Diagnostic Strategy Used To Establish Etiologies of Encephalitis in a Prospective Cohort of Patients in England

Abstract

The laboratory diagnostic strategy used to determine the etiology of encephalitis in 203 patients is reported. An etiological diagnosis was made by first-line laboratory testing for 111 (55%) patients. Subsequent testing, based on individual case reviews, resulted in 17 (8%) further diagnoses, of which 12 (71%) were immune-mediated and 5 (29%) were due to infection. Seventy-five cases were of unknown etiology. Sixteen (8%) of 203 samples were found to be associated with either N-methyl-d-aspartate receptor or voltage-gated potassium channel complex antibodies. The most common viral causes identified were herpes simplex virus (HSV) (19%) and varicella-zoster virus (5%), while the most important bacterial cause was Mycobacterium tuberculosis (5%). The diagnostic value of testing cerebrospinal fluid (CSF) for antibody was assessed using 139 samples from 99 patients, and antibody was detected in 46 samples from 37 patients. Samples collected at 14 to 28 days were more likely to be positive than samples taken 0 to 6 days postadmission. Three PCR-negative HSV cases were diagnosed by the presence of virus-specific antibody in the central nervous system (CNS). It was not possible to make an etiological diagnosis for one-third of the cases; these were therefore considered to be due to unknown causes. Delayed sampling did not contribute to these cases. Twenty percent of the patients with infections with an unknown etiology showed evidence of localized immune activation within the CNS, but no novel viral DNA or RNA sequences were found. We conclude that a good standard of clinical investigation and thorough first-line laboratory testing allows the diagnosis of most cases of infectious encephalitis; testing for CSF antibodies allows further cases to be diagnosed. It is important that testing for immune-mediated causes also be included in a diagnostic algorithm.

INTRODUCTION

Encephalitis is a multifactorial syndrome and is a rare complication of many infections. There is no simple model of exposure leading to disease, but invasion of the central nervous system (CNS) by a pathogen (primary infection or reactivation) is considered the most common mechanism. Immune-mediated etiologies are also now increasingly recognized (6, 7, 10, 17, 23). Herpes simplex virus (HSV) and varicella-zoster virus (VZV) DNA sequences may be detected by PCR more often than other virus infections in cerebrospinal fluid (CSF) specimens from individuals with encephalitis (20). However, the etiology is unclear in many patients, despite recent advances in diagnostic testing (5, 8, 9, 22, 24). One survey estimated that up to 60% of the cases in England were of unknown etiology (4), while a meta-analysis of the literature showed that worldwide up to 85% of the cases were reported to have unknown causes and that there was great variation from region to region in both the recognized causes and the number of undiagnosed cases (11). Some of this variation is attributable to the geographic area in which the encephalitis occurs and which viruses are endemic to that area. For example, the occurrence of tick-borne encephalitis (TBE), West Nile fever, and Japanese encephalitis (JE) is dependent on the ecological distribution of the insect vector harboring the specific arbovirus. For other viruses, their involvement in neurological disease is only now becoming apparent, e.g., hepatitis E virus (HEV) (18).

In an attempt to understand more about the etiology of encephalitis in England, a prospective cohort study was started in 2005 (11, 12). While the main conclusions of this study have been described elsewhere (10), we now report the diagnostic strategy that was used for testing and that was carried out at local and referral centers. This extensive testing was aimed at reducing the number of cases of unknown etiology, to help establish what might be the best approach for a laboratory diagnostic algorithm, and also to investigate whether any unrecognized pathogens or other causes previously unrecognized as potential causes of encephalitis could be implicated in any of the patients recruited to this study.

MATERIALS AND METHODS

Study outline and sample collection.

The specimens used in this prospective study were collected for a 2-year period from 24 hospitals in England between 2005 and 2008. There were two stages of patient recruitment. Initially, any patient in whom a clinical suspicion of encephalitis was indicated was recruited. Subsequently, patients were included based on a syndromic case definition that required admission to a hospital with encephalopathy, defined as an altered level of consciousness persisting for more than 24 h and including lethargy, irritability, or a change in personality and behavior, and two or more of the following: fever, seizures and/or focal neurology, CSF pleocytosis, and electroencephalography or neuroimaging abnormalities in keeping with encephalitis (10). Approximately 1,500 samples, including CSF, blood or serum, urine, feces, throat swabs, and postmortem samples were collected (Fig. 1) and archived at −80°C at the coordinating laboratory (Centre for Infections, Health Protection Agency).

Fig. 1.

