Mammalian Genome

, Volume 18, Issue 5, pp 361–372

Quantitative trait loci for peripheral blood cell counts: a study in baboons

Authors

    • Department of GeneticsSouthwest Foundation for Biomedical Research
  • Michael C. Mahaney
    • Department of GeneticsSouthwest Foundation for Biomedical Research
    • Southwest National Primate Research CenterSouthwest Foundation for Biomedical Research
  • Laura A. Cox
    • Department of GeneticsSouthwest Foundation for Biomedical Research
    • Southwest National Primate Research CenterSouthwest Foundation for Biomedical Research
  • Jeffrey Rogers
    • Department of GeneticsSouthwest Foundation for Biomedical Research
    • Southwest National Primate Research CenterSouthwest Foundation for Biomedical Research
  • John L. VandeBerg
    • Department of GeneticsSouthwest Foundation for Biomedical Research
    • Southwest National Primate Research CenterSouthwest Foundation for Biomedical Research
  • Carlo Brugnara
    • Children’s Hospital BostonHarvard Medical School
  • Orah S. Platt
    • Children’s Hospital BostonHarvard Medical School
Article

DOI: 10.1007/s00335-007-9022-8

Cite this article as:
Bertin, A., Mahaney, M.C., Cox, L.A. et al. Mamm Genome (2007) 18: 361. doi:10.1007/s00335-007-9022-8

Abstract

Increasingly, baseline peripheral blood cell counts are implicated as risk factors for common complex diseases. While genetic influences on these hematologic parameters are firmly established, the genetic architecture of the blood counts is still poorly understood. In this article we used data from 582 healthy pedigreed baboons and variance components methods to localize quantitative trait loci (QTLs) influencing complete blood count variables. Besides performing genome-wide linkage scans for each trait individually, we conducted bivariate linkage analyses for all pairwise trait combinations to also identify pleiotropic QTLs influencing several blood counts. While significant and suggestive QTLs were localized throughout the genome (LOD range: 1.5–3.5), chromosomal regions associated with the expression of various hematologic parameters stand out. In particular, our results provide significant and consistent evidence for a QTL on the orthologous human chromosome 1p that is shared by several blood counts, mainly erythrocyte parameters. In addition, multiple suggestive evidence of linkage was detected on the orthologous human chromosomes 10 (near the q-terminus) and 19 (centromeric section). Future studies should help identify the genes responsible for these QTL and elucidate their role on baseline variation in hematologic indicators of health and disease.

Introduction

Complete blood count tests are performed routinely as part of a medical examination. They inform about current general health, revealing specific pathologies and blood genetic disorders. Baseline peripheral blood counts can also be used to identify those at greater risk for common complex diseases. Recent epidemiologic studies recognized white blood cell count (Hoffman et al. 2004; Madjid et al. 2004), platelet count (Gregg and Goldschmidt-Clermont 2003), mean platelet volume (Bath et al. 2004; Kilicli-Camur et al. 2005), red blood cell count (Puddu et al. 2002), mean corpuscular volume (Haltmayer et al. 2002; Mueller et al. 2001, 2002), hematocrit (Kobayashi et al. 2001), and hemoglobin (Kobayashi et al. 2001) as potential markers of future cardiovascular events. Such increasing evidence of associations between baseline blood characteristics and disease susceptibility should motivate studies designed to achieve a comprehensive understanding of the sources of hematologic variation among healthy individuals (Evans et al. 1999) to develop effective strategies for controlling and preventing diseases of public health interest such as cardiovascular diseases.

Epidemiologic and genetic studies indicate that peripheral blood phenotypes result from the interplay of environmental and genetic influences (Bain 1996; DelValle and Taniguchi 1995; Evans et al. 1999; Garner et al. 2000; KristalBoneh et al. 1997; Mahaney et al. 2005). The genetic bases of the interindividual variation in blood count parameters are particularly well supported, in both humans and animals, by the results of quantitative genetic analyses of these traits (DelValle and Taniguchi 1995; Evans et al. 1999; Garner et al. 2000; Mahaney et al. 2005). However, our comprehension of the genetic etiology underlying normal variation in hematologic variables is still limited due to the scarcity of detailed investigations. For instance, quantitative trait loci (QTLs) analyses of the complete blood count traits have been conducted only in the mouse (Peters et al. 2005, 2006), and studies in other species have focused on only one or a few hematologic traits (Evans et al. 2004; Lin et al. 2005; Wattrang et al. 2005). In addition, most of these analyses have not been designed to detect and localize pleiotropic QTLs that influence the phenotypic expression of multiple hematologic parameters (but see Lin et al. 2005), even though genetic correlations among blood count parameters are well documented (Lin et al. 2005; Mahaney et al. 2005). Linkage studies designed to locate genes with pleiotropic effects are thus likely to uncover QTLs associated with peripheral blood cell counts that have not been identified yet. So doing, they might bring new insights into the biological interrelations of these traits.

In this article we present a QTL mapping study of baseline blood traits in pedigreed baboons. Because of their close hematologic similarity with humans, nonhuman primates have become a model of choice for hematologic studies (Hampton and Matthews 1966; Hanson and Harker 1987; Herodin et al. 2005). In spite of that, genetics of peripheral blood counts in these mammals remain poorly understood. In this study our specific aims were (1) to fill this gap by providing the first QTL report of a whole-genome linkage screen for blood count parameters for a nonhuman primate and (2) to extend our knowledge of the baseline blood counts through extensive QTL mapping analyses. In these analyses we considered all measures of the complete blood count and we performed both independent and joint linkage mapping analyses of these traits to localize QTLs shared by several blood count traits and to address the possibility that they result from pleiotropic gene effects.

