The Role of Cerebrospinal Fluid Hypocretin Measurement in the Diagnosis of Narcolepsy and Other Hypersomnias FREE

NARCOLEPSY IS characterized by sleepiness and abnormal rapid eye movement (REM) sleep events such as cataplexy, hypnagogic hallucinations, and sleep paralysis.^{1 - 3} Prevalence is in 0.02% to 0.05% of the population,^{4 - 5} with onset generally occuring in adolescence.^{1 - 3} Quality of life is as impaired as in epilepsy or depression.⁶ Most cases are sporadic, but twin and family studies demonstrate the existence of genetic factors.^{1 ,5} Current treatment involves psychostimulants for sleepiness and antidepressants for cataplexy.^{2 ,7} γ-Hydroxybutyrate is used to consolidate nocturnal sleep and reduce cataplexy.^{2 ,7 - 8}

In the International Classification of Sleep Disorders (ICSD),⁹ narcolepsy is defined by sleepiness plus cataplexy, or bythe polysomnographic documentation of REM sleep abnormalities. The most commonly accepted diagnostic test is the multiple sleep latency test (MSLT),¹⁰ in which nocturnal polysomnography is performed, followed by 4 to 5 daytime naps during which sleep latency is measured.^{9 - 12} Untreated patients display short mean sleep latency (MSL) (≤5 or ≤8 minutes) and 2 or more sleep onset REM periods (SOREMPs).^{9 - 12} Using the polysomnographic definition, narcolepsy contrasts with idiopathic hypersomnia, which is a condition characterized by extended sleep time, no REM-related symptoms, and a short MSL without SOREMPs.^{9 ,13}

The nosology of narcolepsy is controversial. First, narcolepsy can be diagnosed by patient history alone.⁹ This may not always be accurate, as cataplexy overlaps with experiences reported by healthy subjects.^{4 ,14} Second, up to 15% of typical narcoleptic patients test negative on the MSLT.^{11 - 12} Third, controls and patients with other sleep disorders may have short MSL and multiple SOREMPs.¹¹ Finally, sleep paralysis and hypnagogic hallucinations are common in the general population.^{15 - 16} The greatest difficulty is distinguishing narcolepsy without cataplexy from idiopathic hypersomnia. A definition based on the presence (narcolepsy without cataplexy) or absence (idiopathic hypersomnia) of 2 SOREMPs on the MSLT is usually adopted,^{9 - 11} but some investigators suggest a continuum between these entities.^{17 - 18}

The observation that narcolepsy was associated with HLA-DR2 was the first suggestion of etiological homogeneity in narcolepsy.¹⁹ Recent studies have shown that HLA-DQB1*0602 is the main susceptibility allele.^{20 - 22} However, 12% to 38% of controls are DQB1*0602 positive.²² Clinical and HLA typing studies have identified cataplexy as the most specific symptom, with probable disease heterogeneity in patients without cataplexy.^{17 ,21} HLA-DQB1*0602 positivity is 90% to 100% in patients with definite cataplexy, but it decreases with atypical cataplexy or no cataplexy (40%).^{3 ,21} Even in cataplectic patients, substantial differences in HLA association exist (70%-100%), suggesting that cataplexy may be overdiagnosed.^{3 ,21}

The pathophysiology of narcolepsy involves abnormal hypocretin (orexin) neurotransmission. In a canine model, the disorder is caused by hypocretin receptor-2 mutations.²³ Preprohypocretin (the precursor to 2 peptides: hypocretin-1 and hypocretin-2) knockout mice have narcolepsy.²⁴ Human narcolepsy is generally not due to gene mutations, but hypocretin neurotransmission is impaired. The condition is associated with undetectable and, more rarely, elevated hypocretin-1 levels in the cerebrospinal fluid (CSF).^{25 - 28} A selective loss of hypocretin messenger RNA (mRNA) and immunoreactivity has also been reported in the hypothalamus of 6 patients.^{29 - 30} Together with the HLA association, these results suggest that most human cases are caused by an autoimmune-mediated destruction of hypocretin neurons.