Sample combination present in the archive from all 268 recruited cases. In the majority of the cases, there were multiple samples from each patient. For some patients, paired CSF and serum samples were received; for others, only serum or CSF samples were sent. Other* indicates samples different from CSF or serum, for example, tracheal aspirate or feces.

Laboratory testing.

The laboratory diagnosis of viral infections in this study was predominantly by PCR and serology following the standard methods used in the laboratories that carried out the first-line tests. Subsequent investigations also involved PCR and serology, with the exception of bacterial culture for some samples. All of the testing laboratories employed similar quality control systems and were fully accredited by the UK Clinical Pathology Accreditation scheme at the time of this study. Laboratory diagnostic methods used to test samples from patients with encephalitis have recently been reviewed (3, 5, 13, 14, 28, 30, 32). First-line tests involved the more commonly recognized causes of encephalitis (HSV, VZV, enterovirus/parechovirus, adenovirus), with a modified version for immunocompromised patients; viruses that were under surveillance, e.g., West Nile virus (WNV); and infections indicated by the travel history of the patient. The testing algorithm used has already been described (10). The first-line tests were usually carried out at frontline testing laboratories; others were carried out at the HPA Centre for Infections or specialist laboratories after case review by a panel of experts. A pan-enterovirus PCR was used to detect all known types of enterovirus, echovirus, and coxsackie A and B viruses; a separate parechovirus PCR (detecting parechoviruses, including the viruses formerly called echoviruses 22 and 23) was run concurrently. Adenovirus PCR detected all species A to G types. Flavivirus testing (carried out if the patient had an appropriate travel history) included a number of separate assays covering dengue virus, JE virus, WNV, Murray Valley virus, Kunjin virus, TBE virus, and yellow fever virus. Paired acute- and convalescent-phase serum samples, defined as two samples collected a minimum of 2 weeks apart, were used for influenza virus antibody testing, which was performed only if the illness occurred during the influenza season. Patients aged over 50 years with encephalitis of unknown etiology were tested for WNV antibodies, and all deceased patients in whom a cause was not identified were tested for rabies lyssavirus antibodies. Appropriate investigations, including ribosomal DNA PCR assays, were carried out to exclude bacterial, fungal, and parasitic infections.

Review of cases by a multidisciplinary group of experts ensured that all of the patients recruited to this study met the case definition. Further testing for rarer causes of encephalitis in the United Kingdom was proposed, including, for example, parainfluenza virus and norovirus. The parainfluenza virus PCR assay employed detects parainfluenza virus types 1, 2, and 3, while the norovirus PCR used detects genogroups 1 and 2. Cases of acute disseminated encephalomyelitis (ADEM) were identified primarily by neuroimaging. Additional tests, such as encephalitis-associated autoantibody tests, were used in cases where no evidence of infection had been found on initial screening and were carried out by the Neuroimmunology Group, Nuffield Department of Clinical Sciences, Oxford. Factors taken into consideration by the expert panel to guide diagnostic testing included the clinical details, exposures 1 month prior to symptom onset (e.g., vaccination), laboratory tests already carried out, and the amount and type of available specimens. Etiology was assigned based on etiological case definitions designed for the purpose of this study and described by Granerod et al. (13).

Intrathecal antibody testing.

Oligoclonal banding in paired CSF and serum samples from 99 patients, where the appropriate specimens were available, was tested for using isoelectric focusing and nitrocellulose immunoblotting (30, 31). All samples were initially tested for oligoclonal bands with immunodetection of total IgG Fc. This gives oligoclonal band results irrespective of antigen specificity and is termed total oligoclonal IgG banding (tOCB).

CSF samples from 28 patients (where appropriate specimens were available) were also screened for microbe-specific antibodies by immunoblotting against a variety of viral and bacterial antigens (26). The target profile routinely employed comprised cytomegalovirus (CMV), HSV-1 and -2, measles virus, mumps virus, VZV, rubella virus, Toxoplasma gondii, and Mycoplasma spp. On occasion, antigens from T. gondii and rubella virus were not used, and in their place were antigens from Epstein-Barr virus (EBV) and adenovirus.

CSF samples positive for any antigen in the screening test were then tested, using the corresponding paired serum samples, for antigen-specific oligoclonal bands by isoelectric focusing with immunoblotting using antigen-coated nitrocellulose (2, 25). This detected microbe-specific oligoclonal IgG banding, termed sOCB, where s (specific) indicates the particular organism under discussion. Comparison of the sOCB and tOCB patterns indicates their concordance. Strong similarities between the two patterns (in terms of bands and relative intensities) give high concordance and a greater likelihood that the microbe in question is the main antigen in the tOCB pattern. Conversely, a weak concordance (poor correlation between the sOCB and tOCB patterns) implies that additional factors are implicated in the development of the tOCB pattern.

