Epistatic interactions between genetic disordersof hemoglobin can explain why the sickle-cellgene is uncommon in the MediterraneanBridget S. Penmana, Oliver G. Pybusa, David J. Weatherallb,1, and Sunetra Guptaa,1

aDepartment of Zoology, University of Oxford, South Parks Road, Oxford OX1 3PS, United Kingdom; and bWeatherall Institute of Molecular Medicine,University of Oxford, John Radcliffe Hospital, Headington, Oxford OX3 9DS, United Kingdom

Contributed by David J. Weatherall, September 22, 2009 (sent for review July 30, 2009)

Several human genetic disorders of hemoglobin have risen infrequency because of the protection they offer against death frommalaria, sickle-cell anemia being a canonical example. Here weaddress the issue of why this highly protective mutant, present athigh frequencies in subSaharan Africa, is uncommon in Mediter-ranean populations that instead harbor a diverse range of thalas-semic hemoglobin disorders. We demonstrate that these contrast-ing profiles of malaria-protective alleles can arise and be stablymaintained by two well-documented phenomena: an alleviation ofthe clinical severity of �- and �-thalassemia in compound thalas-semic genotypes and a cancellation of malaria protection when�-thalassemia and the sickle-cell trait are coinherited. The complexdistribution of globin mutants across Africa and the Mediterraneancan therefore be explained by their specific intracellularinteractions.

epistasis � host genetics � malaria � thalassemia � human evolution

I n 1949, Haldane (1) proposed the malaria hypothesis toaccount for the high frequency of thalassemic blood disor-

ders observed in Mediterranean populations. He suggestedthat carriers of thalassemic genes might enjoy protectionagainst malaria, which exacted a very high mortality in South-ern Europe right up to the end of the Second World War. Suchmalaria-protective properties have since been demonstratedfor a range of genetic disorders of hemoglobin (Hb), includingsickle-cell (HbS) disease. The frequencies that these hemo-globinopathies have reached in populations living in malariousregions (2) testify to the profound inf luence that malaria hashad on recent human evolution and present a significant publichealth burden in many nations ill equipped to provide ade-quate care (3).

Inherited disorders of hemoglobin are caused by mutations ingenes encoding its �- or �-globin subunits. These mutations mayresult in structural alterations that interfere with function orstability, as exemplified at the �-globin locus by the sickle-cellallele (�S), or may reduce the rate at which one or anothersubunit is produced, leading to either �- or �-thalassemia. Thelatter are extremely heterogeneous: �-thalassemias are subdi-vided into ��, in which one of the linked pair of �-globin genesfound on chromosome 16 is deleted or inactivated by a pointmutation, and �o, in which both genes are lost*. Similarly, the�-thalassemic mutations (�T) are subdivided into �o, in whichthere is no output of �-chains at all, and �� or ���, in whichthere is, respectively, either a moderate to severe or extremelymild reduction in �-chain production (4).

The global distribution of these various globin gene muta-tions poses a number of interesting and important conun-drums. Here we focus on the contrast between subSaharanAfrica and the Mediterranean: The former is dominated by the�S and �� alleles, yet �S is uncommon in the Mediterraneanwhere �T, ��, and �0 have a significant presence. The diverserange of regionally specific �-thalassemic mutations seenacross the Mediterranean, Middle East, and Asia indicates that

they are generated frequently in malarious regions; theirabsence from most of subSaharan Africa implies that the �S

allele has outcompeted all preexisting �-thalassemic alleles atthe �-globin locus. There is good evidence from haplotypicanalyses (5, 6) that �S did reach the Mediterranean fromAfrica, so it remains a mystery as to why it has not spread moreextensively through this region.

Alpha and beta globin must combine to form functionalhemoglobin, so it seems likely that the phenotypic outcome of amutation at the �-globin locus would vary depending on theassociated �-globin genotype and vice versa. There now exist twowell-documented examples of such epistasis, as outlined below,which we believe may provide an answer to the question of whydifferent suites of hemoglobin mutations predominate in par-ticular geographical regions.