The hypocretins were originally believed to be regulators of appetite.^{31 - 32} Despite their discrete location in the lateral and perifornical hypothalamus, hypocretin neurons project widely throughout the brain, including dense excitatory projections to monoaminergic cell groups.³³ Hypocretin neurons are therefore uniquely positioned to drive monoaminergic activity across the sleep cycle.^{34 - 37} Loss of this excitatory input may explain the abnormalities seen in narcolepsy.^{35 - 36} The hypocretins may also link energy metabolism and sleep.^{36 - 39}

In this study, we describe the spectrum of the hypocretin deficiency syndrome by analysis of CSF hypocretin-1 levels, and clinical, polysomnographic, and HLA data in more than 250 patients with various sleep disorders. This is the first study to offer definitive information on the use of CSF hypocretin-1 measurement for the diagnosis of narcolepsy.

This was a prospective study. Table 1 presents categories and inclusion criteria. Patients and controls gave informed consent for the study, and lumbar puncture and venopuncture (for HLA typing) were performed. Patients with a sleep disorder (and their relatives, when applicable) were identified at Stanford University (Stanford, Calif; n = 148), Leiden University Medical Center (Leiden, the Netherlands; n = 66), Charles University (Prague, Czech Republic; n = 23), Zürich University Hospital (Zürich, Switzerland; n = 21), University of Southern Mississippi (Hattiesburg; n = 13), and Trondheim University Hospital (Trondheim, Norway; n = 2). Five patients were contributed by Dr Rye (Emory University School of Medicine, Atlanta, Ga). One case with central hypoventilation syndrome and possible narcolepsy was referred by Dr Halbower (John Hopkins University, Baltimore, Md).

Unrelated controls were healthy subjects without a known sleep disorder (n = 64) recruited at Stanford University (n = 43), Charles University (n = 13), and Leiden University Medical Center (n = 8). Patients with various neurological disorders undergoing a lumbar puncture (n = 228) were also included. These included 16 new patients recruited at St Charles (n = 12) and Emory (n = 4), plus 212 previously described patients.²⁷

Sleep disorders were diagnosed clinically and using sleep studies. Patients were classified based on their primary ICSD diagnosis, blind of HLA and hypocretin-1 results. Seventeen controls had their lumbar puncture during nighttime. All others were studied during daytime (Table 1). This study includes 212 controls with various neurological disorders, 15 healthy controls, 22 hypersomnia patients, and 43 narcolepsy-cataplexy patients previously reported.^{25 - 27 ,40 - 41} Patient ethnicity was 88% white (Europeans and North Americans), 7% Asian (Chinese, Japanese, and Korean Americans), 2% Black (African Americans), and 3% Latino and mixed.

In narcolepsy, the existence of "definite" cataplexy vs "atypical or doubtful" cataplexy was confirmed using interviews and a validated inventory.¹⁴ Most subjects were treated with amphetaminelike stimulants, modafinil, antidepressants, and γ-hydroxybutyrate (Table 1). An effort was made to include patients with unusual forms of narcolepsy, 27 subjects from 9 multiplex families (4 with HLA-negative cases), and 7 HLA-DQB1*0602–negative patients with cataplexy.

Idiopathic hypersomnia was diagnosed by ICSD citeria.⁹ Most patients (83%) had typical (MSL <10 minutes, 0-1 SOREMP) MSLT (MSL ± SE = 7.1 ± 0.7 minutes; mean ± SE SOREMPs = 0.07 ± 0.05). Patients with periodic hypersomnia (n = 3) all had more than 6 previous episodes (by ICSD criteria⁹ ; 2 untreated, 1 treated with modafinil). Those with obstructive sleep apnea (OSA) had a respiratory disturbance index of 10 episodes per hour (73% treated with continuous pressure airway pressure therapy). Subjects with severe restless legs syndrome met international criteria^{9 ,42} ; all but 4 were taking dopaminergic medications. Twelve untreated patients with psychophysiological insomnia⁹ were studied. Eighteen patients had secondary hypersomnia or cataplexy. All but 3 had short MSL on the MSLT (MSL ± SE = 6.4 ± 0.8 minutes; mean ± SE SOREMPs, 0.38 ± 0.36; n = 18). We also included 7 subjects with reports of hypersomnia but without confirmatory sleep testing.

HLA-DQB1*0602 was identified as previously described.²² Cerebrospinal fluid was frozen immediately and stored until radioimmunoassays (Phoenix Pharmaceuticals, Mountain View, Calif) were conducted. We measured hypocretin-1 both in reconstituted aliquots from 1 mL of extracted CSF^{25 - 26} and directly using 100 µL of nonextracted CSF.²⁷ The detection limits of the extracted and direct assay were 40 pg/mL and 100 pg/mL, respectively. A standard CSF sample was used to correct for interassay variation.

