69p4+++−+GII.4GII 70p5++−−−GI.3bStavanger 71p2−−−−−72p6++++−GII.4Bristol 73p6++−++GII.1bMelksham 74p2+−+++GI.7Winchester 75p2++++−GII.776p2+++++GII.3Toronto 77p4+−+++GII.7GI 78p4+++++GII.7GI 79p1−−−−−80p5++−++GII.4Bristol 81p2−−−−−82p6−−−−−83p1+++++GI.2Southampton 84p1+++++GI.3bStavanger 85p1+−+++GI.4Chiba 86p1++−++GII.1Hawaii 87p1+++++GII.3Toronto 88p1++−++GII.4Bristol 89p1++−−−GII.5Hillingdon 90p1+−++−GII.7Leeds 91p1+−++−GIV.1Alphatron