Cystic Fibrosis and the Use of Pharmacogenomics to Determine Surrogate Endpoints for Drug Discovery

Cystic Fibrosis and the Use ofPharmacogenomics to DetermineSurrogate Endpoints for Drug DiscoveryOfer Eidelman, Jian Zhang, Meera Srivastava and Harvey B. PollardDepartment of Anatomy, Physiology and Genetics, and Institute for Molecular Medicine, Uniformed Services UniversitySchool of Medicine, USUHS, Bethesda, Maryland, USA

ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2231. The Problem of Cystic Fibrosis (CF) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2242. The Potential for Application of Pharmacogenomics to CF Drug Discovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2243. Mechanisms of CF Pathogenesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

3.1 Inefficient Trafficking of the Mutant CF Transmembrane Conductance Regulator (CFTR) . . . . . . . . . . . . . . . . . 2253.2 Intrinsic Propensity for Inflammation and Infection in the CF Lung . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2263.3 IL-8 Levels in CF Lung Epithelial Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226

4. Mechanisms of CPX Therapeutic Effects in CF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2265. Other Potential CF Targets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

5.1 CFTR-Interacting Proteins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2275.2 The NFκB Signaling Pathway Regulates IL-8 Expression in Epithelial Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

6. CF Pharmacogenomics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2316.1 Pharmacogenomic Data Can be Used to Mine Out Drug- or Gene-Dependent Signaling Pathways . . . . . . . . . . 2316.2 Application of GRASP and GENESAVER Algorithms to Pharmacogenomic Analysis of Cystic Fibrosis . . . . . . . . . . . 232

6.2.1 The GRASP Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2326.2.2 The GENESAVER Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234

7. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234

Abstract Cystic fibrosis (CF) is caused by a mutation in the CFTR gene, encoding a chloride channel. For themost common mutation, ΔF508, the basis of the deficit is the failure of the mutant CFTR channel proteinto traffic properly to the apical plasma membrane of the affected epithelial cell. The trafficking failureresults in loss of the cyclic adenosine monophosphate (cAMP)-activated chloride channel function of theCFTR protein in the plasma membrane. The lung is the principal site affecting patient morbidity andmortality in CF. The main reason is that the CF airway epithelial cells also secrete high levels of theproinflammatory cytokine interleukin (IL)-8, resulting in massive cellular inflammation, infection, tissuedamage and lung destruction. The relationship between the trafficking defect, the loss of chloride channelactivity, and inflammation is not known. However, gene therapy of CF lung epithelial cells with thewild-type CFTR gene can repair the chloride channel defect, as well as suppress the intrinsic hypersecretionof IL-8. Repair of both defective channels and high IL-8 secretion can also be effected by treatment withthe candidate CF drug CPX, which is in clinical trials in CF patients. CPX acts by binding to the mutantCFTR protein, and helps the protein to mature and gain access to the plasma membrane. CPX alsosuppresses the synthesis and secretion of IL-8 from CF epithelial cells, presumably by virtue of its repairof the trafficking defect of mutant CFTR. To guide pharmacogenomic experiments we have thereforehypothesized that the genomic signature of CF epithelial cells treated with CPX should resemble thesignature of the same cells repaired by gene therapy. We have developed two algorithms for identifyinggenes modified by repair of CFTR defects. The GRASP algorithm uses a statistical test to identify the mostprofoundly changing genes. The GENESAVER algorithm allows us to identify those genes whose patternof expression changes in-phase or out-of-phase with IL-8 secretion by CF cells. For the latter algorithm we

GENOMICS IN DRUG DEVELOPMENT Am J Pharmacogenomics 2001; 1 (3): 223-2381175-2203/01/0003-0223/$22.00/0

© Adis International Limited. All rights reserved.

modified IL-8 secretion from CF cells by treatment with wild-type CFTR, with CPX, or by exposure tobacteria. The results have supported the hypothesis, and have provided a basis for considering the commonpharmacogenomic expression signature as a surrogate endpoint for CF drug discovery. Significantly, thenature of the hypothesis, as well as the algorithm developed for this study, can be easily applied to phar-macogenomic studies with other goals.

1. The Problem of Cystic Fibrosis (CF)

Cystic fibrosis (CF) is the most common inherited autosomalrecessive lethal disease in the United States.[1] Approximately5% of the population carries one mutant cystic fibrosis trans-membrane conductance regulator (CFTR) gene[2-4] and the dis-ease occurs in a frequency of 1 in 2500 live births. Statistically,death occurs in the majority of patients by age 28. Until veryrecently, few afflicted patients survived past early childhood,death often occurring at birth due to intestinal blockage by me-conium ileus. Those who survived later died from nutritionaldeficiencies.[1] The principal reason was that the pancreaticducts became blocked by mucins, thereby preventing pancre-atic enzymes from gaining access to the small intestine. Nutri-tional deficiencies also ensued from blockage of the bile duct,which resulted in inability to recover dietary fats and vitamins.Finally, blockage of the respiratory tract by thick mucous se-cretions led to death from bacterial infections and gross loss ofability to oxygenate the blood.