Analyses were performed using SYSTAT 10.0 (SPSS Inc, Chicago, Ill). Linear or logistic regression modeling was used to assess significance (P ≤ .05), controlling for confounding factors such as sex, age, and ethnicity. Bonferroni corrections were used where applicable. Ranked data were used for hypocretin-1 analysis. Simple correlations were computed using Pearson correlations. Finally, for signal detection analysis, we used quality receiver operating characteristic curves (QROC) to determine hypocretin-1 values giving the best diagnostic sensitivity/specificity ratios (patients with ICSD-defined narcolepsy vs all other subject samples, and healthy controls vs all other subject samples).⁴³

Age, sex, and ethnicity did not differ across groups (Table 1). Higher body mass index (BMI; calculated as weight in kilograms divided by the square of height in meters) was observed in patients with OSA, narcolepsy-cataplexy, atypical or doubtful cataplexy, and idiopathic hypersomnia (Table 2). When compared with controls (18% of 64 subjects), the highest DQB1*0602 values were observed in patients with narcolepsy-cataplexy (93%; odds ratio [OR] = 80; P<.001), followed by those with narcolepsy without cataplexy (56%; OR = 7.0; P<.01), patients with idiopathic hypersomnia (52%; OR = 6.0; P<.01), narcolepsy family members (P<.01), and patients with atypical cataplexy (35%; OR = 3.3; P = .05).

Age, sex, ethnicity, HLA-DQB1*0602, BMI, and time of day did not affect hypocretin-1 levels in healthy controls (Table 2). This confirms data from healthy volunteers using intrathecal catheters in whom lumbar CSF levels were only slightly (15%) higher at night (E.M. and Ron Salomon, MD, unpublished data, 2002).

Directly measured and extracted hypocretin-1 levels were highly correlated throughout the entire sample (r = 0.86, P<.001). Levels were lower in patients with narcolepsy with typical cataplexy, in subjects with atypical cataplexy, and in subjects with neurological disorders (Table 2). A QROC analysis⁴³ was used to determine hypocretin-1 values "diagnostic" for ICSD-defined narcolepsy. An identical hypocretin-1 threshold of 110 pg/mL was obtained for the direct (sensitivity = 60%, specificity = 98%, predictive value for a positive test [PVP] = 94%, predictive value for a negative test [PVN] = 84%; n = 570) and extracted (sensitivity = 60%, specificity = 98%, PVP = 98%, PVN = 83%; n = 379) assays. The QROC analysis indicated a threshold of 200 pg/mL and 150 pg/mL for the direct and extracted assays, respectively, of healthy subjects.

The simplicity and accuracy of the direct assay makes it more attractive for future use; we therefore used direct CSF results in all further analyses. Subjects were classified as having low (≤110 pg/mL; n = 113), intermediate (>110 pg/mL, ≤ 200 pg/mL; n = 40), and normal hypocretin-1 levels (>200 pg/mL; n = 417). Of note, we did not confirm our finding of high hypocretin-1 (extracted >500 pg/mL) values in narcolepsy.^{25 - 26} In this extended series, high values were also observed in nonnarcoleptic subjects.

Using the 110-pg/mL threshold, 106 of 113 patients with low levels were patients with ICSD-defined narcolepsy (Figure 1). Three subjects had "secondary hypersomnia/narcolepsy" (Table 3). Four subjects had a neurological disorder (Figure 1); 3 had acute Guillain-Barré syndrome, and their cases have been reported²⁷ ; one was a 38-year-old man with a subacute history of change in behavior who became comatose. He was later diagnosed with Hashimoto thyroiditis.

Several groups had mostly normal levels (Figure 1). These included patients with idiopathic hypersomnia (all but one with intermediate levels), OSA, restless leg syndrome, and insomnia. Interestingly, 2 patients with OSA and normal levels had residual sleepiness despite successful continuous pressure airway pressure therapy therapy (1 HLA-negative patient, MSLT SL = 3.9, 1 SOREMP; 1 HLA-positive patient, MSLT SL = 8.0, no SOREMP).

