Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2009 Sep 1.
Published in final edited form as: J Invest Dermatol. 2008 Aug 14;129(1):110–118. doi: 10.1038/jid.2008.189

Development of an IgG4-based Classifier/Predictor of Endemic Pemphigus Foliaceus (Fogo Selvagem)

Bahjat F Qaqish 1, Phillip Prisayanh 2, Ye Qian 2, Eugenio Andraca 1, Ning Li 2, Valeria Aoki 3, Gunter Hans-Filho 4, Vandir dos Santos 5, Evandro A Rivitti 3, Luis A Diaz 2; the Cooperative Group on Fogo Selvagem Research
PMCID: PMC2736304  NIHMSID: NIHMS114339  PMID: 18704107

Abstract

Fogo Selvagem (FS) is mediated by pathogenic, predominantly IgG4, anti-Dsg1 autoantibodies and is endemic in Limao Verde (LV), Brazil. IgG and IgG-subclass autoantibodies were tested in a sample of 214 FS patients and 261 healthy controls by Dsg1-ELISA. For model selection, the sample was randomly divided into training (50%), validation (25%) and test (25%) sets. Using the training and validation sets, IgG4 was chosen as the best predictor of FS, with index values above 6.43 classified as FS. Using the test set, IgG4 has sensitivity 92% (95% CI: 82−95%), specificity 97% (95% CI: 89−100%) and area under the curve 0.97 (95% CI: 0.94−1.00). The IgG4 positive predictive value (PPV) in LV (3% FS prevalence) was 49%. The sensitivity, specificity and PPV of IgG anti-Dsg1 were 87%, 91% and 23%, respectively. The IgG4-based classifier was validated by testing 11 FS patients before and after clinical disease and 60 Japanese pemphigus patients. It classified 21/96 normal individuals from a LV cohort as having FS serology. Based on its PPV, half of the 21 individuals may currently have preclinical FS and could develop clinical disease in the future. Identifying individuals during preclinical FS will enhance our ability to identify etiological agent(s) triggering FS.

INTRODUCTION

Both non-endemic pemphigus foliaceus (PF) and its endemic form [Fogo Selvagem (FS)] are characterized by subcorneal epidermal blisters and pathogenic IgG anti-desmoglein 1 (Dsg1) autoantibodies (Beutner and Jordon, 1964; Roscoe et al., 1985; Stanley et al., 1986; Diaz et al., 1989a). FS patients usually live in impoverished rural areas of certain states of Brazil where the disease is endemic (Vieira, 1937; Diaz et al., 1989b). Strikingly, the prevalence of FS in some states, like Sao Paulo (Diaz et al., 1989b) and Parana (Empinotti et al., 2006), has decreased dramatically in recent years. An endemic form of PF has also been described in Colombia and Tunisia (Robledo et al., 1988; Morini et al., 1993; Abreu-Velez et al., 2003). FS exhibits a strong association with the HLA-DRB1*0102, 0404 and 1402 alleles (p<0.005, RR: 14) and affects people of many races and ethnic backgrounds (Moraes et al., 1997). It has been hypothesized that a local environmental agent(s) acting on genetically predisposed individuals triggers a cross-reactive anti-Dsg1 antibody response that leads to FS (Diaz et al., 1989b). Recent studies suggest that exposure to hematophagous insect bites are risk factors for FS (Lombardi et al., 1992; Aoki et al., 2004; Diaz et al., 2004).

IgG autoantibodies in FS are isotype-restricted and the bulk of pathogenic anti-Dsg1 autoantibodies are predominantly IgG4 (Rock et al., 1989; Santos et al., 2001; Warren et al., 2003). In fact, a recent study showed that progression from preclinical to clinical stage of the disease is associated with a dramatic rise in IgG4 anti-Dsg1 autoantibodies as determined by ELISA assays (Warren et al., 2003). Restriction of IgG4 antibody response in humans have been reported in patients with parasitosis (Kurniawan et al., 1993), individuals undergoing hyposensitization therapy for allergies (Larche et al., 2006; Rossi et al., 2007), individuals exposed to bee stings (Aalberse et al., 1983) and patients with autoimmune pancreatitis (Hamano et al., 2001). IgG4 restriction of the autoimmune response has also been reported in other autoimmune blistering diseases, although there is limited information about their pathogenic role in skin disease (Sitaru et al., 2007).

Anti-Dsg1 autoantibodies in PF and FS are routinely detected by indirect immunofluorescence (IF) techniques and are important diagnostic markers of these diseases. The use of PF and FS autoantibodies as predictors of disease has been limited due to the rarity of these diseases and the limitations of the indirect IF assays. The application of ELISA techniques using recombinant Dsg1 has improved the diagnostic accuracy (Amagai et al., 1999) and made it possible to test large numbers of individuals living in communities where FS is endemic (Hans-Filho et al., 1996). For example, we have identified a new focus of FS in Brazil, the Amerindian Terena reservation of Limao Verde, where the prevalence of the disease is ∼3% (Hans-Filho et al., 1996). We have reported that 30 of 31 patients along with 55% of normal subjects (n: 93) living in this reservation possessed anti-Dsg1 antibodies (Warren et al., 2000; Diaz et al., 2008). We have followed this reservation for the last 14 years and documented the progression of FS from the preclinical stage to disease in 11 cases. Serologically, these patients increased their titers of anti-Dsg1 autoantibodies at the onset of their skin disease (Warren et al., 2000) and importantly, the autoantibodies during the preclinical stage bind the EC5 domain of Dsg1, whereas the pathogenic autoantibodies recognized epitopes located on the EC1−2 domain of the molecule during the clinical stage (Li et al., 2003).

In an effort to identify early sensitive and specific serological markers of FS in healthy individuals, we tested IgG and the IgG subclass anti-Dsg1 autoantibody response by ELISA in a large number of patients and normal donors. A rigorous statistical analysis of the data generated has permitted us to develop a novel IgG4 classifier/predictor that separates donors into one group showing immunological features of FS and another with features of healthy donors. The IgG4 classifier is highly sensitive (92%) and specific (97%). In a population with prevalence of 3%, i.e. Limao Verde, this classifier has positive predictive value (PPV) 49% and negative predictive value (NPV) 99.7%. The sensitivity, specificity and PPV of IgG anti-Dsg1 were 87%, 91% & 23% respectively, all lower than the corresponding values for IgG4. This instrument has been validated in two patient populations; a group of 60 Japanese patients with PF and pemphigus vulgaris (PV) and a group of 11 FS patients from Limao Verde where preclinical stage sera were available. The use of this classifier tool may facilitate the identification of individuals during the preclinical stage of FS, thus enhancing the chances of disclosing the etiological agent(s) triggering this human autoimmune disease. Since IgG anti-Dsg1 autoantibodies are detected in a large number of normal individuals from endemic areas of FS (Warren et al., 2000; Warren et al., 2003; Diaz et al., 2008), the high specificity of the IgG4 anti-Dsg1 classifier will outperformed the total IgG assays when used in these seroepidemiological studies.

