Differentiation and Characterization of Enteroviruses by Computer-Assisted Viral Protein Fingerprinting

Abstract

We have developed and standardized a computerized method for the typing and characterization of enteroviruses with radiolabeled viral protein fingerprints. Enteroviral proteins were radiolabeled with [³⁵S]methionine during growth in cell culture and were then separated by polyacrylamide gel electrophoresis. The dried gel was scanned, and from the resulting computer image (which resembled an autoradiogram) protein patterns were computer extracted and stored in a database. The enterovirus database contained community and prototype strains belonging to 20 different enteroviral serotypes. Each serotype has a discrete protein pattern, and the most important pattern differences for determining each type are in the region of the viral capsid proteins VP1, VP2, and VP3. When the database was challenged with 148 clinical enterovirus strains, 144 (97%) were correctly identified by using the correlation coefficient as a quantitative measure of relatedness between two patterns. This method can identify a type in a single test and represents a practical alternative to virus neutralization because it is less expensive, is much faster (3 rather than 10 days), and does not rely on any virus-specific reagents. The results also show that most of the strains currently isolated from the community have protein patterns different from those of their older prototype strains. Viral protein fingerprinting is an evolving, dynamic system for the typing and characterization of enteroviruses. The method is appropriate for use in clinical virology and reference laboratories for the typing of enteroviruses, for the study of the epidemiology of enteroviruses, and for surveillance of enteroviruses.

Data suggest that nonpoliovirus enteroviruses (EVs) are responsible for 10 million to 30 million illnesses each year in the United States (24) and that the population most affected is under 10 years of age (14). EVs are the most common cause of aseptic meningitis and other illnesses ranging from minor respiratory type infections to paralysis and carditis (3).

Presumptive laboratory confirmation of an enteroviral infection is usually by the distinctive cytopathic effect produced on a selection of permissive host cells. Although cell culture still remains the “gold standard” for EV infection diagnosis, the use of reverse transcription-PCR is becoming more common as a rapid and more sensitive way of identifying an EV directly from the patient sample (18, 19). However, when they are used as diagnostic tests, neither cell culture nor PCR can provide an EV type identification.

The classic means of typing of an EV in the laboratory is by virus neutralization with the Lim and Benyesh-Melnick (LBM) antiserum pools, followed by confirmation of the serotype with monospecific antiserum (13). The antisera in these pools were raised against enteroviral prototype strains, and problems have been noted when the pools were used to type new variants because of pronounced intratypic antigenic variation (15). The World Health Organization has recommended conserving the use of the now limited stock of LBM reference pools. These recommendations and the expense of virus neutralization have resulted in many laboratories abandoning serotyping in favor of making an EV identification by cell culture only (2, 20). This approach has resulted in less effective epidemiological surveillance and will consequently limit future knowledge of the etiology of enteroviral disease. Alternative typing methods such as PCR with restriction fragment length polymorphism analysis (10), PCR–single-stranded conformation polymorphism analysis (6), antigen-capture PCR (22), or tests with monoclonal antibodies (28) have potential, but all methods currently in use or in development are based on the detection of enteroviral antigens, antibodies, or genomic material and have two major limitations: (i) the need for virus-specific reagents and (ii) the large number of different enteroviral serotypes, namely, six coxsackievirus B (CBV) types, 23 coxsackievirus A (CAV) types, 30 echoviruses (ECHO), three polioviruses (PVs), and EV types 68 to 71.

A novel viral identification system based on protein fingerprinting has been used to study several viral groups including adenoviruses, herpes simplex viruses, influenza and parainfluenza viruses, and respiratory syncytial virus (4, 26, 27).

In the present investigation we applied the method of viral protein fingerprinting to the typing and characterization of EVs (7). Computer-assisted numerical classification methods (23) are used to evaluate similarities between electrophoretically separated radiolabeled protein patterns (viral protein fingerprints) stored in computer databases. Within the enteroviral subgroups we found that each serotype has a specific pattern and that 97% of the enteroviral isolates studied were correctly identified by computer comparison of their protein patterns. The method is rapid and has the distinct advantage of being free of virus-specific reagents.

MATERIALS AND METHODS

Community strains.

The source, number, and type of the clinical strains isolated from patients in the community and used in this study are indicated in Table 1. They were randomly selected from specimens isolated and identified as EVs by the Medical Center Viral Diagnostic Laboratory of the University of California, San Diego, and by the San Diego County Public Health Laboratory (SDCPHL). In summary, the strains consisted of 44 CBVs, 7 CAVs, 60 ECHOs, and 37 PVs. The serotyping was done at SDCPHL by microneutralization with the LBM pools.