The OCB patterns in CSF and serum samples from the same patient allowed classification into four types. In the first, CSF and serum are both negative (C⁻ S⁻) for oligoclonal IgG. In the second, CSF is positive and serum is negative (C⁺ S⁻), where the IgG present in the CSF compartment is presumed to result from antigenic stimulation within the CNS. This pattern indicates pure intrathecal (local) synthesis of the oligoclonal IgG. In the third, CSF is positive and serum is positive (C⁺ S⁺), with identical band patterns. This is attributable to antigenic stimulation in the peripheral compartment with passive transfer from the serum to the CSF—this is also referred to as the mirror pattern; there is no local synthesis. In the fourth type, CSF and serum both show bands but the CSF contains additional bands, indicating intrathecal synthesis (C⁺ > S⁺). In this study, OCB positivity is defined as either C⁺ S⁻ or C⁺ > S⁺. This relates to tOCB in some cases and to sOCB in others, as indicated below.

Alphaherpesvirus antibody testing.

When sample volumes permitted, serum samples from patients with HSV and VZV encephalitis were tested for antibodies by enzyme immunoassay (EIA) to try to distinguish primary and recurrent infections. Paired serum samples were used to determine whether a primary HSV infection had occurred. In some cases, HSV-1 infection was distinguished from HSV-2 infection by EIA. All cases of unknown etiology and ADEM were also tested for HSV and VZV seroconversion. For VZV, the avidity of the IgG antibody response was measured (19, 21).

HEV testing.

Selected blood and CSF samples were tested for the presence of HEV RNA and antibodies as previously described (15, 16, 18). Briefly, an EIA (Fortress Diagnostics Limited, Antrim, Northern Ireland) was used for antibody testing, and RT-PCR was used for RNA testing (15).

Metagenomics and next-generation sequencing.

Thirty-six samples from patients with encephalitis of unknown etiology were analyzed as previously described (33, 34). In addition, a CSF sample known to contain VZV DNA (confirmed by PCR) was included as a positive control. Briefly, samples were purified by centrifugation and filtration. Any extraneous DNA was removed by DNase treatment (the viral nucleic acid being protected within the capsid), and the viral nucleic acid was extracted using a QIAamp viral RNA extraction kit (Qiagen, Valencia, CA) and subsequently amplified using sequence-independent methods (34). The resulting PCR products were sequenced directly using GS-FLX 454 pyrosequencing, (Roche, Indianapolis, IN). The data were then processed, assembled, and analyzed (34). Contigs were assembled using Sequencher (Gene Codes) software with a threshold of 95% identity over 35 bp and then analyzed against the BLAST databases, with an E value of 10⁻³ being considered significant.

Statistical analysis.

The time (in days) from hospital admission to the first lumbar puncture (LP) was compared between encephalitis patients with known and unknown etiology to assess whether there was a difference in their management that may have resulted in delayed laboratory investigation. Differences in distributions were assessed by the two-sample Wilcoxon rank-sum (Mann-Whitney) test. A P value of less than 0.05 was considered statistically significant. This analysis was repeated by including only cases with an infectious cause diagnosed by PCR as the comparison group with a known etiology to assess whether cases with an unknown etiology may be due to undetected infectious causes but taking of samples at a suboptimal time in the disease process. Patients were stratified by age into groups <18 years and ≥18 years old, and the analysis was rerun. To further investigate oligoclonal banding, times from hospital admission to the first LP were categorized into the following groups: (i) 0 to 6 days, (ii) 7 to 13 days, (iii) 14 to 28 days, and (iv) >28 days. Differences in the proportion of samples positive for OCB (C⁺ S⁻ or C⁺ > S⁺) between categories were assessed by the chi-square test.

RESULTS

Sample data set.

A total of 203 patients were included in this study. Almost 70% (n = 134) were 5 to 64 years old, with slightly more males (n = 109; 54%) than females (n = 94; 46%) (10). Approximately 1,500 clinical samples (including some postmortem samples) were collected, with at least one CSF sample and one blood component sample from 70% of the patients; the other 30% were evenly distributed as either only CSF or only blood component samples (Fig. 1). Paired serum samples were collected from one-fifth of the patients, allowing further analysis, including influenza virus and HSV antibody detection.

Laboratory diagnosis of infectious agents.

