Automated malaria detection by depolarization of laser light

Authors


Professor B. V. Mendelow Department of Haematology, Wits Medical School, 7 York Road, Parktown 2193, South Africa.

Abstract

Anecdotal experience with full blood count (FBC) technology incorporating analysis of depolarized laser light (DLL) for the enumeration of eosinophils showed that malaria infection generated unusual distributions in the white cell channels. The objective of this study was to identify and define criteria for a diagnosis of malaria using this technology. To determine sensitivity, specificity, and positive and negative predictive values, 224 directed samples referred specifically for malaria were used; true positives were defined as those in which malaria was identified by microscopic and/or immunological methods. For the DLL method, positive was defined as one or more large mononuclear cell(s) for which the 90° depolarized signal exceeded the 90° polarized signal. To determine possible utility in a routine haematology laboratory setting, 220 random undirected FBC samples were evaluated for possible malaria infection by the DLL method. Of the 224 directed samples, 95 were malaria positive as determined by microscopic and/or immunological methods, and 129 were negative. For the DLL method, overall sensitivity was 72% (90% in the case of Black Africans), and specificity 96%. Positive and negative predictive values overall were 93% and 82% respectively. In the utility study a single positive result was identified among the 220 samples studied. This was found to be from a patient with malaria. The detection of unexpected malaria by automated screening FBC analysis could substantially lower the mortality and morbidity from unascertained infection, especially in indigenous African peoples.

Of the 300–500 million cases of malaria infection which are estimated to occur annually, nearly half are due to Plasmodium falciparum and the overwhelming majority occur in Africa (Nussenzweig & Long, 1994); approximately 2.7 million of these are fatal. The outcome of P. falciparum infection is very much dependent on the rapid commencement of appropriate therapy. Consequently, any delay in diagnosis has a substantial effect on the risk of mortality (Greenberg & Lobel, 1990). Current diagnostic techniques, whether by microscopic examination of suitably stained thick and thin blood smears or by immunological methods (Beadle et al, 1994), are predominantly directed tests that confirm or refute a clinical suspicion. Although the detection of malaria parasites in the blood film component of a routine full blood count (FBC) may provide an unexpected diagnosis of malaria, this method is also dependent on a high index of suspicion, and is notoriously unreliable in instances of low parasitaemia. Moreover, demands for a rapid turnaround time and the widespread deployment of automated counters that incorporate white cell differentials are limiting still further the role of microscopy in the routine FBC. These factors are especially relevant in the developed world where the populations are malaria non-immune and where physicians have a low index of suspicion of malaria (Kyriacou et al, 1996). As a further factor, emerging resistance patterns are limiting the effectiveness of prophylactic chemotherapy (Longworth, 1995), which, if taken at all, often merely delays the onset of symptoms and lowers even more the index of suspicion. By the time the first-world tourist who has visited a malaria-endemic area presents with a pyrexia of unknown origin, all too often the geographic history is forgotten, the automated screening FBC is powerless to provide the diagnosis, and the outcome is often fatal.

The screening FBC has become throughout the world a rapid and inexpensive yet valuable component in the diagnostic investigation of any patient, and particularly the patient with a pyrexia of unknown origin. In such patients the white cell count and differential are especially useful, as the patterns of white cell abnormality are often indispensable clues as to the nature of infecting micro-organisms. A variety of techniques have been employed to enumerate the various white cell populations. The most recent of these relies on multiple angle polarized light scattering patterns to subdivide the white cell categories (Goossens et al, 1996; Tatsumi et al, 1996). Because of the characteristic optical properties of their granules, eosinophils are uniquely capable of depolarizing and scattering a laser light beam at 90° to its original path. In this report we describe how malaria pigment located within white cells fortuitously shares this property with eosinophils, probably because of the optical characteristics of malaria pigment (Lawrence & Olsen, 1985). This finding could have a major role in large-scale diagnostic screening and surveillance programmes, including those of relevance to hospital and blood-transfusion services, especially in the malaria-endemic areas of Africa. It is noteworthy that automated FBC technology is available in many parts of Africa. As the instrument was not programmed to record depolarization of laser light due to red cell contents, it is feasible that with appropriate software modifications this technology could also exert a major impact on the timely and therefore effective diagnosis of malaria in first-world travellers returning from third-world countries. Because such patients are almost invariably non-immune to malaria, the effects on mortality reduction could be substantial.

METHODS

For the DLL method, the Abbott CELL-DYN 3500 cell counter was used. Details of this technology have been published (Goossens et al, 1996; Tatsumi et al, 1996). Briefly, white cell populations are initially subdivided into mononuclear and polymorphonuclear fractions, based on differential 10° (complexity) and polarized 90° (lobularity) light scatter properties. The 90° depolarized light scatter channel was designed to separate eosinophils from neutrophils among the polymorphonuclear fraction. In this study the unique and unexpected capacity of products of malaria parasites to depolarize laser light reflected from particles designated as large mononuclear cells was exploited to identify malaria parasites. A positive result was defined entirely objectively as one or more large mononuclear cell(s) for which the 90° depolarized signal exceeded the 90° polarized signal. Red cells and platelets are not analysed in the 90° depolarized channel.