TABLE 1.

Source and type of isolates

Virus and serotype	No. of isolates from the following sourcesa
	CSF	Rectal	Throat	NP/Rhino	Liver	UK	Total
CBV-1		4	1				5
CBV-2	4	3	1	1		1	10
CBV-3	2						2
CBV-4	4	1	1				6
CBV-5	18		2				20
CBV-6	1						1
CAV-9	1						1
CAV-21			5	1			6
ECHO 2	1	1					2
ECHO 4		1	2				3
ECHO 6	14		1		1		16
ECHO 7		4					4
ECHO 9	2	1	2			1	6
ECHO 11	6	6	4				16
ECHO 25		2	1				3
ECHO 30	2	3	3			1	9
ECHO 31		1					1
PV 1		8		1		2	11
PV 2		16	1	1		1	19
PV 3		7					7
Total	55	58	24	4	1	6	148

^aAbbreviations: CSF, cerebrospinal fluid; NP/Rhino, nasopharynges or nose; UK, unknown source. 

Reference viruses.

The following prototype viruses were obtained as infected tissue culture supernatants from the American Type Culture collection: CBV type 1 (CBV-1), Conn-5; CBV-2, Ohio-1; CBV-3, Nancy; CBV-4, JVB; CBV-5, Faulkner; CBV-6, Schmitt; CAV-9, Bozek; CAV-21, Kuykendall; ECHO type 2 (ECHO 2), Cornelis; ECHO 4, Morrisey; ECHO 6, D’Amori; ECHO 6′, D-1 (Cox); ECHO 6", Burgess; ECHO 7, Wallace; ECHO 9, Hill; ECHO 11, Gregory; ECHO 25, JV-4; ECHO 30, Bastianni; ECHO 31, Caldwell; PV type 1 (PV 1), Brunhilde; PV 2, Lansing; and PV 3, Leon. The Sabin polio vaccine strains (Lederle) were obtained from SDCPHL.

Virus stock.

Viruses were passaged in either human rhabdomyosarcoma (RD) or HeLa cells with Eagle’s minimal essential medium containing Earle’s salts and 10% fetal bovine serum and were harvested when the cytopathic effect was 3 to 4+. After three freeze-thaw cycles, the cell debris was removed by centrifugation at 1,000 × g for 15 min at room temperature. The viral supernatants were frozen and stored at −20°C. The titer of the viruses ranged between 10⁴ and 10⁸ 50% tissue culture infective doses/ml.

Radiolabeling of viral proteins.

RD or HeLa cells were grown to confluent monolayers in tubes. The medium was replaced with 1 ml of Eagle’s methionine-free minimal essential medium containing Earle’s salts, without serum, prior to inoculation with 50 to 100 μl of prepared stock supernatant containing EV. After incubation for 5 to 6 h at 35°C in 5% CO₂, 15 to 20 μCi of [³⁵S]methionine (>1,000 Ci/mmol; Tran³⁵S-label; ICN Radiochemicals) was added to each tube, and incubation was continued for an additional 16 to 18 h before protein extraction.

Protein extraction.

The contents of the tubes were transferred to microcentrifuge tubes, and the tubes were spun at 14,000 × g for 2 min at room temperature. The supernatant was removed and the pellets were resuspended in 100 μl of electrophoresis sample buffer containing 2% sodium dodecyl sulfate and 5% 2-mercaptoethanol. After heating at 100°C for 2 min the tubes were vortexed and spun for 30 s, and the supernatant was used immediately for electrophoresis or was stored frozen at −20°C.

Electrophoresis.

One-dimensional sodium dodecyl sulfate-polyacrylamide gel electrophoresis was a modification of the method described by Laemmli (11) adapted to a horizontal format. Gels consisted of a 12.5% (wt/vol) acrylamide resolving gel with a 5% stacking gel and were cast on GelBond PAG (FMC BioProducts). A total of 10 μl of sample was applied to each lane, and 18 samples could be separated along with the control on each gel.

Gel scanning and data acquisition.

The dried gels were scanned directly (AMBIS 100 Radioanalytical Imaging System; Scanalytics Inc., Billerica, Mass.), resulting in images resembling autoradiograms. Each gel lane was delineated, and the protein pattern was extracted and automatically converted to a digital format for computer storage. With MicroPM, version 2.12 (Scanalytics Inc.), the protein patterns were normalized to compensate for slight variations in gel composition and the length of the electrophoresis run and were filed in a database by fast Fourier transformation of the data (9).

