Promoties Geneeskunde

Korte samenvatting:

In het kader van een onderzoek naar oorzaken van atherosclerose paste promovenda Arja Kreeft verschillende DNAtechnieken toe (SAGE =Seriële Analyse van GenExpressie en twee typen DNAmicroarrays) om in levers van genetisch gemodificeerde hyperlipemische muizen (APOE3Leiden- en ApoE-/-muizen) genen te identificeren die betrokken zijn bij vetmetabolisme. Ze onderwiep deze proefdieren aan hoogvetdiëten en aan cafestol, een bekende cholesterolverhoger uit koffie en vond honderden nieuwe en bekende genen (o.a. Cyp7a1, het gen dat de omzetting regelt van cholesterol naar galzouten). Veel genen blijken tevens betrokken te zijn bij door kernreceptoren gereguleerde detoxificatieprocessen en ontstekingsreacties. Haar gegevens vormen een belangrijke basis voor vervolgstudies

Samenvatting:

Hart- en vaatziekten worden veroorzaakt door verstoringen in fysiologische processen. Deze verstoringen kunnen het gevolg zijn van genetische defecten en/of van bepaalde omgevingsfactoren zoals voeding. De belangrijkste oorzaak voor het ontstaan van hart- en vaatziekten is atherosclerose, een chronische ontstekingsreactie van de vaatwand. Atherosclerose wordt een acuut probleem als dit in ernstige mate optreedt, wat uiteindelijk kan resulteren in een hartinfarct. Atherosclerose is een multi-factoriële ziekte, bepaald door talrijke gen-gen en gen-omgevings interacties. Epidemiologische en dierstudies hebben inmiddels talrijke omgevings- en genetisch bepaalde risicofactoren aangetoond, die de frequentie en de mate van ernst van atherosclerose bepalen. Belangrijke risicofactoren zijn onder andere een hoogvet dieet en verhoogde atherogene plasma lipoproteinen spiegels (hyperlipidemie). Op dit moment zijn er slechts enkele genen ontdekt welke ten grondslag liggen aan een verstoord lipoproteine metabolisme en de vorming van atherosclerose. Zodoende is het nog onmogelijk om de genetische oorzaak van de meeste patiënten met hyperlipidemie aan te wijzen.

Kennis van nieuwe genen en regulerende mechanismen geeft nieuwe inzichten in de bestrijding van hart- en vaatziekten. Het doel van dit proefschrift was het identificeren van nieuwe genen en biologische netwerken die geassocieerd zijn met hyperlipidemie en atherosclerose vorming. De studies beschreven in dit proefschrift hebben zich gericht op de lever, omdat dit het belangrijkste orgaan is in het lipidenmetabolisme. Twee muismodellen werden gebruikt, de ApoE knock-out (E-/-) en de transgene APOE3Leiden (E3L) muizen. ApoE deficiëntie in de E-/- muis leidt tot verhoogde cholesterolniveaus door een verstoorde opname van remnant lipoproteine deeltjes uit het plasma. De E3L muis draagt het humane APOE3Leiden - APOC1 gen cluster op een homogene C57BL/6JIco (B6) achtergrond. In mensen is de APOE3Leiden mutatie geassocieerd met Familiaire Dysbetalipoproteinemie (FD). E3L muizen hebben een vergelijkbaar defect in de opname van VLDL remnants uit het plasma als FD patiënten, wat leidt tot verhoogde plasma lipidenspiegels en dieet geïnduceerde veranderingen in het lipidenmetabolisme. Doordat de E3L muis een vergelijkbaar lipidenprofiel laat zien als mensen, is de E3L muis uitermate geschikt voor het bestuderen van de effecten van dieet op het lipidenmetabolisme en het identificeren van dieet gereguleerde genen. Op een mild hoogvet dieet, het zogenaamde “Westerse dieet”, hebben deze muizen duidelijk verhoogde plasmacholesterol en triglyceriden spiegels. Deze verhoging wordt versterkt als diëten worden gebruikt die zowel cholesterol als galzouten bevatten. Deze diëten worden veelvuldig gebruikt voor het induceren van atherosclerose. Verder zijn E3L muizen gevoelig voor cafestol, de cholesterol verhogende factor in ongefilterde koffie.

Aan de E3L muizen werden verschillende diëten gevoerd, waarna de genexpressie in de lever werd bestudeerd door middel van innovatieve technieken zoals Serial Analysis of Gene Expression (SAGE) en microarrays. Met deze technieken is het mogelijk om de genexpressie van duizenden genen te bestuderen in één enkel experiment. Door de experimentele conditie te vergelijken met de controle conditie kunnen verschillen in genexpressie tussen beide condities worden gedetecteerd. Dit geeft informatie over welke processen verstoord zijn en is de fundamentele basis voor het vinden van nieuwe kandidaatgenen voor bepaalde ziekten.

In hoofdstuk 2 hebben we met behulp van SAGE, het effect bestudeerd van het APOE3Leiden transgen op het lipidenmetabolisme. Hiervoor, werden de genexpressie profielen bestudeerd van E3L muizen en B6 muizen op een standaard chow dieet. Door middel van SAGE konden meer dan 18.000 levertranscripten van beide muizen bestudeerd worden en kon worden vastgesteld van welke genen de expressie verschilde tussen de twee groepen muizen. De 18.000 levertranscripten vertegenwoordigden meer dan 9400 genen. Van 175 genen was de expressie verschillend tussen de E3L en B6 muizen (p<0.05). Van de meerderheid van deze genen was de functie onbekend, maar diverse van deze genen behoorden tot bekende biologische processen zoals het lipidenmetabolisme, glycolyse en onstekings- en detoxificatieprocessen.

Om een vollediger beeld te krijgen van genen die worden beïnvloed door het genotype, hebben we in hoofdstuk 3 en hoofdstuk 6 E3L en B6 muizen op chow dieet bestudeerd door middel van cDNA en oligonucleotide arrays. De cDNA arrays (GEM 2.03) bevatten 9552 genes/ Expressed Sequence Tags (ESTs) en de oligonucleotide arrays 6595 genes/ ESTs (Mu11KB). Voor de cDNA en oligo arrays werd hetzelfde levermateriaal gebruikt van de E3L en B6 muizen, waardoor we de data van de cDNA en oligo arrays met elkaar konden vergelijken. De software tool GeneHopper werd gebruikt om de gegevens van de verschillende expressie platforms aan elkaar te koppelen (hoofdstuk 5). Op deze manier vonden we een set van genen die in beide platforms differentieel tot expressie kwamen (hoofdstuk 6).

Opmerkelijk was dat de differentieel tot expressie komende genen gevonden met deze microarray studies tot dezelfde biologische processen behoorden als de genen gevonden in de SAGE studie. We vonden een duidelijk effect op het lipidenmetabolisme, detoxificatie processen en de acute fase/afweer respons. Deze data werden bevestigd in twee andere experimenten. E-/- muizen op chow dieet en E3L muizen die een hoogvet, hoogcholesterol (HFC) dieet gevoerd kregen werden vergeleken met B6 muizen op hetzelfde dieet. We vonden dat dezelfde functionele klassen van genen betrokken waren. Dus, de modificatie van APOE in zowel de E3L als de E-/- muizen resulteerde in de regulatie van dezelfde netwerken van genen. Interessant is dat dit duidt op een vergelijkbaar onderliggend mechanisme dat respondeert op veranderingen in het lipidenmetabolisme.

