PTX3 Genetic Variation and Dizygotic Twinning in The Gambia: Could Pleiotropy with Innate Immunity Explain Common Dizygotic Twinning in Africa?


Giorgio Sirugo, MD, PhD, Centro di Ricerca, Ospedale San Pietro Fatebenefratelli, Via Cassia 600, 00189, Rome, Italy. Tel: +39–6-33585872; Fax: +39–6-33251278; E-mail: or Scott M. Williams, PhD, Dartmouth Medical School, Department of Genetics, 7400 Remsen, Hanover, NH 03755. Tel: +1 603-646-8171; E-mail:


Dizygotic (DZ) twinning has a genetic component and is common among sub-Saharan Africans; in The Gambia its frequency is up to 3% of live births. Variation in PTX3, encoding Pentraxin 3, a soluble pattern recognition receptor that plays an important role both in innate immunity and in female fertility, has been associated with resistance to Mycobacterium tuberculosis pulmonary disease and to Pseudomonas aeruginosa infection in cystic fibrosis patients. We tested whether PTX3 variants in Gambian women associate with DZ twinning, by genotyping five PTX3 single nucleotide polymorphisms (SNPs) in 130 sister pairs (96 full sibs and 34 half sibs) who had DZ twins. Two, three and five SNP haplotypes differed in frequency between twinning mothers and those without a history of twinning (from P= 0.006 to 3.03e-06 for two SNP and three SNP haplotypes, respectively). Twinning mothers and West African tuberculosis-controls from a previous study shared several frequent haplotypes. Most importantly, our data are consistent with an independently reported association of PTX3 and female fertility in a sample from Ghana. Taken together, these results indicate that selective pressure on PTX3 variants that affect the innate immune response to infectious agents, could also produce the observed high incidence of DZ twinning in Gambians.


It has been known for a long time that “with mankind a tendency to produce twins runs in families” (Darwin, 1871, citing Sedgwick, 1863) and there is evidence that familial dizygotic (DZ) twinning is at least, in part, genetically determined (Parisi et al., 1983; Schmidt et al., 1983Hoekstra et al., 2008). Although the fundamental biological phenomenon in di- and multizygotic pregnancies is multiple ovulations, the underlying molecular basis of the trait is not understood.

In some areas of West Africa, the twinning frequency is three to four times the rate seen in Europeans, most notably in the Yoruba population of South-Western Nigeria. At 45.1 twin pairs per 1000 births, the reported twinning rate in Yoruba is four times that of women in Europe and America (10/1000) and more than nine times the twinning rate (5/1000) of some Asian countries (Nylander, 1978; Bowman, 1990; Hoekstra et al., 2008). Trizygotic triplet pregnancies (1.6/1000) are 16 times more common in the Yoruba than European or Asian populations (Nylander, 1969; Nylander, 1971; Nylander, 1978; Pison, 1992; Vogel & Motulsky, 1997). A recent report has supported the notion that twinning is more common throughout most of sub-Saharan Africa than the rest of the world. However, the actual rates are somewhat lower than previous reports, and Benin, not Nigeria, is argued to have the highest frequency (Smits & Monden, 2011). It is not known why DZ twinning is so common in Africa, although some have suggested that the high frequency of twinning/multiple ovulation in West Africans has been the result of a selective reproductive advantage for twinning mothers compared to mothers of singletons (Sear et al., 2001). It is obvious that any selective pressure model would also imply a genetic predisposition.

To identify a DZ twinning predisposing locus, we recruited sisters who had DZ twins from the Gambia, hence enriching for putative genetic factors and compared them to ethnically matched women with no known family histories of DZ twinning. The Gambia is a unique country for studying the genetics of DZ twinning in West Africa because of a very high twinning rate ranging from ∼1.5% to ∼3% of which more than 75% are DZ twins (Jaffar et al., 1998; Sear et al., 2000; Sirugo et al., in preparation). Importantly, there is no reported significant use of fertility drugs or consumption of phytoestrogen-containing foods, minimizing the effects of known environmental agents that increase twinning frequency (Newman & Luke, 2000).