Materials and methods

Pedigreed baboons: breeding and sample characteristics

The pedigreed baboons used in this study belong to the breeding colony that is maintained by the Southwest National Primate Research Center located at the Southwest Foundation for Biomedical Research (San Antonio, TX). All animals from this colony are housed outside in social group cages. They are fed standard monkey chow diet (Purina Monkey Chow, Ralston Purina Co., St. Louis, MO, and Teklad Primate Diet, Teklad, Madison, WI.) to which they have ad libitum access. Animal care personnel and staff veterinarians assure daily maintenance and routine and emergency health care in accordance with the Guide for the Care and Use of Laboratory Animals (National Research Council 1996). All procedures related to the conduct of the present study were approved by the Institutional Animal Care and Use Committee in accordance with the established guidelines.

A sample of 410 females and 172 males consisting of olive baboons (Papio hamadryas anubis), yellow baboons (Papio hamadryas cynocephalus), and their hybrids was selected for this study. All chosen individuals were healthy and were at least 3.5 years old (i.e., equivalent to 12–13 years old in human) at the time of blood collection. These animals are distributed among 11 distinct pedigrees. They represent a large number of relative pairs consisting of 227 parent-offspring, 290 siblings, 23 grandparent-grandchild, 59 avuncular, 4422 half-siblings, 923 half-avuncular, 3 first cousins, 145 half-first cousins, 40 half-first cousins once removed, 1 half-sibling and first cousin, 40 half-siblings and half-first cousins, 6 half-siblings and half-avuncular, and 5 double half-avuncular.

Phenotyping methodology

Before blood collection sedative solutions were administered to the animals (i.e., intramuscular ketamine and intravenous xylazine hydrochloride, atropine, acepromazine, and ketamine) to ensure their relaxation. Blood samples were drawn from the femoral vein into EDTA tubes (ethylenediaminetetraacetic acid) and stored on ice. The complete blood count tests were executed within 4 h after bleeding using a Coulter JT (Beckman Coulter, Fullerton, CA). The primary measurements performed included white blood cell (WBC) count, red blood cell (RBC) count, hemoglobin (Hb), mean corpuscular volume (MCV), platelet count (Plt), and mean platelet volume (MPV). The following blood count measurements derived from the primary measurements presented above were also assayed: hematocrit (Hct, the product of RBC and MCV), mean corpuscular hemoglobin (MCH, the ratio of Hb to RBC), mean corpuscular hemoglobin concentration (MCHC, the ratio of Hb to Hct) and red cell distribution width (RDW, calculated as standard deviation of red cell volume × 100 ÷ MCV).

Baboon linkage map and genotyping

QTL scans for the hematologic traits were conducted using the sex-averaged linkage map of the baboon genome developed originally by Rogers et al. (2000) and mainly constructed from human microsatellite markers. The genome map we used included 284 markers spanning all 20 baboon autosomes with an average marker interval of 8.9 cM. This version of the genome map includes 22 intermaker gaps greater than 20 cM; the largest of them being approximately of 37 cM. Information about the most recent version of the linkage map and the number of informative meioses can be found in Cox et al. (2006). Marker properties are provided at http://www.snprc.org/baboon/genome/index.html. Complete details of the genotyping procedures can be found elsewhere (Cox et al. 2006; Rogers et al. 2000).

Statistical genetic analyses

For all genetic analyses we used variance decomposition methods using maximum likelihood techniques as implemented in SOLAR (Almasy and Blangero 1998). Variance decomposition approaches allow the simultaneous analysis of all familial information, thus providing more power per study subject than designs that focus on smaller familial units (Blangero et al. 2000; Blangero et al. 2001). Based on simulation studies, we estimated that the pedigree sample that we used confers 80% power to detect QTLs accounting for 21% of the phenotypic variance and genetic correlations equal to 0.45 at an alpha level of 5%.

The data of the present study were originally analyzed by Mahaney et al. (2005) who provide heritability estimates of the blood parameters and their pairwise phenotypic and genetic correlations (Note: the present sample excludes 18 of the 600 previously studied baboons who had not been genotyped at the time of these analyses). However, to ensure avoiding the detection of erroneous linkage due to the sensitivity of likelihood-based variance decomposition methods to trait departures from a normal distribution (Allison et al. 1999), we recalculated and report the heritability estimates and genetic correlations for the transformed traits (see below).

Heritability and pairwise genetic correlations

Decomposition of the phenotypic covariance between pairs of relatives was performed to estimate of the additive genetic ( \( \sigma ^{2}_{g} \)) effects. To improve estimation of these effects, we controlled for eventual confounding influences of environmental factors. For each hematologic parameter we evaluated the effects of a set of covariates (age, age2, sex, and the interaction terms between age * sex and age2 * sex) on their phenotypic variance. Significance of the covariate effects was estimated, independently for each covariate, through likelihood ratio (LR) tests (Edwards 1992) that compare the goodness of fit to the data of the polygenic model estimating the effects of all covariates to that of a restricted model where the mean effect of one covariate at a time is set to zero. The LR statistic, which is equivalent to twice the difference of the ln likelihoods of the two competing models, in this case follows a \( \chi ^{2}_{1} \) distribution (i.e., df is the number of constrained parameters in the restricted model) (Edwards 1992). Because our primary goal was not to identify covariates with significant effect on trait variance but to increase the detected additive genetic effects, we controlled for the influence of any covariate showing a borderline significance (p < 0.10). All subsequent analyses, including re-estimation of \( \sigma ^{2}_{g} \), were then performed using the residual trait values to which we applied an inverse Gaussian normalization (Allison et al. 1999). Heritability of the phenotypes (or residual heritability \( h^{2}_{r} \)) was calculated as the proportion of the phenotypic variance unexplained by the covariates that is attributable to the additive genetic effects.

To estimate the extent to which the phenotypic variance of the hematologic traits is influenced by shared additive genetic effects, we used multivariate analyses. Besides partitioning the phenotypic covariance of each trait, these analyses also involve decomposing the phenotypic correlation of the traits; this latter being partitioned into the genetic (ρg) and environmental (ρe) correlations (Mahaney et al. 1995). Using such bivariate decomposition, we calculated all the pairwise genetic correlations between the blood counts.