Database construction.

The database consisted of community and prototype strains of the serotypes listed in Table 1. Also included were the Sabin PV vaccine strains and protein patterns from mock-infected RD and HeLa cell lines. The proteins spanned a molecular mass range of 20 to 40 kDa.

Identification with the database.

With the AMBIS compare software, version 2.12, (Scanalytics Inc.), all 148 clinical strains listed in Table 1 were used to challenge the database. The matching algorithms work in the following way. When the database is challenged with an unknown pattern, all reference patterns in the database are rapidly screened by fast Fourier transformation and the 20 best spectral matches are selected. A second algorithm, the Pearson product-moment correlation coefficient (23), is then used as a quantitative means of determining similarities between the challenge pattern and the 20 best spectral matches. The entire search of the database takes about 15 s.

Match criteria.

The method used to define a match was as follows. One isolate was radiolabeled on 4 different days and was run on 13 gels to give 57 lanes. Some of these 57 lanes were added to the database and the database was challenged with all 57 patterns. When the lane found itself in the database (correlation, 1.0) the next match result was taken. All 57 lanes matched each other, and the average r value was 0.972 with a standard deviation (SD) of 0.018. Fifty-three (93%) lanes fell within 2 SDs of the mean, and all 57 lanes fell within 3 SDs of the mean; i.e., r was 0.92 or greater. A match would be considered correct if the best correlation coefficient was 0.92 or greater. By contrast, the match with a different strain in the database gave a mean correlation coefficient of 0.720 with an SD of 0.047. A mismatch would be 3 SDs above the mean, i.e., r ≤ 0.86. A match of between 0.86 and 0.92 would be considered correct if the patterns matched well by visual inspection.

Specificity and reproducibility.

To test the ability of the database to identify a viral protein pattern consistently, reproducibly, and correctly, three closely related ATCC prototype viruses, ECHO 6, ECHO 6′, and ECHO 6", were prepared in replicate experiments and were run on gels to give six protein patterns for each virus. One pattern for each virus was entered into the database, which was then challenged with the patterns for all 18 lanes.

Protein pattern analysis.

With the compare software, the protein patterns for some isolates and their respective prototypes were matched by using the Pearson correlation coefficient calculation and were clustered into a dendrogram by the unweighted pair group method with arithmetic averages (UPGMA) clustering technique (23). The dendrogram is a graphical display of similarity coefficients and shows the relationship between the strains on the basis of phenotypic similarities which may or may not reflect an evolutionary grouping.

RESULTS

Specificity and reproducibility.

The results showed that when six lanes with patterns for each of the three ECHO 6 prototype strains were used to challenge the database, each matched with itself with a mean best correlation coefficient of 0.97 (0.02). The best match with an EV of another type was a mean r of 0.64 (0.15).

Coxsackieviruses.

The database results in Table 2 indicate the best correlation coefficient match with another strain from the community (referred to here as a community strain) and, for comparison, the match with the prototype strain. In summary, 50 of 51 strains (98%) matched correctly. The majority of strains matched more closely another community strain of the same serotype than the respective prototype strains. Between the coxsackievirus types the matches were all 0.87 or less (data not shown). The dendrogram in Fig. 1 shows the relationship between some community strains, prototype strains, and mock-infected HeLa cells.

TABLE 2.

Results of coxsackievirus database matches

Virus and serotype	No. of strains	Best r match (mean [SD]) with the following strains:
		Another community strain							Prototype strain
		CBV-1	CBV-2	CBV-3	CBV-4	CBV-5	CBV-6	CAV-21
CBV-1	5	0.98 (0.01)							0.84 (0.04) with Conn-5
CBV-2	10		0.95 (0.04)						0.86 (0.03) with Ohio-1
CBV-3	2			(0.82)a					0.85 (0.01) with Nancy
CBV-4	6				0.97 (0.02)				0.94 (0.02) with JVB
CBV-5	20					0.95 (0.08)			0.49 (0.07) with Faulkner
CBV-6	1						—b		0.92 with Schmitt
CAV-9c	1								0.73 with Bozek
CAV-21	6							0.96 (0.03)	0.83 (0.04) with Kuykendall

^aBest match with prototype strain. 

^b—, no other strain in database. 

^cBest match with ECHO 5 prototype strain (r = 0.92). 

FIG. 1.

Dendrogram showing the relationship between some community CBV strains and their respective prototype strains on the basis of protein similarity coefficients. The molecular mass range of the proteins is 20 to 40 kDa. The UPGMA grouping method was used with MicroPM, version 2.12.