Only 10 subjects in our sleep disorder series had intermediate levels (Figure 1 and Figure 2). Three also had intermediate levels using the extracted assay (≤150 pg/mL) and are most likely to have partial hypocretin deficiency. The first patient was HLA positive with periodic hypersomnia in the midst of an episode. The second had idiopathic hypersomnia and was the son of a patient with narcolepsy-cataplexy (Figure 2, "VER" family); this subject had sleepiness, hypnagogic hallucinations, and sleep paralysis, but a negative MSLT (MSL = 9 minutes, 0 SOREMP). The third is HLA positive with a 9-year history of narcolepsy-cataplexy, sleep paralysis, and hypnagogic hallucinations (MSLT was atypical, with an MSL of 5.4 minutes, but no SOREMP). The other 7 subjects included 5 subjects with narcolepsy-cataplexy (3 HLA negative, 2 with atypical cataplexy), 1 with idiopathic hypersomnia, and 1 with hypersomnia undocumented by polysomnography.

Thirty of 228 subjects in our neurological disorder survey had intermediate levels. These previoulsy described patients²⁷ included subjects with head trauma, encephalitis, and Guillain-Barré syndrome. None of the 74 healthy controls had intermediate levels.

Table 4 compares clinical data in narcoleptic patients with (n = 106) and without (n = 65) low CSF hypocretin-1 levels. Note that these subjects do not represent a random sample of narcoleptic patients, and include a higher number of subjects without cataplexy and/or HLA-DQB1*0602. We found that subjects with low hypocretin-1 levels had more typical cataplexy, more abnormal MSLTs, and were more frequently HLA positive. Other characteristics were generally similar. Typical cataplexy was a better predictor of a low hypocretin-1 level than abnormal MSLT. Of the 90 subjects with low hypocretin-1 levels and available MSLT, 77 (86%) had abnormal MSLT, but 11 (12%) had borderline MSLT (SL<8 minutes, 0-1 SOREMPs vs SL>8 minutes, 1 SOREMP), and 2 (2%) had normal MSLT (SL>10 minutes, no SOREMPs). This contrasts with the highly predictive value of cataplexy. Of the 106 narcoleptic subjects with low levels, 97 (92%) had typical cataplexy, 6 (6%) had atypical/doubtful cataplexy, and only 3 (3%) had no cataplexy. We explored the value of low hypocretin levels in narcoleptic patents solely selected on the basis of typical cataplexy (Table 1). The sensitivity and specificity of low hypocretin-1 levels in 101 randomly selected patients with typical cataplexy vs 292 controls was extremely high (87% and 99%, respectively [PVP = 96%, PVN = 96%]). Of note, in 3 subjects with cataplexy and low CSF levels, disease duration was 6 months, 8 months, and 9 months, respectively, but levels were below the detection limit in both assays.

All 3 subjects without cataplexy who had low hypocretin-1 levels were HLA positive. Two patients with low but detectable hypocretin-1 levels (eg, >40 pg/mL and ≤110 pg/mL in the extracted assay) were young narcoleptic subjects (16 and 25 years old, respectively). One case, that of a 16-year-old girl (direct assay = 63 pg/mL; extracted assay = 98 pg/mL), was somewhat atypical; she had narcolepsy without cataplexy following an undetermined viral illness and generalized lymphadenopathy. Remarkably, in the case of a 58-year-old patient with a 45-year history of narcolepsy, there were mild hypnagogic hallucinations and no sleep paralysis. This thoroughly interviewed patient had undetectable levels using both assays in samples from 2 lumbar punctures.

Only 3 narcolepsy subjects with low levels were HLA-DQB1*0602 negative. One subject had a previously reported preprohypocretin gene mutation (included because of HLA negativity and unusually early onset at 6 months).²⁹ The second subject had mild cataplexy, no ancillary symptoms, and a positive MSLT (undetectable levels using both assays in 2 independent lumbar punctures). Onset was unremarkable at 11 years of age, but hypocretin mutation screening was negative (data not shown). The third had mild/atypical cataplexy, no ancillary symptoms, and MSLT indicating borderline sleepiness (MSL = 10.6 minutes, 1 SOREMP).

Nine multiplex families were studied (Figure 2). The results correlated with HLA typing. In 2 families with cataplexy and abnormal MSLT cases (the "EIC" and "RIC" families, respectively), hypocretin-1 levels were normal. One family ("TER") included an HLA-negative subject with normal levels, typical cataplexy, multiple SOREMPs, but no subjective sleepiness. Of special interest was the study of the "DAN" family, a large African American lineage with 7 affected subjects, 6 of whom reported typical cataplexy. The 3 eldest subjects with narcolepsy-cataplexy were hypocretin-deficient, while the other 4, generally younger, had normal CSF hypocretin-1 levels (Figure 2).