The pathophysiology of the thalassemias derives largely fromthe accumulation of unpaired globin subunits produced by theunaffected gene, which have a severely detrimental effect onerythrocytes and erythrocyte precursors. An extensive literaturesuggests that the coinheritance of both �- and �-thalassemia canameliorate the severity of these hemoglobin disorders by reduc-ing the degree of imbalance in globin chain synthesis (4, 7). Thereis also growing evidence that a negative epistatic interactionexists between ��-thalassemia and the sickle-cell trait (8, 9),such that the coinheritance of both traits cancels out theirmalaria-protective effect. This cancellation may be because themalaria protectiveness of HbS is directly linked to its concen-tration in erythrocytes, which is lowered in individuals possessingboth the sickle-cell trait and �-thalassemia, possibly because ofthe smaller pool of �-globin preferentially binding to normal�-globin (10).

Results and DiscussionEpistatic Interactions Generate Contrasting Allelic Profiles. To ex-plore the population-level implications of these two intracel-lular interactions, we developed a simple population geneticframework (see Methods) with three alleles at the �-globinlocus (WT, ��, and �0) and four alleles at the �-globin locus(WT, the sickle allele �S, �0, and ��). Each genotype wasassigned a mortality rate that was the sum of the mortalityassociated with blood disorder severity (q) and the mortality

Author contributions: O.G.P. and S.G. designed research; B.S.P. performed research; B.S.P.,O.G.P., and D.J.W. analyzed data; and B.S.P., O.G.P., D.J.W., and S.G. wrote the paper.

The authors declare no conflict of interest.

Freely available online through the PNAS open access option.

* �� and �0 are used here to represent different types of deletion, inactivating either oneor both of the �-globin genes present on chromosome 16. The genotype ��� canalternatively be represented as ‘‘�-/��’’; ���� as ‘‘�-/�-’’; �� �0 as ‘‘�-/- -’’; �0 �0 as ‘‘--/--’’,etc.

1To whom correspondence may be addressed. E-mail: liz.rose@imm.ox.ac.uk orsunetra.gupta@zoo.ox.ac.uk.

This article contains supporting information online at www.pnas.org/cgi/content/full/0910840106/DCSupplemental.

21242–21246 � PNAS � December 15, 2009 � vol. 106 � no. 50 www.pnas.org�cgi�doi�10.1073�pnas.0910840106

due to death from malaria (r); these could be modified toinclude a cancellation of malaria protection for those whocoinherit �-thalassemia and the sickle-cell trait and an ame-lioration of the blood disorder for those with both �- and�-thalassemia (see SI Appendix). Our analysis indicates thattwo evolutionary outcomes are possible within this dynamicalsystem. When the sickle allele (�S) enters the population early,it can displace �0 and ��, leading to an equilibrium state where�S and �� are the only variant alleles present, and negativeepistasis between the two prevents �� reaching unrealisticallyhigh frequencies, as shown in Fig. 1A.† However, if the entryof �S is delayed, the population may be driven toward analternative stable state in which �S is lost and �� dominates;here �� and �0 eventually drive out WT � (Fig. 1B). In essence,there are two attractors in the phase space of this evolutionarydynamical system as represented by points A and B within the3D construct shown in Fig. 1C. We propose that the present-day situation in Africa [where �� coexists with �S but ��

frequencies do not exceed 50% (2)] lies very close to theattractor A, whereas the Mediterranean scenario [e.g., 17%

�-thalassemia and 14% �-thalassemia in Cyprus, with welldocumented �0 (2)] is a point on the trajectory towardattractor B. Fig. 2 illustrates proposed subSaharan and Med-iterranean behaviors over the short term, differentiated by thetime of entry of the sickle-cell allele.