The results of the first-line testing allowed a diagnosis to be made for 111 of the patients (Fig. 2). Of these cases, 55% were caused by more commonly recognized viral and bacterial infections while the remainder were caused by other pathogens. The review panel evaluated all of the cases and their laboratory results; recommended additional testing for other pathogens and also for N-methyl-d-aspartate receptor (NMDAR) and voltage-gated potassium channel (VGKC) complex antibodies, and classified some cases as ADEM. Finally, after the results of the first-line and additional tests were reported and an expert review was carried out, a cause was assigned for 128/203 (63%) of the patients (Fig. 2).

Fig. 2.

Diagnostic flowchart.

In all, evidence for 16 different viral infections was found, but these were not all necessarily considered to be responsible for encephalitis (10, 13). These were HSV-1 and -2, VZV, EBV, HHV-6 and 7, enterovirus, influenza virus A and B, parainfluenza virus, RSV, HIV-1, norovirus, JC virus, measles virus (subacute sclerosing panencephalitis [SSPE]), rotavirus, and adenovirus infections. WNV infection was detected in a tourist who had been infected outside the United Kingdom. Tests were carried out for other viruses with negative results for mumps virus, CMV, lymphocytic choriomeningitis virus, parvovirus B19, flaviviruses, HEV, human T-cell leukemia virus I, parechovirus, rabies virus, and Toscana virus. Among the nonviral infections, 12 different bacteria and one fungus were identified, namely, Mycobacterium tuberculosis, Mycoplasma pneumoniae, Streptococcus pyogenes, T. gondii, Streptococcus pneumoniae, Neisseria meningitidis, Streptococcus agalactiae, Coxiella burnetii, Enterococcus faecium, Campylobacter jejuni, Listeria monocytogenes, Pseudomonas aeruginosa, and Cryptococcus neoformans, and tests were carried out for a further 10 pathogens, including rickettsiae, fungi, and parasites, and were negative for Bartonella henselae, Chlamydia trachomatis, Leptospira spp., Rickettsia rickettsii, Bacillus anthracis, Chlamydophila psittaci, Legionella pneumophila, Borrelia burgdorferi, Salmonella spp., and Histoplasma capsulatum.

ADEM.

Twenty-three patients in this study were classified as having ADEM (10); only eight had evidence of a preceding infection, including influenza virus, M. pneumoniae, S. pyogenes, and C. jejuni.

Other immune-mediated cases.

If an infectious cause was not identified and if there was a sufficient volume of serum, tests for NMDAR and VGKC-complex antibodies were carried out (n = 46). As a result, 16 cases were identified, of which 7 had antibodies to VGKC complexes and 9 had antibodies to NMDAR. Three of the 203 cases were ascribed to other immune-mediated etiologies, namely, multiple sclerosis, systemic vasculitis, and paraneoplastic disease (10).

Cases of unknown etiology.

In all, 86 patients were diagnosed with an infectious cause and a further 42 had an immune-mediated cause. Seventy-five patients remained with no evidence of a recognized etiology for their encephalitis (Fig. 2). For a third of these patients, exhaustive testing for all likely causes was possible, whereas the sample volumes were insufficient for two-thirds. Pathogen discovery was attempted by the metagenomic techniques of Delwart and colleagues for samples from 36 of these patients considered to be the most likely candidates (33, 34). VZV sequences were observed in the positive-control sample, confirming that the experiments were technically successful. No known or novel pathogen sequences were identified in sufficient detail to be considered to indicate a possible etiology when compared to sequences in the BLAST databases.

Intrathecal antibody testing: intrathecal production of IgG oligoclonal bands.

To investigate whether intrathecal IgG synthesis in the CSF is a useful diagnostic marker of encephalitis, 139 samples were tested for tOCB. Samples from 46 patients with known etiologies (including 12 with HSV, 7 with VZV, and 1 with T. gondii), 13 patients with ADEM, and 40 with unknown etiologies were tested. For 20 patients (8 with a known etiology and 12 with an unknown etiology), OCB testing was carried out with multiple samples from the same patient.

Forty-two (36%) of 118 samples with information on when the sample was collected were tOCB positive. Four further samples for which there was no information on sample timing were tOCB positive, and these were excluded from the analysis. The highest proportion of tOCB positivity occurred among samples collected 14 to 28 days postadmission (10 of 16; 62.5%), whereas the lowest proportion occurred among samples collected 0 to 6 days postadmission (16 of 59; 27.1%). This difference was statistically significant (P = 0.01). Twenty percent of the cases with an unknown etiology were tOCB positive. The proportion of samples positive for tOCB from cases of unknown etiology increased from 13% (3 of 23) in samples taken 0 to 6 days postadmission to 57% (4 of 7) in samples taken 14 to 28 days postadmission (Fig. 3).

