Iron homeostasis is one of the most critical functions in living systems. Too little iron can lead to anemia and tissue-specific disorders, such as splenomegaly. Excessive systemic iron is characteristic of hemochromatosis and is implicated in the brain in Parkinson's disease. With the exception of some single gene diseases like hemochromatosis, we know little about genetic-based, individual differences in iron-related parameters and their impact on biology. To model genetic control of iron homeostasis, we measured liver, spleen, and plasma iron concentrations, hematocrit and hemoglobin, transferrin saturation, and total iron-binding capacity in several BXD/Ty recombinant inbred mouse strains derived from C57BL/6 and DBA/2 progenitors. At 120 days of age, the animals were killed for iron analysis. All measures showed genetic-based variability consistent with polygenic influence. Analysis of principal components of the seven measures revealed three factors that we named availability, transport, and storage. Quantitative trait loci (QTL) analysis revealed one suggestive QTL on chromosome 5 for availability, two suggestive QTL (one on chromosome 1 and the other on chromosome 7) for transport, and one weak QTL on chromosome 2 for storage. The results show that iron homeostasis is a complex trait and is influenced by multiple genes.
- quantitative trait loci
- iron binding capacity
biologically, iron is essential for a number of cellular, molecular, and physiological activities and acts in oxygen transport and as cofactor in metabolic pathways. Reservoirs of iron include both essential iron (hemoglobin, myoglobin, iron-sulfur enzymes, cytochromes, etc.), as well as nonessential pools (ferritin and hemosiderin) (2, 3). Nearly two-thirds of the iron in the body is found in the red cell mass. The other portion of the essential iron pool comprises 5–10% of body iron. The storage pool of iron in ferritin and hemosiderin is available to provide iron to the plasma iron transport system when iron absorption fails to meet the needs of daily iron requirements. The consequences of iron deficiency over a period of time results in tissue iron deficiency despite the action of homeostatic mechanisms in gastrointestinal cells to increase iron absorption (2). Iron overload can also be problematic and is linked to diseases, such as hemochromatosis and Parkinson's disease. Thus it is evident that iron requires tight regulation. Accordingly, because there are many proteins involved in iron absorption, elimination, and change in oxidative states, iron homeostasis is quite complex and influenced by multiple genes.
The role of genetic-based mechanisms in iron homeostasis has been elucidated largely through the use of transgenic animal models with amplified (63) and nullified genes (53, 29). These models have been directed primarily toward understanding the genetics of hemochromatosis. Also, great inroads in understanding cellular regulation of iron management proteins have been made through studies elucidating the role of iron in altering both transcriptional and translational events (3, 28). Additionally, mouse models of thalassemia comprise an important aspect of the genetics of iron-related proteins (70). The production of null mutations in the Hfe genes in mice has been the mainstay model of human hemochromatosis (43, 56).
Observations of genetic-based differences in ferritin and transferrin suggest that this variation plays a major role in the ability of individuals to maintain homeostasis in the face of fluctuations in dietary iron. Importantly, the prevalence of idiopathic anemia is substantial and may be attributed to genetic-based individual differences in iron-related homeostatic mechanisms (31). The genetics underlying individual differences in these iron-management proteins is underexplored; however, a recent study of twins showed that > 40% of the variance in serum ferritin and nearly 50% of the variance in transferrin saturation was genetically based (71).
Genetic dissection of complex traits using genetic reference populations of rats and mice has yielded valuable information about variability in biological systems and related genetic underpinnings for more than two decades. LeBoeuf et al. (42) showed that among eight strains of mice, there were nearly twofold differences in serum iron, transferrin saturation, and hepatic iron. Subsequently, we observed highly significant effects of mouse strain on the sequelae of iron deficiency in brain and behavior in developing mice (47). Others showed that the background strain of mouse in a gene knockout model of hereditary hemochromatosis influenced the rate of iron accumulation (23, 66), suggesting that other genes besides Hfe influence iron metabolism. We previously reported on the utility of quantitative trait loci (QTL) analysis to identify potential candidate genes that affect one or more iron-related phenotypes (37). In this approach, one can index DNA polymorphisms throughout the genome and correlate those polymorphisms with behavioral or biological phenotypes. Using 15 of the BXD/Ty recombinant inbred strains, we identified chromosomal regions associated with variations in liver iron, as well as ventral midbrain iron concentrations. The results showed liver and midbrain iron concentrations to be largely independently and polygenically regulated.