Even though CF is a ‘single gene’ disease, it seems to havepleiotropic physiological consequences. These consequencesinclude effects on a multitude of different cellular functionsincluding the transmembrane transport of Cl- and Na+, Na/Hexchange, mucin secretion, and abnormal inflammatory signal-ing. CF therapy has traditionally been geared towards treatingthe symptoms of the disease. Presently, surgery is often used totreat meconium ileus, and pancreatic enzyme supplements haveprovided an effective mechanism for limiting digestive defects.In addition, physical therapy, DNAse I,[5] ibuprofen,[6] and an-tibiotic therapies have been developed to help treat the obstruc-tive disease in the lung. Whole or partial lung transplants havealso been helpful, although there is a limited transplantationsource. Nonetheless, at the present time the respiratory diffi-culties and ensuing complications of inflammation and lunginfection are directly responsible for the eventual death of over90% of patients with CF at an average age of 28 years.

The principal basis of CF pathology in the lung seems to bean intrinsically activated proinflammatory process, which isindependent of subsequently acquired infections. The CF air-way shows a great increase in proinflammatory signaling mol-ecules, including interleukin (IL)-8, tumor necrosis factor(TNF)-α, and others (see section 3.2 and references 7-10 for

details). These molecules appear to come initially from the lungepithelial cells, but later come from other cellular constituentsof the inflammatory milieu. The molecular basis of this intrin-sic proinflammatory propensity is not presently known. Thereare also a number of other mutation-dependent defects in lungfunction,[1] including sodium conductance, mucociliary clear-ance and bacterial killing which we will not address here. How-ever, available data make it clear that reintroduction of a wild-type CFTR molecule by gene therapy is sufficient to abrogatethe defective CF intracellular signals, whatever they are (seereferences 11-14 and section 5 for details).

2. The Potential for Application ofPharmacogenomics to CF Drug Discovery

In an effort to use the principles of pharmacogenomics todiscover pharmaceuticals that might be useful in treating CF,we have hypothesized that an efficient CF pharmaceuticalshould cause the functional genomics of the CF cell to resemblethat of the wild-type CFTR-repaired CF cell.[15,16] This phar-macogenomic pattern of repair would thus constitute a surro-gate endpoint for drug discovery. It should be emphasized thatsuch an anticipated result has had plenty of predecessors in thegeneral field of pharmacogenomics. Both Bailey et al.[17] andFerrar[18] have predicted that the genetic consequences of ef-fective gene or drug therapy should be assessed at the cellularlevel in terms of specific patterns of global gene expression.Graever et al.[19] have suggested that at the very simplest level,the expectation is that specific patterns of gene expression willbe discerned that could be employed as biologically relevantsurrogate endpoints when searching for more effective drugsor gene targets.

In its simplest manifestation, this approach should providea composition list for a ‘gene-chip’ containing the signaturegenes, which, in their ensemble, would be capable of indicatingthe repair of CF defects by a candidate CF drug. One tacticaladvantage to this approach to CF pharmacogenomics is that wealready possess several ‘lead’ compounds, including the aden-osine A1 receptor antagonists CPX, DAX and certain otherxanthines.[20-22] CPX is currently in clinical trials, and is there-fore available for proof-of-concept experiments. Our initialstudies have identified genes from the TNFα receptor (TNFαR)/

224 Eidelman et al.

© Adis International Limited. All rights reserved. Am J Pharmacogenomics 2001; 1 (3)

NFκB pathway, among others, as important for the proinflamma-tory condition of CF.[16] Yet, not all the delineated genes have anobvious relationship to either CFTR function or to proinflamma-tory processes.

In this review, we shall first discuss the molecular details un-derlying the function of CFTR in airway epithelial cells, and theactions of CPX on the function of mutant CFTR. We presume thatknowledge of such details may help in interpreting the identifiedgenes and in discovering new ones. In the last part of the review,we shall describe two algorithms, gene ratio analysis paradigm(GRASP) and gene space vector (GENESAVER), which wehave developed in order to mine out the potentially dysfunctionalgenes in a hypothesis-driven manner. We believe both algorithmsmay prove to be of general utility to the pharmacogenomic com-munity.

3. Mechanisms of CF Pathogenesis

3.1 Inefficient Trafficking of the Mutant CFTransmembrane Conductance Regulator (CTFR)

In order to understand how mutant CFTR might cause theintrinsically proinflammatory phenotype of cystic fibrosis, atten-tion should be focused on the exact function of CFTR, and what

changes occur to CFTR function when the gene is mutated.

Briefly, the CFTR gene product functions as a cyclic adenosine

monophosphate (cAMP)/protein kinase A (PKA)/adenosine tri-

phosphate (ATP)-activated chloride channel.[11,23-26] The CFTR

channel is normally located in the apical membrane of epithelial

cells.[27] The principal mutation responsible for cystic fibrosis isΔF508, which is found in approximately 90% of CF chromo-

somes.[4,12,28,29] CFTR is a member of the ATP-binding cassette

(ABC) transporter gene family, and this mutation is located

within CFTR’s first nucleotide binding fold domain (NBF-1).

This domain has been previously assumed to reside in the cytosol,

although current functional and crystallographic data suggest a

partial membrane location.[30,31] The mutation causes only mod-

est changes in chloride channel activity per se.[27,32,33] Rather, the

ΔF508 mutation causes the mutant CFTR to be retained in the

endoplasmic reticulum and rapidly degraded by a proteosomal

mechanism. Consequently, mutant CFTR fails to traffic properlyto the golgi and the plasma membrane.[34-37] The biochemical

evidence for this trafficking arrest is that the mutant CFTR re-

mains core-glycosylated, and fails to acquire complex N-linked

oligosaccharides typical of the medial golgi activity.[34,38-41] It

has therefore been presumed that this trafficking failure is in some

way responsible for the proinflammatory properties of the CF

205kDa

C

B

118kDa

wt ΔF508 0.1% DMSO 0.4 2.0 10 50

Controls CPX (μM)

