The assembled library of sources was reviewed for population surveys of G6PD variants and those meeting a specific set of inclusion criteria (Figure  1) were identified. First, only surveys that could be geopositioned to at least the national level were included, and these were mapped to the highest spatial resolution available, ideally as point locations (e g, villages). Second, to ensure that survey samples were widely representative of the communities being studied, only surveys which provided apparently unbiased prevalence estimates were included: all case studies and patient groups, malaria patients (to avoid underestimating frequencies of G6PD variants which may provide a protective effect against malaria), family studies, and samples selected according to ethnic background were therefore excluded. Third, only surveys using molecular diagnostics were included to avoid the diagnostic uncertainty of surveys reliant on biochemical methods (see Additional file 1).

Surveys meeting these criteria were of two types depending upon the nature of the population samples surveyed and were mapped as two separate series:

Studies examining individuals who had been previously identified as phenotypically G6PDd from a population screening survey and were being followed-up for molecular diagnosis of underlying mutations were included in map series 1. Pie charts were chosen to display the relative proportions of the different variants identified in each of the surveys, with each colour-coded segment proportional to the variant’s relative frequency across the study sample. Sample size was reflected in the relative sizes of the pie charts, which was transformed on a square-root scale to allow clear visualization.

It is important to note that studies did not always attempt to identify all the variants represented in the legends of the maps. For instance, if only G6PD Mediterranean was tested for among a sample of deficient individuals, those individuals who tested negative would be included in the “Other” category, regardless of whether they expressed one of the other variants listed in the legend or a completely different one. The absence of a specific variant from a sample can only be inferred if no “Other” samples are reported.

Surveys which investigated G6PD variants in cross-sectional population samples without prior phenotype screening were included in map series 2. These studies estimated the allele frequencies of selected G6PD variants at the population level and were mapped spatially using bar charts to represent the allele frequencies. This visualization conveyed the important concept that frequency estimates are only available for variants that were included in the diagnoses, underlining the importance of knowing which variants had been tested for. The variants tested for were named in the vertical bars of the bar charts. Empty spaces along the x -axis indicate that the named variant was tested for but not identified in the population sample. The size of the plots meant that in some cases, significant spatial uncertainty was introduced to their positioning on the map (Additional file 1).

The G6PD gene’s X-linked inheritance means that allele frequencies correspond to frequencies among males. Heterozygous inheritance in females makes allele frequency estimates from female samples harder to calculate confidently as not all studies consistently distinguished heterozygotes from homozygotes. For reliability, therefore, only data from males were included in these variant allele frequency maps.

Given the diversity of G6PD variants, for the purposes of this mapping study it was necessary to identify those variants that presented a primary public health concern. This study focussed on Type 2 variants which (i) express significantly reduced enzyme activity (<50%); (ii) are clinically significant by predisposing individuals to haemolytic anaemia; and, (iii) reach polymorphic allele frequencies at the population level (Table  1). Although 77 Type 2 variants have so far been described 26 , the database of population survey data compiled in this study indicated that most variants were reported fewer than ten times (most fewer than three) and, when identified, were found to have low prevalence. A clear subset of variants was responsible for the majority of investigated G6PDd cases (Additional file 1); these were defined for the purposes of this study to be those variants reported from a minimum of ten localities across the malaria-endemic region. This ensured that the spatial distributions of the most significant variants could be legibly visualized in the maps. Fifteen variants met these criteria: G6PD A-, Canton, Chatham, Chinese-5, Coimbra, Gaohe, Kaiping, Kalyan-Kerala, Mahidol, Mediterranean, Orissa, Quing Yan, Union, Vanua Lava, and Viangchan (Figure  1).

A total of 18,939 bibliographic sources were identified from keyword searches, of which 2,176 were considered likely to include spatial information about G6PD variants. Following their detailed review, 93 published and unpublished sources were identified which reported surveys eligible for inclusion in the maps (listed in the Additional file 2). From these sources (which could each report several surveys from multiple locations), 138 population surveys could be geolocated and met the criteria for map series 1, as these were of community samples which had undergone prior screening for phenotypic G6PDd. Map series 2 was populated with 24 geolocated population surveys of community samples providing variant frequency estimates. Map series 1 data were predominantly from populations in Asia (123/138 surveys; 89%), while map series 2 data were almost exclusively from the Africa + region (Africa, Yemen and Saudi Arabia 53 : 20/24 surveys, or 83%). These database descriptors are summarized in Table  2.