Data from these subjects (n = 65) are reported in Table 4. These subjects were younger and without (26%) or with (35%) atypical or doubtful cataplexy. Most were HLA negative (66%), and many had a family history of illness (26%). Importantly, these patients did not have recent disease onset, making it unlikely that they will develop cataplexy. Delays between onset and hypocretin-1 measurements were similar in subjects with low and normal levels.

Eighteen subjects were studied (Table 3, Figure 1). Three had low hypocretin-1 levels — a child with central hypoventilation syndrome and narcolepsylike symptoms, a patient with a probable hypothalamic pathological illness, and a patient with Prader-Willi syndrome. Notably, a subject with Niemann-Pick disease type C and severe cataplexy had normal levels.

In this large study, low CSF hypocretin-1 levels were consistently observed in those with narcolepsy-cataplexy. Two approaches were used — a direct radioimmunoassay and measurements after extraction. Both techniques were reliable and highly correlated. A cut-off of 110 pg/mL (representing 30% of the normal mean value in the direct assay) was determined to have the best sensitivity/specificity ratio. Since the direct assay is more cost effective and less labor intensive, we propose using this test for diagnostic purposes. All subjects with a sleep disorder who tested positive by this criteria had narcolepsy (Figure 1). Importantly, a subset of patients without positive MSLT, without the classic HLA marker, and/or without cataplexy also tested positive.

We found that 3 subjects with secondary hypersomnia (Table 3) had levels below 110 pg/mL, indicating the existence of genuine cases of secondary hypocretin deficiency. This extends on 3 previously reported cases with secondary narcolepsy-cataplexy—one with autosomal dominant cerebellar ataxia, deafness, and the other 2 with large vascular and tumoral hypothalamic lesions.^{44 - 46} Of note, the new cases reported in this study may also have included impaired hypothalamic function. Prader-Willi syndrome, a syndrome characterized by obesity, hypotonia, and daytime sleepiness (not always explained by OSA), has long been suspected to involve hypothalamic abnormalities.^{47 - 49} Similarly, late-onset central hypoventilation syndrome is often associated with hypothalamic dysfunction.⁵⁰ These cases illustrate the value of the CSF test in patients otherwise difficult to diagnose.

This series was also used to describe the clinical spectrum of the narcolepsy with hypocretin deficiency syndrome (Table 4). Almost all subjects with low levels were HLA-DQB1*0602 positive with typical cataplexy. Eighty-five percent had positive MSLT, while 63% reported sleep paralysis, and 86% reported hypnagogic hallucinations. These subjects also had moderately increased BMI (Table 2), as previously reported in humans^{2 - 3 ,38 - 39} and mice with hypocretin cell loss.⁵¹ Some patients had recent onset of narcolepsy (eg, 6 months), indicating that hypocretin deficiency is already established at presentation.

As a diagnostic test, CSF hypocretin-1 measurements could complement and even replace the MSLT in some instances. In cases of definite cataplexy, specificity (99%), and sensitivity (87%) for the CSF test are extremely high. Multiple sleep latency test data in large control samples are unavailable, and the MSLT is influenced by drugs and other sleep disorders. Moreover, 16% of hypocretin-deficient narcoleptic patients did not test positive on the MSLT (Table 4). In contrast, we found that only 4 of 292 controls (all with neurological disorders) had low hypocretin-1 levels. Three had quadriplegic Guillain-Barré syndrome.²⁷ In 1 case, levels increased after recovery, suggesting transitory changes. The fourth subject had myxedema coma secondary to Hashimoto thyroiditis. These severe neurological problems are unlikely to be confused for narcolepsy. Interestingly, both Guillain-Barré syndrome and Hashimoto's thyroiditis occur secondary to autoimmune attacks, which is the proposed cause of narcolepsy. Hypocretin-1 levels were not significantly influenced by the use of psychotropic medications (antidepressants, dopamine agonists, hypnotics, stimulant compounds) or by the presence of other sleep disorders. Hypocretin-1 testing may therefore be most useful in cataplectic patients in whom treatment has been initiated, and/or when other sleep disorders may confound the MSLT (eg, insufficient sleep, OSA).