Fig. 1. Western blot analysis of CPX-dependent CFTR expression in HEK cells expressing wild-type CFTR (band C; 180kDa) and [ΔF508]-CFTR (band B; 150kDa).Samples of cells, cultured as described by Srivastava et al. 1999,[42] were homogenized in a medium containing a cocktail of protease inhibitors and other reagents,[21]

and 50μg aliquots of protein run on 6% SDS-PAGE gels. The antibody against the CFTR C-terminal 4 residues (monoclonal from Genzyme, Boston, MA, USA) is usedto detect CFTR antigen in the gels, and the complexes are imaged by enhanced chemi-luminescence. kDa = kiloDaltons

Pharmacogenomics of Cystic Fibrosis 225


airway. As described in section 4, CPX is efficacious in correct-ing the trafficking defect of mutant CFTR.

3.2 Intrinsic Propensity for Inflammation and Infection inthe CF Lung

The CF lung has been described as microscopically normal atbirth, with subtle abnormalities in mucus secretion appearingvery early in life.[5] Bacterial infection and objective evidence ofinflammation occur at later times, with a clear temporal evolutionof different principal bacterial pathogens. For example, Staphy-lococcus aureus and Hemophilus influenzae take up residence inthe CF airway early, the mean age of positive culture being 12.4months.[43] By comparison, Pseudomonas aeruginosa infectionfollows at a substantially later time, the mean age of first positiveculture being 20.8 months. Persistent colonization by P. aerugin-osa characterizes the older CF patient, and profound, persistentcellular evidence of inflammation accompanies persistent infec-tion as the patient approaches the terminal phases of the disease.Nonetheless, inflammatory signals have been detected soon afterbirth,[44] suggesting that the proinflammatory status occurs evenbefore bacterial challenge.

The proinflammatory cellular phenotype of the CF lung ischaracterized by elevated levels of polymorphonuclear leuko-cytes, macrophages, monocytes, lymphocytes and eosinophils.Presumably these cells are attracted from the circulation into theairway by the high levels of interleukin-8 (IL-8) and other pro-inflammatory factors such as IL-1β, IL-6, leukotriene B4,RANTES, and TNFα, which are reported to mark the CF lumenalmilieu.[7,8] Among these factors, IL-8 ranks as the most prevalentand potent. IL-8 is an 8 kiloDalton (kD) chemokine protein whichis a principal chemotactic agent for neutrophils and T cells.[45]

This chemokine is of specific importance for cystic fibrosis be-cause it is profoundly elevated in bronchoalveolar lavage fluids,sputum, and serum from CF patients. [9,46,47,48] It had been con-sidered possible that high IL-8 levels might be secondary tochronic or persistent infections. However, both IL-8 message andprotein are elevated in bronchoalveolar lavage fluids from infantswith CF as early as 4 weeks of age.[44] Importantly, hypersecre-tion of IL-8 occurs prior to objective evidence of infection byviruses, fungi or common CF pathogenic bacteria.[44]

The concept of the generality of a proinflammatory state forCF epithelia is further manifest by the fact that fecal IL-8 levelsin children with CF are approximately 1000-fold elevated overthose of non-CF individuals.[49] Fecal IL-8 levels are inverselycorrelated with lung function (FEV1, forced expiratory volume inone second), and only to some extent with established Pseudom-onas infection. The origin of IL-8 in meconium (newborn feces)

may include swallowed lung secretions, as well as other GIsources. A recent study with bronchial biopsies from CF patientsundergoing lung transplant has demonstrated consistent up-reg-ulation of IL-8 expression in submucosal gland cells.[50] Thus,based on these clinical criteria, high IL-8 levels would appear tobe intrinsic to the CF lung.

3.3 IL-8 Levels in CF Lung Epithelial Cells

Consistently, airway epithelial cells isolated from CF patientssecrete more IL-8 than do cells cultured from patients withoutCF.[10,51,52] Interestingly, cells cultured from much higher in theairway, such as those from the nasal epithelium, do not show thisdisparity between control and CF patients.[53] In addition, CFrespiratory epithelial cells are hyper-responsive in terms of IL-8secretion to Pseudomonas cells and toxins,[52,54] or to a combina-tion of TNFα and INFγ.[55] CFTR levels in human lung are high-est in submucosal glands, and high levels of IL-8 messenger RNA(mRNA) and protein have been shown in these tissues from CFpatients, both in vitro and in vivo.[52] Other proinflammatorycytokines such as IL-1β and IL-6 were unaffected by the CF con-dition in this study.[52]

The high levels of IL-8 production by CF epithelial cells havebeen proposed to be due to retention of mutant CFTR in the endo-plasmic reticulum, which, by an unknown mechanism, activatesNFκB via activation of IκB.[52] Attention is drawn to the NFκBsystem because it is known that transcription of the IL-8 gene isactivated in normal epithelial cells when activated NFκB mi-grates from the cytosol to the nucleus, and binds to the IL-8 pro-motor. Relevantly, an adenovirus overexpressing IκBα has beenemployed to suppress IL-8 secretion both from the ΔF508 homo-zygous CF cell line CFTE, as well as the mouse IL-8 homologueMIP2 from mouse lung, when instilled simultaneously with aninfectious dose of Ps. aeruginosa.[14,56] These data therefore sup-port the concept that the IL-8 signaling pathway is central tocystic fibrosis pathology with regards to epithelial cell dysfunc-tion.