Clinical studies were carried out in two phases. To determine sensitivity, specificity, and positive and negative predictive values, 224 directed samples referred to the microbiology laboratory specifically for malaria were used. With the exception noted below, truth was defined by microscopic examination of Giemsa-stained thick and thin blood films (Dacie & Lewis, 1995), performed independently of the DLL analysis, but on the same specimens. Microscopy and DLL analysis were carried out by different observers in different laboratories and without initial knowledge of the other's result. Samples were also analysed by the immunological Para-sight F method (Beadle et al, 1994), which detects the P. falciparum specific histidine-rich protein antigen, PfHRP-2. DLL positive, microscopy negative cases with a history of positive microscopy within the previous 3 months, and a positive immunological reaction, were regarded as true positives.

In a second phase of clinical study, designed to determine possible utility in a routine haematology laboratory setting, 220 random undirected FBC samples were evaluated for possible malaria infection by the DLL method.

RESULTS

Light scatter patterns of normal blood cells are depicted in Fig 1. In contrast, typical malaria-positive results are shown in Fig 2. The presence of one or more particles identified as large mononuclear cells (purple dots), with 90° depolarized signal in excess of the 90° polarized signal, was found to be strongly associated with a diagnosis of malaria. Although small mononuclear cells (blue dots), with 90° depolarized signal in excess of the 90° polarized signal, were also seen, particularly in malaria, they were not as strongly associated with the diagnosis of malaria. Other features commonly encountered in malaria-positive cases were thrombocytopenia and abnormally distributed particles identified by the instrument as eosinophils (green dots).

Figure 1.

. Scattergram from a sample of normal blood. White cells are analysed initially with respect to lobularity (90° polarized), and complexity (10°). This identifies mononuclear and polymorphonuclear populations. The polymorphonuclear population is then divided into neutrophils and eosinophils on the basis of granularity (90° depolarized) and lobularity (90° polarized), and colour coded for all subsequent analyses into orange and green dots respectively. The mononuclear population, meanwhile, is segregated into lymphocytes (blue), monocytes (purple) and basophils (black), on the basis of differential size (0°) and complexity (10°). These populations are combined finally into a composite scattergram, representing each population as a group of appropriately coloured dots. In normal blood, or blood without malaria infection, there are no instances of large mononuclear cells (purple), which scatter more depolarized than polarized light at 90°.

Figure 2.

. Malaria was diagnosed by the DLL method by the presence of one or more particles identified by light scatter at angles of 0° (size), 10° (complexity) and 90° polarized (lobularity) as large mononuclear cells, which the programme identifies as purple dots, with 90° depolarized signal in excess of the 90° polarized signal (above the 45° line). Scanty (left) and abundant (right) diagnostic events were from patients with equivalent degrees of parasitaemia (<0.5% in each case).

Of the 224 directed samples from 175 different patients, 95 samples were malaria positive as determined by microscopic and/or immunological methods, and 129 were negative. The results are summarized in 1Table I. Among the positive samples, all but two were identified on microscopy as P. falciparum. Parasitaemias ranged from <0.1% to 25%. Eight samples showed >10% parasitaemia. There appeared to be no correlation between the percentage parasitaemia and the number of large mononuclear particles depolarizing laser light. Overall sensitivity and specificity of diagnosis were 72% and 96% respectively. Positive and negative predictive values were 93% and 82% respectively, with a prevalence of 42%. This reflects the actual prevalence from the perspective of the microbiology laboratory processing samples referred for malaria diagnosis. In a subset of nine cases where there was discordance between the DLL analysis, which was positive, and the microscopy, which was negative, the immunological test was positive and they were assigned to the true positive category. This was considered justified on the grounds of a true positive history within the previous 3 months. In one case DLL analysis was positive yet both the microscopic and immunological tests were negative. Further scrutiny of this initially apparent DLL false positive disclosed that the patient had been diagnosed as positive on microscopy 4 d previously, although the immunological result was also negative on that occasion. Accordingly, the true result for the subsequent test was amended to positive. Five false positive cases were recorded in the directed sample study. One of these was associated on one occasion with a weakly positive immunological test result, but this was not reproducible on a repeat test, and the result of a false positive was retained. Of the other four false positive cases, two were from the same patient. Of the two non-P. falciparum cases, the DLL result was positive in one (P. ovale).

Table 1. Table I. Patients overall.Thumbnail image of

False negatives were recorded in 27 cases. They were encountered in indigenous African subjects in only six out of 139 samples from 103 different patients (Table II). Of these samples, 53 were true positives. False negatives, curiously, were more frequent in white patients (Table III), in whom they occurred in 20/81 samples from 68 patients. Of these samples, 15 were true positives. Sensitivities for black and white subjects were therefore 90% and 43% respectively, and specificities were 96% for both groups. For prevalences of 42% (black patients) and 43% (white patients), positive predictive values were 95% and 88%, and negative predictive values 93% and 69%, respectively. There were four other patients (Asian) whose results (three true negative and one false negative) were included in the overall data, but do not appear in 2Tables II or 3III.