Significance of \( h^{2}_{r} \) and ρg was estimated separately through LR tests. These tests entail constraining the focal parameter to zero in the restricted model. The LR statistic allowing testing the significance of ρg follows a \( \chi ^{2}_{1} \) distribution. In contrast, the LR statistic used for assessing the significance of \( h^{2}_{r} \) (which involves setting \( \sigma ^{2}_{g} \) to 0 in the constrained model) is distributed asymptotically as a 0.5:0.5 mixture of \( \chi ^{2}_{1} \) distribution and a point mass at 0. Such distribution is actually expected when one parameter is held constant in the restricted model and constrained at a value lying at the boundary of the parameter space (Self and Liang 1987). Besides these tests, we also evaluated the possibility of complete pleiotropy (i.e., 100% of the genetic variance in each trait is due to shared genetic effects). We compared the performance of the bivariate polygenic models estimating ρg with that of their nested model, where ρg was constrained either to 1 or −1 depending on the sign of the estimated ρg. The LR statistic of this test is also distributed asymptotically as a 0.5:0.5 mixture of \( \chi ^{2}_{1} \) with a point mass at zero.

Significance levels were adjusted for multiple comparisons. However, because Bonferroni correction is too conservative with nonindependent tests, the adjustment we applied considered the global genetic relatedness of the traits which was measured as the average proportion of the genetic variance shared by the traits. Significance levels were then obtained by dividing the desired alpha level by an effective number of tests, obtained as
$$ {1 + (N - 1)(1 - \overline{{\rho ^{2}_{g} }} )} $$
where N is the number of tests performed and \( \overline{{\rho ^{2}_{g} }} \) is the mean value of the squared pairwise genetic correlations.

Linkage analyses

Univariate linkage

QTL mapping was performed using multipoint variance component linkage analyses. In the univariate linkage model, the phenotypic covariance between pairs of relatives is further decomposed to partition the additive genetic effects that are due to a QTL linked to a specific chromosomal location (\( \sigma ^{2}_{q} \)) and those that are caused by other QTLs, unlinked to this genetic position (\( \sigma ^{2}_{{gr}} \)) (Amos 1994). The specific heritability of a QTL (\( h^{2}_{q} \)) can then be easily derived as the proportion of the phenotypic variance that is attributable to its additive genetic effects.

Estimating \( \sigma ^{2}_{q} \) involves modeling the phenotypic covariance as a function of the expected proportion of alleles that are, between relative pairs, identitical-by-descent (IBD) at the chromosomal location of interest (Almasy and Blangero 1998). Marker locus-specific IBD probabilities were assessed by means of pairwise likelihood-based procedures (Curtis and Sham 1994). We then calculated IBD probabilities at 1 cM intervals using the Markov chain Monte Carlo (MCMC) approach implemented in the computer package Loki (Heath 1997; Heath et al. 1997). This MCMC method yields multipoint IBD estimates that more closely approximate those obtained by exact methods that cannot be applied conveniently to large, extended pedigrees. Using the multipoint IBD matrices, we performed for each hematologic trait a LOD score evaluation at 1 cM intervals along each chromosome.

Tests for significant linkage were performed at each linkage point through LR tests; those involved constraining \( \sigma ^{2}_{q} \) to zero in the restricted model. This LR statistic is then distributed as a 0.5:0.5 mixture of \( \chi ^{2}_{1} \) with a point mass at zero. A LOD score, equivalent to that classically reported in linkage analysis, was obtained by dividing the LR estimate by 2 ln 10 (Ott 1988). We considered LOD = 2.7 as the threshold for genome-wide significant evidence of linkage at α = 0.05. This threshold was estimated using a modification of the method of Feingold et al. (1993), incorporating the finite marker density of the baboon linkage map and the mean recombination rate among the baboons as measure of pedigree complexity. Similarly, we estimated that a LOD = 1.5 was likely to occur, by chance, one time for each full-genome scan. As proposed by Lander and Kryglyak (1995), nonsignificant LOD scores reaching this limit were regarded as suggestive evidence of linkage, even though such linkage signals will often be false positives (Lander and Kruglyak 1995).

A sequential procedure was adopted to document the effects of multiple QTL. When a genome screen provided suggestive or significant evidence for genetic linkage, the QTL with the highest LOD score was retained in the model and an additional full-genome scan, conditional on this QTL, was performed to investigate other genetic linkage (Almasy and Blangero 1998).

Multivariate linkage analyses

To identify chromosomal regions influencing simultaneously the phenotypic expression of two blood count parameters, we conducted bivariate linkage analyses for all the pairwise trait combinations. This approach involves the estimation of the additive genetic variance due to the QTL (\( \sigma ^{2}_{q} \)) for both traits and two correlations between the two traits: one due to the shared additive genetic effects of the QTL (ρq) and the other due to the shared additive effects of genes other than the QTL (ρgr).

Tests for significant linkage were performed by comparing the performance of the estimated bivariate models with that of their equivalent nested models, where \( \sigma ^{2}_{q} \) was constrained to zero for both traits. The bivariate LOD scores we report were adjusted based on the conversion with 0.5\( \chi ^{2}_{1} \), 0.25\( \chi ^{2}_{2} \), and 0.25 point mass at zero to be equivalent to the univariate ones (Amos et al. 2001). We also used a sequential procedure when conducting our bivariate analyses.

Shared QTL effects identified through bivariate linkage analyses can reveal either coincident linkage of genes affecting the phenotypic expression of the traits or pleiotropic effects of the same genes or group of genes on the trait variances (Almasy et al. 1997). In an attempt to distinguish between these two alternative hypotheses, we report the tests of coincident linkage (ρq is set to zero in the constrained model) and that of complete pleiotropy (ρq is set to 1 or −1 in the constrained model) (Almasy et al. 1997).