4. Mechanisms of CPX Therapeutic Effects in CF

CPX was first identified by its ability to activate the defectiveCl- conductance of mutant [ΔF508]-CFTR.[20-22,56] We have morerecently learned that CPX both promotes trafficking of mutantCFTR[57] and suppresses IL-8 secretion from CF cells.[16] Wehave also found that CPX binds to the nucleotide binding fold(NBF-1) domain of CFTR in the vicinity of the F508 residue.[58]

Thus, there is a possible connection between reduction of theproinflammatory state and chemical correction of the traffickingdefect of mutant CFTR. The effect of CPX on trafficking of

226 Eidelman et al.


[ΔF508]-CFTR has been studied in recombinant human andmouse cells. As shown in figure 1, Western blot analysis ofHEK293 cells indicates that CPX induces proper trafficking ofthe mutant protein. Properly trafficked wild-type CFTR migratesin an SDS gel as a protein of approximately 180kD, termed the‘C’ band. Mutant [ΔF508]-CFTR migrates as the core-glyco-sylated ‘B’ band (approximately 150kD). However, CPX treat-ment of the cells expressing [ΔF508]-CFTR reveals a dose-de-pendent increase in expression of the higher molecular weight Cband. The CPX effect on CFTR channel function is biphasic,[20]

and an equivalent biphasic dose-response curve is apparent fromthe reduced trafficking in the presence of 50μM CPX.

The same kind of result has been observed in recombinantmouse L cells expressing either wild-type or [ΔF508]-CFTR. Asshown in figure 2, immunoreactive CFTR increases as a functionof CPX concentration, approaching wild-type CFTR control lev-els at 20μM CPX. The lower set of superimposed images showcombined phase contrast and fluorescence data. These data indi-cate that the immunoreactivity is extranuclear in both native andCPX-supplemented cells expressing mutant CFTR. Laser scan-ning cytometry data for the same system are shown in figure 3.Of about 2000 counted cells in random fields, most CPX-treated

[ΔF508]-CFTR cells appear to be responding by making moreCFTR antigen than untreated mutant cells. CPX also somewhatincreases the levels of CFTR in cells expressing wild-type CFTR.On the basis of circular dichroism and fluorescence studies, CPXappears to cause the mutant ΔF508-NBF-1 domain to fold morelike the wild-type domain, and thereby avoid lipid- or protein-based surveillance by chaperones. A similar conclusion has pre-viously been proposed for the low affinity xanthine IBMX.[59]

Thus, the observation that drug therapy with CPX and gene ther-apy with wild-type CFTR have similar genomic consequences inauthentic CF cells[16] may be explained by the fact that the druginduces an increase in functional [ΔF508]-CFTR mass.

5. Other Potential CF Targets

5.1 CFTR-Interacting Proteins

On the basis of the preceding discussion, it is apparent thatthere is a definite connection between mutant CFTR and the pul-monary proinflammatory phenotype of CF. However, the molec-ular basis of this connection remains ill-defined. Returning to firstprinciples, we might anticipate some insight from consideration

Untreated 0.2μM CPX 2μM CPX 20μM CPX Wild-typeCFTRcontrolΔF508-CFTR

Fig. 2. Effect of CPX on [ΔF508]-CFTR expression in recombinant mouse L-cells. Mouse L-cells, expressing either wild-type CFTR or [ΔF508]-CFTR, were exposed toCPX or carrier 0.1% DMSO for 2 days in culture, and imaged by conventional fluorescence microscopy (upper tier of images) or superimposed on phase images (lowertier of images). The antibody used to detect the CFTR antigen is the same Genzyme product described in figure 1. In the presence of 20μM CPX, the levels of expressed[ΔF508]-CFTR are similar in amount and distribution to L-cells expressing wild-type CFTR.



of other molecules with which CFTR has been shown to interact.For example, in addition to intrinsic chloride channel functions,wild-type CFTR appears to interact either directly or indirectlywith other apical membrane ion channels such as the epithelialsodium channel (ENaC)[60,61] and the outward rectifying chloridechannel.[62-66]

Other proteins have been identified recently which interact func-tionally with the N- and C-terminal domains of CFTR. For exam-ple, syntaxin 1 interacts with the amino-terminal tail of CFTR andsuppresses CFTR chloride channel activity.[67,68] Alternatively, anintact PDZ-binding domain in the C-terminal region of CFTR isrequired for export from the endoplasmic reticulum.[69,70] Na+/H+

exchange regulatory factor (NHERF) also binds to the C-terminaldomain through multiple PDZ domains on the NHERF mole-cule.[71] The NHERF interaction may be important physiologi-

cally since we have previously shown that CF cells have a defect

in Na+/H+ exchange that can be repaired by either wild-type CFTR

or the trafficking-rescue drug CPX.[72] The PDZ-domain-contain-

ing protein E3KARP has also been shown to crosslink EZRIN to

the C-terminal domain of CFTR.[73] The EBP50 (EZRIN Bind-

ing phosphoprotein 50) also crosslinks EZRIN to CFTR.[74-77]

Finally, Lamprecht et al.[78] have reported that NHERF and E3KARP

bind both to NHE3 (the apical Na+/H+ exchanger) and to EZRIN.