Ook werd de respons van E3L muizen op verschillende hoogvet diëten bestudeerd. Twee verschillende diëten werden gebruikt; een mild hoogvet dieet W (0.25% cholesterol) en een sterk hoogvet dieet N dat ook galzout bevat (1% cholesterol, 0.5% cholaat). Het lipidenmetabolisme, de onstekings- en detoxificatie- processen werden ook beïnvloed door dieet W en N. Naast genen die geassocieerd zijn met het lipidenmetabolisme bleek dieet N ook genen uit andere functionele klassen te reguleren zoals cytoskelet organisatie/biogenese en cel-cel communicatie. De E3L en B6 muizen vertoonden een vergelijkbare response op dieet N, dat resulteerde in de differentiële expressie van 800-1000 genen. Opmerkelijk is dat het plasma niveau van E3L muizen ongeveer negen keer hoger is dan van B6 muizen, maar dat deze hogere gevoeligheid voor dieet geïnduceerde hyperlipidemie niet resulteert in een proportioneel hoger aantal differentieel geëxpresseerde genen. Dit is de eerste studie die dit laat zien en het is interessant om te weten of dit ook voor andere muismodellen geldt.

Verdere gedetailleerde onderverdeling van het lipidenmetabolisme en detoxificatieprocessen in metabole netwerken op basis van de Kyoto Encyclopedia of Genes and Genomes (KEGG) database liet de repressie van genen zien die betrokken zijn bij het galzuur-, sterol-, steroïde-, vetzuur- en detoxificatiemetabolisme. Nucleaire receptoren worden geactiveerd door binding van cholesterol, galzuren, vetzuren en hun metabolieten. Onze data werden vergeleken met gepubliceerde data die betrekking hebben op de genregulatie via deze nucleaire receptoren. We vonden een substantiële rol voor een heel scala aan genen die onder invloed staan van nucleaire receptoren (nucleaire receptor targetgenen). Veel nucleaire receptoren werden gereguleerd onder hoogvet condities zoals de farnesoid X receptor (FXR), pregnane X receptor (PXR), constitutive androstane receptor (CAR), liver X receptor (LXR), peroxisome proliferating receptors a/d (PPARs a/d), hepatic nuclear factor-4a (HNF-4a) en de sterol regulatory element-binding proteins (SREBPs). De effecten die teweeggebracht werden door dieet N waren sterker dan de effecten die teweeggebracht werden door dieet W. Dieet N reguleerde de meerderheid van de nucleaire receptor targetgenen. Het is mogelijk dat de effecten van dieet N, mede komen door het galzout in het dieet. Behalve de directe regulatie van de FXR en de PXR/CAR receptor door galzouten, zijn er nog geen andere receptoren bekend, die door galzouten worden gereguleerd in de lever. De gezamenlijke regulatie van vele nucleaire receptor targetgenen betrokken in het lipidenmetabolisme en detoxificatieprocessen, suggereert dat een mechanisme ingeschakeld wordt ter bescherming van de cel tegen toxische endogene lipiden en galzouten. Opvallend was dat verscheidene nucleaire receptor targetgenen gezamenlijk gereguleerd werden gedurende alle experimentele condities. Dit kan erop duiden dat deze genen een sleutelrol spelen in regulerende processen die belangrijk zijn in het lipidenmetabolisme. Een van deze genen was Abcg5. Een verstoorde f

Een ander deel van het onderzoek richtte zich op het werkingsmechanisme van cafestol, de cholesterolverhogende factor in koffie (hoofdstuk 4). Cafestol verhoogt plasma niveaus van LDL en VLDL in mensen. Of mensen cafestol tot zich nemen hangt af van de manier waarop de koffie gezet wordt: cafestol wordt uit de koffie gefilterd door het papieren filter. Maar ongefilterde koffie, zoals "French press" koffie, cafetière koffie, Scandinavische kookkoffie en in mindere mate espresso, bevatten cafestol en verhogen cholesterolspiegels. Het drinken van veel ongefilterde kookkoffie heeft bijgedragen aan de hoge cholesterolspiegels en een hoge mate van hart- en vaatziekten in Scandinavië. Een omschakeling van kook- naar filterkoffie was verantwoordelijk voor 30% van de daling van 1 mmol/l in cholesterolniveaus in de periode 1972-1992 in Finland. Alhoewel een aantal studies het onderliggende mechanisme van cafestol op het lipidenmetabolisme bestudeerd hebben, was het onderliggende moleculaire mechanisme nog niet in detail bekend. Het opdoen van kennis van nieuwe genen en netwerken die gereguleerd worden door cafestol kan nieuwe genen identificeren die betrokken zijn bij het reguleren van lipiden niveaus door voedingscomponenten.

In hoofdstuk 4 wordt een studie beschreven dat als doel had het identificeren van genen en netwerken die beïnvloed worden door cafestol. Voor deze studie werden E3L muizen gebruikt, omdat ze een vergelijkbare cholesterolverhoging laten zien als mensen in respons op cafestol. Om het mechanisme te bestuderen hoe cafestol het lipidenmetabolisme beïnvloedt werden er "genome wide" expressie studies uitgevoerd, waarvoor levers van cafestol gevoerde muizen bestudeerd werden. In cDNA array experimenten werden levers van vier individueel cafestol gevoerde muizen bestudeerd, die een verhoogd cholesterolniveau lieten zien in respons op cafestol. Als controle werd een groep muizen gebruikt op controle dieet. Van 648 transcripten was de expressie beduidend verschillend (p<1x10-6) ten aanzien van de controlegroep in tenminste één muis. Deze genen werden geclusterd op basis van gelijkheid van genexpressie. Deze analyse gaf twee clusters die genen bevatten die in grote lijnen een zelfde regulatie vertoonden. Om meer inzicht te verkrijgen in welke genen betrokken waren in het lipidenmetabolisme, werden deze genen in klassen ingedeeld op basis van hun functie met behulp van de Gene Ontology (GO) database. Aan de hand van deze functionele analyse vonden we 42 lipidengenen. Door alle genen van de chip op dezelfde manier te classificeren, waren we in staat het aantal verwachte lipidengenen te bepalen. We vonden dat het geobserveerde aantal lipidengenen aanmerkelijk hoger lag dan het verwachte aantal lipidengenen. Dit toonde aan dat cafestol een significant effect had op het lipidenmetabolisme. Een interessante observatie was dat een van de sterkst gereguleerde genen, in alle vier muizen, Cyp7a1 was. Cyp7a1 is een sleutelenzym bij de omzetting van cholesterol naar galzouten in de lever. Eerdere studies hebben ook laten zien dat cafestol de aanmaak van galzout remt in ratten hepatocyten en in E3L muizen door remming van cyp7a1. De remming van cyp7a1 kan indirect leiden tot verhoogde serum cholesterolniveaus in E3L muizen door repres