The maps reveal conspicuous distinct geographical patterns in the distribution and prevalence of G6PD variants across regions. The two series of maps represent both the relative proportions of the variants responsible for phenotypic G6PD enzyme deficiency (Figures  2, 3, 4, 5, 6, 7) and the allele frequencies of selected variants (Figure 8 and Additional files 3, 4 and 5). Together, these maps revealed a number of clear patterns further described below: (i) a relatively low diversity of G6PD variants reported from populations of the Americas and Africa + regions, among whom variants of the G6PD A- phenotype are predominantly searched for and reported; similarly, G6PD Mediterranean was predominant in west Asia (from Saudi Arabia and Turkey to India); (ii) a sharp shift in variants identified east of India, showing no admixture with the common variants of west Asia; (iii) high variant diversity in east Asia and the Asia-Pacific region, with multiple variants co-occurring and no single variant being predominant.

Relatively few surveys were available from the Americas (nine in map series 1, one in map series 2). Figure  2A indicates that among G6PDd individuals, the predominant variant was G6PD A- (in the majority of cases encoded by the G6PD A- ^202A mutation, though the G6PD A- ^968C variant was common in some samples), identified in 91% of deficient individuals surveyed across the region (579 of 636 total G6PDd individuals surveyed). Other variants identified included G6PD Mediterranean ^563T and Seattle ^844C (the latter was classified as part of the “Other” category as it was only reported twice, both instances from Brazil where it reached a prevalence of 13% (n = 2/15, [Additional file 2: S63] and 5% (n = 9/196, [S30])). A survey of Mexicans reported 61% of deficient cases being due to various G6PD A- variants, but did not test for any other variant (e g, G6PD Mediterranean), explaining the large proportion of “Other” variants [Additional file 2: S85].

A single community screening survey could be identified which investigated G6PD variant allele frequencies. This was in Acrelândia, Brazil, and searched for only one single-nucleotide polymorphism (SNP): G6PD A- ^202A . An allele frequency of 6% (n = 509, [Additional file 2: S14]) was reported (see Additional file 3).

Very few studies (n = 6) investigating the variants of known G6PDd individuals were identified from the Africa + region (Figure  3). Where available, both in West Africa and the western Indian Ocean islands, surveys indicated G6PD A- ^202A to be the predominant variant; in Saudi Arabia, the G6PD Mediterranean variant was responsible for 78% of G6PDd cases in two surveys in the western coastal city of Jeddah [Additional file 2: S6,S28].

Surveys investigating population-level frequencies of specific G6PD variants were more common (Figure  8 and Additional file 4). These consistently looked for SNPs of G6PD A-, notably focussed on the G6PD A- ²⁰² locus. Frequencies of this SNP were over 20% in surveys from Nigeria and Côte d’Ivoire in West Africa, and found at 19% in coastal Kenyan populations. The G6PD A- ^968C mutation encodes a similar phenotype to G6PD A- ^202A and was found to be the more common of the two in West African populations (in The Gambia and Senegal, for instance [Additional file 2: S21,S25]); despite this, only six of 20 surveys across Africa + included the G6PD A- ⁹⁶⁸ SNP in their diagnoses. Large population surveys in East Africa (n = 311 in Uganda [Additional file 2: S38] and n = 2,756 in Kenya [Additional file 2: S76]) found no evidence of the G6PD A- ^968C mutation despite frequencies of 12 and 19% of the G6PD A- ^202A mutation, respectively.

Most population surveys (70%) in the Africa + region tested the G6PD A- ³⁷⁶ locus. Although this mutation has barely any clinical effect, it is commonly co-inherited with mutations at other loci, together encoding the range of G6PD A- variants (Figure  8). The G6PD A- ^376G variant had very high prevalence among sub-Saharan African populations, and in all cases affected over 10% of individuals surveyed; further, in a subset of ten out of 16 of these surveys, its frequency reached 25-50%. Exceptions to this high G6PD A- ^376G prevalence were two surveys (n = 46 and 55) from northern Sudan, which reported low (<5%) prevalence [Additional file 2: S41]; the G6PD A- ^202A mutation was not detected in these samples.

Greatest G6PD variant diversity globally was across the Asia and Asia-Pacific regions (Figures  4, 5, 6, 7) where up to ten variants were reported to co-occur within single populations. Furthermore, significant proportions of “Other” cases were frequently reported in the variant proportion maps (map series 1), indicating that genetic diversity is greater than represented by the pie charts in these maps.

From Turkey to Pakistan (Figure  5), G6PD Mediterranean was predominant, identified in 728 of 895 G6PDd individuals examined (81%). Two variants, G6PD Kalyan-Kerala ^949A and Orissa ^131G , were reported exclusively from Indian populations. On the Indian sub-continent, these two variants and the G6PD Mediterranean variant represented the majority (88% of n = 555 G6PDd individuals tested) of deficiency cases, though notable proportions (up to 80%) of “Other” cases were also reported from eastern and southern India.

