Cell-mediated and serology-based tests for Mycobacterium ulcerans disease: A systematic review and meta-analysis.

Buruli ulcer (BU) is a subcutaneous necrotic infection of the skin caused by Mycobacterium ulcerans. It is the third most common human mycobacterial disease after tuberculosis (TB) and leprosy. The available methods for detection of the bacilli in lesions are microscopic detection, isolation and cultivation of the bacterium, histopathology, and polymerase chain reaction (PCR). These methods, although approved by the World Health Organization (WHO), have infrastructural and resource challenges in medical centres and cell-mediated immunity (CMI) and/or serology-based tests have been suggested as easier and more appropriate for accurate assessment of the disease, especially in remote or underdeveloped areas. This study systematically reviewed and conducted a meta-analysis for all research aimed at developing cell-mediated immunity (CMI) and/or serology-based tests for M. ulcerans disease. Information for this review was searched through PubMed and Web of Science databases and identified up to June 2019. References from relevant articles and reports from the WHO Annual Meeting of the Global Buruli Ulcer Initiative were also used. Twelve studies beginning in 1952, that attempted to develop CMI and/or serology-based tests for the disease were identified. These studies addressed issues of specificity and sensitivity in context of antigen composition as well as study heterogeneity and bias. The two main types of antigenic preparations considered were pathogen-derived and recombinant protein preparations. There was slight difference in test performance when M. ulcerans recombinant proteins [positivity: 67.5%; 32.5%] or pathogen-derived [positivity: 76.0%; 24.0%] preparations were used as test antigens among BU patients. However, pathogen-derived preparations were better at differentiating between patients and control groups [odds ratio (OR) of 27.92, 95%CI: 5.05-154.28]. This was followed by tests with the recombinant proteins [OR = 1.23, 95%CI: 0.27-5.62]. Overall, study heterogeneity index, I2 was 92.4% (p = 0.000). It is apparent from this review that standardisation is needed in any future CMI and/or serology-based tests used for M. ulcerans disease.


Introduction
Buruli ulcer (BU) is a disease caused by the bacteria Mycobacterium ulcerans and has been reported in the tropics and sub-tropics of over 33 countries [1], with very few cases reported in temperate areas [2,3]. Disease progression is marked by destruction of the subcutaneous skin layer, which sometimes damages nerves and blood vessels [4,5]. The destructive nature of the disease has been attributed to mycolactone, a macrolide toxin produced by the bacillus which causes apoptosis of cells [6,7]. Although a recent report indicates a 64% reduction in BU cases globally in the last 9 years [1], the incidence in Australia has increased 248% within the same time period [1]. A steady decline has been observed in endemic areas of West Africa (Ghana, Benin and Cote d'Ivoire) except in Nigeria where cases appear to be on the rise [1,8].
World Health Organization (WHO)-approved laboratory methods for detecting bacilli in lesions include: microscopic detection of acid-fast bacteria (AFB), isolation and cultivation of M. ulcerans, histopathology and polymerase chain reaction (PCR) for M. ulcerans insertion sequence IS2404 [9]. Among these methods, IS2404 PCR is considered the gold standard and is routinely used in laboratories for BU confirmation [9]. Microscopy is also used but lacks sensitivity and specificity, and histopathology is labour intensive (11). Nevertheless, both have been used for validity checks. The only method that detects viable bacilli is the isolation and cultivation of M. ulcerans on Löwenstein-Jensen (LJ) and Middlebrook media at 29-33˚C [9]. However, this requires a duration of 9 to 12 weeks or even longer [9,10]. Due to the complexity of the disease and difficulty with its diagnosis, WHO recommends two laboratory tests for BU confirmation in endemic settings [9].
A previous systematic review by Sakyi et al., 2016 [11] provided an in-depth review on various clinical and laboratory methods used to screen for M. ulcerans infection or to diagnose BU. Techniques discussed included microscopy, culture of the bacilli, histopathology, targeted PCR (including dry reagent based, real-time PCR and nested PCR) and loop mediated isothermal amplification (LAMP assay), thin layer chromatography for mycolactone detection, and serology [11]. All these tests require trained staff with access to well-equipped laboratory facilities [9].
Furthermore, there are difficulties in the accurate diagnosis of BU in endemic localities with the current diagnostic tools [12], resulting in diagnosis delay that inevitably leads to severe forms of BU and long hospitalisation [12,13]. There is a call to develop an easy to use, reliable and rapid test for BU diagnosis and surveillance in "high-risk communities" to aid timely and effective treatment [12]. Methods based on CMI and serology are thought to be reliable in principle for effective diagnosis. However, previous studies have found that in both approaches cross-reactivity with other mycobacterial infections was difficult to overcome [14,15]. In addition, immunosupression caused by mycolactone appears to inhibit the full expression of CMI and humoral immune responses in infected individuals [16,17], that could have been important as biomarkers for BU screening. Cytokine response techniques (CMI-based diagnosis) are still used to screen individuals suspected of infection with M. tuberculosis, the causative agent of the highly fatal disease, tuberculosis (TB) [18]. Although there has been cross-reactivity with the TB bacilli in M. tuberculosis infected patients, CMI and serologybased methods for BU screening have been explored by some researchers [15,19].
Unlike TB, where latent form has been described [20,21], latent form of M. ulcerans disease has not been defined. M. ulcerans has not been isolated from an individual without clinical symptoms. It appears that subclinical BU infection and active BU disease cannot be differentiated by serological or immunological testing at present.
The aim of this study is not to justify the use of CMI or serology-based methods over currently available ones. Our objective is to systematically review and conduct a meta-analysis on original research studies that aimed to develop CMI and/or serology-based tests for M. ulcerans disease, from various countries, and with different antigen preparations. We provide meta-analysis on the diagnostic potential of different preparations. We also discuss the prospects and challenges of these methods and provide perspectives on future research in this area.

