Introduction

Leishmaniasis presents as a spectrum of diseases and is caused by obligate intracellular protozoa of the genus Leishmania which is transmitted by sand flies of the genus Phlebotomus. The three main clinical forms of the disease are the cutaneous (CL) form, mucocutaneous (MCL) and the potentially fatal visceral form, visceral leishmaniasis (VL), also called kala-azar [1]. Human VL is present worldwide, but over 90% of cases are found in a few highly-endemic countries (Brazil, Ethiopia, India, Kenya, Somalia, South Sudan and Sudan) [2]. VL is caused by members of the L. donovani complex, more specifically the closely related species L. infantum and L. donovani, the latter being responsible for kala-azar in East Africa.

Patients with kala-azar present with fever, splenomegaly, and weight loss. Clinical diagnosis of VL is inaccurate and can be difficult in endemic settings as several causes of febrile splenomegaly exist, notably malaria [3]. A direct parasitological diagnosis of VL is obtained through microscopy, culture and PCR on tissue aspirates. Obtaining tissue aspirates is invasive and well-trained health professionals are needed to perform this high-risk sampling. However, microscopy lacks sensitivity whereas culture entails weeks to yield a final result and is vulnerable to bacterial contamination. In recent years, molecular techniques have been shown to have the highest diagnostic sensitivity in aspirate or tissue samples, and to be highly specific. However, in resource-limited settings, the use of the PCR remains largely restricted to referral hospitals and well-equipped research centers [4]. Alternative diagnosis by urinary antigen detection or antibody detection through direct agglutination test (DAT), immunofluorescence assays (IFA) and enzyme-linked immunosorbent assays (ELISA) require experienced laboratory technicians, specific laboratory equipment and is time consuming. As many VL patients are poor and live in remote areas, easy and rapid tools for the diagnosis of VL should be selected in accordance with WHO ASSURED criteria (affordable, sensitive, specific, user-friendly, rapid, equipment free and deliverable where required) [5]. Rapid diagnostic tests (RDTs) for Leishmania antibody detection have been implemented in endemic settings for over a decade. In combination with clinical symptoms, RDTs provide high diagnostic accuracy in suspected first-time episodes of VL [6, 7]. Performance of commercially available assays differs throughout regions depending on the antigen used, age of the affected population and the immune status of the patient. Experience with RDT mainly consists of RDTs with antibody detection specifically against recombinant K39 (rK39), a 39-amino acid repeat that is part of a kinesin-related protein whose presence is conserved within the L. donovani complex. A limitation of the rK39 based assays is its lower sensitivity in East Africa compared to the Indian subcontinent [1,8, 9]. A new synthetic polyprotein was developed, rk28 which is a fusion molecule of L. donovani haspb1, haspb2 and LdK39 tandem repeat regions [10]. Modest improvement in sensitivity of rK28 RDTs over rK39 RDTs was reported in East Africa [11].

VL is highly endemic in both East Africa and the Indian subcontinent but relatively more VL-HIV coinfections are prevalent in East-Africa. Prevalence of coinfection in Northwest Ethiopia varies greatly over different studies. A recent meta-analysis estimated the VL-HIV coinfection prevalence to range from 20.88% to 24.86% [12]. Few studies have assessed the performance of serological assays in the VL-HIV coinfected population in East Africa, but current data suggests a lower sensitivity in VL-HIV coinfected patients [13]. Our objective was to assess the performance of different antibody tests and one antigen test for the diagnosis of new VL cases and compare within and between HIV negative and HIV coinfected patients in an VL- endemic region in Northern Ethiopia.