The rarity of the sickle-cell trait in the Mediterranean isgenerally ascribed to its recent import from Africa into popu-lations that already contained thalassemia (5, 11). However, ourresults indicate that the observed patterns are more robustlyexplained by the active exclusion of �S through intracellularinteractions between globin mutants rather than by invoking arecent entry of �S into the population. Fig. 3 illustrates howepistatic interactions involving � thalassemia can place popula-tions starting with relatively low levels of �� on a trajectorywhere �S is excluded, thus permitting �S to have arrived early onin the history of malaria in the Mediterranean and yet only beat negligible levels today. As indicated by the loop in Fig. 3D, ��

may suffer a decline before recovering to drive out �S. Fig. 4shows the relative frequencies of �� and �� required to prevent�S invasion, illustrating the dramatic reduction in these frequen-cies that epistasis can bring about. Finally, the range of plausible�� alleles (in terms of their clinical severity and the degree ofprotection they offer against malaria) which could resist a �S

invasion is greatly increased by the inclusion of epistasis (see Fig.S1 of the SI Appendix).

†This concurs with a simpler model considering only the effect of negative epistasisbetween ��-thalassemia and the sickle-cell trait in subSaharan Africa (8). As noted in thatpaper, this result offers a compelling explanation for the observation that frequencies of��-thalassemia in subSaharan Africa do not exceed 50%.

Fig. 1. Equilibrium states arising from a mortality framework that includes both positive and negative epistasis. (A and B) Examples of allele frequency dynamicsof populations reaching one of two equilibria associated with the genotype mortality framework given in Table SI of the SI Appendix (parameters identical tothose in Fig. 2). (C) Here we have used a 3D extension of a De Finetti diagram (26) to visualize the two possible equilibria (A and B). Within a triangular cross-sectionof the prism, the distances of a point to each side, designated �s, ��, and WT � respectively, represent the frequencies of these alleles in the population. Theheight from the base of the prism represents the frequency of the �� allele. Equilibrium A represents indefinite maintenance of sickle-cell and ��-thalassemia,and B represents indefinite maintenance of ��-, �-0, and ��-thalassemia.

Penman et al. PNAS � December 15, 2009 � vol. 106 � no. 50 � 21243

EVO

LUTI

ON

Maintenance of the �0 Allele in the Mediterranean. A second featureof Mediterranean populations under malaria selection is thatthe �0 deletion appears to have been maintained in the

population alongside ��. It has been suggested (12) that �0 isa ‘‘fugitive allele’’ because if the �-globin locus is consideredin isolation, the deleterious properties of HbH disease (the

Fig. 2. Allele frequencies achieved over time scales corresponding to likely duration of malaria selection in subSaharan Africa and the Mediterranean. (A andB) Here �S was introduced (at a frequency of 0.0001) after 500 years (A) and after 2,500 years (B), as indicated by the arrows. Initial allele frequencies were 0.0001for ��, �0, and �0, and 0.005 for ��; � � 0.04 years�1, �m � 0.005 years�1; qij and rij are given in Table S1 of the SI Appendix.

Fig. 3. Phase plane plots showing the effect of epistasis. Trajectories toward attractor A from Fig. 1 are represented with blue arrows; those towardattractor B are shown in red. (A and B) The �� homozygote has a mortality rate of 0.08 years�1 and �-thalassemic genotypes offer 50% protection againstdeath from malaria. (C and D) The equivalent parameters are 0.06 years�1 and 40% protection. (B and D) Where epistasis has been included, �� had aninitial frequency of 0.03. All other parameters were as given in Table S1 of the SI Appendix, and malaria was assumed to exert an excess mortality of0.005 yrs�1.

21244 � www.pnas.org�cgi�doi�10.1073�pnas.0910840106 Penman et al.