Fig. 3.

Proportion of OCB-positive samples in samples from cases of known versus unknown etiology within each timing category. Times from hospital admission to the first LP were categorized into the following groups: 0 to 6 days, 7 to 13 days, 14 to 28 days, and >28 days. OCB positive refers to C⁺ S⁻ or C⁺ > S⁺.

However, no apparent correlation between etiological status and the oligoclonal antibody banding pattern was found. Importantly, no predominant tOCB pattern could be identified for the 40 cases with unknown etiology.

For the serial samples from patients with proven etiology, two became positive after days 5 and 77 and one became negative after day 118. For the serial samples from patients with unknown etiology, two became positive after days 2 and 11 and one became negative after day 51.

Intrathecal antibody testing for sOCB.

By immunoblotting, 3 of 12 samples from patients with HSV were positive for HSV antibodies (sOCB) in the CSF. HSV DNA was negative in all of these, so the diagnosis was made on the basis of these sOCB results (13). The positive sOCB results were obtained with CSF samples collected 10, 17, and 22 days postadmission; these same specimens were PCR negative. VZV antibody was not detected by immunoblotting in any CSF samples from patients infected with VZV. SSPE was confirmed by demonstrating intrathecal production of measles-specific IgG. Mycoplasma-specific intrathecal IgG production was found in one patient with ADEM.

Alphaherpesvirus infections.

The alphaherpesviruses HSV and VZV were the most common infectious agents identified in this study. Analysis of the clinical data and the laboratory results, which included tests for herpesvirus DNA in CSF by PCR, HSV IgG and IgM antibodies, intrathecal HSV-specific IgG production (sOCB), and virus by culture, allowed the infections to be classified as confirmed, probable, or possible (10). The majority (88%) were considered confirmed. For 3 of 10 patients from whom paired (acute- and convalescent-phase) serum samples had been collected, a primary infection was demonstrated by the detection of HSV antibodies in the convalescent-phase, but not in the acute-phase, sample. These three adults were 43, 45, and 65 years old; two were HSV-1 infected, and the other was HSV-2 infected and was also positive for HSV DNA in the CSF by PCR. HSV IgM was detected in a serum sample from a neonate from whom HSV-1 was also isolated via a swab sample (origin unknown) and was considered to indicate a primary infection. HSV IgG was detected in 24 of 29 serum samples from patients with HSV-1 infections where serum was available for analysis. Based on this finding, most of the patients with HSV encephalitis in this study were considered to have viral reactivation rather than primary infections.

For 8 out of 10 patients with VZV encephalitis, there was a sufficient volume of serum for VZV IgG tests, in addition to the testing of CSF for VZV DNA. One patient had a primary VZV infection, as VZV IgG was not detected in the baseline serum sample, in contrast to the others that were VZV IgG positive. There were two sets of paired serum samples in which VZV IgG was detected in both acute- and convalescent-phase samples. For 6 patients, there was sufficient sample volume to allow measurement of VZV IgG avidity. Three had VZV IgG of low avidity, and three had VZV IgG of high avidity. Therefore, considering these results together with the clinical data, the infections of three patients were primary and three were due to viral reactivation. Three patients had chickenpox (varicella) 1 week prior to admission; their ages were 6, 8, and 34 years. The three patients with herpes zoster were immunosuppressed; two were HIV positive, and the other was on chemotherapy.

CSF samples from 12 HSV-infected patients were tested for intrathecal antibodies. Five were C⁺ S⁻, one was C⁺ S⁺ (the mirror pattern), two were C⁻ S⁻, one gave conflicting results, and for three, the banding pattern was insufficiently clear to be interpretable. CSF samples from seven VZV-infected patients were also tested for intrathecal antibodies. One was C⁺ S⁻, three were C⁺ S⁺, and three were inconclusive. Only one CSF sample from the three primary VZV infections was available for intrathecal antibody testing, and it was C⁺ S⁺.

Antialphaherpesvirus serological testing in cases of unknown etiology and ADEM.

Serum samples from 38 patients classified as having encephalitis of unknown etiology were tested for evidence of HSV infection by detection of specific antibodies. One sample was found to be positive for HSV IgM, and there was evidence of seroconversion to HSV in another pair of samples. Based on the clinical data, however, both patients were considered by the expert review panel to be possibly rather than definitively HSV infected (10). Three-quarters of the samples of unknown etiology tested were positive for HSV IgG, but these were considered to be past infections not relevant to the encephalitis cases. For the samples tested from patients with ADEM, HSV IgG was detected in 5, indicative of a past infection, and there was no evidence of primary HSV infection.