Because of the large number of linkage analyses performed, we evaluated the number of expected false positives. Because our phenotypes and analyses were not independent, we first calculated the number of effective independent analyses performed using the regression-based method advocated by Camp and Farnham (2001). Overall, 76 nonindependent genome scans were performed (14 univariate genome scans and 62 bivariate scans), which were estimated to correspond to 14 independent linkage analyses. Consequently, a LOD score of 2.7 was expected less than one time by chance alone (i.e., 0.7 time). Our analyses revealed nine significant linkage signals (Tables 4 and 5), a finding that clearly surpasses the expectations under the null hypothesis of no linkage. In the same way we can estimate that 14 LOD scores of 1.5 were expected just by chance (Lander and Kruglyak 1995) and we detected 19 suggestive signals (Tables 4 and 5). This indicates that some suggestive signals reported in this study should correspond to true linkages even though, as expected, most of them likely will not (Lander and Kruglyak 1995).

Results

Phenotypic variation, heritability, and pairwise genetic correlations of the blood counts

The descriptive statistics calculated to quantify the distributions of the complete blood counts are presented in Table 1. In spite of a slight difference in age distributions between males and females in our baboon sample (Table 1), the general patterns of phenotypic variation of the hematologic traits are fairly similar for both sexes. Overall, the blood traits exhibit low levels of phenotypic variation, the average coefficient of variation (CV) being 9.85% for males and 10.98% for females. However, the hematologic traits obviously differ in their degrees of variation. For instance, amplitude of variation of MCHC is very low as demonstrated by the narrow phenotypic range of this trait and CVs that do not exceed 2.35%. In contrast, other hematologic parameters such as WBC and Plt are much more variable, exhibiting CVs that are up to 17 times higher than observed for MCHC.
Table 1

Descriptive characteristics of the phenotypic dispersion for age and the hematologic traits per sex

Variable

Males (n = 172)

Females (n = 410)

Range

Mean ± SD

CV (%)

Range

Mean ± SD

CV (%)

Age (yr)

7.46–22.85

12.26 ± 3.77

 

3.87–30.00

15.39 ± 5.10

 

WBC (× 103/μl)

3.40–28.20

10.55 ± 3.47

32.90

4.90–24.60

11.80 ± 4.54

38.47

RBC (× 106/ml)

2.94–5.86

4.93 ± 0.38

7.63

4.00–6.06

4.78 ± 0.40

8.38

Hb (g/dl)

8.20–15.40

12.92 ± 1.00

7.75

10.15–15.60

12.42 ± 1.08

8.71

Hct (%)

26.20–45.40

39.15 ± 2.83

7.22

31.85–48.10

37.51 ± 2.95

7.85

MCV (fl)

69.10–87.10

79.52 ± 2.58

3.25

70.30–87.90

78.44 ± 2.84

3.62

MCH (pg)

22.40–30.40

26.23 ± 1.05

4.02

22.20–29.10

25.98 ± 1.20

4.63

MCHC (g/dl)

30.50–36.00

32.98 ± 0.63

1.91

31.50–34.50

33.08 ± 0.77

2.34

RDW (%)

11.10–17.40

12.74 ± 0.98

7.67

10.80–17.00

13.00 ± 0.99

7.65

Plt (× 103/ml)

69.00–741.00

271.90 ± 65.15

23.96

72.00–663.00

320.53 ± 86.48

26.98

MPV (fl)

6.70–12.30

8.40 ± 0.92

11.00

6.65–11.10

8.98 ± 1.00

11.14

WBC = white blood cell count; RBC = red blood cell count; Hb = hemoglobin; Hct = hematocrit; MCV = mean corpuscular volume; MCH = mean corpuscular hemoglobin; MCHC = mean corpuscular hemoglobin concentration; RDW = red cell distribution width; Plt = platelet count; MPV = mean platelet volume

Whatever the amplitude of their phenotypic variation, variance decomposition of the blood counts demonstrates that part of their expression is under genetic control. Indeed, after accounting for the influences of environmental covariates, we detected significant additive genetic effects for all the hematologic parameters (Table 2). However, the proportion of the phenotypic variance explained by these effects is also largely variable as indicated by the broad range of values observed for the \( h^{2}_{r} \) estimates which range from 0.29 for MCHC to 0.65 for RBC (Table 2).
Table 2

Quantitative genetic analyses of hematologic variables in pedigreed baboons: Proportions of the phenotypic variance due to the effects of significant covariates and the additive effects of genes

Trait

Covariates included in the final model

C2

Residual heritability (h2r)

Estimate

SD

WBC

Age, Age * Sex

0.03

0.39***

0.11

RBC

Age, Sex, Age * Sex

0.03

0.65***

0.09

Hb

Age, Sex, Age * Sex

0.06

0.61***

0.09

Hct

Age, Sex, Age * Sex

0.07

0.55***

0.10

MCV

Age

0.07

0.54***

0.10

MCH

Age

0.04

0.47***

0.10

MCHC

 

0.00

0.29*

0.09

RDW

 

0.00

0.41***

0.10

Plt

Sex

0.07

0.33***

0.09

MPV

Sex

0.06

0.59***

0.09

WBC = white blood cell count; RBC = red blood cell count; Hb = hemoglobin; Hct = hematocrit; MCV = mean corpuscular volume; MCH = mean corpuscular hemoglobin; MCHC = mean corpuscular hemoglobin concentration; RDW = red cell distribution width; Plt = platelet count; MPV = mean platelet volume

C2 represents the proportion of the total phenotypic covariance explained by the covariates and \( h^{2}_{r} \), the residual heritability, is the proportion of the residual phenotypic variance due to the additive effects of genes (calculated after applying an inverse Gaussian transformation to the residuals, see text). Significance of covariates and heritability was assessed by likelihood ratio tests (see Materials and methods)