RESULTS

Development of the classifier

We used data on sera from 214 FS cases (45%) and 261 healthy controls (non-cases) (55%). The results of the IgG subclass and total IgG anti-Dsg1 ELISA of FS cases and controls were expressed as index values. Table 1 gives descriptive statistics of IgG index values.

Table 1.

Descriptive Statistics for IgG & IgG subclass Index Values

Fogo Selvagem Cases Healthy Controls
N Mean Median IQR SD N Mean Median IQR SD
IgG1 214 232.1 73.2 233.1 419.6 261 50.3 0 2.4 206.1
IgG2 214 50.7 0.3 49.9 113.7 261 26.5 2.5 29.7 59.6
IgG3 214 10.6 5.5 9.6 15 261 4.5 1.7 3.5 9.2
IgG4 214 76.5 69 79.1 53.7 261 1.6 0 1.2 4.6
IgG 209 79.6 88 35.8 32.5 250 2.4 0 0 11.7
Open in a new tab

IQR = Inter-quartile range

SD = standard deviation

Note: Negative index values were replaced by 0 before obtaining these summaries

Model selection and estimation of the accuracy of the chosen model were carried out using a rigorous procedure by splitting the data at random into training (50%), validation (25%) and test sets (25%). This avoids the overly optimistic estimates of area under the curve (AUC), sensitivity and specificity often obtained when the same data are used for model selection as well as for final estimation of prediction accuracy (Hastie et al., 2001). Model parameters were estimated from the training set, and the corresponding AUC was estimated from the validation set.

Results from the model selection procedure, using only the training (n=239) and validation (n=118) sets, are summarized in Table 2. The AUC was used as the criterion for model selection as it is a summary of the whole receiver operating characteristic (ROC) curve for a given model and is not affected by choice of a cut point as is the case for sensitivity and specificity. The model with only x4 (=log(1+IgG4) has an estimated AUC of 0.961 (from the validation set). Using additional predictors has a negligible effect on the AUC. Thus, the model with only IgG4 was chosen as the final model because of its parsimony and high AUC value. The classification rule thus developed is to declare IgG4 index values above 6.43 as cases, and values 6.43 and below as non-cases.

Table 2.

Prediction Models Applied to the Training Set (AUC, Sensitivity and Specificity derived from Validation set)

Predictors Area under the curve (AUC) Sensitivity Specificity
IgG1 0.78 0.74 0.78
IgG2 0.51 0.45 0.54
IgG3 0.81 0.81 0.68
IgG4 0.96 0.89 0.94
IgG 0.92 0.81 1.00
IgG1 & IgG2 0.80 0.72 0.78
IgG1 & IgG3 0.84 0.72 0.78
IgG1 & IgG4 0.97 0.91 0.95
IgG1 & IgG 0.91 0.81 1.00
IgG2 & IgG3 0.81 0.85 0.68
IgG2 & IgG4 0.96 0.91 0.95
IgG2 & IgG 0.94 0.83 1.00
IgG3 & IgG4 0.96 0.89 0.92
IgG3 & IgG 0.95 0.81 1.00
IgG4 & IgG 1.00 0.83 1.00
IgG1, IgG2 & IgG3 0.84 0.72 0.78
IgG1, IgG2 & IgG4 0.96 0.92 0.94
IgG1, IgG2 & IgG 0.94 0.83 1.00
IgG1, IgG3 & IgG4 0.97 0.91 0.94
IgG1, IgG3 & IgG 0.94 0.81 1.00
IgG1, IgG4 & IgG 1.00 0.92 0.94
IgG2, IgG3 & IgG4 0.97 0.83 1.00
IgG2, IgG3 & IgG 0.95 0.83 1.00
IgG2, IgG4 & IgG 1.00 0.87 1.00
IgG3, IgG4 & IgG 1.00 0.85 1.00
IgG1, IgG2, IgG3 & IgG4 0.97 0.92 0.95
IgG1, IgG2, IgG3 & IgG 0.95 0.83 1.00
IgG2, IgG3, IgG4 & IgG 1.00 0.87 1.00
IgG1, IgG2, IgG4 & IgG 1.00 0.83 1.00
IgG1, IgG3, IgG4 & IgG 1.00 0.87 1.00
IgG1, IgG2, IgG3, IgG4& IgG 1.00 0.87 0.98
Open in a new tab

*Each model was estimated from the training set (n=239).

*Parameter estimates were then applied to the validation set (n=118) to estimate the model's AUC, sensitivity and specificity

Final estimates of AUC, sensitivity and specificity of the IgG4 classifier, were obtained from the test set (n=118). The estimated AUC of IgG4 is 0.97 (95% CI: 0.94−1.00), sensitivity is 92% (95% CI: 82−98%), and specificity is 97% (95% CI: 89−100%). The IgG4 classifier was developed entirely in the training and validation sets, yet it performed extremely well in the test set. The AUC for a test is the probability that a random subject from the disease group (patients) has a higher value than a randomly selected subject from the disease-free group (healthy controls). Our results indicate with 95% confidence that the AUC for IgG4 is higher than 0.94. This constitutes strong evidence that IgG4 is a reliable predictor.

Figure 1 shows smooth density plots of IgG and IgG subclass anti-Dsg1 index values. FS sera (red line) and healthy donor sera (blue line). The cut point of 6.43 chosen for IgG4 corresponds to a value of 2 on the x-axis. It is clear that IgG4 anti-Dsg1 autoantibody index values produce the best separation between control and FS sera.

Figure 1.

Figure 1

Open in a new tab

Smooth density plots of index values of anti-Dsg1 IgG Subclasses and IgG in FS patients (red lines) and healthy donors (blue lines). Negative Index values were converted to zero before applying the log transformation.