East of India, a completely different set of variants emerged (Figure  6). G6PD Mahidol ^487A predominated across Myanmar, with 98 of 117 G6PDd individuals (84%) diagnosed across 14 locations as carrying this variant. Variants common among G6PDd individuals in southeast China were largely unique to these populations. Of the 2,883 G6PDd cases from China, the commonly diagnosed variants included G6PD Kaiping ^1388A (total G6PDd cases: 827; 29% across China), Canton ^1376T (792 cases; 27%), Gaohe ^95G (265 cases; 9%), Chinese-5 ^1024T (104 cases; 4%) and Quing Yan ^392T (87 cases; 3%). The distribution of G6PD Viangchan ^871A was diffuse, reportedly common from Laos (where examination of 15 G6PDd individuals all carried this variant [Additional file 2: S34]) and Cambodia (reported from 61 out of 64 G6PDd individuals [Additional file 2: S43]) to Papua New Guinea (where a sample of 13 G6PDd individuals included nine with this variant [Additional file 2: S33]).

Variants across the Asia-Pacific region were also highly heterogeneous (Figure  7). The common pattern emerging from the 36 surveys of deficient individuals identified from this region (n_indivs = 635) was the diversity of variants and the heterogeneity of their prevalence among different populations. Furthermore, it should be noted that 20 of these surveys tested fewer than ten G6PDd individuals, limiting the diversity that would be captured and thus the representativeness of these reports. The greatest diversity identified in a single survey was from the Malaysian neonatal screening programme in Kuala Lumpur, which recorded ten variants from an investigation of 86 G6PDd newborns [Additional file 2: S4]. The short spatial-scale heterogeneity of this co-occurring variant diversity is illustrated by a set of surveys across the East Nusa Tenggara province of Indonesia (Figure  7B). Amid this co-occurring variant heterogeneity, the G6PD Vanua Lava ^383C variant was the most common variant reported from Indonesian studies (identified in 84 out of 249 G6PDd individuals tested).

Only three allele frequency surveys (map series 2) could be identified from the Asia region (of 24 globally) [Additional file 2: S66,S83]. G6PD Mahidol was most commonly tested for, with allele frequencies of 12% and 18% (n_tested = 353 and 11, respectively; Additional file 5). Two surveys in Thailand tested for a broader range of variants; one found five variants (G6PD Canton, Kaiping, Mahidol, Union and Viangchan; n_tested = 84) at allele frequencies of 1-2%; and the other, with a smaller sample size (n_tested = 11), identified only the G6PD Mahidol variant [Additional file 2: S66].

The G6PDd phenotype has long been considered homogenous across African populations, with much of the pre-molecular era literature reporting G6PD A- from electrophoresis assays across the continent 34 . The early investigations of “primaquine sensitivity” among G6PD A- volunteers found primaquine-induced haemolysis to be mild and self-limiting 1 30 64 67 68 69 . Nevertheless, the haemolytic susceptibility of this “mild” variant has been observed through severe reactions requiring transfusion 70 , as well as through the failure of the Lapdap (chlorproguanil-dapsone) anti-malarial trials 71 72 , and as haemolysis induced by the ingestion of fava beans 73 , a pathology previously thought to only be triggered by more severe variants 74 . Haemolysis associated with G6PD A-, while perhaps less severe than with other variants in some circumstances or settings, ought not be universally characterized as “mild” as this risks leading to misplaced confidence in a relatively cavalier application of the drug.

Molecular analyses of G6PD variants among Africans revealed a diversity of SNPs within the umbrella of the G6PD A- phenotype 75 . The data, here mapped spatially, indicated a transition of SNP predominance from west (G6PD A- ^968C polymorphism) to east (G6PD A- ^202A widespread and G6PD A- ^968C apparently absent). However, expansion of the relatively small number of surveys (n = 20) may reveal more complex distributions. Furthermore, the true diversity of G6PD variants cannot be discerned from the assembled surveys; it is not possible to know what proportion of genetic diversity is reflected in the map (Figure  3) because the diagnoses are limited to assessing the presence or absence of a limited number of anticipated SNPs, without any prior identification of phenotypically deficient hosts (as in map series 1). Full gene sequencing is the only reliable way of identifying all gene variants in the population. Genetic diversity within the umbrella G6PD A- phenotype may explain the spectrum of expression levels observed 70 : inheritance of a single SNP may not be sufficient to explain the observed phenotype. If these patterns of co-occurring SNP diversity encoding variable G6PD A- phenotypes are widely observed, a haplotype-based approach which allows for the co-inheritance of multiple SNPs may be required if the use of molecular diagnostics is to the scaled-up, as seems to be increasingly common 75 . Despite its obvious limitations to field-based applications, a gene-wide haplotype perspective would also allow increased compatibility with the rapidly expanding human genome sequencing efforts which are cataloguing human genetic diversity 76 77 .