*p < 0.05, ***p < 0.001 after adjusted for multiple comparisons (see Materials and methods)

The genetic correlations confirm previous findings (Mahaney et al. 2005) that the additive genetic variance of some hematologic traits at least partly results from the action of shared genes. Nine of the pairwise genetic correlations differ significantly from zero (Table 3). Of particular interest are the pairwise genetic correlations of RBC count, Hb, Hct, and MCHC, which all exceed 0.70 (Table 3). Tests of complete pleiotropy further support the close genetic relatedness of the triplet Hb, Hct, and MCHC because none of the pairwise genetic correlations among these traits was found to differ from 1 at an adjusted alpha level of 5% (Table 3). Similarly, the hypothesis of complete pleiotropy could not be rejected for WBC and MPV, albeit the assessed genetic correlation between these two traits was only 0.63 and was not found to differ significantly from zero after correction for multiple comparisons (Table 3). In all other cases, the genetic correlations differ significantly from 1 or −1, even for pairs of traits such as MCV/MCH that were found to be highly genetically correlated (Table 3).
Table 3

Pairwise additive genetic correlations between the hematologic variables

 

WBC

RBC

Hb

Hct

MCV

MCH

MCHC

RDW

Plt

MPV

WBC

 

0.18

0.09

0.18

−0.20

−0.23

−0.06

0.35

−0.19

0.63

RBC

  

0.88***

0.91***

−0.56**

−0.21

0.70**

0.18

0.11

0.20

Hb

   

0.99***

−0.08

0.28

0.91***

−0.06

0.15

0.16

Hct

    

−0.15

0.21

0.88***

0.05

0.18

0.24

MCV

     

0.93***

0.16

−0.46

0.2

−0.12

MCH

      

0.52

−0.59*

0.18

−0.18

MCHC

       

−0.53

0.05

−0.09

RDW

        

0.23

−0.09

Plt

         

−0.37

MPV

          

WBC = white blood cell count; RBC = red blood cell count; Hb = hemoglobin; Hct = hematocrit; MCV = mean corpuscular volume; MCH = mean corpuscular hemoglobin; MCHC = mean corpuscular hemoglobin concentration; RDW = red cell distribution width; Plt = platelet count; MPV = mean platelet volume

Bold values indicate genetic correlations that did not differ from 1 or −1 at an alpha level of 5% adjusted for multiple comparisons

*p < 0.05, **p < 0.01, ***p < 0.001 after correction for multiple comparisons (see Materials and methods)

QTL mapping

Results of the univariate genome-wide linkage scans are summarized in Table 4. These analyses provide significant evidence of linkage for only one hematologic trait, namely, MCV, which had the highest LOD score of the genome scan located on chromosome 1 at 9 cM. Nevertheless, univariate maximum multipoint LOD scores reached the threshold of suggestive evidence for linkage in four other genome-wide scans: one for RDW on chromosome 9 at 127–132 cM, two for Hb on chromosome 19 at 21–32 cM and chromosome 20 at 101–103 cM, and one for Hct in the same region of chromosome 19.
Table 4

Univariate linkage analyses of hematologic variables in pedigreed baboons: Characteristics of QTL with suggestive or better evidence (LOD > 1.5)

Trait

LOD

PHA chromosome (location in cM from the p-ter marker)a

Orthologous human physical locationb

\( h^{2}_{q} \)

% (\( h^{2}_{r} \))

MCV

3.5

1 (0–44)

1p34.2–p36.23

0.37

61

RDW

1.9

9 (127–132)

10q26.3

0.23

56

Hb

1.9

19 (21–32)

19p13.12–q13.11

0.24

40

Hb

1.6

20 (101–103)

16q23.2–q24.3

0.35

58

Hct

1.6

19 (20–53)

19p13.12–q13.41

0.26

47

MCV = mean corpuscular volume; RDW = red cell distribution width; Hb = hemoglobin; Hct = hematocrit

\( h^{2}_{q} \) is the QTL-specific heritability and % (\( h^{2}_{r} \)) is the proportion of the residual heritability explained by the QTL

aThe chromosomal regions reported correspond to the interval of suggestive LOD scores

bHuman orthologous locations of the flanking markers were determined with the UCSC genome browser (http://www.genome.ucsc.edu) using March 2006 assembly of the human genome

Twenty-three bivariate scans yielded maximum multipoint LOD scores that exceed the threshold of suggestive evidence of linkage (Table 5). Evidence for QTL affecting the phenotypic expression of two traits is particularly well supported when power to detect the linkage is increased by the joint analysis of the two traits, which we considered to be the case when the bivariate LOD score was higher than the corresponding univariate ones at the same location. This situation occurred for 13 of the bivariate genome scans (see LOD scores marked with a superscript c). In most of these cases, ρq and ρgr were of opposite sign (i.e., in 11 of the 13 cases) and the overall genetic correlations between the traits were low to moderate: i.e., |ρg| ranged from 0.06 to 0.46 for 12 of these pairs (Tables 3 and 5). With the exception of RBC/Hct (ρg = 0.91), our bivariate linkage analyses did not provide increased linkage evidence for trait pairs showing the highest additive genetic correlations (Tables 3 and 5).
Table 5

Bivariate linkage analyses of hematologic variables in pedigreed baboons: Characteristics of QTL with suggestive or better evidence (LOD > 1.5)

Trait 1

Trait 2

Maximum LOD

Baboon chromosome (location in cM from the p-ter marker)a

Orthologous human locationb

ρgr

\( h^{2}_{q} \) trait 1

\( h^{2}_{q} \) trait 2

%(\( h^{2}_{r} \)) trait 1

%(\( h^{2}_{r} \)) trait 2

ρq

RBC

Hct

3.5c

1 (0–34)

