Microarray studies of liver and duodenum from Hfe^-/- mice identified 1,343 transcripts that were upregulated or downregulated in liver of either B6 or D2 Hfe^-/- mice, as compared with wild-type mice of the same genetic backgrounds. Much fewer genes, namely 370, were upregulated or downregulated in the duodenum of these mice. A list of the transcripts differentially regulated between Hfe-deficient and wild-type mice is provided in Additional data files 1 (liver) and 2 (duodenum). As shown in Figure 1, more transcripts were regulated in Hfe-deficient D2 mice than in B6 mice, and this difference was particularly striking in duodenum.

In liver, clustering analysis detected groups of transcripts that were similarly regulated in response to Hfe disruption in B6 and D2 mice (specifically, they were either downregulated [Figure 2, cluster 4] or upregulated [cluster 5] in both strains). However, most of the transcripts modulated after Hfe disruption had expression patterns that were strain specific (regulated only in D2 mice [clusters 1 and 6] or only in B6 mice [clusters 3 and 8]).

In duodenum of B6 mice, the expression of fewer than 20 genes was significantly modified by Hfe deficiency (Figure 1). Consequently, clustering analysis was essentially based on expression changes in D2 mice. Two main clusters were therefore identified in duodenum, one with genes upregulated (cluster 1, Additional data file 2) and the other with genes downregulated (cluster 3, Additional file 2) in response to Hfe disruption in D2 mice.

The Database for Annotation, Visualization, and Integrated Discovery (DAVID) annotation tool was used to search for over-representation of functional categories within the different gene clusters from Figure 2. Categories found to be enriched within the clusters of genes similarly regulated in the liver of Hfe^-/- compared with wild-type mice are summarized in Table 1. As detailed below, they mainly concern detoxification mechanisms in response to oxidative stress, fatty acid β-oxidation, cholesterol catabolism, and circadian rhythm.

As shown in Table 4, there was no clearly enriched functional categories among the 177 genes (cluster 1) that were induced in #Hfe^-/- D2 mice. Conversely, there was significant enrichment of genes involved in the immune defense among the 131 genes that were repressed in the same mice (cluster 3), particularly for genes involved in apoptosis (Casp4, Cdca7l, Ifit1 and Ifit2, Oasl2, and Scotin), innate antiviral or antimicrobial activity (Defcr4, Ddx58, and Lzp-s), and B and T cell mediated immune response (Mpa2l, Psme1, Trfrsf13b, and Tnfrsf17). This suggests a link between the control of iron metabolism and the immune system that should be explored.

Although mRNAs for duodenal iron transporters were not found to be significantly upregulated, expression levels of other metal ion transporters were increased in duodenum of Hfe^-/- D2 mice, most notably the zinc transporters Slc39a4 and Slc39a14. The copper transporter Slc31a1 and, more anecdotally, the sodium-dependent vitamin C transporter Slc23a2 (previously observed to be increased in response to dietary iron deprivation 17) were also induced in D2 mice lacking Hfe. In addition, Hfe^-/- D2 mice had increased expression of the mucin (Muc3) and spermin synthase (Sms) genes, which encode proteins that both may modulate iron uptake 1819.

