Topic Article:
Development of an IgG4-Based Predictor of Endemic Pemphigus Foliaceus
Bahjat F Qaqish, Phillip Prisayanh, Ye Qian, Eugenio Andraca, Ning Li, Valeria Aoki, Gunter Hans-Filho, Vandir dos Santos, Evandro A Rivitti and Luis A Diaz for the Cooperative Group on Fogo Selvagem Research
Journal of Investigative Dermatology (2009) 129, 110-118; doi:10.1038/jid.2008.189

Who Will Develop Pemphigus Foliaceus?

Julia Tzu ¹, Yvonne Romagosa ¹ and Robert S. Kirsner ¹

Journal of Investigative Dermatology (2009) 129, 6. doi:10.1038/jid.2008.389

Pemphigus foliaceus (PF), an autoimmune blistering disorder, is characterized by the development of superficial cutaneous blisters in a seborrheic distribution, without mucosal involvement. Autoantibodies target the desmoglein 1 (Dsg1) antigen expressed on the surface of keratinocytes located in the superficial levels of the epidermis (Mahoney et al., 1999), and both sporadic PF and endemic PF can occur. Endemic PF, referred to as fogo selvagem (FS), is endemic to specific areas of the world, most notably Brazil, although it has also been described in Tunisia and Colombia (Abreu-Velez et al., 2003). Both genetic and environmental factors appear to play a role in the development of FS. Certain human leukocyte antigen (HLA) types appear to be at greatest risk: the HLA-DRB1*0102, HLA-DRB1*0404, and HLA-DRB1*1402 alleles (P < 0.005, relative risk, 14; Moraes et al., 1997). It has been hypothesized that the disease may be triggered by a local environmental agent(s) such as Simulium (a hematophagous insect), causing a cross-reactive anti-Dsg1 antibody response that leads to FS (Diaz et al., 2004).

Endemic PF is of special interest, beyond its clinical impact on those affected, because it can serve as a model for organ-specific immune disease with both genetic and environmental components. FS appears to be caused by isotype-restricted, pathogenic anti-Dsg1 autoantibodies that are predominantly IgG4 (Warren et al., 2003) and that increase in concentration in serum upon progression from preclinical to clinical disease.

Development of a predictive test for PF would benefit those at high risk, and it could also serve as an investigative tool. In an effort to develop such a test, Qaqish et al. (2009) used Dsg1 enzyme-linked immunosorbent assay in a study of 214 patients with FS and 261 healthy controls, randomly divided into training (50%), validation (25%), and test (25%) sets. IgG4 was found to be the best predictor of FS, with a sensitivity of 92% and specificity of 97%. The positive predictive value of this test in the endemic region of Brazil (which has an FS prevalence of 3%) was 49%. This was then validated by testing 11 patients with FS before and after clinical disease onset, as well as 60 Japanese patients with PF.

Through the following questions, we examine this paper in greater detail.

QUESTIONS
1. Describe pemphigus foliaceus.
2. How is a predictive test developed, and how does the population affect the development and utility of these tests?
3. What were the findings of this study?
4. What may be the clinical implications of this article?
5. What further studies could be performed?

ANSWERS
1. Pemphigus foliaceus (PF) is an autoimmune blistering disorder characterized by the development of superficial cutaneous blisters, commonly in a seborrheic distribution, with no mucosal involvement. Because of its superficial location, these blisters are fragile and easily sheared by application of lateral shearing pressure (Nikolsky’s sign), and they appear as crusted erosions on an erythematous base. Histologically PF is characterized by clefting/blister formation in the granular or upper spinous layer of the epidermis, with acantholytic cells detaching from the roof of the blister. Blister cavities may contain numerous neutrophils, and the dermis may contain a mild inflammatory infiltrate (Rapini, 2005). The superficial appearance of the blisters, as observed both clinically and histopathologically, can be explained by the presence of circulating IgG autoantibodies that, in the skin, target the desmoglein 1 (Dsg1) antigen expressed on the surfaces of keratinocytes located in superficial levels of the epidermis (Amagai et al., 1999; Mahoney et al., 1999). Hence, indirect and direct immunofluorescence displays intense staining of the IgG antibodies, most prominently in areas of acantholysis and intercellularly within the superficial layers of the epidermis (Rapini, 2005).