Two recent studies report QTL related to systemic iron. In the first of these, Bensaid et al. (12) conducted their study of QTL related to liver iron concentration in an F2 intercross between C57BL/6 and DBA/2 inbred mouse strains made deficient in the Hfe genes. Mutation of this gene in humans has been shown to be associated with hereditary hemochromatosis. Johannes et al. (36) conducted QTL analysis of hematocrit in nonmutant C57BL/6 and DBA/2 mice and their derivative recombinant inbred strains. This latter work also reports how identified QTL may change across the life span.
In this paper, we report an analysis of seven markers for systemic iron status, viz., hematocrit, hemoglobin, serum iron concentration, transferrin saturation, total iron-binding capacity (TIBC), liver and spleen iron contents in 30 strains of the well-known [and the same used by Johannes et al. (36)] BXD/Ty recombinant inbred strain panel. These seven measures were then subjected to principal components analysis to reveal overt and latent combinations of factors important for iron homeostasis. The factors identified were then subjected to QTL analysis to reveal chromosomal regions harboring genes that influence these factors. Recombinant inbred strains, when densely mapped for genetic polymorphisms, are especially well-suited for discovery of multiple genes that influence biological and behavioral phenotypes (73). Our major hypothesis was that indices of iron homeostasis are complex traits and are influenced by multiple genes. Moreover, we propose that several of these traits covary and form latent factors that better define the mechanisms by which this important metal is controlled biologically.
MATERIALS AND METHODS
Animals, treatment, and housing.
In this study, we used male and female mice from the BXD/Ty recombinant inbred strain panel. This group comprised 29 strains for the males and 28 for the females and included the C57BL/6 and DBA/2 parental strains. All mice came from our breeding colony at The Pennsylvania State University. At 21 days after birth, mice were separated by sex and were housed 2–4/cage. Breeders and offspring had ad libitum access to water and food (Purina Mills Lab diet #5001; St. Louis, MO) containing ∼270 ppm Fe. Ambient housing conditions were controlled for temperature (23 ± 2°C) and humidity (40%). The animals were maintained on a 12:12-h light-dark cycle (lights on at 0600 h). At death, the mice were 120 ± 3 days of age and each strain included between 4 and 25 mice per sex. All experimental protocols were conducted in accordance with The National Institutes of Health Animal Care guidelines and were approved by the Pennsylvania State Institutional Animal Care and Use Committee.
Hematological and liver iron determination.
For blood and tissue collection, all mice were weighed and killed by CO2 suffocation between the hours of 0900 and 1200. Blood was collected by cardiac puncture; the plasma was separated and frozen at −20°C until use. Plasma iron and TIBC were determined by standard methods (21, 57). Hemoglobin values were determined photometrically by using cyanmethemoglobin standard solution (Sigma Aldrich, St Louis, MO), and hematocrit values were calculated after centrifugation of blood samples in heparinized microcapillary tubes. Livers and spleens were rapidly removed from each mouse, weighed, and then frozen at −80°C for photometric assessment of iron content, modified from Cook et al. (14).
The data are expressed as means ± SE. A two-way ANOVA was performed on the raw data with strain and sex as between-subjects variables. We report main effects of strain and sex and not strain by sex interactions because of the large number of strains. Correlational analysis was determined by Pearson's r using the BXD published phenotypes database within WebQTL (htt://www.genenetwork.org). Following univariate analyses, we performed a principal components analysis using varimax rotation of the seven parameters to examine our data for latent variables. QTL analysis was performed using WebQTL (http://www.genenetwork.org). WebQTL gives the user a choice between likelihood ratio statistics or logarithm of the odds ratios (LOD) as indices of the significance of the QTL. WebQTL performs 1,000 or more permutations of the strain mean data and significance is defined by a QTL LOD score of correctly ordered data exceeding all other permutations 95% of the time, i.e., the 0.05 alpha level [Williams (72)]. Lander and Kruglyak (41) suggest LOD scores of 2.8 and 4.3 to be suggestive and significant, regardless of obtained QTL values. Permutation testing moves these values around slightly. Narrow-sense heritability estimates were calculated as SSstrain/SStotal (6).