Table 2. Table II. Black African patients.Thumbnail image of
Table 3. Table III. White patients.Thumbnail image of

In the utility study of the undirected samples referred to the haematology laboratory for routine FBC, a single positive result was identified among the 220 samples studied; this was found to be from a patient with malaria.

DISCUSSION

This study has demonstrated the feasibility of automated malaria diagnosis. That this approach is part of an existing unmodified strategy, applied in the automated white cell differential component of a full blood count, is an extremely fortunate coincidence, because FBC itself is a mandatory investigation in the investigation of any patient with a pyrexia of unknown origin. Accordingly, it is likely that this technology has the potential to ascertain at least some patients in whom the specific diagnosis of malaria has not, for whatever reason, been considered clinically. The utility study of this report was a successful attempt to simulate this scenario, from the perspective of the haematology laboratory. As failure to establish an initial diagnosis has been identified as the cause of 40% of the mortality associated with malaria in the developed world (Greenberg & Lobel, 1990; Lobel et al, 1985), the benefit of this approach could be substantial. Although there were too many false negatives for the technique to be regarded as an alternative to existing methods, it must be stressed that the technology is completely unmodified from that designed for an entirely different application, that of white cell differential counting, and there are preliminary indications that the capacity to detect malaria by this method could be substantially improved by design features aimed at optimizing this unanticipated application. One simple example of this could be software modification to flag samples meeting the diagnostic criteria described in this report. More fundamental design modifications to identify depolarizing red cells would have the potential to improve sensitivity of diagnosis dramatically, as malaria diagnosis limited to white cell scrutiny could never compare with what might be achieved by the equivalent analysis of red cells. These studies are under way at present.

More than 10 years ago malaria pigment was found to be birefringent, and it was predicted that this property could be used to automate the detection of malaria (Lawrence & Olsen, 1985). Data from in vitro studies (not shown) indicated that the DLL method probably detects malaria pigment. We believe pigment-laden monocytes, which have ingested parasites and therefore malaria pigment, to be the major component of the defining population in our study. The number of such cells, and of pigment-containing neutrophils, has been identified as an adverse prognostic indicator, of much greater value than the percentage parasitaemia (Hoan Phu et al, 1995). This raises the intriguing possibility that the diagnostic approach proposed in our study could also be useful prognostically. This is supported by the unusual distributions of what the instrument identifies as eosinophils in certain malaria-positive cases. From the parameters displayed by these populations, they appear to be indicative of pigment-laden neutrophils. In our study we have not attempted to correlate such populations with outcome. Another source of positive signals could be gametocytes or schizonts, which would be expected to share many of the optical properties of mononuclear cells, and their comparatively large store of pigment would be a very effective depolarizer of laser light.

The diagnostic events we describe do not appear to be due to the ring forms which dominate the blood picture viewed by conventional microscopy. Whether they are caused by gametocytes, schizonts or by pigment-laden monocytes, or a combination of these, no correlation with percentage parasitaemia is to be expected, and none was found. From this it is clear that this approach cannot be used to monitor therapy. For this reason, and also to identify Plasmodium species, blood film examination is still required for clinically suspected cases and as a necessary sequel in clinically unsuspected cases identified by this method.

The very large difference in the sensitivities applicable to indigenous Black African peoples (90%), as opposed to White patients (43%), is of interest. We have not determined the reason for the differences found. Possibilities could include polymorphisms such as red cell antigens, MHC determinants, or variations in tumour necrosis factor sequences, all of which are known to impact on malaria pathogenesis (Field et al, 1994; Udomsangpetch et al, 1993; Miller et al, 1976; Hill et al, 1991; Mcguire et al, 1994). Other genetic factors could also be considered. For example, it has been suggested that parasitaemic load might be determined by an unidentified gene (Abel et al, 1992). It has also been reported that Black patients may respond differently to infection than Whites (Powell, 1989). Alternatively, previous exposure or varied patterns of presentation could account for some of the differ-ences found. Whatever the reason for the difference, it is clear that deployment of this technology, even without modification, in malaria endemic areas is likely to be particularly effective in ascertaining unexpected malaria patients.

In conclusion, we have described a novel approach to malaria diagnosis whose main utility is in the fact that it has the potential to make an unexpected diagnosis as part of the screening FBC. Although the sensitivity leaves much room for improvement with hardware and software design modifications, which are in progress, the unmodified technology already has the ability to identify many patients whose diagnosis would otherwise be missed.

Acknowledgements

The authors thank Dr John Frean, of the South African Institute for Medical Research, for helpful advice on malaria pigment. This work was supported by grants from the South African Institute for Medical Research and the Medical Research Council of South Africa.

Ancillary