From this analysis, it follows that there could be an important

functional interaction between CFTR and the Na+/H+ exchanger,

and that EZRIN plays an important role in the physical interaction

between the two molecules. This interests us in the present context of

pharmacogenomics because our work indicates that EZRIN is one

of the genes whose message is significantly regulated by both the

1 2

0

2

5

8

11

14

Control

1

0

17

35

52

70

88

5μM CPX

2

2

0

10

21

32

43

54

1

104 105 106 107

2

0

32

64

66

128

161

1

104 105 106 107

FITC integral

Cel

l cou

nt p

er c

hann

el

WT

ΔF508

Fig. 3. Cell sorting analysis by laser scanning cytometry of immunoreactive [ΔF508]-CFTR in recombinant L-cells treated with CPX. Cells cultured in each of 4 conditionswere analyzed in terms of relative expression of either wild-type CFTR or immunoreactive [ΔF508]-CFTR following treatment with 5μM CPX for 2 days. Among cellsexpressing wild-type CFTR (WT/Control), only a small proportion of the total cells express low levels of CFTR (black peak), while most are expressing high levels ofCFTR (blue peak). Upon treatment with CPX (WT/5μM CPX), even this small proportion of underexpressing cells is converted to high levels of CFTR expression. Incells expressing [ΔF508] CFTR (ΔF/Control) most cells are expressing low levels of antigen, while upon treatment with CPX (ΔF/5μM CPX), the expression profilestrongly resembles the wild-type CFTR case. At least 2000 total cells were counted in each of the 4 conditions.

228 Eidelman et al.


proinflammatory state of the CF cell and the trafficking state ofCFTR (see table I).

5.2 The NFκB Signaling Pathway Regulates IL-8 Expressionin Epithelial Cells

The NFκB signaling pathway is frequently mentioned from theperspective of IL-8 expression.[79] However, the number of actual

or potential components in this pathway is enormous. Thus, usingconventional experimental approaches, the complete set of com-ponents has been difficult to consider in its entirety. However,from the viewpoint of modern functional genomics it is possibleto consider the complete known set, because cDNA microarrayshave the capacity to yield data on the expression of each of theactual and potential genes. However, no matter how selective thedata mining algorithm might be, it is a frequent experience in

Table I. Genes selected by gene space vector (GENESAVER) algorithm using IL-8 as a physiological variable. Data are from Eidelman et al., 2001[16]

Gene description Positive correlation with IL-8expression (0-60°)

Inverse correlation with IL-8 expression(120-80°)

Not correlateda (60-120°)[location, SD, angle]

[location, SD, angle] [location, SD, angle]

Associated with TNF receptorsand NFκB pathway

TNFαR [C1c, 0.8, 37°] NFκB-p105 [D1a, 1.3, 128°]

CRAF-1 (TRAF) [C5d, 1.0, 22°]

TRADD [C4a, 1.6, 49°]

RIP [C3k, 0.3, 43°]

FAS antigen [C2b, 1.7, 52°]

FAN [C5b, 1.6, 50°]

CASPASE 7 [C5f, 1.1, 42°]

CASPASE 10 [C4n, 1.6, 48°]

FLICE 2 [C4n, 1.6, 48°]

CD70 [C2a, 1.1, 53°]

TRAMP [C5c, 1.4, 39°]

bcl-2 [C4b, 2, 56°]

Other Inflammatory RANTES [F2j, 1.4, 52°] Interferon consensus BP [D1j, 2.3, 147°]

TGFβ [C1f, 1.0, 23°]

Known to bind CFTR EZRIN [A2f, 1.2, 143°]

Other (top 5% by length) GRB2 isoform [C2i, 3.8, 21°] Ets-like gene [A1j, 2.1, 160°] IGF-I R [C1a, 2.3, 67°]

BMP-2B [F2b, 2.7, 30°] GATA-binding protein [D1b, 1.7, 150°]

c-myb λ [A1c, 2.2, 116°]

Ku (p70/p80) subunit [C6a, 1.9, 47°] DNA-binding protein [D2f, 2.4, 143°]

SLAP [B6a, 2, 70°]

Cyclin G1 [A7a, 1.8, 55°] Global transcription activator [D1i, 2.2, 148°] Ca-ATPase 2 [B1a, 1.9, 89°]

DNA-repair XRCC1 [C6d, 1.7, 39°] Guanine nucleotide-BP G-s α [D3k, 2, 149°]

i NOS [C1b, 1.9, 63°]

PKA-RI-β [B5m, 1.5, 56°] B-myb [A1f, 2, 131°] NIP1 (NIP1) [C4c, 1.8, 63°]

Transcription factor LSF [D5n, 1.9, 143°]

c-erbA [A1g, 1.8, 133°]

DNA-binding protein PO-GA [D2h, 1.6, 151°]

c-kit proto-oncogene [A1d, 1.6, 124°]

DNA-binding SMBP2 [D1h, 1.6, 149°]

IGF-RII [A1m, 1.6, 157°]

Transcription factor AREB6 [D2I, 1.6, 149°]

a Genes from the ‘Not correlated’ column are from the top 5% by SD.

SD = standard deviation.



pharmacogenomic studies that many of the genes identified ap-pear to have no obvious connection to the disease or process inquestion. In our experience, this has also been true for the CFproblem. However, as noted from the review of the CFTR proteinabove, the number of demonstrated interacting gene products istruly remarkable. This is most spectacularly true for the appar-ently disparate NFκB system, which is known to be intimatelyinvolved with the induction in IL-8 production in non-CF cells.