1p34.2–p36.23

1.00

0.09

0.17

14

31

0.76

MCV

MPV

3.4

1 (0–30)

1p34.2–p36.23

−0.24

0.26

0.13

49

22

0.13

MCV

Hb

3.2

1 (0–41)

1p34.2–p36.23

−0.69

0.36

0.20

67

32

0.47

MCV

RBC

3.2

1 (0–41)

1p34.2–p36.23

−1.00

0.36

0.10

67

16

−0.08

MCV

Hct

3.2

1 (0–41)

1p34.2–p36.23

−0.96

0.37

0.18

69

32

0.54

MCV

Plt

3.1

1 (0–30)

1p34.2–p36.23

0.08

0.33

0.03

61

08

0.81

MCV

WBC

3.0

1 (0–32)

1p34.2–p36.23

0.07

0.29

0.04

54

10

−1.00

MCV

RDW

2.8

1 (0–33)

1p34.2–p36.23

−0.35

0.34

0.05

64

12

−1.00

MCV

MCHC

2.6

1 (0–32)

1p34.2–p36.23

0.14

0.32

<0.01

60

<1

1.00

MCV

MCH

2.5

1 (0–32)

1p34.2–p36.23

0.87

0.31

0.22

58

45

1.00***

MCH

MCHC

2.5c

1 (0–29)

1p34.2–p36.23

0.63

0.21

<0.01

45

<1

1.00

MCH

MPV

1.5c

1 (0–5)

1p34.2–p36.23

−0.41

0.18

0.14

39

23

0.34

MCV

RDW

1.5c

13 (93–97)

2q11.1–q12.1

1.00

0.16

0.14

30

35

−1.00**

RDW

Plt

1.6c

2 (121–128)

3p23–p22.3

−0.36

0.10

0.19

25

57

1.00

WBC

MCHC

2.0c

5 (122–129)

4q34.3–q35.1

0.40

0.15

0.06

37

25

−1.00

RBC

MPV

1.6c

8 (1–13)

8p21.3–p22

0.44

0.06

0.15

8

24

−1.00

MCV

MPV

1.7c

15 (113–119)

9q34.2–q34.3

1.00

0.20

0.10

37

17

−1.00**

Plt

MPV

2.0c

9 (117–132)

10q26.3

−0.74

0.08

0.15

23

25

1.00

RDW

Plt

1.5

9 (130–132)

10q26.3

−1.00

0.23

0.02

56

7

1.00

RDW

MPV

2.0c

9 (127–132)

10q26.3

−1.00

0.19

0.09

47

16

0.04

Hb

RDW

2.0c

20 (98–103)

16q23.2–q24.3

0.23

0.39

0.25

65

62

−0.19

Hb

Plt

2.2c

19 (20–55)

19p13.12–q13.41

−0.25

0.24

0.08

40

25

1.00

RDW

MPV

2.4c

10 (7–33)

20p12.1–q11.22

−0.84

0.25

0.16

62

28

0.80

WBC = white blood cell count; RBC = red blood cell count; Hb = hemoglobin; Hct = hematocrit; MCV = mean corpuscular volume; MCH = mean corpuscular hemoglobin; MCHC = mean corpuscular hemoglobin concentration; RDW = red cell distribution width; Plt = platelet count; MPV = mean platelet volume

\( h^{2}_{q} \) is the QTL-specific heritability; % (\( h^{2}_{r} \)) is the proportion of the residual heritability explained by the QTL; ρq is the genetic correlation between the pair of traits at the chromosomal location; and ρgr is the residual additive genetic correlation

*p < 0.05, **p < 0.01, ***p < 0.001. Bold values indicate QTL correlations not different from 1 or −1 at α = 5%

aThe chromosomal regions reported correspond to the interval of suggestive LOD scores

bHuman orthologous locations of the flanking markers were determined with the UCSC genome browser (http://www.genome.ucsc.edu) using March 2006 assembly of the human genome

cBivariate LOD score greater than the corresponding univariate ones at the same location

The bivariate analyses confirm and improve upon many of the linkage results obtained by the single-trait mappings. All trait combinations involving the erythrocyte parameter MCV revealed a QTL located in the telomeric region of chromosome 1p that accounts for a large proportion of MCV’s heritability (estimated between 49% and 69%, Table 5). The bivariate analyses also provide further support for the linkage for RDW and Hb, reported above. Suggestive evidence of linkage for RDW on chromosome 9 at 127–132 cM is corroborated by the results of two bivariate analyses (RDW/MPV and RDW/PLT, Table 5), and linkage signals suggested by the joint analyses of Hb/RDW and Hb/Plt correspond to those identified by the univariate variance decomposition of Hb (Tables 4 and 5).

Genes contributing to variation in the blood counts are distributed throughout the genome. Linkage signals were detected on ten different chromosomes (Tables 4 and 5). However, chromosomal regions related to the expression of multiple hematologic parameters were also identified. This is the case of the telomeric region of chromosome 1p. More than half of the bivariate genome-wide scans demonstrating or suggesting linkage point to that location (Table 5). While some of these results might be primarily driven by the strong effect of a gene or genes in this chromosomal region on MCV, we found support for linkage from genome scans involving five other traits. The pair RBC/Hct exhibits the strongest evidence of linkage in this region (LOD = 3.5, Table 5). The importance of this region on blood count traits is further sustained by the genome-wide scans of MCH/MCHC and MCH/MPV (Table 5). In most cases our analyses were not powerful enough to distinguish between the complete pleiotropy and coincident linkage hypotheses for the QTL found in the telomeric region of chromosome 1p (Table 5). Only the QTL correlation between MCV and MCH differed significantly from zero but not from one (Table 5), suggesting complete pleiotropy for these two traits. In contrast, the hypothesis of complete pleiotropy was rejected in two cases, both involving MPV (Table 5).