Table 1 presents the summary statistics for all of the measures. For hemoglobin, the effect of strain was significant (F29,788 = 6.82, P < 0.001) and the narrow-sense heritability estimate was 0.19. There was no significant effect of sex. For hematocrit, the effect of strain was significant (F29,783 = 7.89, P < 0.001) and the narrow-sense heritability estimate was 0.21. There was no significant effect of sex. For plasma iron concentration, we observed significant effects for strain and sex (F29,771 = 7.53 and 21.57, P < 0.001, respectively). Narrow-sense heritability estimate was 0.21. Overall, males showed higher values for this measure than did females (178.14 ± 2.14 μg/dl vs. 167.10 ± 1.94 μg/dl). For TIBC, the effect of strain was significant (F29,775 = 4.02, P < 0.001) and the narrow-sense estimate of heritability was 0.13. There was no significant effect of sex. For transferrin saturation, we observed significant strain and sex effects (F29,769 = 5.32 and 4.08, P < 0.001, P < 0.05, respectively). The estimate of narrow-sense heritability was 0.16. Males showed slightly higher values than did females (39.33 ± 0.76% vs. 37.58 ± 0.66%). For liver iron concentration, the effects of strain and sex were significant (F29,778 = 20.57 and 425.56, P < 0.001, respectively). The narrow-sense heritability was estimated to be 0.31. Overall, females evinced higher liver iron concentrations than did males (120.39 ± 1.76 vs. 83.01 ± 1.65 μg/g tissue). For spleen iron concentration, the main effect of strain was significant as was the effect of sex (F29,751 = 17.16 and 198.13, P < 0.001, respectively). Narrow-sense heritability was estimated to be 0.33 and, as with liver, females showed much higher concentrations of iron in the spleen (970.79 ± 22.07 vs. 589.05 ± 18.04 μg/g tissue). Strain distribution pattern histograms are available to those interested at http://www.genenetwork.org.
Genetic correlational analysis between measures.
Using strain means as index, we observed significant correlations (all P < 0.01) between transferrin saturation and TIBC, between transferrin saturation and plasma iron concentration, and between hemoglobin and hematocrit in both sexes (Table 2).
Genetic correlations between sex by measure.
For liver, spleen, and plasma iron concentrations the correlations between males and females were 0.75, 0.76, and 0.53, respectively (all P < 0.01). The correlation between males and females for transferrin saturation was 0.6 (P < 0.01); for hemoglobin, 0.4 (P < 0.05); for TIBC, 0.36 (P < 0.06); and for hematocrit, 0.01 (P > 0.5).
Principal components analysis and QTL analysis.
For principal components analysis, we combined the data from both sexes by strain to explore for major factors involved in iron homeostasis. We revealed three factors that together explain about 70% of the total variance in the set of seven variables. Because of their loadings, the factors were named iron availability, iron transport, and iron storage. Using WebQTL (http://www.genenetwork.org), we identified QTL, i.e., polymorphic areas on mouse chromosomes that correlated with each of the factors.
Factor 1: iron availability.
Variables loading this factor with eigenvalues greater than 0.5 were Hb, Hct, and TIBC. This factor accounts for about 33% of the total variance in the set of variables. Figure 1 shows the strain distribution of z-scores. For this factor, we identified a QTL on chromosome 5 near the centromere (at 4 Mb) with a LOD score of 3.3. The bootstrap statistic for this QTL was 41% with a width of 5 Mb.
Factor 2: iron transport.
This factor loaded transferrin saturation and plasma iron and accounts for ∼25% of the total variance in the set of variables. The strain distribution of z-scores is illustrated in Fig. 2. We observed two QTL for this factor, one on chromosome 1 at ∼146 Mb (LOD = 2.2) and the other on chromosome 7 at ∼48 Mb (LOD = 3.8). For the former, the bootstrap statistic was 23% with a width of ∼5 Mb and for the latter, the bootstrap statistic was 50% with an approximate width of 3 Mb.
Factor 3: iron storage.
This factor loaded liver iron uniquely and accounts for ∼15% of the total variance among the set of variables. The strain distribution of z-scores is illustrated in Fig. 3. QTL analysis revealed a peak at 79 Mb on chromosome 2. The LOD score was 1.9. The bootstrap statistic was 22, and the width was ∼3 Mb.
Surprisingly, spleen iron did not load on any of the factors; however, when subjected to QTL analysis, it showed a peak on chromosome 1 at 170 Mb with a bootstrap statistic of 28 and an estimated width of 3 Mb.
Genetic correlations between the factors and other published phenotypes.
One of the advantages of using a panel of recombinant inbred strains is that because the mean is used as the phenotypic index, investigations using the same strains in different laboratories yield results that are comparable. Genetic correlational analysis is based upon mean phenotype indices among laboratories and can point to phenotypes that have genetic-based mechanisms in common. Table 3 presents correlations between factor 1 (iron availability) and phenotypes obtained from other laboratories, and similarly, Tables 4 and 5 present correlations between published phenotypes for factors 2 (iron transport) and 3 (iron storage).