In our study, we tested for association between SNPs in the Pentraxin 3 (PTX3) gene and DZ twinning. The PTX3 gene maps on chromosome 3q25 (MIM ID 602492) and belongs to the Pentraxin superfamily that is highly evolutionarily conserved. The superfamily members are divided into Long and Short Pentraxins, where C-reactive protein and serum amyloid are Short Pentraxins produced in the liver and PTX3 is an example of a Long Pentraxin. PTX3, as other Long Pentraxins (PTX4, NP1, NP2, and NPR) is characterized by a 174 amino acid-long amino-terminal domain and it is expressed on dendritic cells as well as macrophages, following toll-like receptor activation and inflammatory cytokine production (e.g., IL-1, TNF-α). Of significance to our study, PTX3 has an important role in female fertility, i.e., in the delivery of the cumulus oophorus–oocyte complex to the oviduct as well as in determining successful fertilization (Varani et al., 2002). Further biological evidence of the role of PTX3 in reproduction is supported by a mouse model, where matrix-embedded PTX3 can direct and facilitate entrapment of spermatozoa and hence the fertilization of eggs (Salustri et al., 2004). Our study involved two analyses for association, one where we compared cases to controls using only a single twinning sister per sibship (92 cases vs. 95 ethnically matched controls) and another where we used all twinning sisters (195 total cases).

Materials and Methods

In this report, we present information on a sample of 130 affected sister pairs each of whom had DZ twins and 95 healthy Gambian control women with no evidence of twin deliveries. With the exception of five sets of twins who were of the same sex and for which we determined zygosity by genotyping SNPs from unrelated autosomal loci, all the twin sets in the sample were of different sex. In total, we recruited 195 women who had DZ twins from 92 families for a total of 130 sibpairs, 96 of whom were full sisters and 34 were half-sisters. The pairs were recruited from across The Gambia and their age, family relationships, parity, food consumption habits, and ethnicity (traced back to the grandparents) were assessed using a specific study questionnaire. Unrelated controls (n= 95) were recruited from the same sites and matched for ethnicity. Written informed consent was obtained from all subjects. The study was approved by the combined Gambian Government/MRC National Ethics Committee.

PTX3 genotypes for five intragenic SNPs (TaqMan assay, ABI, Foster City, CA, USA) were determined in genomic DNA samples, obtained by a standard salting out method. SNPs were the same as reported in Olesen et al. (2007). Marker positions and allele frequencies are shown in Figure 1 and Tables 1 and 2.

Figure 1.

PTX3 gene structure. Gene map with 3′ UTR region, exons, and introns. SNPs position and intermarker distances are indicated.

Table 1.  Gene and SNP information.
Gene and chromosomeSNPPosition (relative to reference assembly)Gene RegionAmino acid change
Pentraxin 3rs2305619158637555Intron 1 (boundary)
 Chromosome 3q25rs3816527158638008Coding exon 2Ala→Asp
 rs1840680158638723Intron 2
 rs3845978158642388Intron 2
 rs26141586436933′ UTR
Table 2.  Case-control single locus association based on case resampling.
SNPAlleleAllele FrequencyHWE P valueCase v Control P value mean (SD)
Mean (SD) caseControlMean (SD) caseControlAllelicGenotypic
rs2305619C0.45 (0.02)0.440.70 (0.25)0.530.79 (0.18)0.72 (0.19)
rs3816527C0.25 (0.02)0.280.67 (0.28)1.000.57 (0.22)0.71 (0.19)
rs1840680T0.30 (0.02)0.290.56 (0.30)0.460.78 (0.18)0.50 (0.22)
rs3845978T0.25 (0.02)0.240.53 (0.30)0.390.72 (0.21)0.81 (0.16)
rs2614T0.12 (0.01)0.110.72 (0.28)1.000.80 (0.19)0.78 (0.21)

We performed a case-control analysis by selecting one twinning sister from each family as a case (92 total cases per analysis). To ensure that our results were not biased by the “case” selection process we repeated this random selection 1000 times. The 1000 case datasets were compared to the same controls, and all analyses were repeated 1000 times.