ECHO.

The results in Table 3 indicate the best correlation coefficient match and the prototype strain match. In summary, 57 of 60 strains (95%) matched correctly. As with the coxsackieviruses the majority of ECHOs matched more closely with another community strain than with their respective prototype strains. A dendrogram of a selection of community types and their prototypes is shown in Fig. 2.

TABLE 3.

Results of ECHO database matches

Virus and serotype	No. of strains	Best r match (mean [SD]) with the following strains:
		Another community strain									Prototype
		ECHO 2	ECHO 4	ECHO 6	ECHO 7	ECHO 9	ECHO 11	ECHO 25	ECHO 30	ECHO 31
ECHO 2	2	0.98									0.98 with Cornelis
ECHO 4	2		0.96								0.77 with Morrisey
ECHO 4	1		0.75								0.63
ECHO 6	9			0.97 (0.02)							0.83 (0.06) with Cox/D’Amori
ECHO 6	6			0.97 (0.01)							0.93 (0.03) with Cox/D’Amori
ECHO 6	1			(0.91)a							0.98 with D’Amori
ECHO 7	4				0.97 (0.02)						0.62 (0.06) with Wallace
ECHO 9	6					0.91 (0.07)					0.65 (0.07) with Hill
ECHO 11	16						0.95 (0.02)				0.77 (0.08) with Gregory
ECHO 25	3							0.94 (0.04)			0.47 (0.04) with JV-4
ECHO 30	6								0.95 (0.02)		0.82 (0.05) with Bastianni
ECHO 30	1								(0.82)a		0.94
ECHO 30	1						0.78				0.54
ECHO 30	1									0.78	0.62
ECHO 31	1								0.78	—b	0.56 with Caldwell

^aBest match with prototype strain. 

^b—, no other strain in database. 

FIG. 2.

Dendrogram showing the relationship between some community strains of ECHOs and their respective prototype strains on the basis of protein similarity coefficients. The molecular mass range of the proteins is 20 to 40 kDa. The UPGMA grouping method was used with MicroPM, version 2.12.

PVs.

All PV strains were vaccine related, and database match results are presented in Table 4. In summary for all 37 strains (100%) matches with another community strain of the same serotype and with their respective Sabin vaccine strains were correct. Between the PV types the matches were all 0.85 or less (data not shown). A dendrogram of selected lanes is in Fig. 3.

TABLE 4.

Results of poliovirus database matches

Virus and serotype	No. of strains	Best r match (mean [SD]) with the following strainsa
		Another community strain			Sabin strains			Prototype strains
		PV 1	PV 2	PV 3	1	2	3	P/T 1	P/T 2	P/T 3
PV 1	11	0.98 (0.01)			0.96 (0.02)			0.83 (0.02)
PV 2	19		0.98 (0.01)			0.96 (0.02)			0.87 (0.05)
PV 3	7			0.97 (0.01)			0.94 (0.03)			0.90 (0.03)

^aP/T, prototype strains; for PV 1 the prototype is Brunhilde, for PV 2 the prototype is Lansing, and for PV 3 the prototype is Leon. 

FIG. 3.

Dendrogram showing the relationship between some clinical strains of PVs, PV prototype strains, and Sabin vaccine strains on the basis of protein similarity coefficients. The molecular mass range of the proteins is 20 to 40 kDa. The UPGMA grouping method was used with Micro M, version 2.12.

DISCUSSION

Protein gel electrophoresis has been used for the identification, characterization, and differentiation of microorganisms since the introduction of the technique in the 1960s. Although largely superseded by nucleic acid-based methods, protein electrophoresis nevertheless remains a powerful tool for the indirect investigation of the microbial genome by the study of gene products. Computer analysis of the protein profiles with numerical taxonomy is essential and allows the rapid, objective study of hundreds of protein banding patterns by automatic matching and clustering.

Standardization of method protocols is important for ensuring the reproducibilities of the protein patterns. For sample preparation we used RD and HeLa cells for viral replication, although in our experience and in the experience of others the expression of viral proteins is independent of the host cell (17). The use of a radiolabel provides a means of visualizing freshly synthesized viral proteins, and the method takes advantage of the fact that EVs naturally suppress host cell protein synthesis, which begins to decline 2 to 3 h after infection (data not shown). No purification of the virus is necessary before electrophoresis.