Chromosome 9 at 117–132 cM is another region likely to be involved in the control of more than two traits as indicated by the pairwise genome-wide scans of Plt, MPV, and RDW, which were all found to be linked to that region at a suggestive level of significance (Table 5).

Finally, our observation of increased evidence for a Hb-related QTL on chromosome 19 at 20–55 cM in the joint analysis of this trait with Plt (Tables 4 and 5) suggests that in addition to influencing Hct and Hb (Table 4), a gene (or genes) in this chromosomal region also affects variation in Plt. However, the ρq between Hb and Plt differs significantly from one but not from zero (Table 5), suggesting coincident linkage of the genes influencing these two traits.

For each hematologic trait, one to five linkage signals were identified. Because most of the suggestive signals are likely to be false positives, the estimated portion of the overall additive genetic variance restored by these linkage signals is expected to be overestimated in most cases. However, it is interesting to note that this portion is variable among the traits (Tables 4 and 5 or see Table 6 for a summary of all the detected linkage signals per trait). For instance, the detected linkage signals for MCHC can explain only a small fraction of the genetic variance of this trait (Table 5). However, for most of the blood count traits, a substantial fraction of the additive genetic variance might be explained as for Hb, MCV, RDW, Plt, and MPV. Regardless of their number (i.e., two to five linkage signals per trait, Tables 4 and 5), the putative QTLs virtually restore the entire residual heritability estimated for these hematologic parameters (Table 5).
Table 6

Summary of the identified QTLs

Trait

Type of linkage analyses

Significance level

Baboon chromosome (location in cM from the p-ter marker)a

Orthologous human locationb

%(\( h^{2}_{r} \))

WBC

Bivariate

Suggestive

5 (122–129)

4q34.3–q35.1

37

RBC

Bivariate

Significant

1 (0–41)

1p34.2–p36.23

14–16

 

Bivariate

Suggestive

8 (1–13)

8p21.3–p22

8

Hb

Univariate/Bivariate

Suggestive

19 (20–55)

19p13.12–q13.11

40

 

Univariate/Bivariate

Suggestive

20 (98–103)

16q23.2–q24.3

58–65

Hct

Univariate

Suggestive

19 (20–53)

19p13.12–q13.41

40–47

 

Bivariate

Significant

1 (0–41)

1p34.2–p36.23

31

MCV

Univariate/Bivariate

Significant

1 (0–41)

1p34.2–p36.23

49–69

 

Bivariate

Suggestive

13 (93–97)

2q11.1–q12.1

30

 

Bivariate

Suggestive

15 (113–119)

9q34.2–q34.3

37

MCH

Bivariate

Suggestive

1 (0–32)

1p34.2–p36.23

39–45

MCHC

Bivariate

Suggestive

1 (0–29)

1p34.2–p36.23

<1

 

Bivariate

Suggestive

5 (122–129)

4q34.3–q35.1

25

RDW

Bivariate

Suggestive

13 (93–97)

2q11.1–q12.1

35

 

Bivariate

Suggestive

2 (121–128)

3p23–p22.3

25

 

Univariate/Bivariate

Suggestive

9 (127–132)

10q26.3

47–56

 

Bivariate

Suggestive

10 (7–33)

20p12.1–q11.22

28

 

Bivariate

Suggestive

20 (98–103)

16q23.2–q24.3

62

Plt

Bivariate

Suggestive

2 (121–128)

3p23–p22.3

57

 

Bivariate

Suggestive

9 (117–132)

10q26.3

23

 

Bivariate

Suggestive

19 (20–55)

19p13.12–q13.41

25

MPV

Bivariate

Suggestive

1 (0–5)

1p36.23–p34.2

23

 

Bivariate

Suggestive

8 (1–13)

8p22–p21.3

24

 

Bivariate

Suggestive

9 (117–132)

10q26.3

16–25

 

Bivariate

Suggestive

10 (7–33)

20p12.1–q11.22

28

 

Bivariate

Suggestive

15 (113–119)

9q34.2–q34.3

17

WBC = white blood cell count; RBC = red blood cell count; Hb = hemoglobin; Hct = hematocrit; MCV = mean corpuscular volume; MCH = mean corpuscular hemoglobin; MCHC = mean corpuscular hemoglobin concentration; RDW = red cell distribution width; Plt = platelet count; MPV = mean platelet volume

%(\( h^{2}_{r} \)) is the estimated proportion of the residual heritability explained by the QTLs

aThe chromosomal regions reported correspond to the interval of suggestive LOD scores

bHuman orthologous locations of the flanking markers were determined with the UCSC genome browser (http://www.genome.ucsc.edu) using March 2006 assembly of the human genome

Discussion

This study largely confirms and amplifies previous findings concerning the organization of the genetic network of peripheral blood cell counts. Specifically, the QTL mapping analyses corroborate the oligogenic and polygenic inheritance of the baseline blood parameters. None of the QTLs explains completely the additive genetic effects we estimated. In fact, we obtained up to five linkage signals per trait, a finding that highlights the complexity of the genetic architecture of hematologic phenotypes. Such cumulative action of many different genes, each with a small effect, is also supported by the lack of significant linkage evidence for blood counts such as RBC, Hb, and Hct that were found to be highly heritable.

Our results also support pleiotropic genetic effects on blood count traits. Overall, the patterns of high genetic correlation among erythroid parameters are very similar to those previously reported for the same pedigreed baboons (Mahaney et al. 2005). The close genetic interrelationship observed between many of these traits is consistent with complete pleiotropy. In particular, none of the pairwise genetic correlations between Hb, Hct, and MCHC differed significantly from one. These three traits are to a large extent surrogate measures of one another, e.g., a high Hct and/or high MCHC necessarily is highly correlated with high blood Hb. Therefore, the same genes are likely to contribute to the variation of these phenotypes. Our results clearly support this by demonstrating that most, if not all, of the additive genetic variance in each of these traits is due to genetic effects shared with the other traits.