The Affymetrix GeneChip^® Mouse Genome 430 2.0 arrays contain probe sets for the transcripts of all the genes directly or indirectly involved in iron metabolism 20. Significant alterations in their expression in liver or duodenum of Hfe^-/- mice and gene expression differences between wild-type strains are summarized in Table 5. Specifically in the D2 strain, Hfe disruption induces expression of the Cybrd1 gene in duodenum; this gene encodes Dcytb, which converts dietary ferric iron into its ferrous form for transport. In the liver, Hfe-deficient mice of both strains exhibit upregulated expression of the gene encoding the ferritin light chain, which is responsible for cytosolic iron storage, and of the ferroportin gene, which is consistent with the notion that this protein plays a protective role by facilitating the release of excess iron 21. Somewhat unexpectedly, we observed significant downregulation of the sideroflexin gene (Sfxn2) and upregulation of the mitoferrin gene (Slc25a37) and the Bcrp gene (Abcg2), which encode three molecules that are involved in the mitochondrial import/traffic of iron and heme export. Also worthy of mention are several strain-specific modifications of the messengers of some regulators of iron metabolism in Hfe-deficient mice. First, we confirmed that wild-type B6 and D2 diverge in terms of the amounts of the two hepcidin messengers, namely Hamp1 and Hamp2 22, and we observed a downregulation of the two genes in Hfe^-/- D2 mice. Conversely, we observed significant upregulation of the gene encoding the upstream transcription factor Usf2, which was recently found to be involved in the control of hepcidin expression 23, in the B6 strain. Finally, and worthy of note within the context of modifiers of iron loading severity, wild-type D2 mice have significantly lower expression of the Smad4 transcription factor, also involved in the control of hepcidin expression, than wild-type B6 mice.

Quantitative real-time PCR was performed on 21 genes expressed in the liver and four genes expressed in the duodenum. The selection of these genes was based on different criteria. The first group included genes of an enriched functional category identified using the DAVID annotation tool (Aox1, Ftl1, Fpn1, Hmox1, Vnn1, Por, Cpt1a, Aco2, Cyp7a1, and Hsd3b5). The second group of genes encode proteins of iron or heme metabolism, and their expression was either induced or repressed in Hfe^-/- mice (Hfe2, Hamp1, Hamp2, Usf2, Lcn2, Sfxn2, Alas2, Slc25a37, and Abcg2). The third group encode proteins that might modulate iron absorption in the duodenum (Dcytb, Slc39a4, and Muc3). The fourth group includes genes that, although their involvement in iron metabolism regulation cannot be assumed, were highly regulated in liver (Lcn13 and Fmo3) or duodenum (Clca4) of Hfe^-/- mice. A further 20 mice that were not analyzed using Affymetrix arrays (five per genotype/strain combination) were included in the analysis to test the validity of the results.

Concordant results were obtained for 24 out of 25 genes selected. Downregulation of the hemojuvelin gene (Hfe2) in Hfe^-/- B6 mice was not confirmed. Downregulation of Lcn2, Hamp1, and Hamp2 in Hfe^-/- D2 mice was confirmed in the samples used for Affymetrix array hybridizations but not in the additional samples used for validation, although a trend toward downregulation was observed in the validation set for Hamp1 and Hamp2. The upregulation of Usf2 and Slc25a37, originally found only in the liver of Hfe^-/- B6 mice, was observed by quantitative PCR in both strains. Interestingly, Lcn13 and Fmo3 - which had highly significant S-scores of 11.06 and -6.66, respectively, in the liver of Hfe^-/- D2 mice - were confirmed to be regulated by Hfe deficiency in both datasets. Because neither of these two genes is regulated by dietary iron content in wild-type mice (data not shown), these variations appear specific to Hfe disruption and warrant further investigation.

Differences in liver or duodenal expression of specific genes between B6 and D2 wild-type mice could contribute to the divergent phenotypes induced by Hfe disruption in the two strains. We therefore established a list of the 1,538 transcripts with differential expression between wild-type D2 and B6 mice (Additional data file 3). In order to relate genomic results to severity of hemochromatosis, we first identified 210 genes exhibiting differences in basal expression between strains or with expression regulation in response to Hfe disruption, which reside within the five Hfe-modifier regions that we previously mapped on chromosomes 3, 7, 8, 11, and 12 10. To identify those that could be potential candidates for disease severity, we used the WebQTL interface to map the loci that regulate the expression of these genes. The information necessary to map these regulatory loci was available for a subset of 139 of these 210 genes.

We found that two genes on chromosome 3, four on chromosome 7, six on chromosome 8, 17 on chromosome 11, and one on chromosome 12 exhibited highly significant evidence for cis regulation (for regulation by a polymorphic variant between B6 and D2 mice located in the region of the gene itself; Table 6). None of them, except for Hamp, has yet been implicated in iron metabolism.