Verdere karakterisatie van de differentieel tot expressie komende lipidengenen liet zien dat verscheidene genen een rol spelen in het cholesterol-, het galzuur- en het steroïdenmetabolisme. Een unieke observatie was dat de meerderheid van deze genen direct gereguleerd worden door de nucleaire FXR receptor. Deze specifieke nucleaire receptor target genen zijn Cyp7a1, Pltp, ApoA-I en ApoA-V. Deze studies vormen de basis voor verdere in vitro studies die het mechanisme van cafestol via deze receptor verder ontrafelen. Deze studies laten zien dat de omzetting van cholesterol in galzouten in de lever een belangrijk aangrijpingspunt kan zijn in de ontwikkeling van nieuwe diëten en geneesmiddelen die cholesterolniveaus en atherosclerose beïnvloeden. Recent is aangetoond dat ook andere componenten die in de natuur voorkomen de FXR receptor kunnen activeren, zoals guggulsteron. Guggulsteron is een plantensterol afkomstig van de guggul boom dat LDL niveaus in mensen verlaagd. Eveneens verlaagt guggulsteron het levercholesterol in wildtype muizen die een hoogvet dieet gevoerd werden, maar dit gebeurde niet in FXR- knockout muizen.

Cafestol had niet alleen effect op genen die betrokken zijn bij het lipidenmetabolisme, maar liet ook sterke effecten zien op detoxificatie-processen. De effecten op glutathione S-transferases zijn eerder gemeld in vorige studies, maar in deze studie werden vele nieuwe detoxificatiegenen gevonden. Opmerkelijk was dat we ook verscheidene targetgenen voor de nucleaire receptoren PXR/CAR hebben geïdentificeerd. PXR is een receptor die geactiveerd wordt door lichaamsvreemde stoffen (xenobiotica) en die vele detoxificatie enzymen reguleert. Interessant is dat recente studies aantoonden dat Cyp7a1 ook een target gen is voor de PXR receptor.

In hoofdstuk 7 worden de resultaten van deze studies samenvattend besproken. De studies in dit proefschrift bestudeerden de effecten van verschillende diëten op genexpressieniveau, in de lever van verschillende hyperlipidemische muis modellen. Met SAGE en microarrays werd aangetoond dat heel veel genen differentieel tot expressie komen. Naast het lipidenmetabolisme werden de afweerrespons en detoxificatieprocessen sterk gereguleerd door dieet en genotype. Verder vonden we aanwijzingen dat vele nucleaire receptoren een belangrijke rol spelen in deze processen. Een aantal van deze genen bleken interessante kandidaatgenen te zijn voor verdere studies. Deze data versterken ook dat er een nauw verband bestaat tussen hyperlipidemie en ontstekingsprocessen. De expressiestudies die betrekking hadden op cafestol vormen de basis voor verdere studies die het mechanisme van cafestol via de FXR receptor ontrafelen. Verder vonden we in deze studies vele genen die veranderd tot expressie kwamen die nog niet eerder in verband werden gebracht met hyperlipidemie en atherosclerose. Ze vertegenwoordigden een heel scala aan metabole processen. Ook vonden we genen waarvan de functie onbekend was, of het waren nieuw geïdentificeerde genen. De gedetailleerde resultaten, die uit dit onderzoek naar voren komen, vormen een waardevolle basis voor verdere vervolgstudies naar het ontstaan en de preventie van hart- en vaatziekten.

Summary:  

The work described in this thesis was to identify (novel) dietary response genes and pathways associated with hyperlipidemia and the development of atherosclerosis. Two widely used mouse models for studying hyperlipidemia and atherosclerosis were included in these studies: the E3L and the E-/- mice. Mice were fed chow diet and different high fat, high cholesterol diets and the natural dietary compound cafestol, a diterpene present in coffeebeans. To study dietary gene regulation, livers of these mice were used for large scale gene expression profiling, using three different expression profiling techniques, namely Serial Analysis of Gene Expression (SAGE), oligonucleotide (-oligo) arrays and cDNA arrays. This chapter will start with a discussion of several technical aspects of the different expression profiling platforms, and the rest of this chapter will focus on the biological implications and future prospects of these genome wide gene expression profiling studies.

1. Technical aspects

1.1 Technical aspects of SAGE

In chapter 2 the results of a SAGE study were described, in which the liver gene expression profiles of E3L and B6 mice were compared on standard chow diet. Over 18,000 liver transcripts of B6 and E3L mice were analyzed, and when combined resulting in > 37,000 tags representing some 9500 unique genes. Currently, the sequencing efforts of the mouse genomes resulted in estimations of the total number of genes of the mouse; 27,000-30,500 protein-encoding genes 1. Thus, assuming the total number of genes expressed in liver, a SAGE expression profile was generated of about 30% of the total estimated number of mouse genes. Accordingly, in our SAGE expression profile, mostly high abundant genes were identified. These genes encoded plasma proteins involved in binding and transport, protease inhibitors, complement factors and genes involved in several metabolic processes such as the synthesis of lipoprotein particles and detoxification processes. The most abundant genes in liver were serum albumin (4,5 %) and ApoE (2.6 %). On the other hand, low abundant genes such as receptors were not detected or the corresponding tag was detected only once or twice as in the case of the LDL receptor. Currently, only one other study reported a SAGE expression profile of (human) liver and our results were in agreement with results from this study. They sequenced a total of 30,982 tags of which 8,596 were unique gene products 2. In that study they also found serum albumin (3.5%) and several apolipoproteins including APOE (0.94%) as the most abundant transcript in liver. Furthermore, other high abundant genes were protease inhibitors, complement factors, coagulation factors and enzymes associated with lipid and amino acid metabolism and detoxification processes. However, this study was limited to only 1 SAGE expression profile and included no comparisons with other SAGE profiles.

In our SAGE study, 40% of the 9500 unique genes matched with a known mouse sequence from Genbank. Considering the (near) completion of the mouse genome, it may be expected that many more unknown tags could be currently matched to a known sequence. However, to identify novel transcripts, more sequence of the tags that still remain unknown should be obtained, and for this purpose, several techniques have been developed such as PCR by rapid amplification of sequenced SAGE tags and the generation of longer cDNA fragments from SAGE tags for gene identification 3,4.

Subsequently, SAGE expression profiles of B6 and E3L mice were compared and 175 genes showed a differential expression (p<0.05). For confirmation of differential gene expression, we selected 10 genes indicated by SAGE as differentially expressed. The differences in expression levels from 8 out of these 10 genes detected by SAGE could be confirmed by Northern blot analysis. Differences between the observed results may be due to differences between the techniques. SAGE and Northern blot analysis are complementary techniques, PCR-based and hybridization-based respectively, each with their own specific sensitivity and consequences for quantifying differences in expression levels. For instance, the differential expression of cytochrome p-450 naphtalene hydroxylase as indicated by SAGE could not be confirmed by Northern blot analysis. This may be due to the fact that northern blot analysis does not allow to distinguish between members of (large) gene families such as homologous members of the cytochrome p450 family. Moreover, if the current SAGE expression profile would be extended by increasing the number of analyzed SAGE tags, it is possible that the tag count for E3L and B6 mice may fluctuate resulting in an altered fold change. Therefore, the differences between SAGE tags should not be considered as absolute values, but rather as an indication of a significant difference in liver gene expression between mouse strains. Other observed biases between techniques may be caused by sequencing errors, non-uniqueness and non-randomness of SAGE tag sequences 5.