PF has multiple clinical variants (see below). For the purposes of this discussion, we focus on the fogo selvagem variant, also known as endemic PF. Sporadic PF and endemic PF show identical clinical, histologic, and immunologic features (Diaz et al., 1989). However, specific epidemiologic features distinguish sporadic from endemic PF. The latter, as its name suggests, is endemic to specific areas of the world, most notably Brazil, although it has also been described in Tunisia and Colombia (Robledo et al., 1988; Morini et al., 1993; Abreu-Velez et al., 2003). Impoverished rural areas are more likely to be foci for this disease, and urbanization of these areas correlates with its disappearance. It affects children and young adults, compared with the older population affected by sporadic PF. Moreover, the disease is associated with specific HLA subtypes (Moraes et al., 1997). It has therefore been hypothesized that an environmental factor, possibly an infectious agent carried by hematophagous insects such as Simulium, serves as the triggering etiologic agent in genetically susceptible individuals (Diaz et al., 2004; Aoki et al., 2004).

PF can be divided into the following subtypes.

Endemic pemphigus/fogo selvagem. A description for this subtype is provided above.

Pemphigus erythematosus/Senear-Usher syndrome. This subtype represents a localized form of PF in which typical lesions are found in malar areas of the face. Patients exhibit immunologic features of both PF (anti-Dsg1 antibodies) and lupus erythematosus (positive ANA, deposition of IgG and C3 on keratinocyte surfaces and the basement membrane zone, and positive lupus band test in perilesional skin) (American and Ahmed, 1984). Thymoma has been reported to be associated with this variant of PF (Souteyrand et al., 1981).

Drug-induced PF. This subtype of pemphigus is mostly associated with ingestion of penicillamine or captopril, medications with a sulfhydryl group or precursor of the sulfhydryl group, such as a disulfide bond (Brenner et al., 1998). It is thought that the sulfhydryl groups in these medicines may interact with sulfhydryl groups in the desmogleins, altering their antigenicity and inducing autoantibody production. Discontinuation of the medication may not lead to remission in half the cases (Maruani et al., 2008).

Pemphigus herpetiformis. Considered a separate variant of pemphigus or a subtype of PF, pemphigus herpetiformis presents with pruritic, flaccid vesicles in an annular or, as its name suggests, herpetiform pattern. Histology reveals eosinophilic spongiosis with or without acantholysis and pustules filled with eosinophils or neutrophils. IgG autoantibodies circulating in the sera target the keratinocyte surface antigen desmoglein 1 (or desmoglein 3 in the pemphigus vulgaris variant) (Robinson et al., 1999).

Paraneoplastic pemphigus with clinical and immunologic features of PF. A single case has been reported; the patient had underlying non-Hodgkin’s B-cell lymphoma that displayed reactivity characteristics of paraneoplastic pemphigus and clinical/immunologic criteria for PF in spite of mucosal involvement (Chorzelski et al., 1999).

2. The first half of the paper by Qaqish et al. (2009) addressed the development of a predictive model. The investigators used serum IgG and IgG subtype index values derived from 475 patients (214 FS and 261 normal controls) to create multiple prediction models as shown in Table 2 of the paper, based on various IgG and IgG subtype predictors, used individually or in various combinations. With the data divided into training (50%), validation (25%), and test (25%) sets, the authors used the training set to estimate model parameters and the validation set to estimate area under the curve (AUC), sensitivity, and specificity for each model. The authors preferred a model with high AUC and few predictors; hence, the model with the IgG4 predictor was selected on the basis of its high AUC (0.961) and parsimony of predictors. The final AUC, sensitivity, specificity, positive predictive values (PPV), and negative predictive values (NPV) for this model were derived using the test set data and found to be satisfactorily high at 97%, 92%, 97%, 49%, and 99.7%, respectively. Owing to a 3% prevalence rate of FS in Limao Verde (Brazil), the PPV and NPV combination is maximized using biomarker IgG4, confirming the results of the authors’ previous model selection process. The investigators then constructed smooth density plots of index values in anti-Dsg1 IgG and IgG subclasses in controls and FS patients. They found that the IgG4 index smooth value plot provided the best distinction between controls and patients at a cutoff of 6.43, once again confirming the results of their previous model selection process.

Some of the statistical terms discussed above deserve closer attention.

Area under the receiver operating curve. Evaluation of the performance of predictive models may be accomplished using receiver operating characteristic (ROC) curves, a concept introduced during World War II for radar signal analysis. ROC curves have since been used in many fields, including medicine, to provide an estimate of the various possible sensitivities and specificities of a model in predicting a binary outcome over the range of threshold probabilities. The area under the ROC curve measures the discriminatory effectiveness of a biomarker, or its ability to distinguish between two populations. An area of 1 represents a perfect test; an area of 0.5 represents a worthless test (Søreide, 2008). As applied to the predictive test in this article, the area under the curve measures the probability that a random participant from the disease group (patient) has a higher value than a randomly selected participant from the disease-free (control) group.