Although demand for primaquine has historically been low across Africa 78 , it was Ethiopia that reported the highest number of P. vivax cases globally in 2011 (665,813 cases; 79 ). G6PDd prevalence has been estimated at 1.0% in Ethiopia (50% CI: 0.7-1.5) 2 , though no molecular information exists to indicate which variants may be responsible. Furthermore, beyond these areas with acknowledged endemic P. vivax, the status of transmission in human populations previously considered refractory to infection 78 is being brought into question by several lines of evidence 80 81 82 83 84 85 86 . The use of primaquine for P. falciparum control may also increase following the new WHO gametocytocide guidelines 63 . The elimination agenda includes African nations, with most located at the fringes of entrenched holo-endemic transmission 87 88 , and the pressure to apply primaquine both as a hypnozoitocide for P. vivax and Plasmodium ovale, and as a gametocytocide for P. falciparum will increase with further progress towards zero transmission. A more detailed and targeted understanding of African G6PD epidemiology is necessary beyond the opportunistic surveys presented in this paper.

The G6PD Mediterranean variant, appearing predominant across southwest Asia and exceedingly common in much of India (especially western regions), is known to be highly sensitive to AHA induced by primaquine. This wide distribution and dominance of G6PD Mediterranean presents a significant hurdle to the malaria elimination programmes in West Asia, which include Azerbaijan, Georgia, Iran, Iraq, Kyrgyzstan, Saudi Arabia, Tajikistan, Turkey, and Uzbekistan 89 . These nations must consider the relatively high risk of serious harm caused by unintentional dosing of G6PDd patients in need of malaria therapy. Implementation of G6PDd screening prior to primaquine radical cure therapy would greatly mitigate such risks and accelerate progress to malaria elimination in this region.

Although G6PD Mediterranean was common across Indian populations, two other variants were also frequently reported which were indigenous to populations from this subcontinent: G6PD Kalyan-Kerala and Orissa. Little is known of the primaquine sensitivity phenotypes of these two variants. Since almost half of the global population at risk of P. vivax malaria occurs in this single nation 52 79 90 studies characterizing such sensitivity would ultimately prove very useful in planning towards elimination.

The maps of eastern Asia and the western Pacific present the most complex picture of G6PD variants globally, coinciding with regions of heterogeneous and high P. vivax endemicity 52 . Almost all the common variants of public health concern globally were reported from this region. Reasons for this genetic diversity are unclear, but it is interesting that P. falciparum parasites (postulated to be selective agents of G6PDd 16 ) have been found to show a greater degree of population structure with lower genetic relatedness between populations in Asia than across Africa 91 . As well as this overall diversity, the structure of G6PD variant heterogeneity with multiple variants co-occurring is starkly different from other areas where single variants predominate.

Despite large variant diversity across this region where many countries now target elimination (thus require primaquine therapy) 89 , only one truly Asian variant has been examined in relation to AHA sensitivity to primaquine. G6PD Mahidol, common across Myanmar and parts of Thailand (Figure  6), exhibits residual G6PD enzyme activity ranging between 5-32% 42 . A handful of small primaquine sensitivity studies examining this variant have been conducted in Thailand, reporting very mild to moderate sensitivity to both 14-day and eight weekly primaquine regimens 40 92 93 , as well as to single-dose regimens of P. falciparum transmission blocking therapy 94 . However, the G6PD Mahidol primaquine sensitivity phenotype must not be assumed to be representative of the region. As emphasized by the WHO Expert Review Group on primaquine in 2012 62 63 , far greater evidence-based understanding of these phenotypes among the many common Asian variants is urgently needed. The identity and distribution of variants as sensitive as G6PD Mediterranean in the Asia-Pacific region remains unexplored and unknown. This lack of understanding threatens patients and programmes with a more aggressive roll-out of primaquine therapy. In populations of such high heterogeneity, a dual approach of phenotypic screening followed by variant analysis of residual enzyme activity may be required to identify such risks.

The danger of under-diagnosing deficiency by using molecular identification alone (i e, the data informing map series 2) is illustrated by comparison of the two types of maps (Figure  2C and Additional file 5). If only a selected subset of perceived “common” variants are used in population screening surveys, a proportion of deficient individuals may be missed and put at risk from primaquine therapy: only those variants that are looked for can be identified. The maps presented here also repeatedly highlight considerable proportions of “Other” G6PD variants. These correspond to an unknown haemolytic risk, and hint towards an ever greater diversity of variants. Only full gene sequencing will allow full characterization of the diversity in this polymorphic gene.