Materials and methods

Results

Test performance

No invalid test results were obtained for any of the tests. Performance of all evaluated tests in both HIV and non-HIV VL patients and controls are presented in Table 1. Overall, sensitivities ranged from 69.2% for DAT with cut-off ≥ 1/3200 to 92.2% for the Onsite RDT. Sensitivity was lower in the HIV positive group for all tests, though this was only statistically significant for rK39 Serion ELISA (p = 0.0084) and OnSite RDT (p = 0.0159) for which sensitivity in the non-HIV group reached 100%. DAT titers and rK39 Serion ELISA results according to HIV coinfection are illustrated in Fig 1. For DAT there was no significant difference (p = 0.1856) between proportions (number of patients with a given titer or ratio) of the HIV positive and HIV negative group, opposed to rK39 Serion ELISA (p = 0.0270).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Sensitivities and specificities of DAT, rk39 ELISA and six RDTs for VL diagnosis.

https://doi.org/10.1371/journal.pntd.0008963.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. DAT and rK39 ELISA results stratified by VL-HIV coinfection.

(A) rK39 ELISA Serion U/mL results compared between VL-HIV co-infected (black) and non-HIV VL patients (light grey) and (B)DAT titer distribution.

https://doi.org/10.1371/journal.pntd.0008963.g001

There was no statistically significant correlation between the test result proportions and parasite density for any of the tests except for DAT (Pearson Chi square p-value 0.0419 with cut-off ≥ 1/800 and 0.0019 with cut-off ≥ 1/3200). Also at a cut-off ≥ 1/3200, DAT was positive in 55 out of 74 (74.3%) microscopy positive cases, and only in 8 out of 17 (47.1%) microscopically negative cases (confirmed through PCR on slide).

Overall specificities ranged from 20,0% for Kalazar Antigen ICT to 100.0% for IT Leish, and rK39 IgG1. None of the tests were positive in the 20 endemic healthy controls. Out of 10 non-endemic control samples 1 sample was positive with the Kalazar Detect RDT, 1 other sample for Onsite RDT and 8 samples with the Kalazar Antigen ICT. One out of 10 malaria positive samples tested positive with both rK28 IgG1 and Leishmania IgG ELISA, though still within borderline limits. Two other malaria samples were found positive with the Kalazar Detect RDT and eight with Kalazar Antigen ICT and Onsite RDT.

Agreement between two tests was calculated (Table 2). The highest agreement was noted between rk39 Serion ELISA and IT Leish (96.9%) and the lowest between DAT at cut-off ≥ 1/3200 and Kalazar Antigen ICT (56.8%). Median values obtained by rK39 ELISA were compared across qualitative results.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Percentage of concordant results between all tests.

https://doi.org/10.1371/journal.pntd.0008963.t002

Discussion

With this study we aimed to assess diagnostic accuracy of multiple tests for VL diagnosis in HIV-coinfected Ethiopian patients as data are scarce. Our data have shown that sensitivity decreased in the HIV-VL coinfected group with some tests being more affected than others. To date, there is a lack of evidence for the minimum sensitivity and specificity of a single test or algorithm for VL diagnosis. As clinical presentation lacks specificity, and treatment is toxic, laboratory confirmation is necessary. Previously it was recommended to have a minimum sensitivity of 95% [19] and the test should be able to make a distinction between asymptomatic patients, treated patients and acute diseased patients. Current Ethiopian national guidelines on the diagnosis of a first-time VL episode start with a non-invasive anti-rK39 RDT upon presentation of clinical symptoms [16]. If positive, the patient should be considered as a new VL case and receive treatment. If negative, a DAT test is performed with a positive cut-off titer of ≥ 1/3200 and negative ≤1/400. DAT positive patients are referred for treatment, borderline results should be repeated or splenic, bone marrow, lymph node aspiration could be considered if safe. Rarely a DAT negative result would require aspiration. There is no different algorithm for HIV patients, and even with reduced immune response due to HIV infection, sensitivity is still considered acceptable in the national guideline. An algorithm with high sensitivity to rule out or exclude VL infection and a high specificity to avoid these invasive sampling procedures and unnecessary treatment is needed.