Male Hfe^-/- (knockout) mice of the C57BL/6 (B6) and DBA/2 (D2) backgrounds were produced in the Institut Fédératif de Recherche (IFR) 30 animal facility 9. Wild-type Hfe^+/+ controls (wild-type) of the same sex and genetic backgrounds were purchased from the Centre d'Elevage Robert Janvier (Le Genest St-Isle, France). The studied population consisted of 16 wild-type mice (eight B6 and eight D2) and 16 knockout mice (eight B6 and eight D2). Three mice in each of the four genotype/strain groups were used for genome-wide expression profiling, and five for validation of microarray results. Wild-type and knockout mice were housed in the IFR30 animal facility and had free access to water and R03 diet (UAR, Epinay-sur-Orge, France) containing 280 mg Fe/kg. All mice were analyzed at 7 weeks of age and fasted for 14 hours before they were killed. Experimental protocols were approved by the Midi-Pyrénées Animal Ethics Committee. Liver and duodenum were dissected for RNA isolation, rapidly frozen, and stored in liquid nitrogen. Nonheme iron was quantified as described previously 10. Mean ± standard deviation iron concentrations were 304 ± 50, 456 ± 68, 946 ± 110, and 2,937 ± 282 μg/g dry weight in liver of B6 wild-type, D2 wild-type, B6 knockout, and D2 knockout mice, respectively. Mice fed an iron-deficient or an iron-supplemented diet were obtained as described previously 40. Liver and duodenum samples were used to compare gene expression variations resulting from lack of functional Hfe with those induced by secondary iron deficiency or iron overload.

Total RNA was extracted and purified using the RNeasy Lipid Tissue kit (Qiagen, Courtaboeuf, France). RNA quality was checked on RNA 6000 Nano chips using a Bioanalyzer 2100 (Agilent Technologies, Palo Alto, CA, USA). RNA samples used for chip experiments all had RNA Integrity Numbers 41 ranging from 9 to 10. Double-stranded cDNA and biotin-labeled cRNA were synthesized using the Affymetrix cDNA synthesis and IVT Labeling kits. Fragmented cRNAs (15 μg) were hybridized to 24 GeneChip^® Mouse Genome 430 2.0 arrays (Affymetrix, Santa Clara, CA, USA), in accordance with the standard protocol of the manufacturer. The arrays were scanned with a GeneChip^® Scanner 3000 (Affymetrix) and raw image files were converted to probe set data (*.CEL files), using the Affymetrix GeneChip^® Operating Software. Expression microarray data have been submitted to the National Center for Biotechnology Information's Gene Expression Omnibus repository (accession number Genbank: GSE7357).

All the analyses were performed using Bioconductor, an open source software for the analysis of genomic data rooted in the statistical computing environment R 42. Arrays were normalized to have the same target mean intensity of 100. Quality control metrics were first obtained using the simpleaffy Bioconductor package 43. Average background and the number of genes called present (42% to 48% in liver and 50% to 55% in duodenum) were similar across all chips. All arrays had a scale factor lower than 1.4-fold away from the average scale factor for all samples, a GAPDH (glyceraldehyde 3-phosphate dehydrogenase) 3':5' ratio at around 1 and a β-actin 3':5' ratio of under 2.2. Furthermore, plots of mean intensity per probe position averaged over all probe sets had very similar slopes for the different arrays, permitting valid comparisons within genes across arrays. Genes that were not reliably detected in at least three liver samples or three duodenum samples, in accordance with the Affymetrix detection call algorithm, were excluded from further analysis 44. Of the 45,101 probe sets represented on the GeneChip^® Mouse Genome 430 2.0 arrays, 28,031 were retained for assessing changes in gene expression between groups of mice.