The MSLT remains the test of choice in patients without cataplexy since very few subjects (n = 3) in this category had low CSF hypocretin-1 levels. A similar result was recently reported in another study.²⁸ Another limitation of the CSF test is the occurrence of post–lumbar puncture headaches in 5% to 10% of the cases. Lumbar punctures are, however, more generally available and may be cheaper than the 2-day MSLT. The MSLT and CSF tests are also difficult to compare, as the MSLT measures a physiological process (daytime sleepiness), while the CSF test measures a neurochemical abnormality. These 2 tests should, therefore, be considered complementary in most cases.

The finding that cases with secondary hypersomnia, periodic hypersomnia and/or atypical narcolepsy have intermediate or low levels is noteworthy. Periodic hypersomnia, often associated with increased appetite and hypersexuality, has been suspected to involve hypothalamic abnormalities,^{52 - 53} and our finding supports this hypothesis. In our survey, 30 of 228 patients with various neurological conditions had intermediate levels. These subjects include patients with head trauma, encephalitis, and Guillain-Barré syndrome, which may alter hypothalamic function,^{27 ,54} result in sleepiness,^{55 - 56} and more rarely, narcolepsy-cataplexy (eg, in the context of the encephalitis lethargica epidemic).^{56 - 60} Intermediate levels may thus point toward a secondary hypothalamic dysfunction. Alternatively, these levels may reflect changes in CSF flow or blood-brain barrier permeability and should be interpreted in the clinical context.

The majority of patients with idiopathic hypersomnia and narcolepsy without cataplexy had normal hypocretin-1 levels. A parsimonious explanation may be that the etiology of these cases does not involve hypocretins, and that narcolepsy with typical cataplexy is a distinct entity. It also indicates that differentiating narcolepsy without cataplexy and idiopathic hypersomnia may not be justified etiologically, as others have suggested.^{17 - 18} Surprisingly, however, HLA-DQB1*0602 frequency was increased in noncataplexy cases (Table 2), paralleling previous reports.^{17 ,21} A small number of patients with typical cataplexy also had normal levels, especially multiplex and HLA-negative subjects (Figure 1 and Figure 2). Hypersomnia and narcolepsy-cataplexy may thus be part of the same disease continuum. In this model, some narcolepsy patients have hypocretin deficiency in projection areas important for sleep regulation, but they still have normal CSF levels, especially younger subjects without HLA-DQB1*0602. Partial hypocretin deficiency would be associated with less severe symptomatology and less frequent cataplexy.

Accepted for publication April 29, 2002.

Author contributions: Study concept and design (Drs Mignot, Lammers, Nevsimalova, Bassetti, and Schrader); acquisition of data (Drs Mignot, Lammers, Nevsimalova, Vancova, Black, Harsh, Bassetti, and Nishino, Mss Ripley and Okun, and Mr Overeem); analysis and interpretation of data (Drs Mignot and Lammers, and Mr Overeem); drafting of the manuscript (Drs Mignot, Lammers, and Harsh and Ms Ripley); critical revision of the manuscript for important intellectual content (Drs Mignot, Lammers, Nevsimalova, Vancova, Black, Bassetti, Schrader, and Nishino and Mr Overeem); statistical expertise (Dr Mignot); obtained funding (Drs Mignot, Nevsimalova, Black, and Bassetti); administrative, technical, and material support (Drs Mignot, Lammers, Nevsimalova, Vancova, Harsh, Bassetti, Schrader, and Nishino, Ms Okun, and Mr Overeem); study supervision (Dr Mignot).

This study was supported by grants NS 23724, NS 33797, and MH40041 from the National Institutes of Health (Bethesda, Md); and grants GA UK 56/99 and MSM 1111 000 1 from the Czech Republic. Mr Overeem was supported by the foundation De Drie Lichten (Rotterdam, the Netherlands) and Hersenstichting Nederland (Den Haag, the Netherlands).

We thank David Rye, MD, for providing cerebrospinal fluid samples; Paul Stowers, MS, and David Brooks for assistance in organizing and performing lumbar punctures; Ling Lin, MD, PhD, for HLA typing; Anna Voros for technical assistance; Goeffrey Harwig for collection of the clinical data in a large African American lineage; Shahrad Taheri, MD, PhD, and Andrew Ray, PhD, for editing the manuscript; Jamie Zeitzer, PhD, for editing and preparation of the figures; and Helena Kraemer, PhD, for statistical assistance with the quality receiver operating characteristic analysis.

Corresponding author and reprints: Emmanuel Mignot, MD, Center for Narcolepsy, Stanford University, 701B Lower Welch Rd, Palo Alto, CA 94304-5742 (e-mail: mignot@leland.stanford.edu).