Initially independent pathways involving TNFα, IL-1 and bac-terial lipopolysaccharides (LPS) are hypothesized to converge onNFκB interacting kinase (NIK), or on members of the MAP6Kfamily, which then activate a complex of IκB kinases (IKKα,β,&γ). The NFκB complex, composed of p65 and p50 components,sits inactive when complexed with IκB. However, when the acti-vated IKKs phosphorylate IκB, the proteosome attacks the phos-pho-IκB, releasing the residual p65/p50 complex. This NFκBheterodimer enters the nucleus and binds to κB sites on promotersfor IL-8, κB, TNFα, IL-1, and several other inflammatory signal-ing molecules. In this manner, IL-8 is transcribed and eventuallysecreted. Once within the nucleus, the NFκB complex is removedfrom the κB sites by a fresh IκB molecule from the cytosol, whichthen leaves the nucleus as an p65/p50/ IκB complex, ready forfuture activation. The p50 component is synthesized as a largerp105 precursor, which is able to bind to p65. When bound to p65,p105 acts not only as a p50 ligand to p65, but also as an inhibitoryIκB-like ligand. In this latter role, the IKKs have no activatingeffect on the NFκB complex. As shown in table I, our genomicexperiments indicate that the inhibitory p105 mRNA expressiondeclines in precise synchrony with physiological or pathologicalelevation of IL-8 export from IB3 cells.

Activation of NFκB can also be accomplished by phosphory-lation of IκBα via pathways controlled by Ca2+-dependent andCa2+-independent protein kinase C (PKC) isozymes and bycalcineurin.[80-82] Specific PKC isoforms have been implicated asfollows: PKC-ζ associates with IκBα kinase in resting cells toform PKCK2.[83,84] This kinase can phosphorylate the Pro-Glu-Ser-Thr (PEST) domain of IκB. PKC-α activates the NFκB/IκBaautoregulatory feedback loop, thereby suppressing the nucleareffects of NFκB.[85] PKCα also activates the IKK complex inresponse to tissue plasminogen activator (TPA).[86] PKC-θ phys-ically associates with activated IKK complexes, at least in acti-vated T cells, relocating the complex to membrane bound lipidrafts which are then capable of phosphorylating IκBα.[87] Homo-cysteine stimulates NFκB activity by reducing the expression ofIκBα mRNA.[88] Thus the different PKC isoforms can affectNFκB activation by direct or indirect actions on IκBα integrity.

The TNFα receptor (TNFαR)-1 transduces the TNFα signalto the IκB kinase system through a complex of gene products

including TNFαR-associated death domain (TRADD) protein,TNFαR-associated factor-2 (TRAF2), TNFαR1-interacting pro-tein (RIP), and C inhibitor of apoptosis protein 1 (IAP1/2). Asthe names imply, this system is closely connected to regulationnot only of inflammation but also of the apoptotic pathway. Forexample, the TRADD adapter also transduces interactions be-tween TNFαR and downstream apoptotic components such ascaspase 7, FLICE, FAS antigen, FAN, and TRAMP. In the con-text of the simplistic ‘divide-or-die’ alternatives available tocells, this cluster of TNF-associated gene products is thought tobe initially used by cells for both purposes. However, it is impor-tant to appreciate that in the case of CF cells, the potential forcross-talk between proinflammatory and pro-apoptotic pathwaysmay be more than idle speculation. For example, a 10-fold eleva-tion in TdT-mediated dUTP biotin nick-end labeling (TUNEL)for fragmented DNA has been reported in crypt enterocytes ob-served in duodenal biopsies of CF patients.[89] In addition, FASand FAS ligand expression, markers of apoptosis as well as in-flammation, are markedly increased over non-CF controls in bi-opsies of CF bronchial epithelium and cultured CF tracheal celllines.[90] Disordered regulation of apoptosis has also been ob-served in heterologous C127 cells expressing ΔF508-CFTR.[91]

Other gateways into the NFκB signaling pathway, includingthose associated with interaction with bacteria, involve the IL-1receptor (IL1R). When IL1β binds to the receptor, the adapterprotein IL1R-associated kinase (IRAK) is activated, which inturn activates TNFα-receptor associated factor, variants 2 and 6(TRAF 2 and 6). TRAF2 is a common link between TNFα andIL-1 systems. Bacteria, per se, are also known to be potent acti-vators of the NFκB signaling pathway. However, the latter mech-anisms are not very well understood. Toll-like receptors (TLRs)are known to be able to bind lipopolysaccharide (LPS) compo-nents of the Gram-negative cell wall as a complex with the lipo-polysaccharide binding protein (LBP) and CD14. The conservedintracellular domain of the TLR is homologous with that of IL-2receptors, and the common adapter proteins MyD88, IRAK andTRAF2,6 may therefore feed into the common NFκB activationcascade.[92] However, intact bacteria do not apparently interactwith the TLRs. Rather, they bind to the ganglioside GM1 on theepithelial cell surface, and in some fashion activate IL-8 secre-tion. This productive interaction can also be mimicked by theaddition of an antibody to asialo-GM1, and by purified bacterialligands such as pilin or flagellin. The bacterial interaction alsocauses IP3 receptor-mediated intracellular Ca2+ release, andequivalent effects on IL-8 secretion can be detected by simplyadministering epithelial cells with Ca2+ releasing agonists suchas thapsigargin.[93] Thus, the mechanistic relationship between

230 Eidelman et al.


bacterial attack on epithelial cells and the NFκB pathway is pres-ently unclear.