Positive and negative predictive value of the IgG4-based classifier

The accuracy of predictions generated by a given classifier in a population depends not only on its sensitivity and specificity, but also on the prevalence of disease in that population (Medina, 1999). Two important measures in that regard are the positive predictive value (PPV) and the negative predictive value (NPV). The PPV is the probability that a subject classified as a case does indeed have the disease. The NPV is the probability that a subject classified as a non-case is actually disease-free. Table 3 shows how PPV and NPV depend on prevalence for a diagnostic test with sensitivity 92% and specificity 97%, which are the estimates for the IgG4-based classifier developed above. In a population such as LV with 3% prevalence, a randomly drawn subject has a 3% chance, or prior probability, of having FS. However, if that subject tests positive, the disease probability goes up to nearly 49%, the PPV. Similarly, a random subject from LV has a 97% of being disease-free. If that subject tests negative, the probability of being disease-free goes up to 99.7%, the NPV.

Table 3.

Positive Predictive Value and Negative Predictive Value for IgG4, IgG1, IgG and IgG1+IgG4 anti-Dsg1 tests with different sensitivities and specificities applied to populations with various prevalence of disease (The Specificty & Sensitivity derived from the test set [25% of the total data set] as described in Methos & Results)

Prevalence IgG (Sens=87%, Spec=91%) IgG1 (Sens=72%, Spec=80%) IgG4 (Sens 92%, Spec=97%) IgG1+IgG4 (Sens=89%, Spec=94%)
PPV NPV PPV NPV PPV NPV PPV NPV
0.1% 0.00958 0.99986 0.00359 0.99965 0.02978 0.99992 0.01463 0.99988
1% 0.08896 0.99856 0.03509 0.99648 0.23650 0.99917 0.13031 0.99882
3%A 0.23016 0.99560 0.10019 0.98929 0.48677 0.99746 0.31449 0.99639
10% 0.51786 0.98438 0.28571 0.96257 0.77311 0.99092 0.62238 0.98716
20% 0.70732 0.96552 0.47368 0.91954 0.88462 0.97980 0.78761 0.97158
80% 0.97479 0.63636 0.93506 0.41667 0.99191 0.75194 0.98343 0.68116
90% 0.98864 0.43750 0.97006 0.24096 0.99639 0.57396 0.99257 0.48705
99% 0.99896 0.06604 0.99720 0.02805 0.99967 0.10911 0.99932 0.07946
99.9% 0.99990 0.00696 0.99972 0.00285 0.99997 0.01199 0.99993 0.00848
Open in a new tab

PPV: Positive predictive value

NPV: Negative predictive value

A

The Amerindian reservation of Limao Verde, Brazil exhibits a prevalence of Fogo Selvagem of ∼3%

In order to compare IgG4 to other potential markers, sensitivity and specificity for IgG, IgG1 and IgG1+IgG4 were estimated from the test set (n= 118). These sensitivity and specificity estimates differ from those in Table 2 that were obtained from the validation set. Table 3 shows the corresponding PPV and NPV for IgG1 (sensitivity: 72%, specificity: 80%), IgG (sensitivity: 87%, specificity: 91%) and the combination of IgG1+IgG4 (sensitivity: 89%, specificity: 94%) for the population of Limao Verde. The PPV for IgG1= 10%, for IgG =23% and for the combination IgG1+IgG4= 31% are lower than those of IgG4 anti-Dsg1.

Assessing the performance of the IgG4-based classifier

The performance of the classifier was evaluated in three additional sets of individuals:

a) FS cases from Limao Verde during the preclinical and clinical phases of the disease

We have collected 11 FS cases for whom sera were available before they developed frank clinical FS. Some donors had several preclinical samples. The IgG4-based classifier found that 5 of the 11 cases (45%) had serological features of FS before developing clinical disease (cases 1, 3, 6, 9, and 11 of Table 4). The duration of the preclinical follow up in each of the 11 cases is also presented in Table 4. The transition from preclinical to disease stage lasted 1 year (one case), 2 years (three cases), 3 years (one case), 4 years (one case), 7 years (three cases) and 10 years (two cases). The total IgG anti-Dsg1 index values are also included in Table 4. In 8 of the 11 cases the IgG anti-Dsg1 index values were positive. The 5 IgG4 positive cases were also positive for total IgG anti-Dsg1. The sera of 3 cases (#7, # 8 and # 10) were positive for IgG anti-Dsg1 but negative for IgG4, indicating the different sensitivities and specificities of the total IgG and IgG4 anti-Dsg1 assays.

Table 4.

IgG and IgG4 anti-Dsg1 autoantibodies in 11 FS cases during the Preclinical and Clinical Stage of the Disease

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Preclinical
Case 1 IgG 94.34 Unavailable 69.92 2 y
FS-27 IgG4 95.73 Unavailable 87.66
Case 2 IgG 50.08 Unavailable Unavailable 70.76 3y
FS-29 IgG4 1.49 Unavailable Unavailable 31.5
Case 3 IgG 56.63 Unavailable 27.38 2y
FS-31 IgG4 9.2 Unavailable 17.75
Case 4 IgG 108.03 Unavailable 104.18 2y
FS-32 IgG4 3.2 Unavailable 24.79
Case 5 IgG 30.02 Unavailable Unavailable Unavailable 50.76 4y
FS-33 IgG4 1.29 Unavailable Unavailable Unavailable 59.12
Case 6 IgG 68.99 121.01 Unavailable Unavailable 27.83 Unavailable nd Unavailable 45.07 −45.85 4.17 10y
FS-45 IgG4 6.87 7.04 Unavailable Unavailable 3.99 Unavailable 0.65 Unavailable 26.36 16.74 18.84
Case 7 IgG −8.88 Unavailable Unavailable −19.98 Unavailable −25.87 Unavailable nd 49.9 −31.10 103.42 10y
FS-46 IgG4 1.33 Unavailable Unavailable 0.22 Unavailable 12.56 Unavailable 0.17 0.7 32.83 28.76
Case 8 IgG −17.66 Unavailable Unavailable Unavailable Unavailable Unavailable Unavailable 37.52 7y
FS-37 IgG4 1.38 Unavailable Unavailable Unavailable Unavailable Unavailable Unavailable 12.42
Case 9 IgG 73.06 Unavailable Unavailable Unavailable 84.31 Unavailable Unavailable 136.63 7y
FS-38 IgG4 33.29 Unavailable Unavailable Unavailable 96.02 Unavailable Unavailable 137.83
Case 10 IgG −88.01 Unavailable Unavailable Unavailable Unavailable Unavailable Unavailable 27.21 7y
FS-39 IgG4 −0.46 Unavailable Unavailable Unavailable Unavailable Unavailable Unavailable 11.26
Case 11 IgG 71.89 169.5 1y
BA1−1 IgG4 16.09 82.48
Open in a new tab