As previously described we have sequenced > 18,000 tags for the B6 and E3L mice on chow diet. Thus, our SAGE expression profile gave a reliable overview of the expression of high abundant genes that are expressed in mouse liver. However, to retrieve exhaustive information about more genes and to get a better indication of differential gene expression, one would have to sequence more tags, as the sensitivity of SAGE increases with the accumulation of tags. It has been calculated that ~1,200,000 tags have to be sequenced per condition to identify at least one tag for a given transcript, when assuming 300,000 transcripts per cell 6. Sequencing such high amounts of tags is unrealistic in practice, because of time and resource. Therefore, to extend our SAGE study and retrieve more information on a genome wide scale on differentially expressed genes in livers of B6 and E3L mice on chow, microarray experiments were performed as described in chapter 3 and 6.

1.2 Technical aspects and comparison of oligonucleotide and cDNA arrays

Using cDNA arrays, only the expression of genes that are present on the array can be assessed. To retrieve optimal information in our studies two Gene chips were used, an oligo microarray containing 6500 genes/ESTs from Affymetrix (Mu11KB), and a cDNA array containing 9900 genes/EST from Incyte Genomics (GEM 2.03). Currently, due to the rapid developments in the microarray field and sequencing efforts of the mouse genome, there are Genechips available containing the whole sequenced mouse genome interrogating approximately 39,000 full-length mouse genes and EST clusters from the UniGene database (Genechip Mouse Expression Set 430) (Affymetrix, Santa Clara, CA, USA). For the SAGE and microarray experiments, starting material was used derived from the same livers enabling us to perform direct comparisons between the gene expression patterns obtained with these different techniques (technical replication), as described in chapter 6. Currently, biological and technical replication is often limited due to practical reasons like the available quantity of RNA and high experimental costs 7. Although SAGE data are easily comparable between SAGE experiments, a comparison between different microarray platforms is complicated because different probe identifiers for the same genes are often used. To circumvent the problem of the use of different probe identifiers for the same genes in different microarray platforms, a software tool called GeneHopper was developed. Using Genehopper we were able to link the cDNA array probes with the corresponding Affymetrix probes through their Genbank accession number. This and other bioinformatic tools delivered by Genehopper were described in Chapter 5. A similar high-throughput web-based database for the annotation and comparison of commonly available microarray resources, RESOURCERER, was developed at The Institute for Genomic Research. This database allows comparisons between resources from the same species using either the TIGR Gene Indices or UniGene and between species using the TIGR

Next to using different expression profiling platforms, application of different software packages that often apply different statistics for expression analysis, makes direct comparison of microarray data even more difficult 9. Currently, for the analysis of oligo array and cDNA array data it is not possible to use a similar software package, applying similar statistics, for analyzing the data. Thus we performed the same statistical test (z-test) on the replicate expression intensities retrieved from the different software packages to compare the results from both techniques. The z-test is a statistical test quite similar to the t-test. However, in contrast to the t-test, the z-test achieves a sufficient sensitivity by virtue of a more precise error estimation with only two replicates 10. The cDNA array probes (=10176) and oligo array probesets (n=6595) were linked by their accession numbers using GeneHopper and 2261 genes were shared in the datasets, which was a sufficiently large number of overlapping clones to compare the different datasets.

Several differences between datasets generated with oligo and cDNA arrays were detected. Firstly, it appeared that for the cDNA arrays ~90% of all the clones were reliably detected, which was much lower for the oligo arrays (~30%). This result was similar for another study in which liver material from fish-oil fed B6 and control mice was studied using similar Mu11KB oligo arrays, indicating a similar percentage (~27%) of reliably detected clones as in our experiments 11. Additionally, for the 2261 clones in the shared dataset, 2200 clones (97.3 %) could be reliably detected by cDNA arrays that were only 1110 (49%) for oligoarrays. Because similar material was used for both techniques, our results indicated that cDNA arrays were more sensitive than oligo arrays in detecting expression signals, including signals from the lower abundant genes such as receptors. Secondly, less variability between the replicates for cDNA arrays was found when compared with oligo arrays, indicating a more precise measurement of differential gene expression by cDNA arrays than by oligo arrays. Thus, our results indicated that cDNA arrays provided a more consistent estimate of fold change, and a greater sensitivity in detecting gene expression signals. Subsequently, fold changes even as low as 1.3 as indicated by the cDNA arrays, could be confirmed using RT-PCR. Considering this, we expected a better estimation of the number of significantly changed genes by the cDNA array experiments than by the oligo experiments. However, more regulated genes were found by oligo arrays than by cDNA arrays, which could be an indication of a higher rate of false positive calls indicated by the oligo arrays (chapter 6). It appeared that the overlap of significant differentially expressed genes detected by cDNA and oligo arrays was rather low, only ~32% (chapter 6). However, the results were based on two replicates and an increase in the number of replicates might increase the overlapping number of genes 12.

The differences between oligo arrays and cDNA arrays may be due to the differences in length of the used probes attached to the array surface. For the oligo arrays a probeset of 20 oligos (~25 mer) were present designed over the whole gene, and for cDNA arrays long cDNAs (~500-~5000 basepairs) were used as probes. The use of short oligos in comparison with long cDNAs may result in less specific hybridization and reduced sensitivity 13. Recent developments have resulted in arrays containing pre-synthesized longer oligos (~50-100 mers) to counteract the disadvantages of the shorter oligos 14, and it would be interesting to compare the performance of these oligoarrays and cDNA arrays and see if the overlap of differentially expressed genes would improve. On the other hand, a major advantage of the use of oligos is that probes can be designed to represent the most unique part of a given transcript, making the detection of closely related genes and splice variants possible 15, which is a limitation when using long cDNAs. Therefore, cDNA arrays may be prone to cross-hybridization of homologous genes that may results in the masking of true differences 16.

An interesting outcome of our comparison study was that the overlapping significantly changed genes detected by the oligo and cDNA arrays were all in the medium to higher signal ranges, indicating that the detection of differences in gene expression were most reliable for the more abundant genes for both platforms.

Our results could be compared with only one recent study comparing cDNA with oligo arrays, indicating that there is a lack of comparative data using both global screening platforms 17. In this study, results from Affymetrix oligo arrays (U95Av2) and Incyte cDNA arrays (GEM V 2.0) were compared. They also found less regulated genes using cDNA arrays. Only one differentially expressed gene using cDNA arrays was found, while hundreds of differentially expressed genes were detected by oligo arrays. Thus, there was hardly any overlap of significantly changed genes when using different platforms. However, they did not calculate the cut-off values for significant genes on the basis of variability within replicates (which was large for oligo arrays in their experiments), but based the cut off values on arbitrary values (oligo arrays; ³ 3 FC, cDNA arrays ³ 1,7 FC), which made the results from the different platforms difficult to compare. This study is an example that the determination of differentially expressed genes is still often based on non-statistical heuristics, influencing differential expression judgments and judgments on the comparability of different techniques. For a good comparison of microarray datasets from different techniques, data should be treated in a similar statistical manner to give a reliable picture of the comparability of the different datasets.