Overall, none of the tests evaluated here reach an individual sensitivity of 95% nor sufficiently high specificity. When applying the current algorithm to these results with the most sensitive RDT (Onsite RDT), DAT and aspiration if required, still 6 out of 90 (6.7%) cases (all HIV positive patients) went undetected and 95% sensitivity was not reached either in our study. As per previous study protocol, 5 of these 6 patients underwent splenic or bone marrow aspiration and obtained an aspirate positive (grades 2 to 6) result but they may not have been tested according to the algorithm because of a negative DAT result. Replacing DAT by the second most sensitive RDT (IT-Leish), would result in detection of one additional case. Remarkably, DAT was less sensitive in comparison with previous reported East African results, ranging from 83.5% to 99.1% [11,20–27]. Of note, sensitivities of DAT are often compared against a parasitological diagnosis obtained through microscopy which already lacks sensitivity. This was also shown here as 17 additional cases were afterwards detected through PCR on slides reported negative by microscopy. There are only limited data on the sensitivity of PCR on Giemsa-stained slides compared to microscopy on VL tissue aspirates, however it seems to have improved sensitivity [28, 29]. Assays detecting total antibodies against rK39 all showed a comparable sensitivity ranging from 89.0% to 91.2%. There was no significant difference in this study with sensitivity from the Onsite RDT detecting antibodies against rK28, though previously this has been reported to be higher in East African patients compared to rK39 antigen based RDTs [10, 11]. Presumably the reduced sensitivity of rK39 RDTs in East Africa may be a reflection of molecular diversity of the kinesin gene [30], or by human genetic differences moreover as VL is mainly a pediatric disease in the Indian subcontinent compared to East Africa where all age groups are affected [11]. Not only throughout continents, but also differences in sensitivities within a country may occur, as recently reported by a group from Brazil [31]. Albeit the Onsite RDT exhibits a high sensitivity, a problem implementing this as a screening test is its low specificity in a malaria endemic setting, as 8 out of 10 acute malaria cases reacted false positive (Table 1). Malaria is highly endemic in Ethiopia with an estimated of almost 3 million new cases in 2016 [32]. Prevalence of VL/malaria co-infection in Ethiopia remains largely unknown, but in the Metema hospital in Ethiopia a prevalence of 4.3% has been reported [33]. Aside from coated antigens, other elements influence performance, e.g. the format (ELISA, RDT, lateral-flow vs dipstick…), conjugate used, concentrations of components, sample type (serum, plasma or whole blood) and sample volume applied, etc. The Onsite RDT requires a higher sample volume for testing compared to the other RDTs. Dilution of sera has previously been successful in eliminating cross-reactivity against malaria [25], however this may negatively impact sensitivity. An rK39 antigen used in RDT format showed lower sensitivity in Sudan compared to Indian patients, but the same antigen in an ELISA format, results in comparable sensitivities across countries [25]. Here, sensitivity for the anti-rK39 ELISA was comparable to the anti-rk39 RDTs detecting IgG antibodies. None of the RDTs intrinsic quality suffered from invalid results, inter-reader variability or reproducibility problems.

Though these RDTs are implemented in diagnostic settings, they do not distinguish between an acute and treated infection as they detect either IgG or total antibodies. The only antigen detecting test evaluated here did not prove valuable because of its very low specificity. Previously, IgG1 was found to be the predominant antibody present in VL infection and has been explored as a potential biomarker of post-chemotherapeutic relapse [34, 35]. Therefore it was hypothesized that IgG1 detection may also be useful in predicting progression to VL infection among asymptomatic patients which was previously supported by some limited data [35]. Rates of asymptomatic Leishmania infection in endemic regions in Ethiopia are estimated to be 10–20% [36] and there is a need for biomarkers to predict progression. Data on the IgG1 titers and sensitivity in VL diagnosis and especially HIV/VL coinfected patients are still sparse [35]. Here, the 2 prototype IgG1 detection antibody tests against rK39 and rK28 antigens showed lower sensitivities compared to total IgG antibody detection when used as a diagnostic assay. Sensitivity in Indian VL patients for the rK39 IgG1 RDT was previously reported to be 94,7%–100% depending on the concentration of antigen used [37]. The lower sensitivity obtained here could be due to different population characteristics between continents. However, samples in the study by Mollett et al were selected based on a rK39 total IgG positive test and microscopy at diagnosis, which could contribute to the higher sensitivity they found. The observed results confirms that the IgG1 RDT’s are not intended for use as a VL diagnostic assay but as a test of cure or relapse. Moreover these sensitivities may be improved by increasing reagent concentrations, combining rK28 and rK39 antigens or using Leishmania lysate as previously described [35].