Positive IgG4 index value: >6.43

Positive IgG index value: >11.34

nd: serum sample unavailable, test not done

The 11 cases that underwent the transition of pre-clinical to clinical FS were further analyzed by comparing their serological features (IgG and IgG4 anti-Dsg1 autoantibodies) with a group of age and sex matched normal individuals from Limao Verde, known to have IgG anti-Dsg1 autoantibodies for the last 3−5 years and normal skin. The IgG4 classifier was positive in 4 of these individuals, 3 of them are relatives of one FS patient and maybe genetically predisposed. We are following these individuals closely for evidence of FS.

b) Sera from Japanese patients with PV and PF

Sera from 20 mucosal PV (mPV) patients, which were known to contain only anti-Dsg3 autoantibodies, were classified as normal sera by the IgG4 anti-Dsg1 classifier. On the contrary, the sera of 17/20 mucocutaneous PV (mcPV) and 18/20 PF were classified as having features of FS since they posses anti-Dsg1 autoantibodies. The sera of Japanese PF patients contain anti-Dsg1 autoantibodies, predominantly IgG4 (Data not shown). Moreover, the sera of mcPV patients contain a combination of anti-Dsg1 and anti-Dsg3 autoantibodies.

c) Sera from three Limao Verde Cohorts (2005 data)

The IgG4-based classifier was applied to the sera of 96 individuals, members of the 3 cohorts from the Limao Verde Reservation under study since 2005, when the initial evaluation was performed and serum samples were obtained. The IgG4 classifier identified 21 individuals (22%) with serological features of FS: 6/34 individuals (17.6%) in cohort 1 were positive, 6/39 (15.3%) in cohort 2 were also positive as well as 9/24 (37.5%) in cohort 3 (Table 5). Additionally, the same donors show the following percentages of total IgG anti-Dsg1 autoantibodies: cohort 1, 1/34 (2.9%), cohort 2, 3/41 (7.3%) and cohort 3, 7/24 (29%) (Diaz et al., 2008). From the 21 cohort members showing positive IgG4 anti-Dsg1 autoantibodies, 12 had IgG4 autoantibodies alone (57%) and 9 had a combination of IgG and IgG4 autoantibodies (42.8%). Only 2 individuals outside the 21 were positive for IgG anti-Dsg1 alone. The remaining 75 healthy cohort members exhibited negative IgG anti-Dsg1 autoantibodies. These results demonstrate the higher sensitivity of the IgG4 classifier to detect anti-Dsg1 autoantibodies. It is interesting that a member of Cohort 3 classified as FS by our IgG4 predictor (individual JDM) has developed clinical FS in 2005.

Table 5.

Individuals from Limao Verde, Brazil identified as FS cases by the IgG4 classifier

Cohort 1 (Ages 5−10) Age IgG4 Index Values
MSS 5.4 7.32
MOD 5.6 11.17
ACdM 6.3 6.43
LAD 7.7 7.09
FCdM 8.2 15.17
GFS 9.7 8.19
Cohort 2 (Ages 11−15)
ADF 10.7 130.62
DFS 11.1 34.88
GFS 11.5 40.87
LD 11.8 12.36
AMdS 12.4 31.12
JFD 14.2 57.55
Cohort 3 (Ages 16−20)
EMC-F 15.1 47.15
DCP 15.5 25.63
LPC 15.5 19.89
AMdS 16.8 34.09
EMC-M 17.3 121.73
RDC 17.9 21.29
JDM* 18.4 56.75
RPC 19.8 52.39
LTM 20.3 34.68
Open in a new tab
*

Developed FS during the course of study

DISCUSSION

The development of an autoimmune disease is one of the fundamental enigmas of immunology. Many genetic and environmental factors appear to play a role, making it difficult to fulfill Koch's criteria for any of these diseases. Some of the advantages that an organ-specific autoimmune disease such as FS (which is epidermal-specific) offers to research in autoimmunity are related to the fact that anti-Dsg1 autoantibodies are pathogenic and the self-antigen, Dsg1, is fully characterized. Dissection of the underlying mechanisms involved in autoantibody formation is a difficult task because in many instances it is not possible to differentiate between the pathogenic autoantibody response and a normal background (natural) immune response.

Importantly, IgG autoantibodies against Dsg1 are detected not only in FS patients but also in a significant number of healthy individuals living in an endemic region such as Limao Verde (Hans-Filho et al., 1996; Diaz et al., 2008). In this settlement we have witnessed the transition from preclinical to clinical stage of the disease (Warren et al., 2000; Li et al., 2003; Warren et al., 2003; Diaz et al., 2008). The challenge therefore is to develop assays to detect the earliest serological markers of FS during the preclinical stage. As shown in this paper we have developed a classifier/predictor by analyzing the IgG and the four IgG subclass anti-Dsg1 autoantibody response in a large data set derived from FS patients and healthy controls (Table 1) using a sensitive and specific Dsg1 ELISA for IgG and each IgG subclass.

A set of 475 serum samples were tested for IgG and IgG subclass anti-Dsg1 were divided at random into a training set (N=239), a validation set (N=118) and a test set (N=118) in order to select a prediction model and to assure unbiased estimation of classification accuracy. The best classifier involves only IgG4 with an estimated AUC of 0.97 (95% CI: 0.94−1.00), sensitivity of 92% (95% CI: 82−98%), and specificity of 97% (95% CI: 89−100%) (Table 3 and Figure 1). The sensitivity and specificity for the IgG assay for anti-Dsg1 autoantibodies were 5% and 6% lower than those of IgG4 predictor. The PPV and NPV of the classifier, when applied in a population such as the Limao Verde reservation with 3% prevalence, were calculated. It is estimated that an individual classified as positive has a 49% chance of having FS while a subject classified as negative has a 99.7% probability of being disease-free (Table 3). The PPV for IgG1 (10%), IgG (23%) and IgG1+IgG4 (31%) anti-Dsg1 autoantibodies were lower than IgG4. Although the IgG anti-Dsg1 test also performed well as a serological marker of FS (Table 2 and Figure 1), the AUC of IgG (0.92) was lower than the IgG4 AUC (0.96). Additionally, using the sensitivity and specificity for the IgG and IgG4 assays, and the 3% prevalence of FS in Limao Verde, we estimated that 8.5% of the inhabitants of this reservation would have a false positive IgG anti-Dsg1 assay. The false positive tests using the IgG4 anti-Dsg1 assay however would be only 2.9%. Finally, since it has been reported that normal inhabitants of endemic areas of FS, i.e. Limao Verde, possess anti-Dsg1 autoantibodies (Warren et al., 2000; Diaz et al., 2008), it use as a classifier in these human settlements would be very limited. On the contrary, IgG4 anti-Dsg1 autoantibodies are detected in the sera of individuals developing FS or during recurrences (Li et al., 2003; Warren et al., 2003). For similar reasons the IgM anti-Dsg1 autoantibodies (Diaz et al., 2008) did not performed well (data not shown).