Methodology
This review was conducted according to the Preferred Items for Systematic Reviews and Meta-Analysis (PRISMA).

Data sources and search strategy
The following online reference databases were searched for relevant articles: PubMed (1930 to June 2019), Web of Science / ISI Web of Knowledge (1930 to June 2019) and the Buruli ulcer disease database maintained by WHO in Geneva, Switzerland [22]. The search terms used were: "Buruli ulcer and/or Mycobacterium ulcerans" in combination with the following terms: "cell-mediated screening test", "serological screening test" and "diagnostic tests". The details of conducted searches are included in the supplement (S1 Text).

Inclusion and exclusion criteria
We included original human and animal studies with full text (or abstract if full text was not available). There was no language restriction set for the searches, however all relevant studies were in English.

Data extraction and analysis
Study rationale, study design, year of study, M. ulcerans strain, participant description, country of the study, as well as sensitivity and specificity results were extracted from selected articles to a standard table. Data collation followed the guidelines for review structure in the PRISMA checklist [23] (S1 PRISMA Checklist). The extracted data was cross-checked with a second reviewer (TMN) and any discrepancies were resolved by a senior author (NTW).

Quality of the included studies
The Newcastle-Ottawa Scale (NOS) was used to assess the quality of the included studies (S1 Data) [24]. NOS is used to assess the quality of non-randomised studies, and scored as follows: Good quality score: 3 or 4 stars in selection domain and 1 or 2 stars in comparability domain and 2 or 3 stars in outcome/exposure domain. Fair quality score: 2 stars in selection domain and 1 or 2 stars in comparability domain and 2 or 3 stars in outcome/exposure domain. Poor quality score: 0 or 1 star in selection domain or 0 star in comparability domain or 0 or 1 star in outcome/exposure domain [24].

Statistical and meta-analysis
Studies that provided sufficient information, effect size (diagnostic odd ratio) with standard error (SE) were calculated and assigned weight at 95% confidence interval (CI). STATA statistical tool was then used to conduct a random-effects model meta-analysis to assess study heterogeneity and bias. The meta-analysis results were graphically displayed as forest and funnel plots (p< 0.05).

Results
Out of 689 records identified through database searching (n = 689), a total of 119 publications were selected for full text review. After completion of full text review, 107 records were excluded because no defined protein preparation was used in those studies, or the study did not aim to develop CMI or serology-based tests for the disease. Twelve original research studies met the inclusion criteria (Fig 1 and Table 1), with the earliest from 1952 [25] and the most recent from a WHO conference abstract in 2019 presented by Avumegah [8], which was also  part of a PhD thesis [26]. Yeboah-Manu et al. [15] study in 2012 was reported in our systematic review but we could not include it in the qualitative analyses due to differences in timelines of participant group recruitments. A summary of these 12 studies are presented in Table 1. Nine out of the 12 CMI and/or serological studies identified were conducted in Africa, specifically Benin [14], Cameroon [31], Cote d'Ivoire [29], Zaire (now Democratic Republic of Congo, DRC) [27], Ghana [15,19,[30][31][32]34] and Uganda [27]. The remaining three studies, were conducted in Australia [8,26,28].
All the studies that used BU patient samples for their assay also included healthy controls [8,14,15,19,26,27,29,31,32]. It was only Stanford et al. [27], Dobos et al. [29] and Okenu et al. [30] that sought to address the cross-reactivity challenge by including patients other than BU with TB and leprosy, in their assays.