1.3 Comparison SAGE and microarrays

A technical difference between SAGE and microarrays, is the large amount of input RNA (2.5-5 &#956;g poly A+ RNA) that is required for SAGE in comparison with microarrays (250 ng – 1 &#956;g). This can be a drawback for tissues or cell types isolated from complex heterogeneous tissues from which only small amounts of RNA can be obtained. For our studies this was not a problem, while sufficient amounts of RNA could be derived from liver. However, by using laser capture microdissection, it has become possible to analyze gene expression profiles of a specific cell population. A recent study used laser capture microdissection to analyze gene expression in macrophages from heterogeneous atherosclerotic lesions 18.

Other technical differences made the overlap between the differentially expressed genes according to SAGE and microarrays difficult to assess. Microarrays can only assess the differential expression of genes present on the array, while SAGE is not limited to ‘beforehand’ defined genes. This difference resulted in many genes detected by SAGE that were not present on the microarray. On the other hand, many SAGE tags did not match with a single gene, but with several genes, and these were excluded from the comparison 5. This correction resulted in several genes reliably detected by SAGE and microarrays of which some genes showed a similar direction of regulation (chapter 6). However, the SAGE data for overlapping genes often consisted of 1-5 tags representing the lower abundant genes, which made a reliable comparison of differential gene expression difficult. Sequencing of more tags would increase this reliability 6.

1.4 Concluding remarks

In our studies we used three widely used expression profiling platforms; SAGE, oligo- and cDNA microarrays. This allowed us to perform a thorough comparison between SAGE and microarrays, and the advantages and limitations of SAGE and microarrays were discussed. SAGE does not require prior knowledge of the sequences to be analyzed, and therefore using SAGE we obtained a good impression of the higher abundant genes expressed in liver. Moreover, using SAGE we were able to identify novel transcripts. However, SAGE performed at this scale lacked the sensitivity to detect lower abundant genes. Therefore, we extended our studies with cDNA and oligo microarrays. Comparison of the results derived from oligo and cDNA arrays showed that the cDNA arrays were more sensitive in detecting genes, including lower abundant signals of receptors, and showed less variability between replicates than the oligo arrays. On the basis of our results, we used cDNA arrays for further experiments.

In our study we were able to fully exploit and compare the data generated by SAGE, oligo- and cDNA arrays. Comparison of all three techniques led us to the conclusion that when judging gene expression data by one technique, one should be aware that observed expression differences might not always be true. This still necessitates the need for validation of gene expression by alternative techniques such as Northern blot analysis, in-situ hybridization and quantitative RT-PCR.

2. Biological Implications

2.1 Dietary gene regulation

The mechanisms of diet induced hyperlipidemia and atherosclerosis have been widely studied by delineating the role of candidate genes in transgenic and gene targeted mouse models. However, the events following high fat feeding represent a complex process that is only partly understood. Studies described in this thesis were aimed at delineating these events at the level of gene expression by performing genome wide expression analysis. First, the effect of genetically altering the lipid metabolism in mice was studied using SAGE (chapter 2) and cDNA arrays (chapter 3). Using SAGE, liver expression profiles of E3L mice were compared to B6 mice on standard chow diet. This study provided an accurate inventory of gene expression in the livers of B6 and E3L mice. Analysis of > 18,000 transcripts per strain revealed altered expression of 175 genes. Interestingly, clustering by function of differentially expressed genes revealed many lipid and inflammatory/acute phase (defense) response genes. To create a more complete overview of genes that were influenced by the genotype, gene expression profiles were compared of the E3L and the B6 mice on chow diet using cDNA arrays. Although the use of microarrays in this study precluded a direct comparison of the genes identified with SAGE, it is remarkable that the E3L mice (as compared to B6 mice on chow) showed differentially regulated genes involved in similar processes, i.e. lipid metabolism, detoxification and the defense response. The latter data were confirmed in two additional comparisons. The profile of the E-/- mice on chow diet and the profile of E3L mice fed a high fat, high cholesterol (HFC) diet as compared to B6 mice on the similar diet, showed also regulation of similar functional classes of genes (chapter 3). Thus the different modifications of APOE in both the E3L and E-/- mice resulted in similar pathways that were affected. Interestingly, the related pathways detected in these comparisons indicated a similar compensatory mechanism responding to the changes in li

Early studies already indicated the close role between hyperlipidemia and inflammation 19. The effect of a HFC diet was already studied in B6 mice. B6 mice fed the more severe atherogenic diet containing 0.5% cholate exhibited elevated LDL/VLDL and reduced HDL cholesterol levels 20. In B6 mice, this diet induced inflammatory and oxidative stress response genes such as serum amyloid A, monocyte chemotactic protein-1, colony stimulating factors and heme oxygenase 21. However, this study focused only on several beforehand defined markers of inflammation. With the recent development of microarrays it has become possible to study the role of many genes in a single experiment. A very recent study of Vergnes et al.22, examined liver gene expression profiles of B6 mice fed different high fat, high cholesterol diets for 3 weeks using Mu11K oligonucleotide arrays (~11,000 clones). They showed that the more severe HFC diet containing cholesterol (1.25%) and cholate (0.5%) induced widespread changes in hepatic gene expression levels with ~1300 regulated genes. In our studies we also found major effects on liver gene regulation induced by the HFC diet in B6 mice (~800 regulated genes). They found many genes associated with lipid metabolism and inflammatory-related processes such as immune/defense response genes and extracellular matrix proteins. Studying diets with only cholesterol or cholate, they found that inflammatory gene activation was dependent on the presence of cholesterol in the diet, whereas the extracellular matrix proteins were specifically induced by cholate. These results corresponded with our results; diet W also significantly affected lipid and inflammatory related processes, while diet N affected these pathways and many genes encoding for extracellular matrix proteins. Other biological processes regulated by diet N were the immune response, membrane proteins, cell communication and amino acid metabolism (chapter 3). The results described in this thesis confirmed and extended this work by also examining th

We further delineated this complex response, by sub-classifying the classes of genes involved in lipid metabolism and detoxification processes according to their function based on the KEGG database. These data were integrated with published data regarding gene regulation through nuclear receptors that become activated upon binding to cholesterol, bile acids, fatty acids or their metabolites. We found a major role for a whole variety of nuclear receptor target genes, with HFC diets affecting many nuclear receptors: PXR, CAR, LXR, PPAR-a/d, HNF-4a and SREBPs. The effects modulated by diet N were stronger than the effects modulated by diet W, with diet N affecting most of the nuclear receptor target genes. It is likely that these effects are partly due to the cholate present in diet N. Except for the direct regulation by bile acids of the FXR 23 and the detoxification receptors PXR/CAR 24, no other nuclear receptors were known to be directly regulated by bile acids in liver. The common regulation of many nuclear receptor target genes involved in lipid and detoxification processes as found in this study, suggested a defense mechanism involving many nuclear receptors to protect against the accumulation of toxic endogenous lipids and bile acids. Previous studies already indicated the effects of lipids on immune and inflammatory responses involving nuclear receptors such as the PPARs and the LXRs 25,26. However, findings described in chapter 3 also implicated roles for other nuclear receptors in this process. Intriguingly, several nuclear receptor target genes were commonly regulated during the different conditions, which may implicate that they are key genes in regulatory pathways underlying lipid metabolism. One of these genes Abcg5 was already shown to be the cause of the lipid disorder sitosterolemia 27. Intriguingly, more strongly regulated genes such as Cyp4a10 and Cyp4a14 were found which are not previously associated with lipid disorders. This makes these genes interesting candidate genes for further studies.