Single site allele frequency, genotype frequency, and Hardy–Weinberg Equilibrium analyses were performed using PLINK and the means and standard deviations for the results are presented (Purcell et al., 2007). Statistical significance was determined using χ2 tests. An alternative analysis, Generalized Estimating Equations (GEE) that had the capacity to adjust for relatedness was also performed, in which we included all twinning mothers from a family to test whether our sampling biased the association findings (Hancock et al., 2007). GEE is powered to detect associations in datasets containing concordant sibpairs, and adjusts correlation because of shared genetic and environmental factors between cases by modeling the covariance structure of the correlated measurements. Simulation studies have shown that GEE estimates are robust to misspecification of the covariance matrix. STATA 11.0 statistical software (College Station, TX, USA) was used for this analysis.

Haplotype analyses to test for linkage disequilibrium (LD) and to determine haplotype frequencies were performed using the Powermarker software on the 1000 randomly selected case sets and controls (Liu & Muse, 2005). This program uses an EM algorithm to determine haplotype frequency distributions when phase is unknown and it was run both using a sliding window of two–three SNPs, as well as the complete set of five SNPs. The Powermarker haplotype trend analysis performed is a regression approach to test haplotype-trait association for a dichotomous or continuous trait. The test for association then uses an F test for a specialized additive model.

An alternative haplotype analysis method, CCREL version 3.0 (Browning et al., 2005) was also employed because CCREL is optimized for haplotype association testing in study designs containing related cases and unrelated controls with haplotype phase unknown. Therefore, CCREL was able to be run using the full dataset (n= 195 cases, 95 controls) and was used to perform two–three locus sliding window haplotype analyses as well as single locus genotypic association tests; four–five SNP haplotypes were not assessed because of potentially increased Type 1 error beyond three SNP haplotypes. CCREL accounts for the correlations between related case individuals because of IBD sharing by calculating an optimal “weight” for each individual based on their unique IBD sharing probability. These “weights” are then utilized to construct a composite likelihood, which is then maximized iteratively to form likelihood ratio tests for haplotype and single-marker association testing. The likelihood ratio test is asymptotically equivalent to a χ2 test of association using the aforementioned “weighted” counts of each haplotype/allele in cases and controls. To reduce the degrees of freedom of the likelihood ratio tests, and to optimize the overall efficiency of the test, rare haplotypes, defined as having ten or fewer expected observations, were pooled together with the next larger haplotype group by using the combined threshold = 10 option. However, the pooling per se did not drive the results, therefore justifying this approach. Previous studies, on both actual and simulated data, show that CCREL is more powerful than methods that employ χ2 testing after selecting one member of each pedigree (Browning et al., 2005). We also used CCREL to assess single SNP associations in our data.

Pairwise LD was characterized and standard summary statistics D' and r2 were calculated using the HaploView statistical software (Devlin & Risch, 1995; Barrett et al., 2005).

Results and Discussion

No single SNP associations were found between DZ twinning and PTX3, and results across analytical methods were highly consistent (Tables 2 and 3). However, both the sliding window haplotype and the entire five SNPs-haplotype analyses demonstrated significant association with PTX3 using the resampling procedures (Table 4). Of note, the analyses using related cases provided significant evidence for an effect with the same three SNP haplotype (rs3816527-rs1840680-rs3845978) as the resampling method, with P at the 10−6 level (Tables 4 and S1). The two SNP haplotype, rs3816527-rs1840680, was also significant using CCREL (P= 0.012) and almost significant using our resampling method (P= 0.07; Tables 4 and S2).