The sensitivity of a test is its ability to correctly identify true positives. As applied to this paper, this is the percentage of people with a disease that is correctly classified. The equation for its calculation is:
Sensitivity = number of true positives/(number of true positives + number of false negatives)

The specificity of a test measures its ability to correctly identify true negatives. As applied to this paper, this is the percentage of people without a disease that is correctly classified. The equation for its calculation is:
Specificity = number of true negatives/(number of true negatives + number of false positives)

The positive predictive value (PPV)of a test measures the proportion of all reported positives that are true positives, or the likelihood that a positive result is true. As applied to this paper, this would mean the likelihood that a person classified as having the disease actually has the disease. The mathematical equation for its calculation is:
PPV = Number of true positives/(number of true positives + number of false positives)

The negative predictive value (NPV)of a test measures the proportion of all reported negatives that are true negatives, or the likelihood that a negative result is true. As applied to this paper, this would mean the likelihood that a participant classified as not having a disease actually does not have the disease. The equation for its calculation is:

NPV = Number of true negatives/(number of true negatives + number of false negatives)

The PPV and NPV depend on prevalence of the disease. In fact, PPV is directly proportional to the prevalence of the disease according to the formula:

PPV = (sensitivity) (prevalence)/[(sensitivity) (prevalence)+(1 – sensitivity) (1 – prevalence)]

Therefore, the higher the prevalence of the disease, the higher the PPV and the lower the NPV. The lower the prevalence of the disease, the lower the PPV and the higher the NPV.

3. The article’s results are organized into two sections: the development of the predictive model and assessment of its performance. A full description of predictive model development is provided above. The remainder of the results described the performance of the predictive model when applied to three populations.

First, the authors examined the sera of 11 known FS patients from Limao Verde and 11 controls. Because these patients were followed before they developed frank disease, preclinical sera were also available. Five of these 11 FS patients (45%) were found to exhibit serologic changes of FS using the IgG4 predictor model. Four of the 11 normal individuals, three of whom are relatives of FS patients, were also found to have serologic changes of FS using the IgG4 predictor model. The authors state that they are still following the progress of the 11 normal individuals to see whether they will eventually develop the disease.

Although this is not noted in the discussion, according to these numbers, the sensitivity of the predictive test would be 45%, which is not in accord with the authors’ projected sensitivity of 97%. It is important to bear in mind that n = 11, so a larger sample may be needed to prove 97% sensitivity.

The investigators then examined the sera of 40 Japanese patients with pemphigus vulgaris (PV) or PF. They found that the IgG4 predictive model classified 17 of 20 PV patients and 18 of 20 PF patients as positive, whereas it classified 20 of 20 mucosal PV patients as negative. According to the authors, this was further evidence in support of test performance, as patients with diseases involving only Dsg1 autoantibodies were more likely to be classified as positive, and those with only Dsg3 autoantibodies were all classified as negative.

In light of the fact that the predictive model was developed specifically for FS patients, it is not entirely clear why the predictive model could be applied to a broader subset of pemphigus patients and why this application should serve as proof of its predictive performance for FS. This is not addressed in the discussion.

Finally, the authors examined sera from 96 control Limao Verde residents, whom they divided into three cohorts based on age. The IgG4 predictor model classified 21 individuals (22%) as positive. The investigators are still following these individuals to track whether they develop FS.

4. The ultimate goal of predictive models constructed through clinical datasets is to provide physicians with enhanced decision-making tools. The direct clinical implication of having an IgG4 classifier is the identification of individuals living in endemic areas who are more likely to eventually develop FS. It is possible that by identifying these individuals, medical or environmental intervention can be instituted at an early stage to prevent or mitigate their disease. This tool may lead to the development of other predictive models using autoantibodies as classifiers for the eventual development of non-skin autoimmune disorders.

The article mentions that this predictive model may ultimately help us identify the etiologic agent causing FS, because identification of patients who are likely to develop this disease can lead to an assessment of their environmental risk factors. This reasoning may represent a logical leap, as those who are identified by their IgG4 levels have already been exposed to whichever etiologic agent is causative of the condition. Identifying environmental risk factors in individuals who will go on to develop clinical disease may not elucidate the etiologic agent any more than does identifying the environmental risk factors of those who already have the disease.

5. Written by a group that has done much work in FS over the past decades, this article describes the development of a quantitative model to predict the development of an autoimmune disorder that has long been speculated to have an environmental trigger. The data concerning this model’s predictive performance fail to prove that the model performs well. First, in their comparison of preclinical and clinical IgG4 index levels of the 11 FS patients and controls, the authors did not achieve the projected high sensitivity of the test. This could be secondary to the limited sample size, which needs to be increased. In their third example, the investigators need to demonstrate that those with IgG4 index levels >6.43 indeed go on to develop clinical disease. This has not been prospectively demonstrated in their Limao Verde group, and the authors claim that this is still in progress. Additional data from these studies may strengthen the authors’ claim. In addition, testing this predictive model on non-Brazilian populations with FS will illustrate generalizability of this model to all populations with FS. If this model demonstrates strong performance, it is conceivable that predictive models for other autoimmune disorders using serum autoantibodies as classifiers may be constructed.