The composition of the gel has also been standardized, and any remaining slight differences between gel batches and/or the lengths of gel runs are corrected by computer software. After drying, the gel is scanned directly so no autoradiography is required. Gel lanes are extracted from a computer image to give a histogram of patterns which reflects the intensity of each protein and its relative position in terms of its molecular mass, providing the basis for automated qualitative and quantitative comparisons of any two viral strains.

The aim of this study was to develop a simple method for the typing of viruses that have been identified in the clinical laboratory as EVs but that would not normally be typed because of the problems associated with serotyping. Our initial intention was to use a database of EV prototype strains, but it became apparent quite quickly that recently isolated EVs had protein patterns which were clearly different from those of their reference prototype strains. Consequently, the database consists of patterns for both community and prototype strains. The database contains patterns for about one-third of the known EV serotypes representing the strains most commonly isolated from the San Diego community and also reported by surveillance groups of the Centers for Disease Control and Prevention (2). As more serotypes are tested the new patterns will be stored in the database as reference strains. In this way the database becomes an evolving library of protein patterns capable of detecting mutational changes that have occurred and that are occurring in the EV genome.

In theory, most of the enteroviral structural (capsid) and nonstructural proteins between 20 and 120 kDa will be seen in a full gel lane profile. The four structural proteins have approximate molecular masses of 30, 27, 24, and 7 kDa for viral proteins (VPs) 1, 2, 3, and 4, respectively (16). For the database we choose to use a molecular mass range of 20 to 40 kDa, and discrimination between and within types was based mainly on the changes in the positions of VPs 1, 2, and 3 (at 7 kDa, VP4 is too small to be seen). Other proteins that fall into the molecular mass range of the database are VP0 (precursor for VP2 and VP4) and nonstructural proteins 2C, 2C^pro, 3C′, and 3D′ (5, 12). By using a full gel lane database profile (molecular mass range, 20 to 120 kDa) the results were similar, although the correlation coefficients were slightly lower (data not shown).

The results for the non-PV EVs indicate that community strains more closely match another community strain of the same serotype than prototype strains. These differences are not surprising considering the reputed high mutation rate (10⁻³ to 10⁻⁴) of RNA viruses (8) and the fact that the majority of the prototype strains were isolated between 1947 and 1957. The relationship between prototype strains and some of the clinical non-PV EVs is illustrated in the dendrograms in Fig. 1 and 2 and is based on similarity coefficients derived when the positions and intensities of the proteins are compared. The points at which the similarity levels join have high coefficients for the duplicate strains, but the relationship of the duplicate strains to their respective prototype strains shows various degrees of distance, as seen by lower similarity coefficients.

Seven strains of ECHO 6 and strains belonging to serotypes ECHO 2 and CBV-4 have high levels of matches with their prototype strains. ECHO 2 is not commonly isolated (nor is CBV) (2, 24), and this may be the reason why these strains are still prototype-like (i.e., not often subjected to selective pressures of repeated infections or replications and/or frequent exposure to circulating antibodies). However, ECHO 6 and CBV-4 are frequently isolated (2, 24), and why some of these strains are still prototype-like is unclear and suggests that EVs do not all evolve at the same rate.

The results of the PV matches are exactly as expected. The last case of poliomyelitis associated with wild-type PV isolation in the Americas was in 1991 (1), and only vaccine strains are circulating in the community, hence the very high levels of matches with the Sabin strains.

Four strains in this present study were misidentified, and an incorrect best match was recorded. These were two strains of ECHO 30 and single strains of ECHO 31 and CAV-9. The ECHO 30 strains recorded a best match of an r of 0.78 and were considered to be antigenic variants of ECHO 30. The ECHO 30 group has been studied further and will be the subject of a future report. ECHO 31 recorded a match of an r of 0.78 with an ECHO 30 strain and CAV-9 a match of an r of 0.92 with an ECHO 5 prototype strain. More strains of ECHO 31 and CAV-9 need to be tested to check the validity of these results.

The method of protein fingerprinting is simple, rapid, inexpensive, objective, and free of organism-specific reagents. It can be used in general clinical microbiology laboratories as well as virology and reference laboratories for identification, typing, and molecular epidemiology. The technique has been used successfully (i) in epidemiological studies and for the identification of various bacterial species (25), (ii) for the identification of ECHO 22 variant strains from a collection of untypeable EVs from the Viral and Rickettsial Disease Laboratory, California Department of Health Services, Berkeley (21), and (iii) as a preserotyping screening method for SDCPHL, resulting in considerable savings in time and resources. In the present application it has proved to be specific, reproducible, and 97% accurate. It is ideally suited for characterizing variations within EV strains and is an appropriate tool for rapid epidemiological surveillance.