Table 3.  Analyses of twinning mothers including related cases using GEE and CCREL.
95% CI
SNPOR1Lower1Upper P value1 P value2
  1. 1 P values from single marker genotypic test of association using GEE

  2. 2 P values from single marker genotypic test of association in CCREL v 3.0

Table 4.  Haplotype sliding window association.
HaplotypeMean P value (SD)1LR P value2
  1. *In bold are statistically significant P values (P≤0.05)

  2. 1 P-values based on 1000 random samples from the case families. Standard deviation of P values are in parentheses

  3. 2 P-values are from CCREL weighted likelihood ratio haplotype tests, combined threshold = 10; N/A: not applicable because of method instability (see main text)

rs2305619-rs38165270.69 (0.17)0.577
rs3816527-rs18406800.07 (0.07) 0.012
rs1840680-rs38459780.22 (0.14) 0.006
rs3845978-rs26140.76 (0.16)0.53
rs2305619-rs3816527-rs18406800.20 (0.14) 2.42e-03
rs3816527-rs1840680-rs3845978 0.04 (0.04) 3.03e-06
rs1840680-rs3845978-rs26140.23 (0.13) 7.12e-04
rs2305619-rs3816527-rs1840680-rs3845978-rs2614 0.05 (0.06)N/A

Our study provides strong evidence that PTX3 variation has a significant association with DZ twinning in The Gambia. From a functional point of view, it is well known that PTX3 is a physiological downstream target of GDF9 (Varani et al., 2002) and GDF9 mutations have been associated with both DZ twinning and increased ovulation rate (Montgomery et al., 2004; Palmer et al., 2006). PTX3 expression and secretion in the periovulatory cumulus oophorus has a key function in the assembly of the hyaluronic-rich extracellular matrix, known to facilitate fertilization (Russell & Salustri, 2006). However, in our sample twinning associates with PTX3 genetic variation per se, that is regardless of any transacting control, such as GDF9, that could also contribute to the twinning phenotype.

Although our results support a strong association of PTX3 with DZ twinning in our population sample, it is not clear whether DZ twinning is of benefit in terms of reproductive fitness. Specifically, the fitness role of DZ twinning may be related to the level of resources in a given location and time (Lummaa et al., 1998; Helle et al., 2004); in resource limited environments twinning is associated with an overall decrease in reproductive fitness (Lummaa et al., 1998; Helle et al., 2004). In addition, in some West African countries (e.g., Guinea-Bissau) it has been shown that twins are breastfed six months longer than singletons, potentially limiting the number of pregnancies (P. Aaby, 2002, personal communication). More importantly, the risk of maternal mortality is three–five times higher in twin pregnancies and infant mortality is higher in twins than singletons (Hoj et al., 2002), so it is far from obvious that twinning itself represents a reproductive or fitness advantage in sub-Saharan Africa, where resources are often limited. This argument is contrary to an advantage of DZ twinning per se in our population and would lead to the prediction that twinning should not be so frequent in The Gambia. Therefore, we provide an alternative explanation for its high frequency, based on studies of PTX3 demonstrating other roles for the gene.