REFERENCES
Abreu-Velez AM, Hashimoto T, Bollag WB, Tobon Arroyave S, Abreu-Velez CE, Londono ML et al. (2003) A unique form of endemic pemphigus in northern Colombia. J Am Acad Dermatol 49:599–608

Amagai M, Tsunoda K, Zillikens D, Nagai T, Nishikawa T (1999) The clinical phenotype of pemphigus is defined by the anti-desmoglein autoantibody profile. J Am Acad Dermatol 40(2 Pt 1):167–70

American ML, Ahmed R (1984) Pemphigus erythematosus: presentation of four cases and review of literature. J Am Acad Dermatol 10:215–22

Aoki V, Millikan RC, Rivitti EA, Hans-Filho G, Eaton DP, Warren SJ et al. (2004) Environmental risk factors in endemic pemphigus foliaceus (fogo selvagem). J Investig Dermatol Symp Proc 9:34–40

Brenner S, Bialy-Golan A, Ruocco V (1998) Drug-induced pemphigus. Clin Dermatol 16:393–7

Chorzelski TP, Hashimoto T, Amagai M, Sysa-Jedrzejowska A, Choczaj A, Nousari HC et al. (1999) Paraneoplastic pemphigus with cutaneous and serological features of pemphigus foliaceus. Br J Dermatol141:357–9

Diaz LA, Sampaio SAP, Rivitti EA et al. (1989) Endemic pemphigus foliaceus (fogo selvagem). I. Clinical features and immunopathology. J Am Acad Dermatol 20:657–69

Diaz LA, Arteaga LA, Hilario-Vargas J, Valenzuela JG, Li N, Warren S et al. (2004) Anti-desmoglein-1 antibodies in onchocerciasis, leishmaniasis and Chagas disease suggest a possible etiological link to fogo selvagem. J Invest Dermatol 123:1045–51

Mahoney MG, Wang Z, Rothenberger K, Koch PJ, Amagai M, Stanley JR (1999) Explanations for the clinical and microscopic localization of lesions in pemphigus foliaceus and vulgaris. J Clin Invest 103:461–8

Maruani A, Machet MC, Carlotti A, Giraudeau B, Vaillant L, Machet L (2008) Immunostaining with antibodies to desmoglein provides the diagnosis of drug-induced pemphigus and allows prediction of outcome. Am J Clin Pathol 130:369–74

Moraes M, Fernandez-Vina M, Lazaro A, Diaz Filho L, Friedman H, Rivitti E et al. (1997) 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 49:35–40

Morini JP, Jomaa B, Gorgi Y, Saguem MH, Nouira R, Roujeau JC et al. (1993) Pemphigus foliaceus in young women: an endemic focus in the Sousse area of Tunisia. Arch Dermatol 129:69–73

Qaqish BF, Prisayanh P, Qian Y, Andraca E, Li N, Aoki V et al. (2009) Development of an IgG4-based predictor of endemic pemphigus foliaceus (fogo selvagem). J Invest Dermatol 129:110–8

Rapini RP (2005) Practical Dermatopathology. Elsevier Mosby: Philadelphia, 86–87

Robinson ND, Hashimoto T, Amagai M, Chan LS (1999) The new pemphigus variants. J Am Acad Dermatol 40(5 Pt 1):649–71; quiz, 672–3

Robledo MA, Prada S, Jaramillo D, Leon W (1988) South American pemphigus foliaceus: study of an epidemic in El Bagre and Nechi, Colombia 1982–1986. Br J Dermatol 118:737–44

Søreide K (2008) Receiver-operating characteristic (ROC) curve analysis in diagnostic, prognostic and predictive biomarker research. J Clin Pathol e-pub ahead of print 25 September 2008

Souteyrand P, Berthier-Boachon M, Thivolet J (1981) Pemphigus and thymoma: review and etiopathogenesis. Ann Dermatol Venereol 108:457–67

Warren S, Arteaga L, Rivitti E, Aoki V, Hans-Filho G, Qaqish B et al. (2003) The role of subclass switching in the pathogenesis of endemic pemphigus foliaceus. J Invest Dermatol 120:104–8

¹ Department of Dermatology and Cutaneous Surgery, University of Miami Miller School of Medicine, Miami, Florida, USA