The IgG4-based classifier was validated further by analyzing two groups of patients; a group of 11 FS patients during the preclinical stage of the disease and a group of 60 Japanese PV and PF patients. In the first group, the classifier predicted FS in 5/11 individuals (45%) during the preclinical stage and in all samples during the clinical stage of FS (100%). It must be emphasized that this classifier identifies subjects with serological features of FS regardless of the presence of active skin disease. We propose that this IgG4-based classifier is a serological marker of FS during the preclinical and clinical stages of the disease. During the preclinical stage this classifier may show variations over time due to fluctuations in environmental antigenic stimulation as observed in cases 6 and 7 in Table 4. However, once a subject exhibits skin disease, the classifier will be positive with high probability, provided the serum sample is obtained before initiation of systemic therapy. Eight of the eleven patients exhibited IgG anti-Dsg1 autoantibodies during the pre-clinical stage of FS, in agreement with the high prevalence of these IgG autoantibodies in this human settlement. Similarly, a control group from Limao Verde composed of normal individuals with positive IgG anti-Dsg1 autoantibodies, matched by age and sex with the 11 patients of Table 4, includes 4 normal subjects with positive IgG4 anti-Dsg1. Three of these normal individuals were first degree relatives of a patient with FS. These individuals have been under close observation for 3 to 5 years since the collection of serum samples. We shall continue following them for any evidence of clinical FS.

In the group of 60 Japanese patients, 20 patients with mPV, possessing only anti-Dsg3 autoantibodies, were classified as normal donors (because the absence of anti-Dsg1 autoantibodies). The IgG4-based classifier performed well in a group of 20 PF patients, of whom 18 were identified as cases. In a group of 20 mcPV patients, possessing anti-Dsg1 and anti-Dsg3 autoantibodies, the classifier predicted the disease in 17 cases. Hence the IgG4-based classifier performed well in the Japanese group of patients with PF and mcPV since both groups of patients possess anti-Dsg1 autoantibodies.

The IgG4-based classifier was also used to test the members of an ongoing prospective cohort in the Limao Verde reservation including 96 donors, aged 5 to 20 on their first evaluation in 2005. The IgG4 classifier identified 21 individuals (22%) with serological features of FS: 6/34 individuals (17.6%) in cohort 1 were positive, 6/39 (15.3%) in cohort 2 were also positive as well as 9/24 (37.5%) in cohort 3 (Table 5). The same donors show the following percentages of total IgG anti-Dsg1 autoantibodies: cohort 1, 1/34 (2.9%), cohort 2, 3/41 (7.3%) and cohort 3, 7/24 (29%) (Diaz et al., 2008). Interestingly, one member of the third cohort (JDM), classified as a case, has developed FS in the course of the study. According to the PPV of the IgG4 classifier, it is estimated that about 50% of these positive subjects from Limao Verde have FS in the preclinical stage and are at risk to develop clinical disease if the conditions are appropriate. Similarly, each of the 75 subjects identified as normal by the classifier have a 99% chance of being disease-free. Forecasting active clinical disease in individuals of both groups (positive and negative) using the classifier is the subject of current investigation in our laboratory. An ongoing prospective study of these cohorts will validate further this immunological instrument not only as an identifier of current FS serology but also as a predictor of future disease. These cohorts are evaluated clinically every 4 months and serologically every 2 years.

Organ-specific and non-organ-specific autoantibodies have been reported as diagnostic aids in diseases such as type I diabetes, thyroiditis, adrenalitis, myasthenia gravis, systemic lupus erythematosus and rheumatoid arthritis (Leslie et al., 2001; Scofield, 2004). Moreover, in some of these diseases the respective autoantibodies have been detected years before the onset of clinical disease (Notkins and Lernmark, 2001; Arbuckle et al., 2003). The value of autoantibodies as predictors has been reviewed (Arbuckle et al., 2003; Scofield, 2004) and is well documented in diabetes (Notkins and Lernmark, 2001). In diabetes, the presence of autoantibodies against glutamic acid decarboxylase (GAD), protein tyrosine phosphatase-like molecule (IA-2) and insulin in healthy individuals could predict the development of disease in over 70% of first-degree relatives of the cases in the course of 2 to 10 years (Notkins and Lernmark, 2001). This report moves FS into the group of organ-specific autoimmune diseases where epidermal-specific autoantibodies can be used not only as diagnostic markers of FS but also as predictors of disease as shown in six of eleven FS cases where preclinical sera was available.

Finally, we are reporting an IgG4-based classifier that is able to identify a serum as exhibiting features of FS or features of normal serum. Following the ELISA anti-Dsg1 assay methodology to test for IgG subclasses reported in this paper, we found that IgG4 index values above 6.43 are sufficient to classify a serum sample as having features of FS. Its usefulness in the evaluation of PF patients or in endemic forms of PF in other regions of the world must be validated since FS and the endemic regions of FS are unique. However, our study suggests that the IgG4-based classifier can be extremely useful in identifying individuals during the preclinical stage of FS. HLA typing and the IgG4-based classifier would become powerful tools for the selection of individuals to undergo close clinical and serological surveillance. Moreover, as the environmental risk factor(s) can also be assessed among potential FS patients, these immunological markers may enhance our ability to identify these factor(s) involved in triggering the autoimmune disease in FS.