It would appear from Table 2 that the majority of correlates for factor 1 include indices of immune system and response. Next is general behavioral spontaneous activity followed by Morris water maze performance.
Factor 2 indicates central nervous system dopamine neurochemical and behavioral correlates including effects of cocaine, amphetamine, and ethanol. The brain concentrations of zinc and iron are also related to dopamine neurobiology as is PKC. Neuronal and astrocyte proliferation are represented and, as with factor 1, some indices of general immune biology. Again we observe a correlation between this factor on Morris water maze performance.
Factor 3, like factor 1 appears to have several immune system/function correlates. Also, two different studies of seizure susceptibility give very similar correlations. Ethanol, cocaine, and now, benzodiazepine, pharmacology covariates are seen with this factor.
Table 6 presents genomic information concerning some of the better-known iron management proteins. Chromosomal locations in the mouse and human genomes are presented together with gene expression (transcript abundance) data from microarray analysis (http://www.genenetwork.org). With the possible exceptions of our QTL on chromosomes 1 and 7, we observed no QTL in the vicinity of the other protein locations.
Systems genetics analysis capitalizes on genetic-based differences in any chosen biological parameter and then reveals biological mechanisms through related genetic-based differences in DNA sequence, gene expression patterns, and other physiological or biochemical parameters. Moreover, when several parameters with common genetically related mechanisms can be combined through factor analysis, the result is a heuristic that is more stable and more powerful than offered by single-measure parameters. In the present work, through principal components analysis, we identified three factors related to peripheral iron homeostasis among seven estimated parameters.
The factor that explained most of the variance, factor 1 (iron availability), combined TIBC, hemoglobin, and hematocrit, and thus may be considered the most important of the three factors purely in terms of its impact on total variance in the system. Our QTL analysis revealed two genes in the vicinity of the QTL for this factor, Fzd1 and Pftk1. The former gene is involved in the receptor system of the canonical Wnt signaling pathway and the latter is involved in the regulation of thyrotropin-releasing hormone activity through NO-cGMP.
The second factor, iron transport, combined transferrin saturation and plasma iron. We observed two QTL for this factor, one on chromosome 1 at ∼146 Mb and the other on chromosome 7 at ∼48 Mb. The former QTL is close to one of the RIKEN genes, B830045N13, whose function is unknown at this time, and the second QTL is near Luzp2, Leucine Zipper, a transcriptional enhancer, which is located in a dense SNP area.
Iron storage, our third factor, was based on iron concentration in the liver. QTL analysis for iron storage revealed a weak peak at 79 Mb on chromosome 2. The LOD score was 1.9, and the nearest candidate gene was identified as Frzb. We report this possible candidate gene because of its association with the Wnt pathway and the fact that Wnt and β-catenin play a major role in liver zonation (11). Moreover, a search of genetic correlations between Frzb gene expression in hematopoetic stem cells and published phenotypes revealed a Pearson r of 0.52 between transcript abundance and plasma iron concentration in the BXD/Ty recombinant inbred (RI) mouse panel (http://www.genenetwork.org).
Interestingly, spleen iron content was unrelated to any other iron measure. This shows that, while liver and spleen iron content may be correlated in response to iron overload or deficiency, under basal conditions, the genetic-based mechanisms for iron accumulation and retention are different for each organ.
One of the rather striking findings from this work is the apparent differential regulation of various iron parameters between the sexes. For hemoglobin, hematocrit, and TIBC, there were no significant differences between the sexes. For transferrin saturation and plasma iron, males showed slightly higher values than did the females. Liver and spleen iron showed very much larger values for females than for males. In primates, such differences might be explained in terms of relationship to the menstrual cycle; however, rodents have estrous cycles, and periodic blood loss would not explain an apparent adaptive mechanism. The significance of these differences, therefore, awaits further study. The genetic correlations between measures by sex are also of interest. It would seem that liver, spleen, and plasma iron concentrations and transferrin saturation (although showing sex-based differences) share common genetic-based regulatory mechanisms. Hemoglobin, however, seems to be only moderately related between the sexes, and hematocrit shows no correlation at all despite the fact that both measures show genetic-based variation.