Because few studies have examined the genetic interrelatedness of the blood counts, it is not documented to what extent such covariation patterns exist for other species, although it would be surprising if they were drastically different. For human, a similarly high genetic correlation between Hct and Hb has been reported recently (i.e., ρg = 0.85) but was found to differ from one (Lin et al. 2005). This suggests that, as expected, a rather large proportion of the phenotypic variation of Hb and Hct is determined by the same genes in both human and nonhuman primates.

As expected, the mapping strategy undertaken here substantially improves our understanding of the genetic architecture of peripheral blood counts. For instance, we report the first QTL evidence for MCV at a location orthologous to human location 1p34.2–p36.23 in humans. In the same way, we identified four putative QTLs for RDW, a trait that so far has been ignored in genetic analyses of hematologic variation.

More important than the mapping of novel QTLs is our identification of chromosomal regions that likely harbor genes that influence a number of blood count traits. Clearly, some of the detected linkage signals reveal a single underlying QTL that affects several traits. For instance, significant linkage signals for three traits (MCV, RBC count, Hct) and suggestive linkage signals for three others (MCH, MCHC, MPV) were revealed in the telomeric section of chromosome 1p, orthologous to human 1p34.2–p36.23. The tests of significance of the QTL correlation between MCV and MCH suggest that a gene or genes with pleiotropic effects on these two closely related erythrocyte parameters lie in that region. On the other hand, we found coincident linkage for all trait pairs involving MPV. This result indicates that the platelet and erythrocyte linkages to chromosome 1p may result from the actions of different genes.

Whatever the genetic phenomena responsible for the detection of the multiple bivariate linkages at the telomeric region of chromosome 1p, our results reveal a major role for this region in the baseline variation of erythroid parameters and this seems to hold true in other mammal species. Indeed, major loci for RBC count, Hb, and Hct were identified in mice at the orthologous human location 1p34.2–p36.23 (Peters et al. 2006; Smith et al. 2006). This telomeric region of chromosome 1p includes a very large number of genes (using the March 2006 assembly of the human genome of the UCSC genome browser, we listed 952 different known genes on chromosome 1p34.2–p36.23). Peters et al. (2006) recognized the gene encoding the erythroid membrane protein band 4.1 as a viable candidate gene to explain the QTL they report for RBC on the orthologous human location 1p34–p36. This gene has been associated with erythrocyte shape-related disorders such as elliptocytosis, microspherocytosis, and poikilocytosis (Delaunay 2002) and might participate in linkages on chromosome 1p of erythrocyte parameters, especially of erythrocyte morphologic characteristics such as MCV.

The telomeric part of chromosome 9q (HSA location: 10q26.3-qter) corresponds to another region apparently associated with the baseline variation of several blood counts. While such QTLs need to be carefully considered because we found only suggestive evidence of linkages, the results of all the bivariate analyses of MPV, Plt, and RDW are consistent and yield maximum LOD scores at the exact same location. Such QTLs are also supported by the increased linkage evidence provided by the bivariate analyses of Plt/MPV and RDW/MPV compared with the univariate analyses of the traits. Suggestive evidence of linkage on chromosome 10q for nonpathologic variation in Plt count has also been reported previously in humans (Evans et al. 2004). However, the chromosomal region involved (Evans et al. 2004) does not overlap with the QTL we detected here.

Finally, convergent lines of evidence indicate that chromosome 19 may also harbor genes that contribute to the phenotypic variation of several blood counts. The univariate analyses of Hb and Hct both produce suggestive evidence for linkage at the same location on chromosome 19, a result consistent with our finding of complete pleiotropy between these two traits. The joint analysis of Hb and Plt also improves the linkage evidence for Hb, which suggests that the telomeric end of chromosome 19 harbors genes that are related to variation in Plt as well. Likewise, a Plt QTL on HSA chromosome 19q13.12-q13.41 has been identified in humans (Evans et al. 2004) and suggestive evidence for such a QTL has been found in mice (Cheung et al. 2004). Altogether, these results support the existence of genes on chromosome 19 that influence baseline Plt. Evans et al. (2004) considered GP6, the gene encoding the glycoprotein VI, a membrane protein involved in collagen-induced platelet activation (Kato et al. 2003), the best candidate gene lying in the chromosomal region (Evans et al. 2004).

In summary, we have localized several loci of importance for the variation of blood count parameters. The bivariate approach proved particularly advantageous in this process. Twenty-three bivariate scans reached the level of suggestive evidence of linkage in comparison with four in the univariate cases. Interestingly, bivariate linkages were preferentially demonstrated for trait pairs characterized by moderate levels of global genetic relatedness. This is because in most cases the QTLs and the residual genetic correlations occurred in opposite directions. This observation agrees with theoretical studies that predict that multivariate analyses will perform well in such circumstances (Amos et al. 2001; Evans 2002). In contrast, but also in conformity with simulation studies (Amos et al. 2001), the bivariate analyses did not outperform the univariate approach for strongly genetically correlated traits. Altogether, our results highlight the difficulty of predicting in advance the success of multivariate analyses and suggest that the systematic application of such methods to biologically related phenotypes such as the blood count traits might be very helpful for uncovering their genetic etiology.

Acknowledgments

The authors thank D. Winnier and two anonymous reviewers for constructive comments on an earlier version of the manuscript and D. Newman for outstanding technical assistance. This work was supported by grants from the National Institutes of Health (NIH): R01 HL068922 (OSP), R01 HL054141 (MCM), P01 HL028972 (JLV, MCM, JR), P51 RR013086 (JLV, MCM, JR), and R01 RR08781 (JR). This investigation was conducted in facilities constructed with support from Research Facilities Improvement Program Grant Number C06 RR13556 from the National Center for Research Resources, National Institutes of Health.

Copyright information

© Springer Science+Business Media, LLC 2007