Testing of serum samples from patients with encephalitis of unknown etiology revealed one with VZV IgM and one with VZV IgG antibody of equivocal avidity. Neither result was considered diagnostic of VZV encephalitis when all of the clinical data were considered, and so the patients were classified as possibly rather than definitely having VZV encephalitis (10). In total, VZV IgG was detected in serum samples from 26/32 (81%) of those with encephalitis of unknown etiology. This is comparable to the 88% positivity for VZV IgG among the patients previously diagnosed with VZV. Thirteen ADEM patients could be tested for VZV IgG; none had clinical signs of primary infection, and 9/13 (69%) were positive for VZV IgG.

Sample timing, management of patients with known diagnoses versus those with unknown etiologies, and time to first LP.

The time from hospital admission to the date of the first LP did not differ between cases of known etiology and those of unknown etiology (P = 0.19). The median number of days was 1 (range, 0 to 74) and 2 (range, 0 to 74) for cases of unknown and known etiology, respectively (Fig. 4). Over 50% of the cases with an unknown etiology had multiple CSF samples taken, significantly more than the cases with a known etiology (32%; P = 0.02). These data suggested that there was no difference in management (based on clinical sampling) between these two groups on hospital admission, which might have explained the failure to make an etiological diagnosis.

Fig. 4.

Time from hospital admission to date of first LP for all encephalitis cases with known etiology (n = 116) and those with unknown (n = 74) etiology.

Sample timing and PCR.

We investigated whether the timing of the PCR testing influenced the diagnosis by comparing the cases of infectious encephalitis diagnosed by initial PCR (five cases positive in the second CSF were excluded from the analysis) to those of unknown etiology (Fig. 5). The median number of days from admission to the first positive CSF collection for cases of known etiology (n = 47; 2 days [range, 0 to 20 days]) did not differ (P = 0.14) from the median number of days for the cases of unknown etiology (n = 74; 1 day [range, 0 to 74 days]).

Fig. 5.

Time from hospital admission to date of first positive CSF for encephalitis cases of known etiology (n = 47) and time to first LP for cases of unknown etiology (n = 74).

For cases of known etiology, the median time to a PCR diagnosis was 1 day (range, 0 to 13 days) for children (n = 15) and 2.5 days (0 to 20) for adults (n = 32). For the cases of unknown etiology, the median time to the first CSF collection for children (n = 29) was 2 days (range, 0 to 13 days) and for adults (n = 45) it was 1 day (range, 0 to 74 days). There was thus no significant difference between the time to a first positive result for cases with a known etiology and the time to the first CSF collection for cases with an unknown etiology in children (P = 0.88) or adults (P = 0.07).

HEV.

HEV infection has recently been linked to neurological disorders (18), and we therefore tested as many of the samples from patients with encephalitis of unknown etiology as possible for evidence of HEV infection. HEV RNA and/or IgM and IgG tests were carried out for 54 of the 75 patients with encephalitis of unknown etiology.

HEV RNA was not detected in any of the 35 CSF samples tested or in plasma samples from 18 of these patients. Evidence of infection, as indicated by the presence of HEV IgG and/or IgM in serum was investigated. Serum samples from 28 of the patients who had been tested for HEV RNA by RT-PCR and from an additional 17 patients were available. Of these 45 samples, 10 were positive for the presence of HEV IgG, indicating past infection, and 1 was equivocal. HEV IgM was detected in one of the IgG-positive samples, indicating recent infection. The CSF intrathecal IgG tOCB pattern was C⁺ S⁺. However, no HEV RNA was detected in the CSF or serum from this patient, a 71-year-old woman with no record of gastrointestinal symptoms or liver dysfunction.

DISCUSSION

A previous study of encephalitis in England estimated that no cause could be identified for 60% of the patients tested (4). The comparable final value for the work reported herein is 37%. This reduction can be accounted for by four factors: more rigorous first-line testing, case review by multidisciplinary experts (the review panel in this study), exclusion of nonencephalitis mimics, and increased recognition of immune-mediated encephalitis. For the first factor, after the initial laboratory testing, the proportion of cases with unknown etiology was 45% (Fig. 2). This reduction from the expected 60% of the samples from patients with encephalitis of unknown etiology based on the survey by Davison and colleagues (4) may represent technical advances, as well as improved clinical surveillance and the better data quality of a prospective study than the previous retrospective data survey.