compound heterozygote for �� and �0) preclude the possibilityof both types of �-thalassemia coexisting stably at equilibrium.Our results indicate that �0 can be maintained indefinitely(Fig. 1B), provided that the ���0 genotype is strongly protec-tive against malaria and that coinheriting �-thalassemia re-duces the severity of HbH disease without reducing theprotection offered against malaria. There is evidence for apositive interaction between � thalassemia and HbH disease(4) at the level of the blood disorder; few studies haveconsidered the possible protective effect of HbH againstmalaria, but some in vitro evidence points to HbH RBCs beingparticularly refractory to the multiplication of the malariaparasite (13). Additionally, individuals with both HbH diseaseand �-thalassemia, although benefiting from more-balancedglobin chain synthesis and increased (even to normal) RBCsurvival time compared with those with HbH disease alone,still display certain erythrocyte membrane abnormalities as-sociated with full-blown HbH disease (4). Such membraneabnormalities might well have an effect on the suitability ofthese cells for invasion by Plasmodium falciparum and henceprovide a protective mechanism. The presence of �� may thusbe of importance to the maintenance of �0; the lack of �� in

subSaharan Africa, compounded with negative epistasis withthe sickle-cell trait, may explain why �0 is not found there.

Absence of the �� Allele in SubSaharan Africa. Within our frame-work, a population that has experienced an early entry of �S

cannot maintain ��. This finding is consistent with most ofsubSaharan Africa, with the notable exception of parts ofLiberia (14), which harbor a �-thalassemic allele (���) with amuch milder clinical phenotype than those of the Mediterra-nean (15). The sickle-beta thalassemia phenotype in thispopulation is also mild (16). Our analysis shows that such a��� allele would take an extremely long time to be driven outby �S (see SI Appendix) which could account for its persistence.A small number of West African populations also displaysignificant frequencies of the �-globin structural variant thatgives rise to hemoglobin C (HbC). HbC heterozygotes havebeen shown to be protected against cerebral malaria (9, 17) butare unlikely to enjoy as much of an advantage as �S heterozy-gotes, who are strongly protected against all forms of severemalaria (9). However, the homozygous state is relatively mild(as a hematological condition) and appears also to offerprotection against malaria (18, 19). For these reasons, HbC

Fig. 4. The effects of positive and negative epistasis on the levels of the thalassemias required to resist invasion by �S. To remove positive epistasis, all compoundthalassemic genotypes were given the higher of the two mortality rates associated with their constituent individual thalassemia genotypes (parameter valuesgiven in Table S1 of the SI Appendix). To remove negative epistasis, all genotypes heterozygous for �S were given 94% protection against death from malaria.‘‘Noninvasion’’ of �S was defined as follows: after 2,000 generations, �S (initial frequency 0.0001) must be at a frequency below 0.001 and not increasing infrequency. In the case of ‘‘no epistasis’’, �S is only prevented from invading the population when the frequency of �� is 0.9999 and �S is introduced at a frequencyof 0.0001; there is no normal �-globin present.

Penman et al. PNAS � December 15, 2009 � vol. 106 � no. 50 � 21245

EVO

LUTI

ON

may be slowly increasing in frequency (20), but it is currentlyconfined to a very restricted area.

Malaria Protection in Compound Thalassemic Genotypes. An impor-tant property of the intracellular interactions included in ourframework is that although coinheritance of �- and �-thalassemiaameliorates their respective hematological disorders, the com-pound thalassemic genotypes retain malaria protection. The cellu-lar mechanisms involved in the protection against malaria mediatedby the thalassemias are still not clear (21). Thalassemic red cells aresmall and deficient in hemoglobin, and the red cell count isconsiderably elevated; it has recently been suggested that the lattermay offer some protection against the often profound anemiaof acute malaria (22). In compound heterozygotes for �- and�-thalassemia, ���/�T�, the hematological changes are not signif-icantly different from those of �-thalassemia heterozygotes, al-though the genotype ����/�T� is associated with larger and betterhemoglobinized red cells (4). The red cells of � thalassemiaheterozygotes also have abnormal membrane function, includingincreased binding of malarial antibodies (23), presumably reflectingwell-defined metabolic defects (24); very similar abnormalities arefound in compound heterozygotes for �- and �-thalassemia (25).Thus, unlike the cancellation of malaria protection mediated bynegative epistasis between �-thalassemia and HbS, there are noobvious hematological or biochemical differences between the redcells of �-thalassemia heterozygotes and compound heterozygotesfor �- and �-thalassemia that would be likely to have an effect ondiminishing malaria resistance.