In our study we used livers of mice after several weeks of dietary feeding (late endpoints). Studying differential gene expression in microarray experiments under different dietary conditions and during different time points will indicate which genes and pathways are involved in early, medium and late response effects of dietary gene regulation and may reveal most interesting candidate genes. A nice example was the study of Maxwell et al.28, in which they studied temporal patterns of gene expression at 1, 2, 4 and 7 days after feeding a cholesterol rich (0.5%) diet. Using quantitative RT-PCR, they already showed a down-regulation of the cholesterol responsive genes HMG-CoA reductase, HMG CoA synthetase and squalene epoxidase after 1 day up till day 7. These genes were also found downregulated in our studies. Moreover, the inflammatory marker SAA 3 was not upregulated until day 4 of cholesterol feeding. It would be of interest to perform these time course studies in liver on a genome wide scale to determine the spatial-temporal order of events following high fat feeding.

A recent study indicated the Vitamin D receptor as an intestinal bile acid sensor for lithocholic acid 29. In addition, high fat feeding has also major effects in the intestine, and results in an increase in the excretion of fecal bile acids to protect against toxicity. Therefore, it would be of interest to study the effect of high-fat feeding in the intestine to see which pathways are particularly affected. Besides liver and intestine, adipose, muscle and heart tissue are also involved in lipid homeostasis. Thus, it would be of interest to study the effect of high fat feeding in these tissues at defined time points to see which pathways become deregulated. Besides the effects on lipid metabolism, HFC diets were already reported to have an effect on oxidative stress, inflammation and on cell signaling/adhesion in rat skeletal muscle 30. Moreover, oligo arrays were used to assess gene expression in adipose tissue of male B6 given 8 weeks high fat diet. These mice were used as a model for diet-induced obesity. Interestingly, they found similar deregulated pathways as we found in liver namely lipid metabolism, detoxification processes and structural components of the cytoskeleton 31.

On the basis of expression profiling we already obtained a clear impression which pathways are associated with hyperlipidemia. In addition, metabolic studies should be performed to integrate these expression profiles with physiological processes that should give indications of the key affected pathways. Moreover, expression data should be integrated with QTLs associated with dyslipidemic phenotypes from mouse and human studies (Fig.1). These studies will facilitate the identification of novel candidate genes for hyperlipidemia.

2.2 Cafestol

Cafestol is a diterpene present in coffee beans and unfiltered coffee brews. Cafestol potently raises serum levels of LDL and VLDL in humans 32. Whether cafestol is ingested by coffee consumers depends on the brewing method: cafestol is retained by paper filters and therefore paper-filtered coffee does not effect cholesterol levels. However, unfiltered coffee brews such as French press coffee, cafetiere coffee, Scandinavian boiled coffee, and to a lesser extend espresso do contain cafestol and raise cholesterol levels. High intakes of unfiltered boiled coffee contributed to the high levels of cholesterol and high rates of coronary heart disease in Scandinavia, and a switch from boiled to paper-filtered coffee was thought to be responsible for 30% of the 1 mmol/liter fall in cholesterol in Finland in the period 1972-1992 33. Although several studies focused on the regulatory mechanism underlying the effect of cafestol on lipid metabolism 34-38, the molecular mechanism remained to be elucidated in more detail. Knowledge of novel genes and pathways affected by cafestol, may identify novel targets controlling serum lipoprotein levels through diet.

A study described in this thesis was aimed at elucidating the molecular mechanism of genes and pathways by which cafestol affects lipid metabolism (chapter 4). To this end E3L mice were used because they show a similar rise in serum cholesterol as humans do in response to cafestol 34. To elucidate the molecular mechanism of genes and pathways by which cafestol affect lipid metabolism, we performed genome-wide expression profiling using livers of cafestol fed (0.04% wt/wt) E3L mice. In microarray experiments livers of four individual cafestol fed mice exhibiting elevated cholesterol levels in response to cafestol were examined, and hybridized against a pool of control mice on GEMs 2.03 cDNA microarrays. A total of 648 transcripts were identified whose expression changed (p<1´10-6) in at least one mouse as compared to the control group. These genes were then subjected to hierarchical clustering and visualized 39. This cluster analysis revealed 2 large clusters of genes showing an overall similar direction of up- or downregulation. To gain further insight in which genes were involved in lipid metabolism, genes of these 2 clusters were functional annotated based on the Gene Ontology database 40, revealing a total of 42 lipid genes. By identifying all lipid genes present on the chip in a similar manner, we were able to determine the expected number of regulated lipid genes. It appeared that the observed number of lipid genes was significantly higher (~ 3 fold) than expected, showing that cafestol severely affected the lipid pathway. An interesting observation was that one of the strongest regulated genes in all 4 mice was Cyp7a1, a key enzyme involved in the conversion of cholesterol to bile acids in the liver. Previous studies already showed that cafestol suppressed bile-acid synthesis in cultured rat hepatocytes and E3L mice through inhibition of cyp7a1 34,35. The downregulation of Cyp7a1 in cafestol fed mice as found in this study was in line with these findings, confirming that cafestol inhibits bile acid synth

Further characterization of the differentially expressed lipid genes showed that several of these genes were involved in cholesterol, bile acid, and steroid metabolism. A unique observation was that the majority of these cholesterol responsive genes were under the direct regulation of the nuclear receptor FXR. These specific nuclear receptor target genes were Cyp7a1, Pltp and ApoA-I 23,42,43. Also a very recent study indicated an FXR response element in the promoter of ApoA-V 44. In line with this observation, we found that cafestol suppressed ApoAV.

Conversion of cholesterol into bile acids is tightly regulated. High levels of bile acids in the liver cause a decrease in bile acid synthesis that is mediated by the FXR. Activation of FXR by bile acids results in the regulation of FXR target genes such as Cyp7a1, Pltp, ApoA-I and ApoA-V 23,42-44. Cafestol also affected these FXR target genes, which may suggest that cafestol acts as a ligand for the FXR. Recent results, gave already strong indications that cafestol is a direct ligand for the FXR in vitro (Boekschoten M. et al., unpublished). They showed that cafestol induced activity of the bile salt export protein (Bsep) promoter in liver HEPG2 cells. Moreover, in a study using control and FXR knockout hepatocytes Bsep was induced in control hepatocytes but not in the FXR knockout hepatocytes. Further, analysis of the protease digestion patterns of in vitro translated FXR incubated with cafestol or chenodeoxycholic acid (a strong biological ligand for FXR) showed that cafestol directly binds to the FXR. Additionally, it would interesting to know the binding affinity of cafestol with FXR. However, these data should be confirmed in an in vivo experiment using cafestol fed FXR knockout and control mice. The effects observed on lipid parameters and FXR target genes as seen in control mice should not be observed in the FXR knockout mice. In summary, these experiments gave strong indications of the mechanism of action of cafestol through the FXR. Moreover, it also showed that the pathway from cholesterol to bile acids could be an important target in diet- and drug mediated attempts to affect serum cholesterol levels and atherosclerosis. Recently it was shown that other compounds can also activate FXR namely guggulsterone. Guggulsterone is a plant sterol derived from the guggul tree that lowers LDL in humans. Moreover, guggulsterone treatment decreased hepatic cholesterol in control mice fed a high cholesterol diet but was not effective in FXR knockout mice 45.