MATERIALS AND METHODS

Sources of sera

A total of 475 serum samples were tested for IgG subclass anti-Dsg1 in this investigation, 214 from FS cases and 261 from healthy controls (Table 1). From this set there were 459 sera tested to IgG anti-Dsg1 autoantibodies (209 FS and 250 controls) (Table 1). Sera from classic cases of FS were collected from 6 Brazilian hospitals dedicated to treat these patients: Hospital das Clinicas, Sao Paulo (Hospital SP), (n: 49); Hospital de Doenças Tropicaes, Goiania (Hospital GO), (n: 41); Hospital Adventista do Penfigo, Campo Grande (Hospital CG) (n: 27); Hospital Universitario de Belho Horizonte, Minas Gerais (Hospital MG) (n: 47); Hospital Universitario de Brasilia (Hospital Br (n: 47); and Outpatient Clinic, Cascavel, Parana (Parana (n: 3). The FS sera derived from patients with widespread skin disease. Clinically they show superficial blisters and erosions and histologically subcorneal vesicles. The indirect IF studies showed anti-epidermal ICS autoantibodies in titers above 1:80. FS patients admitted to Brazilian hospitals with widespread disease comprised individuals with the generalized forms of the disease as previously described (Hans-Filho et al., 1999). They include the bullous-exfoliative, the exfolitive erythodermic, and forms characterized by generalized keratotic plaques and nodular lesions.

Normal human sera were obtained from blood bank donors from Hospital SP (n: 57), BH (n: 32), Hospital GO (n: 41) and the University of North Carolina Blood Bank (n: 131).

Sera used to validate the IgG4 Classifier/predictor

a) Eleven sera from FS patients before and after the onset of FS. b) Sera from Japanese patients with PF and pemphigus vulgaris (PV). Professor M. Amagai from Keio University, Tokyo, kindly provided us with sera from the following groups of patients: Dsg1 positive PF (n: 20), Dsg3 positive mucosal mPV (n: 20) and Dsg1 positive, Dsg3 positive mcPV (n: 20) and c) sera from three Limao Verde cohorts. Currently we are following 3 cohorts, all inhabitants of the Limao Verde reservation. The first cohort comprise normal donors age 5 to 10 (n: 34), the second, age 11 to 15 (n: 38) and the third, age 16 to 20 (n: 24). The sera were obtained during the first evaluation of the cohorts in May 2005. These investigations were approved by the Institutional Review Boards of the University of North Carolina and the University of Sao Paulo, Brazil

Production and purification of recombinant desmoglein 1

A His-tagged recombinant form of Dsg1, encompassing the extracellular domain of this protein was generated in the baculovirus system and purified by nickel affinity chromatography using the procedure of Ding et al (Ding et al., 1997). Optimum conditions for this expression system were determined empirically as described by Liebman et al (Liebman et al., 1999). The typical protein yield was 10 ug per ml of culture supernatant.

IgG and IgG subclass anti-Dsg1 ELISA assays

ELISA plates were coated with 200 ng/well of purified Dsg1 at 4°C overnight. After washing with Tris-buffered saline containing 3.7 mM Ca++ and 0.05% Tween-20 (TBS/Ca++/T-20), the plate was blocked with 1% BSA in TBS/Ca++/T-20 at room temperature for 1 hour. The plate was then incubated with duplicate 1:100 dilutions (for IgG plates) and 1:50 dilutions (for IgG subclasses) of serum samples for 1 hour at room temperature. Following wash, the plate was incubated with a 1:2000 dilution (for IgG plates) 1:1000 dilution (for IgG subclasses) of a mouse horseradish peroxidase (HRP) labelled monoclonal antibody against human IgG or human IgG1, IgG2, IgG3 and IgG4 subclasses (Zymed, San Francisco, California) (Warren et al., 2003; Diaz et al., 2007). Results were expressed as index value units as reported by Amagai et al. (Amagai et al., 1999) and (Diaz et al., 2007). The index value was defined in terms of optical density (OD) as follows:

Index value=(Test sampleOD)−(Negative control)OD)(Positive controlOD)−(Negative control)OD)×100

Statistical analysis

A logistic regression model (McCullagh and Nelder, 1989) was used to develop a classifier that predicts case-control status based on IgG and the four IgG subclass index values. The predictors were defined as follows: first, negative IgG index values were replaced by 0, then predictors were computed as x=log (IgG+1). This removed much of the skewness in the IgG index values. Thus x1 through x5 were derived from IgG1 through IgG4 and total IgG. For the purpose of developing and evaluating a classification rule, the data set was divided at random into three parts; a training set, a validation set and a test set, containing 50%, 25% and 25%, respectively, of cases and controls. This follows the guidelines given by Hastie, Tibshirani and Friedman (Hastie et al., 2001).

The following procedure was applied in order to choose the best model for prediction. All 31 possible models, containing an intercept and from 1 to 5 predictors were considered. Each model was estimated from the training set. Parameter estimates were then applied to the validation set to estimate the model's area under the curve (AUC). For example, the model “IgG1, IgG2 & IgG3” in Table 2 is a logistic regression model with linear predictor β0 + β1x1 + β2x2+ β3x3. Fitting the model to the training set yields estimates b0, b1, b2, b3. A score was then computed for each subject in the validation set as: score = b0 + b1x1 + b2x2 + b3x3. The score was used to compute the AUC as a measure of the model's predictive ability. AUC was estimated using non-parametric methods (Hanley and McNeil, 1982).

Additionally, the score was transformed to the probability scale by p = 1/{1+ exp (-score)}. For the purpose of computing sensitivity and specificity, if p>0.45 (equivalent to score above −0.2), the subject was classified as a case, otherwise as a non-case. Cut points other than 0.45 were evaluated, but 0.45 was deemed to provide the best tradeoff between sensitivity and specificity. Sensitivity and specificity were estimated as binomial proportions. The analysis was carried out using SAS 9.1 and R 2.5 software.

ACKNOWLEDGMENTS

This project was supported by NIH grants R01-AR30281, RO1-AR32599, and T32 AR07369 (LAD).

Abbreviations

Dsg

desmoglein

PF

pemphigus foliaceus

FS

Fogo Selvagem

PV

pemphigus vulgaris

Footnotes

CONFLICT OF INTEREST

The authors state no conflict of interest.