Second, the extensive testing requested by the review panel also resulted in an improved diagnostic yield for those patients with encephalitis of unknown etiology. These additional diagnoses included viral infections, ADEM, and immune-mediated disease. This demonstrates that noninfectious causes of encephalitis are an important component of the diagnostic algorithm. Therefore, we conclude that comprehensive first-line testing, including tests for immune-mediated noninfectious causes of encephalitis, is a sound basis for an algorithm that would maximize the diagnostic yield and inform patient management.

As far as sample volumes allowed, rigorous, comprehensive testing was carried out for 24 patients with encephalitis of unknown etiology, but no infectious etiologies were identified. It has previously been demonstrated that CSF sampling very early (<3 days after symptom onset) or late (>14 days after symptom onset) in the disease process may reduce the likelihood of a positive PCR result (3) However, our study does not indicate a difference between the cases with a known etiology and the samples from cases with an unknown etiology in terms of sample timing (Fig. 3 to 5). Indeed, more samples were collected from the cases with an unknown etiology, presumably due to difficulty in making a diagnosis. For some of these cases, there might not have been sufficiently high viral loads for DNA or RNA detection by PCR or RT-PCR. Our study showed that the median time to PCR positivity for samples from patients with a known encephalitis cause was 2 days after hospital admission, similar to previous findings, i.e., from 3 days after symptom onset (3).

In one-fifth of the cases with an unknown etiology, tOCB production was found in the CSF, consistent with a localized CNS inflammatory response. Intrathecal oligoclonal IgG production is reported in both infectious and inflammatory disorders of the CNS, including immune-mediated and postinfectious encephalitis. Therefore, the presence of intrathecal production of IgG in one-fifth of the cases of infection with an unknown etiology indicates that noninflammatory syndromes such as stroke and brain tumors, which can often mimic encephalitis, are unlikely to be responsible. Rather, it provides very strong evidence of intrathecal immune activation, although whether this activation results from yet-to-be recognized infectious or immune-mediated processes needs further investigation.

It is striking that NMDAR and VGKC complex antibody encephalitis occurred in 16 patients (10). These antibodies have been identified or recognized as a cause of encephalitis only in the past few years. The spectrum of autoantibodies has yet to be fully characterized, so it is possible that further autoantibodies could be associated with encephalitis (6, 7, 17, 23).

HEV IgG was detected in 22% of the samples tested, which is what would be expected in this subgroup of patients, the majority of whom who were tested for HEV were over 50 years of age (16). HEV IgM was detected in serum from one of the cases with an unknown etiology, compatible with a recent HEV infection as a possible cause of this patient's encephalitis. HEV RNA was not detected in the serum or CSF, but these samples had been frozen and thawed several times, which would have compromised the PCR test.

It is also possible that unidentified pathogens could cause encephalitis, i.e., novel or emerging viruses, an argument that is consistent with OCB being found in one-fifth of the patients with encephalitis of an unknown etiology. Viral metagenomic studies using next-generation sequencing were undertaken to try to identify any novel pathogens present in 36 samples with no other etiology (33). Although no new viral sequences were identified, the possibility of their presence at a low titer in some of the samples cannot be excluded.

We considered whether the failure to identify the etiology of an infection could have been due to sample collection at an inappropriate time. Another CSF sample may need to be taken if the CSF sample collected at presentation is PCR negative. The converse is true of the CSF antibody studies, as the chances of detecting antibody are higher the longer the time since symptom onset. Also, antimicrobial therapy may eliminate the etiological agent below the sensitivity levels of the laboratory test. HSV DNA is rarely detected after 2 weeks of antiviral therapy (1). Ideally, acute-phase CSF and serum samples should be taken prior to treatment, with subsequent samples, especially convalescent-phase serum samples, taken throughout the progression of the disease to demonstrate possible seroconversion (29). We found, however, that delayed sampling did not seem to contribute to cases being undiagnosed (Fig. 3 to 5). Rather, it seemed that the time of sampling for the patients with encephalitis of unknown etiology was earlier than for those with an etiological diagnosis, but the difference was not statistically significant. Furthermore, a greater proportion of those with an unknown etiology had repeat or convalescent-phase samples obtained.

Testing for serum IgM or changes in IgG levels may help clinically, as viremia, and hence the possibility of the detection of nucleic acid by PCR, may occur before symptom onset, as seen in WNV infections (5). Some serological tests may not, however, be sensitive or specific enough to determine the causality of an agent, as they cannot distinguish among a primary infection, a prior vaccination (e.g., against yellow fever virus), or reactivation of an agent. Vaccines given 1 month prior to symptom onset were excluded as a cause in this series.