In northwest Ethiopia, up to 24.% of VL cases are HIV co-infected [12] and this is likely an underestimation as only 17% of VL cases are screened for HIV [38]. Current guidelines recommend HIV screening in VL cases [36]. However, HIV screening should already be performed upon clinical suspicion of VL to guide correct diagnostic algorithms. Sensitivities from all assays decreased upon stratification by HIV status, but the impact of an HIV co-infection was significant for two tests, the rK39 Serion ELISA and the Onsite RDT. Lower sensitivities are a corroboration of previous results [21,31] and likely due to the decrease in CD4⁺ T-cells- with as a consequence impaired B-cell stimulation and reduced antibody response [39]. Gradoni et al [40] suggested that the sequence in which both infections are acquired may impact the seropositivity for VL, unfortunately it is often impossible to retrieve which infection was acquired first. Antibody titers in the rK39 ELISA were significantly lower in the HIV coinfected group, though this was not significant for DAT. In contrast to our results, a meta-analysis proved DAT to be the most effective serological technique in the immunosuppressed by HIV infection [21]. Some tests were 100% sensitive in the non-HIV infected group and could be used to rule out VL infection. However, the current algorithm does not allow to rule out VL among the HIV infected. Agreement of test results between assays were calculated to assess if other algorithms would be of interest or tests could be used in parallel to increase sensitivity and specificity, however with the tests evaluated here this is not possible. Even when combining all antibody detection tests together, still four out of 51 HIV VL coinfected cases would go undiagnosed. Current antibody test cannot reliably rule out VL infection in HIV infected patients. Recently, a test was developed combining the advantages of both DAT and RDTs, with a leishmanial membrane extract used in an RDT format [41]. In Ethiopia this test had 100% sensitivity, but low specificity. It could be a valuable tool to rule out VL in the future, but more data are needed, especially in HIV patients. Other antigens have been explored, like rKLO8 and rKE16 [25,42], but also demonstrated regional differences in performance.

Limitations of this study are (i) it’s relatively small sample size influences the variability of results obtained, (ii) samples were analyzed retrospectively and might have undergone multiple freeze-thaw cycles, with a negative impact on sensitivity, however, antibodies are unlikely to decrease substantially and this would apply to all assays evaluated, (iii) results from patients in control groups were compared against DAT as a reference test as opposed to PCR and/or microscopy on tissue aspirates, which is invasive—this mayintroduce verification bias, (iv) PCR was not systematically performed on all slides; however, microscopy positive results are highly specific and less sensitive compared to PCR and this would unlikely have changed the results, (iii) we have no information on what infection was acquired first, HIV or VL, nor on HIV disease status or HAART, which may influence antibody titers.

Performance of commercial antibody detecting RDTs was variable and sub-optimal. Improvement of sensitivity and specificity with better diagnostic tests is needed. The IgG1-based assays are not appropriate as screening assays due to low sensitivity, but need to be assessed as tools for follow-up of treatment. Assay performance depends on regional differences, associated with parasite divergency and on patient characteristics. Therefore, it is strongly recommended to select the appropriate tests and algorithm to a specific endemic setting and verify performance before implementing it. Separate algorithms for HIV coinfected against non-HIV infected would be valuable. Negative serology does not reliably rule out VL infection in HIV patients. This highlights the need to establish clear guidelines on minimal performance criteria of individual tests and algorithms, but also against which reference these should be compared as parasitological identification through microscopy only has low sensitivity. Large prospective evaluations of both serological and molecular tests taking all these variables into account are required to enable creation of clear guidelines for use of tests in the field.

Supporting information

Acknowledgments

We want to acknowledge all staff from the Leishmania Research and Treatment Centre, Gondar, Ethiopia for their cooperation on this project. We thank Michael Miles and Tapan Bhattacharya for providing the rK28 antigen for development of the IgG1 rapid test.