Cafestol not only affected genes involved in lipid metabolism but showed the strongest effects on detoxification processes (chapter 4). The effect on GSTs were reported by previous studies 36,37, however we identified many more detoxification genes (chapter 4). Intriguingly we identified several target genes for the nuclear receptors PXR/CAR such as several GSTs and Cyp2a 46. PXR is a xenobiotics sensor that regulates enzymes in detoxification processes. Activation of PXR by bile acids, steroids and antibiotics results in detoxification of xenobiotic compounds. Interestingly, recent studies indicated that Cyp7a1 may also be a target gene for the PXR 47,48. In vitro and in vivo studies should be performed to see if cafestol also targets the PXR.

2.3 Concluding remarks

This work examined the effect of dietary feeding at the level of gene expression in liver in different hyperlipidemic mouse models. Using SAGE and microarrays we identified many differentially expressed genes. Besides lipid metabolism, the defense response and detoxification processes were strongly affected by diet and genotype. Moreover, we found indications that many nuclear receptors played instrumental roles in these processes. Several of these genes were commonly regulated during the different conditions, making them interesting candidates for further studies. These data further strengthened the close link between hyperlipidemia and inflammatory processes.

The expression profiling studies investigating the effects of cafestol were the basis for further studies that clarified the mechanism of action of cafestol through the nuclear receptor FXR. These studies showed that the pathway from cholesterol to bile acids could be an important target in diet- and drug mediated attempts to affect serum cholesterol levels and atherosclerosis. Moreover, many differentially expressed genes described in this work were not previously linked with hyperlipidemia and atherosclerosis. They represented a whole spectrum of metabolic processes, were novel or their biological role still remains unknown. In conclusion, the datasets derived from these genome-wide expression profiling studies will provide a detailed basis for future studies to obtain more insight into the basic molecular mechanisms modulated by perturbations in lipid metabolism.

References

1. Consortium MGS. Initial sequencing and comparative analysis of the mouse genome. Nature 2002;420(6915):520-62.

2. Yamashita T, Hashimoto S, Kaneko S, Nagai S, Toyoda N, Suzuki T, Kobayashi K, Matsushima K. Comprehensive gene expression profile of a normal human liver. Biochem Biophys Res Commun 2000;269(1):110-6.

3. van den Berg A, van der Leij J, Poppema S. Serial analysis of gene expression: rapid RT-PCR analysis of unknown SAGE tags. Nucleic Acids Res. 1999;27(17):e17.

4. Chen JJ, Rowley JD, Wang SM. Generation of longer cDNA fragments from serial analysis of gene expression tags for gene identification. Proc Natl Acad Sci U S A 2000;97(1):349-53.

5. Stollberg J, Urschitz J, Urban Z, Boyd CD. A quantitative evaluation of SAGE. Genome Res. 2000;10(8):1241-8.

6. Trendelenburg G, Prass K, Priller J, Kapinya K, Polley A, Muselmann C, Ruscher K, Kannbley U, Schmitt AO, Castell S and others. Serial analysis of gene expression identifies metallothionein-II as major neuroprotective gene in mouse focal cerebral ischemia. J Neurosci 2002;22(14):5879-88.

7. Draghici S, Kuklin A, Hoff B, Shams S. Experimental design, analysis of variance and slide quality assessment in gene expression arrays. Curr Opin Drug Discov Devel 2001;4(3):332-7.

8. Lee Y, Sultana R, Pertea G, Cho J, Karamycheva S, Tsai J, Parvizi B, Cheung F, Antonescu V, White J and others. Cross-referencing eukaryotic genomes: TIGR Orthologous Gene Alignments (TOGA). Genome Res. 2002;12(3):493-502.

9. Hoffmann R, Seidl T, Dugas M. Profound effect of normalization on detection of differentially expressed genes in oligonucleotide microarray data analysis. Genome Biol. 2002;3(7):RESEARCH0033.

10. Nadon R, Shi P, Skandalis A, Woody E, Hubschle H, Susko E, Rghei N, Ramm P. Statistical Inference Methods for Gene Expression Arrays. Bittner MM, editor. In: Microarrays: Optical Technologies and Informatics Proceedings of SPIE. Volume 4266; 2001. p 46-55.

11. Takahashi M, Tsuboyama-Kasaoka N, Nakatani T, Ishii M, Tsutsumi S, Aburatani H, Ezaki O. Fish oil feeding alters liver gene expressions to defend against PPARalpha activation and ROS production. Am J Physiol Gastrointest Liver Physiol 2002;282(2):G338-48.

12. Pavlidis P, Li Q, Noble WS. The effect of replication on gene expression microarray experiments. Bioinformatics 2003;19(13):1620-1627.

13. Schulze A, Downward J. Navigating gene expression using microarrays--a technology review. Nat Cell Biol. 2001;3(8):E190-5.

14. Kane MD, Jatkoe TA, Stumpf CR, Lu J, Thomas JD, Madore SJ. Assessment of the sensitivity and specificity of oligonucleotide (50mer) microarrays. Nucleic Acids Res. 2000;28(22):4552-7.

15. Lockhart DJ, Dong H, Byrne MC, Follettie MT, Gallo MV, Chee MS, Mittmann M, Wang C, Kobayashi M, Horton H, Brown EL. Expression monitoring by hybridization to high-density oligonucleotide arrays. Nat Biotechnol 1996;14(13):1675-80.

16. Evertsz EM, Au-Young J, Ruvolo MV, Lim AC, Reynolds MA. Hybridization cross-reactivity within homologous gene families on glass cDNA microarrays. Biotechniques 2001;31(5):1182, 1184, 1186 passim.

17. Li J, Pankratz M, Johnson JA. Differential gene expression patterns revealed by oligonucleotide versus long cDNA arrays. Toxicol Sci 2002;69(2):383-90.

18. Trogan E, Choudhury RP, Dansky HM, Rong JX, Breslow JL, Fisher EA. Laser capture microdissection analysis of gene expression in macrophages from atherosclerotic lesions of apolipoprotein E-deficient mice. Proc Natl Acad Sci U S A 2002;99(4):2234-9.

19. Slim RM, Toborek M, Watkins BA, Boissonneault GA, Hennig B. Susceptibility to hepatic oxidative stress in rabbits fed different animal and plant fats. J Am Coll Nutr. 1996;15(6):289-94.

20. Paigen B, Morrow A, Brandon C, Mitchell D, Holmes P. Variation in susceptibility to atherosclerosis among inbred strains of mice. Atherosclerosis. 1985;57(1):65-73.

21. Liao F, Andalibi A, deBeer FC, Fogelman AM, Lusis AJ. Genetic control of inflammatory gene induction and NF-kappa B-like transcription factor activation in response to an atherogenic diet in mice. J Clin Invest. 1993;91(6):2572-9.