PTX3 binds a number of infectious organisms, ranging from fungi to bacteria, activates complement and facilitates phagocytes, making PTX3 an important mediator of the innate immune response (for a comprehensive review see: Garlanda et al., 2005; Mantovani et al., 2008; Bottazzi et al., 2009). The link with innate immunity may provide an alternative explanation for the distribution of PTX3 haplotypes in West Africa. Specifically, PTX3 haplotypes that we show to associate with DZ twinning have been shown to confer protection from infectious diseases, and are therefore likely to have been under positive selection for this protective effect (Olesen et al., 2007; Chiarini et al., 2010). This leads us to ask whether “protective” SNPs or haplotypes could also have an effect on fertility, partially contributing to the unusually high frequency of DZ twinning in The Gambia. Specifically, we postulate a direct, independent effect of the same PTX3 mutations, on both immune responses to pathogens and multiple ovulations, i.e., a pleiotropic model where DZ twinning would occur in parallel with and independently from immunological pathways because of the PTX3 function on both innate defenses and on fertility. Specifically, in a previous study of tuberculosis susceptibility it was found that PTX3 haplotypes associated with protection from disease in controls. Of significance for this study, non-DZ twinning haplotypes tracked almost perfectly with the haplotype distribution in TB cases (e.g., A-C-A-C-C, 0.24 in TB cases and 0.25 in non-twinning mothers compared to 0.18 in TB controls and 0.19 in twinning cases) and haplotypes more common in TB controls are also more common in DZ twinning mothers (e.g., G-A-G-C-C, 0.22 in TB controls and 0.28 twinning cases compared to 0.17 and 0.24 in TB cases and non-twinning mothers; Table 5). The very similar “protective”PTX3 haplotypes, observed in Bissau TB controls and in Gambian twinning mothers suggests that the effect is common across West Africans who belong to different ethnic groups. Further strengthening this argument is the observed linkage disequilibrium (LD) pattern in the TB cases, which is almost identical to that in our non-twinning mothers (Fig. 2A–D for r2; Fig. S1 for D'), indicating that they are tagging one or more variants common to both populations that affect both phenotypes or act pleiotropically. Additional support for this conclusion comes from a study investigating the effect of PTX3 genetic variants on fertility in a female Ghanaian population sample that identified an association in the gene region encompassing SNPs rs2305619-rs3816527-rs1840680, corroborating the notion that PTX3 affects reproductive characteristics in West Africans (May et al., 2010). Finally, this hypothesis is further reinforced by another study in which the same haplotype that associates with protection from pulmonary tuberculosis in Bissau and with DZ twinning in The Gambia has been shown to associate with protection from Pseudomonas aeruginosa airway infection in European patients with CF (Chiarini et al., 2010). We would argue that it is highly unlikely to have observed this recurrent haplotype pattern associating with different phenotypes by chance alone because combining the probabilities from these independent studies yields an overwhelmingly significant result (<<0.05); this strongly indicates that the PTX3 haplotypes we studied tag functional variation. LD analyses (Fig. 3) of rs1840680 and adjacent SNPs data from a GWAS of tuberculosis in Gambians (data of the Wellcome Trust Case Control Consortium,, Thye et al., 2010) and from the 1000 genomes project study of Yoruba trios from Nigeria, show significant LD encompassing PTX3 and extending on both sides of the gene (up to 125kb at 3’ end of the gene). This pattern is consistent with neutral variants hitchhiking with one or more beneficial PTX3 mutations on limited length haplotypes resulting from a combination of selection pressure and recombination rate, i.e., a selective sweep across PTX3. Finally, the region encompassed by the SNPs we genotyped includes five previously defined missense mutations (Fig. 4). One of these was genotyped by us (rs3816527), but it is not in LD with any of the other four SNPs in the Yoruba HapMap samples. Although PolyPhen-2 predicts that this is a benign mutation with respect to protein structure and function (; Adzhubei et al., 2010), we cannot rule out the possibility that there is another effect such as mRNA stability differences, which is captured by the haplotype that associates with twinning.

Table 5.  Haplotype frequencies and association for five marker haplotypes and comparison to tuberculosis study haplotype distributions.
Haplotype rs2305619-rs3816527-rs1840680-rs3845978-rs2614Haplotype frequencyMean (SD) P value
Case mean (SD)Controls
A-C-A-T-C0.05 (0.01)0.01 0.03 (0.03)
A-A-G-C-T0.09 (0.02)0.100.70 (0.21)
G-A-G-T-C0.14 (0.02)0.210.11 (0.08)
A-A-G-C-C0.16 (0.02)0.170.68 (0.21)
A-C-A-C-C0.19 (0.02)0.250.15 (0.12)
G-A-G-C-C0.28 (0.02)0.240.32 (0.21)
Haplotype rs2305619-rs3816527-rs1840680-rs3845978-rs2614Haplotype FrequencyHaplotype frequency TB study
Case mean (SD)ControlsControlsCases
  1. *In bold are statistically significant P values (P≤0.05). P values are generated comparing each haplotype to all others

A-C-A-T-C0.05 (0.01)0.01NANA
A-A-G-C-T0.09 (0.02)
G-A-G-T-C0.14 (0.02)
A-A-G-C-C0.16 (0.02)
A-C-A-C-C0.19 (0.02)
G-A-G-C-C0.28 (0.02)
Figure 2.

DZ twinning and non-twinning PTX3 LD structures. Linkage disequilibrium (LD) structures for pairwise r2 between markers characterizing haplotype blocks in PTX3 in non-twinning (A), twinning (B), TB cases (C), and TB controls (D). All figures are oriented 5′ to 3′, left to right. Strong LD is indicated by dark gray, whereas light gray and white indicate uninformative and low confidence values, respectively; r2 (shades of black) is indicated in percentages within squares in the LD plots, with solid blocks without numbers indicating r2= 1. LD Blocks were created with the default algorithm in HaploView program (version 4.1).