ConclusionPositive epistasis between the �- and �-thalassemias has beenwell studied at the molecular and individual level. Here we showthat, provided a degree of malaria protection is retained in thecompound thalassemic genotypes, such epistasis can have pro-found population genetic consequences, enabling populationscontaining a threshold level of �- and �-thalassemia to resistinvasion by the sickle-cell allele. We have also considered therecently described phenomenon of negative epistasis between�-thalassemia and the sickle-cell trait, which can similarly assistin the exclusion of the sickle-cell allele from a thalassemia-containing population but which more importantly enables ourframework to realistically capture subSaharan allele frequencies.In addition to offering a robust solution to the question of whythe highly protective sickle-cell allele has not penetrated Med-

iterranean populations, our demonstration that contrasting setsof hemoglobin disorders can be stably maintained within a singleframework provides a general model for the evolution of distinctsubsets of malaria-protective hemoglobinopathies as observed inother global regions.

We stress that it is only because the phenotypes of globingene variants have been so exhaustively studied that we havebeen able here to link their recognized intracellular interac-tions to observed population genetic patterns; our conclusionsthus strongly underscore the importance of phenotypic infor-mation, without which patterns of genetic variation alone cangive limited insight.

MethodsThe rate of change of frequency of each genotype, yij, with respect to time wasgiven by the following equation:

dyij

dt� �fi��1, �2, �3�fj��1, �2, �3, �4� � �qij� � rij�m�yij

where � is the mortality rate of the WT genotype in the absence of malaria;�m is the extra mortality imposed by malaria on the population; qij is therelative blood disorder associated mortality of genotype ij; and rij is therelative susceptibility to death from malaria of genotype ij. � is the totalbirth rate [equal to �i�1, j�1

i�6, j�10yij�qij� � rij�m�], which is apportioned accordingto the frequencies of allelic combinations at the �- and �-globin loci, whichtogether define genotype ij. �1, �2, �3 and �1, �2, �3, �4 are the frequenciesof all of the possible alleles at the �- and �-globin loci respectively, and thefunctions fi(�1, �2, �3) and fj(�1, �2, �3, �4) describe the frequencies of thedifferent potential genotypes at the alpha ( fi) and beta ( fj) globin loci (seeSI Appendix for details). We used Markov Chain Monte Carlo methods todetermine the ranges of mortality rate parameters that can generate theequilibrium outcomes of interest within our framework (see SI Appendix),given the reasonable assumption that ��-thalassemia is costly in the ho-mozygous state and not as malaria protective as �S. This analysis confirmeda role for negative epistasis between �-thalassemia and the sickle-cell traitand for positive epistasis between �- and �-thalassemia. The values givenin Table S1 of the SI Appendix represent just one plausible set of mortalityrate parameters [where we have used malaria protectiveness estimates forthe �� ��S, ��� ��, and ��� ��S genotypes from the literature (9)].

ACKNOWLEDGMENTS. The authors are grateful to Sean Nee, Adrian Hill,Martin Maiden, Mario Recker, Caroline Buckee, Paul Harvey, Angus Buck-ling, Charles Godfray, and Rodney Phillips for reading the manuscript andproviding comments. B. S. P. was funded by the Christopher Welch Trust,D. J. W. was funded by the Wellcome Trust, and O. G. P. was funded by theRoyal Society.

1. Haldane JBS (1949) The Rate of Mutation of Human Genes. Hereditas Suppl 35:267–273.2. Weatherall DJ & Clegg JB (2001) Inherited haemoglobin disorders: An increasing global

health problem. Bulletin of the World Health Organization 79: 704–712.3. Christianson AL, Howson CP & Modell B (2006) March of Dimes global report on birth

defects March of Dimes Birth Defects Foundation, New York.4. Weatherall DJ & Clegg JB (2001) The Thalassaemia Syndromes, 4th Ed. (Blackwell

Science, Oxford, United Kingdom)5. Ragusa A et al. (1988) � (s) gene in Sicily is in linkage disequilibrium with the Benin

haplotype: Implications for gene flow. Am J Hematol 27: 139–141.6. Boussiou M, Loukopoulos D, Christakis J, Fessas P (1991) The origin of the sickle

mutation in Greece; evidence from �(S) globin gene cluster polymorphisms. Hemo-globin 15:459–467.