In this study, some test results were positive but were not considered to indicate the causal agent by the review panel. Positive results do not necessarily indicate that the pathogen identified is the cause of the patient's encephalitis (13). They must be interpreted in the context of the epidemiological, clinical, and other diagnostic results and findings. The determination of the cause of encephalitis in this study was discussed previously by Granerod and colleagues (10, 13).

To assess the benefit of testing for microbe-specific antibodies in the CSF, extensive intrathecal IgG OCB testing with immunoblotting was carried out with samples selected by the review panel. An etiological cause was identified for five cases (three due to HSV, one due to M. pneumoniae [ADEM], and one due to measles virus [SSPE]). These positive results show that the techniques have some merit. Moreover, a further 20% of the patients with encephalitis of unknown origin that were tested had intrathecal production of IgG, implying the presence of a localized immune response within the CNS. Screening for tOCB may be one way of identifying candidate cases for pathogen discovery, e.g., earlier CSF samples from those who go on to produce tOCB could be used, if available. Ideally, CSF samples collected early in the encephalitic illness should be targeted for metagenomic and next-generation sequencing studies; however, the cases should be selected from those whose convalescent-phase samples show the development of intrathecal IgG production. An alternative strategy is to use a peptide library (27, 36) to seek the putative antigen stimulating the production of intrathecal IgG OCB. This could be a self-antigen (i.e., an autoantibody) or an antigen of a novel microbe.

The proportion of OCB-positive samples among cases of unknown etiology increased from earlier to later postadmission samples (Fig. 3); it can be concluded that an intrathecal immunological process may be present. This supports pursuing further research to identify their specificity, and it is consistent with these cases being real encephalitis cases rather than mimickers. Noninfectious/noninflammatory mimics of acute encephalitis (e.g., strokes, tumors, septic encephalopathy) are not associated with the intrathecal production of OCB.

The patients with HSV encephalitis in whom HSV DNA was detected in the CSF were negative for oligoclonal intrathecal antibodies and by immunoblotting. HSV DNA was not detected in the CSF of those patients whose samples were positive for HSV oligoclonal intrathecal antibody. This suggested that these two tests are complementary and can identify different stages of disease pathology. The same was seen for those with VZV infection, although no tested samples were positive for intrathecal VZV antibodies.

Extensive HSV and VZV antibody testing was carried out with serum samples. Two cases of HSV encephalitis were diagnosed in this way in the first-line testing. Another VZV case was recognized following tests recommended by the review panel. From the serology results, for half of the cases of HSV encephalitis, it was possible to estimate those due to primary versus recurrent infections. About three-quarters of the cases were judged to be due to recurrent infections, while half of the remainder were classified as due to primary infections. From the VZV IgG avidity testing, about half of the VZV cases were judged to be primary (varicella). Thus, the cases of HSV and VZV encephalitis were manifestations of both primary and recurrent infections, with more primary VZV infections than HSV infections being responsible for encephalitic symptoms in this cohort.

No cases of encephalitis attributable to adenovirus, CMV, HHV-7, parechovirus, or mumps virus were diagnosed, and only one probable case of HHV-6 infection was identified in an immunocompromised individual. Primary HHV-6 and -7 infections are both causes of status epilepticus with fever (35) and possibly encephalitis in children below the age of 2 years, but there were insufficient numbers of patients in this age group to pick this up. Adenovirus, CMV, and HHV-6 are all prevalent in the population, so they may be expected as uncommon opportunistic causes of encephalitis in, for example, the immunocompromised, but again the number of immunocompromised patients in our study may have been insufficient to pick this up. Parechovirus infections are uncommon in older children and adults and may not have been circulating within the study population, and the prevalence of mumps has been reduced by vaccination programs.

Based on our findings, we recommend that initial laboratory testing for viral causes should include HSV, VZV, and enteroviruses and should include influenza virus during influenza season. In this study, there were two cases classified as probably caused by influenza virus infections. This number might be expected to be different during epidemic periods and possibly with some strains of influenza virus more than with others. Likewise, when mumps and measles viruses are circulating in the population (or there is a history of travel to areas where they are endemic), they should be tested for, depending on the clinical indications. Similarly, the initial nonviral pathogen laboratory testing should include S. pneumoniae, M. tuberculosis, M. pneumoniae, N. meningitidis, and T. gondii (in immunocompromised patients). Noninfectious causes need to be excluded, and testing for antibodies in CSF may help resolve a few cases. Those CSF samples from patients with encephalitis of unknown etiology that have positive OCB (20% in this study) arguably represent the most promising ones to study in order to identify novel causes of encephalitis.