6. CF Pharmacogenomics

6.1 Pharmacogenomic Data Can be Used to Mine OutDrug- or Gene-Dependent Signaling Pathways

One of the most attractive features of pharmacogenomics isthe promise of using genomic information to develop surrogateendpoints for drug discovery. For ‘discovery-driven’ research, asimple list of genes whose expression levels are changed in agiven disease state might seem to be sufficient for definition ofsuch an endpoint. An efficacious drug might be expected tochange the expression array to a ‘control’ or non-diseased state.However, the more we understand a system, the more this simpleapproach can become confounded by elements as diverse as theclinical course of the patient’s contributing cells, genetic consti-tution of the subject cells, the tissue from whom the cells origi-nated, the mechanism of immortalization (or lack of immortiliza-tion), the clinical history of the patient from which the cells werederived, and many other variables. In addition, the resultant dataoften consist of hundreds of genes, many with no discernibleknown biological relationship to one another. Thus, although thegenes mined out by such a simple approach can be said to havesome relationship to one another in the context of the disease,drug treatment or process, it is difficult to distinguish which genesin the data-set constitute signal and which ones constitute noise.

One approach for elucidating the ‘important’ genes has beento attempt to understand the cause-and-effect relationships be-tween all the interrogated genes by measuring changes in geneexpression over time. In this approach, the set of gene expressionsignals are sampled systematically over an appropriate time do-main. When studying the process of cell division in bacteria[94]

or yeast,[94] the time domain is about 20 to 30 minutes. Using eachtime point as an additional dimension in the analysis, hierarchicalclustering algorithms have been used to identify clusters of simul-taneously changing genes. The resulting clusters of genes can beinterpreted as having some causal or regulatory relationship thatis stronger than a relationship with genes in other clusters. Sincethe microbial cultures continue to divide synchronously, the re-sulting signal even has the approximate appearance of an oscil-lating signal, just as might be found in a simple telecommunica-tions problem. However, most biological problems are far lesstractable. For example, gene expression can be followed over thecourse of central nervous system development in the rat.[96] Sucha complete process happens only once per generation, a time pe-riod of several months. Nonetheless, in one developmental period

the investigators were able to discern 5 basic waves of gene ex-pression in the developing rat CNS. The value of this approachto pharmacogenomics is that such clusters of causally relatedgenes might be targets for drug discovery.

However, as it currently impacts on pharmacogenomics, thereare two fundamental problems in this approach to analysis of geneexpression. One problem is lack of knowledge of the true kinetictrajectory for each of the genes. One can assume that a linearmodel might suffice, or that clusters of genes might be handledas one ‘gene’ in order to simplify the number of genes to beconsidered.[97] However, the molecular biology database gives usmany examples of non-linear transcription events, includinghyperbolic, sigmoid, and bell-shaped processes. To address suchalternatives, others have also been developing non-linear differ-ential equations to deal with the likely event of non-linear geneexpression patterns.[98] However, the kinetic properties of mostgenes are not well established. Furthermore, most of the humangenes are yet to be annotated and are known only as an addressin the genome. Therefore, in order to do any analysis at all, thecurrent default assumption begins with linear models.

A second as yet unsolved problem with the time series ap-proach is based on the fact that, in the experiments used to collectdata, time is divided grossly into experimentally convenient in-tervals. Thus, many intervening time points are lost. Whether thetime points chosen are suitable to the analysis depends on thenature of the processes occurring in the time intervals. Fundamen-tally complicating the analysis is the fact that each of these timepoints constitutes a dimension in the analysis; and since the num-ber of time points is infinite, so are the possible dimensions. Thus,since the available data are granular, the distinctions betweengene expression trajectories are necessarily coarse. By granular,we refer to the magnitude of the experimental subdivisions intime. One recent attempt to bypass this ‘dimension problem’ hasbeen to reverse engineer the time series data from a hypotheticallycausally connected network of genes.[99-101] The hope has been toarrive by an iterative process at a model network, or cluster ofnetworks, that significantly represent the actual data. To be ulti-mately successful, such an approach still needs an increased den-sity of time points to be accurate, as well as more biologicalinformation to determine whether the predicted causal connec-tions in the model are actually valid.

In addition, time domain studies do not lend themselves to thetypical biological questions that are usually encountered in drugdiscovery. For instance, in the case of a chronic genetic diseasesuch as cancer, diabetes or cystic fibrosis, the disease in questioncauses a change in expression of different relevant genes by aprocess in which the time progression of these changes is histor-ically long. Furthermore, a typical question for pharmacogenom-



ics is how these sets of changed genes might be affected by a drug,and whether there are any genes for which the effect of the drugmight herald possible adverse effects. There are indeed situationswhere time domain questions might be relevant, but are opera-tionally impossible.

As an alternative approach to these time domain studies, manyinvestigators choose to collect data at one relevant time point. Insome cases there is no other choice, as is necessarily the casewhen tissue is collected at biopsy, surgery or post-mortem. Inother cases, for example, a high throughput screen of drugs, orcomparisons between numerous cell lines or experiments understeady-state conditions, the choice of one time point seems mostefficient.[102-105] The projects customarily begin with dense arraysof thousands of genes, which must be winnowed down to sparsearrays of hundreds of genes, and then to tractable arrays of 10 to20 genes. The problem facing the pharmacogenomics team is todevelop algorithms which permit this screening process to pro-ceed efficiently. Currently, various types of hierarchical cluster-ing algorithms and self-organizing maps are being used. We havealso developed the GRASP and GENESAVER algorithms, whichwe shall discuss in section 6.2 in the context of the CF problem.