7. Galanello R et al. (1989) Molecular analysis of �o-thalassemia intermedia in Sardinia.Blood 74:823–827.

8. Williams TN et al. (2005) Negative epistasis between the malaria-protective effects ofalpha� thalassaemia and the sickle cell trait. Nat Genet 37:1253–1259.

9. May J et al. (2007) Hemoglobin variants and disease manifestations in severe falcipa-rum malaria. JAMA 297:2220–2226.

10. Bunn H (1987) Subunit assembly of hemoglobin: An important determinant of hema-tologic phenotype. Blood 69:1–6.

11. Barrai I, Schiliro G, Beretta M (1987) Population structure of Sicily: Beta-thalassemiaand HbS. Hum Genet 75:1–3.

12. Wills C, Londo DR (1981) Is the doubly deleted alpha-thalassemia gene a ‘‘fugitive’’allele?. Am J Hum Genet 33:215–226.

13. Ifediba T, Stern A, Ibrahim A, Rieder R (1985) Plasmodium falciparum in vitro: Dimin-ished growth in hemoglobin H disease erythrocytes. Blood 65:452–455.

14. Willcox M, Bjorkman A, Brohult J (1983) Falciparum malaria and �-thalassaemia traitin northern Liberia. Ann Trop Med Parasitol 77:335–347.

15. Willcox MC, Weatherall DJ, Clegg JB (1975) Homozygous beta thalassaemia in Liberia.J Med Genet 12:165–173.

16. Bienzle U, Kappes R, Reimer A (1983) Sickle cell-��-thalassaemia: A haematologicaland clinical study in Liberia. Blut 47:279–285.

17. Agarwal A, et al. (2000) Hemoglobin C associated with protection from severe malariain the Dogon of Mali, a West African population with a low prevalence of hemoglobinS. Blood 96(7):2358–2363.

18. Fairhurst RM, Fujioka H, Hayton K, Collins KF & Wellems TE (2003) Aberrant develop-ment of plasmodium falciparum in hemoglobin CC red cells: Implications for themalaria protective effect of the homozygous state. Blood 101:3309–3315.

19. Modiano D et al. (2001) Haemoglobin C protects against clinical plasmodium falcipa-rum malaria. Nature 414:305–308.

20. Modiano D et al. (2008) Haemoglobin S and haemoglobin C: ‘quick but costly’ versus‘slow but gratis’ genetic adaptations to plasmodium falciparum malaria. Hum MolGenet 17(6):789–799.

21. Weatherall DJ (2008) Genetic variation and susceptibility to infection: the red cell andmalaria. Brit J Haemat 141:276–286.

22. FowkesFJIetal. (2008) Increasedmicroerythrocytecount inhomozygous ��-thalassaemiacontributes to protection against severe malarial anaemia. PLoS Med 5:e56 OP.

23. Luzzi GA et al. (1991) Surface antigen expression on plasmodium falciparum-infectederythrocytes is modified in �- and �-thalassemia. J Exp Med 173:785–791.

24. Knox-Macaulay HH, Weatherall DJ (1974) Studies of red-cell membrane function in het-erozygous beta thalassaemia and other hypochromic anaemias. Br J Haematol 28:277–297.

25. Knox-Macaulay HH, Weatherall DJ, Clegg JB, Bradley J, Brown MJ (1972) The clinicaland biosynthetic characterization of ��-thalassaemia. Br J Haematol 22:497–512.

26. Cannings C, Edwards AWF (1968) Natural selection and the De Finetti diagram. AnnHum Genet 31:421–428.

21246 � www.pnas.org�cgi�doi�10.1073�pnas.0910840106 Penman et al.