REFERENCES

  1. Aalberse RC, van der Gaag R, van Leeuwen J. Serologic aspects of IgG4 antibodies. I. Prolonged immunization results in an IgG4-restricted response. J Immunol. 1983;130:722–726. [PubMed] [Google Scholar]
  2. Abreu-Velez AM, Hashimoto T, Bollag WB, Tobon Arroyave S, Abreu-Velez CE, Londono ML, et al. A unique form of endemic pemphigus in northern Colombia. J Am Acad Dermatol. 2003;49:599–608. doi: 10.1067/s0190-9622(03)00851-x. [DOI] [PubMed] [Google Scholar]
  3. Amagai M, Komai A, Hashimoto T, Shirakata Y, Hashimoto K, Yamada T. Usefulness of enzyme-linked immunosorbent assay using recombinant desmogleins 1 and 3 for serodiagnosis of pemphigus. Br J Dermatol. 1999;140:351–357. doi: 10.1046/j.1365-2133.1999.02752.x. [DOI] [PubMed] [Google Scholar]
  4. Aoki V, Millikan RC, Rivitti EA, Hans-Filho G, Eaton DP, Warren SJ, et al. Environmental risk factors in endemic pemphigus foliaceus (fogo selvagem). J Investig Dermatol Symp Proc. 2004;9:34–40. doi: 10.1111/j.1087-0024.2004.00833.x. [DOI] [PubMed] [Google Scholar]
  5. Arbuckle MR, McClain MT, Rubertone MV, Scofield RH, Dennis GJ, James JA, et al. Development of autoantibodies before the clinical onset of systemic lupus erythematosus. N Engl J Med. 2003;349:1526–1533. doi: 10.1056/NEJMoa021933. [DOI] [PubMed] [Google Scholar]
  6. Beutner EH, Jordon RE. Demonstration of Skin Antibodies in Sera of Pemphigus Vulgaris Patients by Indirect Immunofluorescent Staining. Proc Soc Exp Biol Med. 1964;117:505–510. doi: 10.3181/00379727-117-29622. [DOI] [PubMed] [Google Scholar]
  7. Diaz L, Sampaio S, Rivitti E, Martins C, Cunha P, Lombardi C, et al. Endemic pemphigus foliaceus (fogo selvagem). I. Clinical features and immunopathology. J Am Acad Dermatol. 1989a;20:657–669. doi: 10.1016/s0190-9622(89)70079-7. [DOI] [PubMed] [Google Scholar]
  8. Diaz LA, Arteaga LA, Hilario-Vargas J, Valenzuela JG, Li N, Warren S, et al. Anti-desmoglein-1 antibodies in onchocerciasis, leishmaniasis and Chagas disease suggest a possible etiological link to Fogo selvagem. J Invest Dermatol. 2004;123:1045–1051. doi: 10.1111/j.0022-202X.2004.23438.x. [DOI] [PubMed] [Google Scholar]
  9. Diaz LA, Prisayanh PS, Dasher DA, Li N, Evangelista F, Aoki V, et al. The IgM anti-desmoglein 1 response distinguishes Brazilian pemphigus foliaceus (fogo selvagem) from other forms of pemphigus. J Invest Dermatol. 2008;128:667–675. doi: 10.1038/sj.jid.5701121. [DOI] [PubMed] [Google Scholar]
  10. Diaz LA, Prisayanh PS, Dasher DA, Li N, Evangelista F, Aoki V, et al. The IgM Anti-Desmoglein 1 Response Distinguishes Brazilian Pemphigus Foliaceus (Fogo Selvagem) from Other Forms of Pemphigus. J Invest Dermatol Advance. 2007 October 25; doi: 10.1038/sj.jid.5701121. online publication. 2007; doi:10.1038/sj.jid.5701121. [DOI] [PubMed] [Google Scholar]
  11. Diaz LA, Sampaio SA, Rivitti EA, Martins CR, Cunha PR, Lombardi C, et al. Endemic pemphigus foliaceus (Fogo Selvagem): II. Current and historic epidemiologic studies. J Invest Dermatol. 1989b;92:4–12. doi: 10.1111/1523-1747.ep13070394. [DOI] [PubMed] [Google Scholar]
  12. Ding X, Aoki V, Mascaro JM, Jr., Lopez-Swiderski A, Diaz LA, Fairley JA. Mucosal and mucocutaneous (generalized) pemphigus vulgaris show distinct autoantibody profiles. J Invest Dermatol. 1997;109:592–596. doi: 10.1111/1523-1747.ep12337524. [DOI] [PubMed] [Google Scholar]
  13. Empinotti JC, Aoki V, Filgueira A, Sampaio SA, Rivitti EA, Sanches JA, Jr., et al. Clinical and serological follow-up studies of endemic pemphigus foliaceus (fogo selvagem) in Western Parana, Brazil (2001−2002). Br J Dermatol. 2006;155:446–450. doi: 10.1111/j.1365-2133.2006.07302.x. [DOI] [PubMed] [Google Scholar]
  14. Hamano H, Kawa S, Horiuchi A, Unno H, Furuya N, Akamatsu T, et al. High serum IgG4 concentrations in patients with sclerosing pancreatitis. N Engl J Med. 2001;344:732–738. doi: 10.1056/NEJM200103083441005. [DOI] [PubMed] [Google Scholar]
  15. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982;143:29–36. doi: 10.1148/radiology.143.1.7063747. [DOI] [PubMed] [Google Scholar]
  16. Hans-Filho G, Aoki V, Rivitti E, Eaton DP, Lin MS, Diaz LA. Endemic pemphigus foliaceus (fogo selvagem)--1998. The Cooperative Group on Fogo Selvagem Research. Clin Dermatol. 1999;17:225–235. doi: 10.1016/s0738-081x(99)00014-0. [DOI] [PubMed] [Google Scholar]
  17. Hans-Filho G, dos Santos V, Katayama JH, Aoki V, Rivitti EA, Sampaio SA, et al. An active focus of high prevalence of fogo selvagem on an Amerindian reservation in Brazil. Cooperative Group on Fogo Selvagem Research. J Invest Dermatol. 1996;107:68–75. doi: 10.1111/1523-1747.ep12298213. [DOI] [PubMed] [Google Scholar]
  18. Hastie T, Tibshirani R, Friedman JH. The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Springer; New York: 2001. p. 196. [Google Scholar]
  19. Kurniawan A, Yazdanbakhsh M, van Ree R, Aalberse R, Selkirk ME, Partono F, et al. Differential expression of IgE and IgG4 specific antibody responses in asymptomatic and chronic human filariasis. J Immunol. 1993;150:3941–3950. [PubMed] [Google Scholar]