Reactivity of CMI and serology-based tests among BU patients and controls
Positive immune responses to M. ulcerans recombinantly generated, and pathogen-derived antigens results are presented in Table 2. Positive reactivity among the BU patients ranged from 70% to 85% when using pathogen-derived preparations, but the range of reactivity in controls varied greatly from 3 to 37%. The worst performing recombinant protein was ACP3, showing 47% reactivity among BU patients [34]. Also the panel of proteins used in the CMI study by Loglo et al. [34], could not differentiate BU patients from healthy controls [34], however, the 18 kDa shsp was by far a better test antigen.
The studies by Fenner et al. [25], Pidot et al. [14], Roberts et al. [28], Röltgen et al. [31] and Avumegah [8,26], reported insufficient results to extract meaningful data on the number of positive and/or negative samples and as such were excluded from Table 2. Pidot et al. [14], however, reported an average sensitivity of 69% and a specificity of 88% for their top six antigens [14]. It must also be pointed out that in this study, sensitivity and specificity values were calculated based on ELISA results of "BU endemic" (ELISA absorbance values of both BU patients + healthy endemic controls) against non-endemic controls [14]. This was based on the assumption that healthy endemic controls living in the same area as BU patients could equally be exposed to M. ulcerans without obvious symptoms of disease [14]. Yeboah-Manu et al. [15] shared the same assumption. Avumegah [8,26], who used the same recombinant proteins (ATP, Hsp65, MUP057 and MUL_2232) described by Pidot et al. [14] as test antigens in ELISA and Western blots in an Australian cohort did not find significant difference between BU patients and healthy controls with regards to specificity. The non-heat treated pathogen preparation used as antigen in ELISA was able to discriminate between BU patients from the control group with 90% sensitivity (CI = 55.50-99.75%) and a specificity of 95% (CI = 75.13% -99.87%) [8,26]. This antigen, however, was not used on TB or leprosy patients to check for cross-reactivity. On the other hand, serum antibody responses to the heat treated antigenic preparations were highly variable and lacked specificity [8,26].

Discussion
This systematic review and meta-analysis summarises the results of 12 studies which looked at developing CMI and serology-based assays for M. ulcerans disease. Our review and meta-analysis show that the experimental design for CMI or serology-based tests for BU varies greatly. In addition, there appears to be a significant heterogeneity (I 2 = 92.4%) in the test outcomes. Disparity in outcome was not entirely unexpected, which was also confirmed by the asymmetric funnel plot. It is unlikely this was due to publication bias alone. The heterogeneity and bias in the effects of PDP and RCP as test antigens could be due to differences in M. ulcerans strains used for protein preparations, protein production procedures, study populations and type of samples tested. There is the urgent need for standardisation of study design in BU. It would be appropriate for future studies involving PDP preparation to consider conducting mass spectrometric analysis of protein composition for comparative studies as did Avumegah [8,26]. It  (I 2 ) for PDP and RCP studies were > 80%, p = 0.000. The overall I 2 for entire studies was 92.4%, p = 0.000, and this is shown as^. All 7 studies had an average weight > 13.00% (range: 13.92-14.74).
https://doi.org/10.1371/journal.pntd.0008172.g002 is fair to assume that since pathogen-derived antigenic preparation is a mixture of whole bacterial lysate, it would be nearly impossible to elucidate the most reactive components in cellmediated or humoral immune response experiments without fractionation or purification. The effects of immuno-dominant and highly expressed proteins could curtail the ultimate effect of critical but low expressed proteins.
The only systematic review that is similar to our study was that by Sakyi et al. [11]. There was no discussion on CMI or serology-based assays other than that of Dobos et al. [29]. Therefore, we have provided a detailed overview of CMI and serology-based studies aimed at developing detection assays for M. ulcerans disease.
Immune modulation by mycolactone [16,38,39] could be one of the reasons CMI and serology-based methods in BU have not been popular. However, our review has provided an insight and general overview of studies that have shown promising results and that need to be further validated. There appears to be a lack of research continuity between approaches to arrive at meaningful conclusions. For example, Stanford's skin test study with burulin in Uganda and DRC [27] has not been replicated elsewhere. The same could be said about skin test in Dobos et al. study in Côte d'Ivoire with M. ulcerans cultural filtrates [29]. These two skin tests [27,29], albeit similar, were not comparable as they used different M. ulcerans strains and different protein preparation procedures. We are aware of the limitation of the potential "small sample size effect" (only 12 studies were reviewed) in exaggerating bias in our metaanalysis. However, this was beyond our control and further highlights the need for more research efforts in this area.
It is obvious there is no approved standard or study design for conducting BU CMI or humoral immunity studies. Furthermore, there is also no agreed upon test antigen or M. ulcerans strain to use. We assessed the quality of the individual studies using the Newcastle-Ottawa Scale (S1 Data) [24].