The S-score algorithm, available in the Bioconductor Sscore package 45, was applied to compare hybridization signals between two arrays. It uses the statistical power of all oligonucleotide pairs for a given gene and is thus particularly useful for studies having limited numbers of Affymetrix microarrays 46. S-scores have a normal distribution with mean of 0 and standard deviation of 1, and are correlated with the fold change. Three types of comparisons were made: S-scores were calculated for D2 wild-type versus B6 wild-type samples within each organ to examine basal strain expression differences between D2 and B6 mice; S-scores were calculated for knockout versus wild-type samples within each organ and mouse strain to study responses to Hfe disruption; and control S-scores were calculated between biologic replicates within the different groups. To reduce the contribution of biologic and technical noise, S-scores were divided by the greater of 1 or the standard deviation of control S-scores within each organ. This general approach has been applied previously to microarrays 47 and reduces variance across experimental replicates 48. Statistical analysis of microarrays (SAM) 49, a rank-based permutation method, was carried out to identify genes with S-scores significantly different from 0, using the R samr package. Genes regulated by Hfe deficiency were identified for each strain/organ combination by performing one-class SAM on knockout versus wild-type scores, using a false discovery rate of ≤10% to avoid eliminating genes that may be biologically important and increase our ability to populate functional networks of genes in subsequent bioinformatics studies. Hfe-regulated transcripts identified by SAM were filtered to count transcripts with an average S-score over three observations of ≥2 or ≤-2. Genes that exhibited both significant and reproducible changes were further analyzed for correlated gene expression patterns by application of k-means clustering, as described by Eisen and coworkers 50. Genes differentially expressed between mice strains were identified by one-class SAM on wild-type D2 versus wild-type B6 S-scores, using a false discovery rate of ≤1%. This gene list was further filtered for an average S-score of ≥2.6 or ≤-2.6 over three observations.

DAVID (2007), a functional annotation tool 5152, was used to identify enriched biologic themes and to discover function-related enriched gene groups among clusters, compared with all genes present on the Mouse Genome 430 2.0 array. The following annotation groupings were analyzed for overrepresentation in gene lists: the Protein Information Resource keywords, Kyoto Encyclopedia of Genes and Genomes and BioCarta pathways, and Gene Ontology biological processes and molecular functions. Results were filtered to remove categories with EASE (expression analysis systematic explorer) scores, based on a Fisher exact test, greater than 0.05. Redundant categories with the same gene members were removed to yield a single representative category. The chromosomal location of all genes exhibiting differential basal expression between strains or regulation by Hfe deficiency was superimposed on support intervals for hepatic iron loading modifiers on mouse chromosomes 3, 7, 8, 11, and 12 10, and a list of differentially expressed genes mapping to these intervals was obtained. The WebQTL resource 5354, which includes measures of mRNA expression in livers of 35 adult BXD recombinant inbred male mice obtained with Agilent G4121A microarrays, was used to link expression of the genes in this list to genetic markers and identify potential cis-regulators.

All primers were designed using the Primer Express 2.0 software (Applied Biosystems, Foster City, CA, USA). Quantitative real-time PCR reactions were prepared with M-MLV reverse transcriptase (Promega, Charbonnières-les-Bains, France) and qPCR MasterMix Plus for SYBR^® Green (Eurogentec, Seraing, Belgium), as described previously 9, and run in duplicate. GenBank accession numbers, forward (F) and reverse (R) primers, and measured PCR efficiencies for the genes to be validated are given in Table 8. For each mouse, an expression measure was calculated as E_GoI^{Ct GoI}/E_HPRT^{Ct HPRT}, where GoI is the gene of interest; HPRT is a transcript with stable level between strains and genotypes, quantified to control for variation in cDNA amounts; E is the PCR reaction efficiency associated with either the gene of interest (E_GoI) or the reference gene (E_HPRT); and Ct is the cycle number at which fluorescence reaches a given threshold. Data were analyzed by one-factor (iron-deficient, standard, or iron-supplemented diet) or two-factor (chip/validation experiment and wild-type/knockout genotype) analysis of variance followed by Scheffe post-hoc tests using SAS software (version 9.1.3; SAS Institute Inc., Cary, NC, USA).