22. Vergnes L, Phan J, Strauss M, Tafuri S, Reue K. Cholesterol and cholate components of an atherogenic diet induce distinct stages of hepatic inflammatory gene expression. J Biol Chem. 2003;278(44):42774-84.

23. Makishima M, Okamoto AY, Repa JJ, Tu H, Learned RM, Luk A, Hull MV, Lustig KD, Mangelsdorf DJ, Shan B. Identification of a nuclear receptor for bile acids. Science 1999;284(5418):1362-5.

24. Kliewer SA, Willson TM. Regulation of xenobiotic and bile acid metabolism by the nuclear pregnane X receptor. J Lipid Res 2002;43(3):359-64.

25. Bocher V, Chinetti G, Fruchart JC, Staels B. Role of the peroxisome proliferator-activated receptors (PPARS) in the regulation of lipids and inflammation control. J Soc Biol. 2002;196(1):47-52.

26. Joseph SB, Castrillo A, Laffitte BA, Mangelsdorf DJ, Tontonoz P. Reciprocal regulation of inflammation and lipid metabolism by liver X receptors. Nat Med. 2003;9(2):213-9.

27. Berge KE, Tian H, Graf GA, Yu L, Grishin NV, Schultz J, Kwiterovich P, Shan B, Barnes R, Hobbs HH. Accumulation of dietary cholesterol in sitosterolemia caused by mutations in adjacent ABC transporters. Science 2000;290(5497):1771-5.

28. Maxwell KN, Soccio RE, Duncan EM, Sehayek E, Breslow JL. Novel putative SREBP and LXR target genes identified by microarray analysis in liver of cholesterol-fed mice. J Lipid Res. 2003;In press.

29. Makishima M, Lu TT, Xie W, Whitfield GK, Domoto H, Evans RM, Haussler MR, Mangelsdorf DJ. Vitamin D receptor as an intestinal bile acid sensor. Science 2002;296(5571):1313-6.

30. Sreekumar R, Unnikrishnan J, Fu A, Nygren J, Short KR, Schimke J, Barazzoni R, Nair KS. Effects of caloric restriction on mitochondrial function and gene transcripts in rat muscle. Am J Physiol Endocrinol Metab. 2002;83(1):E38-43.

31. Moraes R, Blondet A, Birkenkamp-Demtroeder K, Tirard J, Orntoft T, Gertler A, Durand P, Naville D, Begeot M. Study of the alteration of gene expression in adipose tissue of diet-induced obesity (DIO) mice by microarray and RT-PCR analyses. Endocrinology 2003;144(11):4773-82.

32. Heckers H, Gobel U, Kleppel U. End of the coffee mystery: diterpene alcohols raise serum low-density lipoprotein cholesterol and triglyceride levels. J Intern Med 1994;235(2):192-3.

33. Pietinen P, Vartiainen E, Seppanen R, Aro A, Puska P. Changes in diet in Finland from 1972 to 1992: impact on coronary heart disease risk. Prev Med. 1996;25(3):243-50.

34. Post SM, de Roos B, Vermeulen M, Afman L, Jong MC, Dahlmans VE, Havekes LM, Stellaard F, Katan MB, Princen HM. Cafestol increases serum cholesterol levels in apolipoprotein E*3- Leiden transgenic mice by suppression of bile acid synthesis. Arterioscler Thromb Vasc Biol 2000;20(6):1551-6.

35. Post SM, de Wit EC, Princen HM. Cafestol, the cholesterol-raising factor in boiled coffee, suppresses bile acid synthesis by downregulation of cholesterol 7 alpha- hydroxylase and sterol 27-hydroxylase in rat hepatocytes. Arterioscler Thromb Vasc Biol 1997;17(11):3064-70.

36. Lam LK, Sparnins VL, Wattenberg LW. Isolation and identification of kahweol palmitate and cafestol palmitate as active constituents of green coffee beans that enhance glutathione S-transferase activity in the mouse. Cancer Res 1982;42(4):1193-8.

37. Lam LK, Sparnins VL, Wattenberg LW. Effects of derivatives of kahweol and cafestol on the activity of glutathione S-transferase in mice. J Med Chem 1987;30(8):1399-403.

38. van Tol A, Urgert R, de Jong-Caesar R, van Gent T, Scheek LM, de Roos B, Katan MB. The cholesterol-raising diterpenes from coffee beans increase serum lipid transfer protein activity levels in humans. Atherosclerosis 1997;132(2):251-4.

39. Ahlberg C. Visual exploration of HTS databases: bridging the gap between chemistry and biology. Drug Discov Today 1999;4(8):370-376.

40. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP, Dolinski K, Dwight SS, Eppig JT and others. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nature Genetics 2000;25(1):25-29.

41. Pullinger CR, Eng C, Salen G, Shefer S, Batta AK, Erickson SK, Verhagen A, Rivera CR, Mulvihill SJ, Malloy MJ and others. Human cholesterol 7alpha-hydroxylase (CYP7A1) deficiency has a hypercholesterolemic phenotype. J Clin Invest 2002;110(1):109-17.

42. Urizar NL, Dowhan DH, Moore DD. The farnesoid X-activated receptor mediates bile acid activation of phospholipid transfer protein gene expression. J Biol Chem 2000;275(50):39313-7.

43. Claudel T, Sturm E, Duez H, Torra IP, Sirvent A, Kosykh V, Fruchart JC, Dallongeville J, Hum DW, Kuipers F and others. Bile acid-activated nuclear receptor FXR suppresses apolipoprotein A-I transcription via a negative FXR response element. J Clin Invest 2002;109(7):961-71.

44. Prieur X, Coste H, Rodriguez JC. The human apolipoprotein AV gene is regulated by peroxisome proliferator-activated receptor-alpha and contains a novel farnesoid X-activated receptor response element. J Biol Chem. 2003;278(28):25468-80.

45. Urizar NL, Liverman AB, Dodds DT, Silva FV, Ordentlich P, Yan Y, Gonzalez FJ, Heyman RA, Mangelsdorf DJ, Moore DD. A natural product that lowers cholesterol as an antagonist ligand for FXR. Science 2002;296(5573):1703-6.

46. Cai Y, Konishi T, Han G, Campwala KH, French SW, Wan YJ. The role of hepatocyte RXR alpha in xenobiotic-sensing nuclear receptor-mediated pathways. Eur J Pharm Sci 2002;15(1):89-96.

47. Staudinger JL, Goodwin B, Jones SA, Hawkins-Brown D, MacKenzie KI, LaTour A, Liu Y, Klaassen CD, Brown KK, Reinhard J and others. The nuclear receptor PXR is a lithocholic acid sensor that protects against liver toxicity. Proc Natl Acad Sci U S A. 2001;98(6):3369-74.

48. Goodwin B, Gauthier KC, Umetani M, Watson MA, Lochansky MI, Collins JL, Leitersdorf E, Mangelsdorf DJ, Kliewer SA, Repa JJ. Identification of bile acid precursors as endogenous ligands for the nuclear xenobiotic pregnane X receptor. Proc Natl Acad Sci U S A 2003;223(8):223-8.