Bensaid et al. (12) investigated liver iron loading in F2 mice derived from B6 and D2 Hfe null mutant parents. They showed iron concentrations in the liver to be significantly associated with markers on chromosomes 7, 8, 11, and 12. Our iron transport QTL on chromosome 7 is at ∼48 Mb, while the marker of Bensaid et al., is near D7Mit246 at 30 Mb. The iron transport protein, hepcidin genes Hamp 1 and Hamp 2 are near D7Mit246, and at this time, because of the large areas of recombination in the BXD mice (this panel was derived from an F2 intercross), cannot be ruled out as possible candidate genes for iron transport in the present study. The recent report by Johannes et al. (36) identified several QTL for basal hematocrit in BXD mice at 150, 450, and 750 days of age. In contrast to their findings, we did not find any suggestive or significant hematocrit-related QTL for either sex, although we did find a weak association for females at about 150 Mb on chromosome 1, near their reported QTL for females. Exactly why we failed to replicate their Hct QTL is unknown at this time. One factor may have been time of day of collection, as Hct is known to show a circadian rhythm in rabbits and rats and (24, 54) and probably in mice as well. In our work, we routinely kill and harvest between the hours of 0900 and 1200, i.e., 3–6 h after lights are turned on in the vivarium. Another reason for the apparent discrepancy is the possibility that minor variations occur in Hct based on fluid balance changes making this a relatively unstable measure over the short term. The lack of correlation between males and females across the RI strains for Hct would support this notion. When we combined Hct with Hb and TIBC in both sexes combined, by principal components analysis into a single factor, we were then able to identify at least one suggestive QTL on chromosome 5 near the centromere.
Our factor loadings are remarkably similar to those identified by Murray-Kolb LE and Beard JL (49) in a multivariate study of iron deficiency in humans. Our iron transport factor reflected the same loadings as did the study by Murray-Kolb and Beard, viz., transferrin saturation and plasma iron concentration. Our third factor, iron storage, related to liver iron, is likely related to the storage component identified by Murray-Kolb and Beard. They showed loading of total body iron on this factor.
Genetic correlational analysis revealed that iron availability showed significant associations with immune functions. Iron transport is related to several neurobiological parameters and iron storage is related to psychopharmacological measures. While genetic correlations alone do not necessarily indicate a direct functional relationship, they can be quite useful in generating hypotheses. Thus the genetic correlations observed in Table 2 between iron transport and dopamine D2 receptor density in the ventral midbrain in BXD mice, between iron transport and iron concentration in the ventral midbrain and caudate-putamen, and between iron transport and sensitivity to the behavioral effects of cocaine and ethanol, fit with previous work showing iron-dopamine neurobiology relationships (5, 37). The next step in this research approach will be to identify the actual genes contained in the QTLs identified and then to identify related gene networks.
The BXD/Ty panel of recombinant inbred mice has been studied extensively for a large number of phenotypes, gene expression, and DNA polymorphisms. Many, if not most, of the phenotypes measured, dating from the 1980s, are available on http://www.genenetwork.org. Because the measures are strain means and not values from individual animals, the findings among a set of phenotypes gathered from one laboratory may be compared by genetic correlation of strain means to phenotypes measured in another laboratory in which the same strains are studied. Thus, we see evidence in the three tables that each peripheral iron factor covaries with somewhat distinct sets of biological and behavioral phenotypes. The BXD/RI panel was most intensively studied for their immune system characteristics and functions early on and then beginning in the late 1980s, a large effort to use these strains to understand the neurobiological underpinnings of drugs of abuse, including alcohol, produced a large literature. In fact, the reader should be aware that this latter effort is overrepresented and as the BXD panel becomes employed to study other systems, the correlates that we see with each of the factors should broaden considerably. Indeed, considering the role of iron in immune functions, the number of related covariates, especially for the factors, Iron Availability and Iron Storage would be expected. The number of highly correlated factors related to central nervous system biology and pharmacology is also not surprising, considering the fact that iron is highly connected to neurochemistry, especially dopamine neurotransmission (4).
In conclusion, we have demonstrated the power of systems genetics analysis to study homeostatically regulated biological systems, how factor analysis can reveal hyperparameters that are stable and heuristically useful, and, finally, how genetic correlational analysis can be applied to explore hitherto unknown relationships among physiological and biochemical systems. This work also leads the way to experimental work to show how, for example, iron deficiency may alter the factors, QTL and correlations and reveal new relationships relevant to and useful in understanding physiological systems in health and in disease.
This work was supported in part by National Institute on Aging Grants AG-021190 and National Institute of Neurological Disorders and Stroke NS-35088.
The authors thank the Restless Legs Foundation and GlaxoSmithKline for gifts to help establish the breeding colony of mice. The authors also thank Dr. Sonia Cavigelli for making helpful suggestions on improving the manuscript.
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
- Copyright © 2007 the American Physiological Society