  20. Larche M, Akdis C, Valenta R. Immunological mechanisms of allergen-specific immunotherapy. Nat Rev Immunol. 2006;6:761–771. doi: 10.1038/nri1934. [DOI] [PubMed] [Google Scholar]
  21. Leslie D, Lipsky P, Notkins AL. Autoantibodies as predictors of disease. J Clin Invest. 2001;108:1417–1422. doi: 10.1172/JCI14452. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Li N, Aoki V, Hans-Filho G, Rivitti EA, Diaz LA. The role of intramolecular epitope spreading in the pathogenesis of endemic pemphigus foliaceus (fogo selvagem). J Exp Med. 2003;197:1501–1510. doi: 10.1084/jem.20022031. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Liebman JM, LaSala D, Wang W, Steed PM. When less is more: enhanced baculovirus production of recombinant proteins at very low multiplicities of infection. Biotechniques. 1999;26:36–38, 40, 42. doi: 10.2144/99261bm05. [DOI] [PubMed] [Google Scholar]
  24. Lombardi C, Borges PC, Chaul A, Sampaio SA, Rivitti EA, Friedman H, et al. Environmental risk factors in endemic pemphigus foliaceus (Fogo selvagem). “The Cooperative Group on Fogo Selvagem Research”. J Invest Dermatol. 1992;98:847–850. doi: 10.1111/1523-1747.ep12456932. [DOI] [PubMed] [Google Scholar]
  25. McCullagh PP, Nelder JA. Generalized Linear Models. Chapman & Hall/CRC.; New York: 1989. Chapter 4. [Google Scholar]
  26. Medina LS. Study design and analysis in neuroradiology: a practical approach. AJNR Am J Neuroradiol. 1999;20:1584–1596. [PMC free article] [PubMed] [Google Scholar]
  27. Moraes M, Fernandez-Vina M, Lazaro A, Diaz Filho L, Friedman H, Rivitti E, et al. An epitope in the third hypervariable region of the DRB1 gene is involved in the susceptibility to endemic pemphigus foliaceus (fogo selvagem) in three different Brazilian populations. Tissue Antigens. 1997;49:35–40. doi: 10.1111/j.1399-0039.1997.tb02707.x. [DOI] [PubMed] [Google Scholar]
  28. Morini JP, Jomaa B, Gorgi Y, Saguem MH, Nouira R, Roujeau JC, et al. Pemphigus foliaceus in young women. An endemic focus in the Sousse area of Tunisia. Arch Dermatol. 1993;129:69–73. doi: 10.1001/archderm.129.1.69. [DOI] [PubMed] [Google Scholar]
  29. Notkins AL, Lernmark A. Autoimmune type 1 diabetes: resolved and unresolved issues. J Clin Invest. 2001;108:1247–1252. doi: 10.1172/JCI14257. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Robledo MA, Prada S, Jaramillo D, Leon W. South American pemphigus foliaceus: study of an epidemic in El Bagre and Nechi, Colombia 1982 to 1986. Br J Dermatol. 1988;118:737–744. doi: 10.1111/j.1365-2133.1988.tb02590.x. [DOI] [PubMed] [Google Scholar]
  31. Rock B, Martins CR, Theofilopoulos AN, Balderas RS, Anhalt GJ, Labib RS, et al. The pathogenic effect of IgG4 autoantibodies in endemic pemphigus foliaceus (fogo selvagem). N Engl J Med. 1989;320:1463–1469. doi: 10.1056/NEJM198906013202206. [DOI] [PubMed] [Google Scholar]
  32. Roscoe J, Diaz L, Sampaio S, Castro R, Labib R, Takahashi Y, et al. Brazilian Pemphigus Foliaceus Autoantibodies Are Pathogenic to BALB/c Mice by Passive Transfer. J Invest Dermatol. 1985;85:538–541. doi: 10.1111/1523-1747.ep12277362. [DOI] [PubMed] [Google Scholar]
  33. Rossi R, Monasterolo G, Coco G, Silvestro L, Operti D. Evaluation of serum IgG4 antibodies specific to grass pollen allergen components in the follow up of allergic patients undergoing subcutaneous and sublingual immunotherapy. Vaccine. 2007;25:957–964. doi: 10.1016/j.vaccine.2006.08.040. [DOI] [PubMed] [Google Scholar]
  34. Santos S, Patrus O, Filgueira A, Diaz L. Perfil evolutivo das subclasses de imunoglobulinas Gama em pacientes de Pênfigo foliáceo endêmico. An bras dermatol. 2001;76:561–574. [Google Scholar]
  35. Scofield RH. Autoantibodies as predictors of disease. Lancet. 2004;363:1544–1546. doi: 10.1016/S0140-6736(04)16154-0. [DOI] [PubMed] [Google Scholar]
  36. Sitaru C, Mihai S, Zillikens D. The relevance of the IgG subclass of autoantibodies for blister induction in autoimmune bullous skin diseases. Arch Dermatol Res. 2007;299:1–8. doi: 10.1007/s00403-007-0734-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Stanley J, Klaus-Kovtun V, Sampaio S. Antigenic Specificity of Fogo Selvagem Autoantibodies Is Similar to North American Pemphigus Foliaceus and Distinct from Pemphigus Vulgaris Autoantibodies. J Invest Dermatol. 1986;87:197–201. doi: 10.1111/1523-1747.ep12695334. [DOI] [PubMed] [Google Scholar]
  38. Vieira J. Contribuição ao estudo de pemphigo no estado de São Paulo. São Paulo, Brazil: Empresa Gráfica da Revista dos Tribunais. 1937 [Google Scholar]
  39. Warren S, Arteaga L, Rivitti E, Aoki V, Hans-Filho G, Qaqish B, et al. The Role of Subclass Switching in the Pathogenesis of Endemic Pemphigus Foliaceus. J Invest Dermatol. 2003;120:104–108. doi: 10.1046/j.1523-1747.2003.12017.x. [DOI] [PubMed] [Google Scholar]
  40. Warren SJ, Lin MS, Giudice GJ, Hoffmann RG, Hans-Filho G, Aoki V, et al. The prevalence of antibodies against desmoglein 1 in endemic pemphigus foliaceus in Brazil. Cooperative Group on Fogo Selvagem Research. N Engl J Med. 2000;343:23–30. doi: 10.1056/NEJM200007063430104. [DOI] [PubMed] [Google Scholar]

RESOURCES