Figure 3.

PTX3 linkage disequilibrium in African populations. The advent of large-scale genetic variant analysis allows for the fine scale calculation of r2 in multiple populations. This figure shows (A) in blue the r2 relationship between rs1840680 and adjacent variants within the genome-wide association study of tuberculosis within Gambians as part of the Wellcome Trust Case Control Consortium ( In addition, (B) in green, all variants found within the 1000 genomes project study of Yoruba trios from Nigeria and the r2 relationship with rs1840680, with the dotted vertical lines representing the core region of LD with variants of r2 > 0.80. Plot provided with the assistance of code from Paul de Bakker (

Figure 4.

PTX3 region LD structure of the Yoruba HapMap samples. Linkage disequilibrium structure for pairwise r2 between markers characterizing haplotype blocks in PTX3 region encompassed by the SNPs we genotyped (in the blue boxes). The figure is oriented 5′ to 3′, left to right. Strong LD is indicated by dark gray, whereas light gray and white indicate uninformative and low confidence values, respectively; r2 (shades of black) is indicated in percentages within squares in the LD plots, with solid blocks without numbers indicating r2= 1. Green and red boxes around markers indicate synonymous and missense mutations respectively. LD blocks were created with the default algorithm in HaploView program, version 4.1.

The interplay between innate immunity and fertility could result in PTX3 variants simultaneously playing a role in resistance against pathogens as well as in self/non self discrimination editing (Rovere et al., 2000; van Rossum et al., 2004). The PTX3 role in twinning could be indirect, via elimination of cellular debris from luteal cell apoptosis, consequently altering steroidogenesis and ovulation (Pate & Landis, 2001) or in abating inflammatory responses generated by dead and dying luteal cells and preserving ovarian tissues from damage. Activated innate immunity pathways modulate tissue wasting and preservation of integrity by sterile inflammation; in the ovary tissue damage and remodeling take place in a controlled fashion and innate immunity seems to play a key role in modulating the overall process (Spanel-Borowski, 2011). In this light, ovulation can be thought of as an inflammation-like process in which PTX3 (produced by cumulus oophorus cells and localized within the cumulus matrix) is a main player (Moalli et al., 2011), a concept supported by the finding that Ptx3−/− mice, generated by homologous recombination, are severely subfertile (Varani et al., 2002; Moalli et al., 2011). Thus, variation in the PTX3 gene may operate at the level of ovulation and fertilization to influence the risk of DZ twinning.

The totality of studies testing for association between PTX3 and infection, taken together, are consistent with the hypothesis that PTX3 haplotypes confer resistance to infections. In The Gambia, this protective effect could partly explain the unusual frequency of DZ twinning, via an indirect selection mechanism. That is, whatever the biological mechanism involved, in Gambians (and possibly West Africans at large) DZ twinning could simply be a by-product of gene-variants (in this case PTX3 alleles) selected primarily for protection from infectious diseases. Our interpretation that DZ twinning is a simple consequence of selection on another complex trait (“susceptibility to infection”), might, if confirmed, represent a working model for the relationships between immunity-related genes that can be under intense selective pressure in the human genome and multiple other seemingly unrelated phenotypes.

Our conclusion is based not only on our own findings as reported here, but also on observations previously published, associating PTX3 variants with protection from pulmonary TB in Guinea-Bissau (Olesen et al., 2007). That report and our current study taken together support the hypothesis that this gene and in particular specific haplotypes affect both traits, providing a more compelling explanation for both data sets.


We gratefully acknowledge the participation of the many Gambian families who made this study possible. We would like to thank Dr Luca Lavra at Ospedale San Pietro FBF (Rome, Italy) for his helpful assistance during editing. Grant support: IP was supported by NIH grant 2T32HL007751–16A2; the twinning study in The Gambia was supported by the Medical Research Council (UK) award G0000690 to GS.