Way forward and research perspectives
There is a clear need for M. ulcerans strain selection from which protein preparations would be made for comparable and consistent CMI and humoral experiments. Screening tests based on pathogen-derived preparation and recombinant protein expression have also showed promise, but experimental optimisation and standardisation are key. The fact that pathogenderived preparations were better at discriminating between BU patients and controls is interesting. This could mean that among the M. ulcerans whole cell lysates, some unique pathogenspecific proteins are expressed and these can serve as lead candidates antigens in CMI or serology-based assays. Recombinant proteins identified could be used in combination with pathogen-derived preparations for screening purposes. A recent publication on M. tuberculosis point-of-care screening for wild animals has successfully used a combination of M. tuberculosis specific antigens as well as precipitated protein derivates (PPD) [40].
It is important to keep in mind that there will be differences in expressed pathogen antigens between M. ulcerans within the host and during in vitro experiments. Once infective within a host, M. ulcerans may potentially up-regulate particular proteins of immunogenic importance that may never be expressed in vitro (culture medium) due to differences in interactions and growth environment. Therefore, it may also be necessary to culture M. ulcerans under different growth conditions and assess the composition of cell-derived protein preparations by mass spectrometric analyses for comparative studies. This proposition is important as M. ulcerans 18 kDa shsp and other heat shock proteins/stress biomarkers are up-or down-regulated during certain adverse conditions to protect cells [41,42]. There is a need for a comprehensive genomic and proteomic evaluation among different M. ulcerans strains and other mycobacterial species to further assess this phenomenon.
In addition, normalisation of total immunoglobulin in sera from study participants could provide an opportunity to compare immune responses across study groups. It is not unusual to develop hypergammaglobulinemia (a condition of elevated immunoglobulin) during active infection with pathogenic bacteria [43], which could result in false positives. Moreover, the establishment of ELISA absorbance ranges or cut-offs for BU or M. ulcerans sero-positive indviduals would be an ideal goal of future studies.
PCR has proven to be one of the most effective diagnostic methods for clinical disease in health settings where it is affordable, however other methodologies became recently available. Two alternative methods based on specific microRNA (miRNA) and mycolactone appear to warrant further investigation. MiRNA are short non-coding RNA sequences usually between 2-22 nucleotides. They are synthesised from longer nucleotides from animals, plants and virus genomes as a result of post-transcriptional repression of gene expression [44]. Recent publications have indicated that miRNA is present in body fluids and might present a new target for infectious diseases screening [45]. A previous study also observed that BU patients have detectable levels of mycolactone in their body fluids (blood/serum) [46], which might provide a new target for the development of a new diagnostic tool. For a potential mycolactone biosensor, applications using graphene template could be explored. Through computer simulation, the surface of graphene could be functionalised as a substrate for biomolecule scavenging for potential biosensor applications [47]. This study also proposes a mycolactone and graphene interaction study using computational simulation for possible application as a biosensor for BU diagnosis.
Lastly, the widespread use of the Bacillus Calmette-Guérin (BCG) vaccine for TB and probable environmental exposure to other mycobacteria has presented research challenges due to cross reactivity in all the BU studies in Africa [15]. By comparison in Australia, which also reports BU disease, there is the potential to develop a CMI and serology-based assays as the population contains a significant proportion of people naïve to BCG vaccination. Due to public health awareness and subsequent national TB eradication program conducted from the late 1940s, Australia currently has one of the lowest notification rates for TB worldwide and mass BCG vaccination has halted since the mid-1980s [48]. This has created a BCG naïve population in the Australian BU cohort, providing an opportunity for further screening to study crossreactivity issues.
These suggestions and recommendations have also been considered elsewhere [26].

Conclusions
Based on this systematic review, pathogen-derived protein preparations appear to perform better at discriminating M. ulcerans infected patients from non-infected study controls when used in CMI and serology-based assays. We identified a need to standardise study design, protocols for reagent generation and the test design in diagnosis for M. ulcerans disease. We found a paucity of published articles attempting to develop CMI and serology-based tests for the disease. We recommend further investigations into the development of M. ulcerans disease specific and sensitive tests.