6.2 Application of GRASP and GENESAVER Algorithms toPharmacogenomic Analysis of Cystic Fibrosis

6.2.1 The GRASP AlgorithmThe gene ratio analysis paradigm (GRASP) algorithm is a sta-

tistical method in which the entire power of the array is harnessedto evaluate the significance of changes in gene expression follow-ing gene or drug therapy.[106] Briefly, this discovery-driven al-gorithm graphically compares the average change in gene expres-sion following drug or gene therapy, and distinguishes thosegenes which change significantly on the basis of number of stand-ard deviations away from the average change of all the genes. Inaddition, by normalizing expression data in a transfected cell ora drug treated cell by calculating a ratio of the expression levelsof each gene to the expression of the same gene in the parentalcell line, GRASP allows one to compare up to 4 different condi-tions on a 2-dimensional graph. This allows the comparison of 2drugs or an experimental drug and gene therapy. By setting oneof the conditions as a ‘cured status’ surrogate, one can first iden-tify a tentative set of genes that are affected by the disease, andthen ask how the expressions of these genes are affected by theexperimental drug. Furthermore, once a drug is selected, GRASPcan allow the identification of dose-dependent effects, both ben-eficial (on genes affected by the disease status) and unfavorable(e.g. adverse effects, such as elevation of a proto-oncogene or

reduction of a tumor suppressor gene). The steps in this algorithmare outlined in Srivastava et al.[42]

When applied to gene expression in HEK293 cells expressingeither wild-type or [ΔF508]-CFTR, approximately 5% of the totalinterrogated genes were sensitive to whether wild-type or mutantCFTR was being expressed by the cell. In this particular study, a588 gene Clontech cDNA microarray was employed as the target.Administration of CPX had mutation-independent and mutation-dependent effects on the expression of specific genes. In partic-ular, CPX caused the expression of protein kinase A (PKA) reg-ulatory subunit II to be hyperexpressed by over 6 standarddeviations (SD) over the average change in expression of all que-ried genes. The potential importance of this observation is thatCFTR is activated by a cAMP signal mediated via phosphoryla-tion by PKA.

This particular approach can also be used in a direct graphicalmanner to identify genes whose expression is being changed to-wards a normal expression level by given pharmaceutical or genetherapy. This is a valuable asset because the expression of suchgenes might constitute elements of a surrogate endpoint. Theprinciple is outlined in figure 4a, and some actual data are shownin figure 4b. As shown in figure 4a, genes expressed at equivalentlevels in both mutant and wild-type expressing cells are on thediagonal. For clarity only two differentially-expressed genes areshown, with their positions on the graph indicating their expres-sion levels in wild-type and mutant cells. When cells expressingthe mutant CFTR gene are treated with CPX, the magnitude anddirection of the change in gene expression of each sample genecan be represented as a ‘genomic drug vector’ of a particularlength and angle. A gene whose expression level moves closer tothe diagonal in response to the drug (e.g. gene α2, fig 4a) mightbe considered ‘corrected’. Ideally, one would like the genomicdrug vectors to be pointing directly up or directly down towardsthe diagonal. In figure 4b, two radial plots summarize 588 geno-mic drug vectors, abstracted from the data of Srivastava et al.[106]

Of course, one does not know whether specific drug vectors are‘good’ or ‘bad’ for CF. The vectors pointing towards the diagonalonly indicate that these genes are clearly candidates to be consid-ered as surrogate endpoints for discovery of drugs that cause geneexpression changes similar to those observed upon transfectionof mutant cells with wild-type CFTR.

The GRASP ratios can also be used as dimensions in a hierar-chical clustering algorithm. We analyzed global gene expressionusing such an algorithm in CF airway IB3 epithelial cells,[107]

which we had found to be secreting massive amounts of IL-8 (seeEidelman et al. [16]). These cells were treated with adeno-associ-ated virus (AAV)-mediated CFTR gene therapy to yield a cor-rected cell line (the IB3-S9 cell line), and both this and the orig-

232 Eidelman et al.


0 20 40 60 800

345

330

315

300285270255

240

225

210

195

180

165

150

135

120105 90 75

60

45

30

15

All

<−1sd

av ± 1sd

1-2sd

>2sd

0 1 2 3 4 5 60

345

330

315

300285270255

240

225

210

195

180

165

150

135

120105 90 75

60

45

30

15

0.01

0.1

1

10

100

0.01 0.1 1 10 100

Expression level in control

Exp

ress

ion

leve

l in

mut

ants

a

b

α1

α2

Fig. 4. Graphical representation of the gene ratio analysis paradigm (GRASP) algorithm. (a) A graph of gene expression levels in cells expressing either wild-type CFTR(horizontal axis) or mutant CFTR (vertical axis). Two genes, denoted α1 and α2, are chosen from many genes in an array. Genes expressed at equivalent levels in bothcell types are on the diagonal. The outer parallel lines indicate 2 standard deviations away from the diagonal. The yellow squares indicate expression levels of 2differentially-expressed sample genes in wild-type and mutant cells in the absence of CPX. The filled black circles indicate the expression of the same genes when CPXis added to the cells expressing mutant CFTR. The red arrows are ‘genomic drug vectors’, indicating the magnitude and direction of the change in gene expression forthe 2 example genes. When CPX was added, gene α1, overexpressed in the mutant cell in the absence of the drug, further increased its expression and moved up thegraph at a 45° angle. The 45° angle indicates that expression was increased equivalently in both mutant and control cells. By comparison, gene α2, underexpressedin the mutant cell in the absence of the drug, moved towards the diagonal at a 135° angle. This indicates that its expression level increased to close to wild-type levels.Therefore gene α2 attracts our attention as a possible surrogate endpoint gene. Each effect of CPX on gene expression is therefore a vector with a length and anangle.[42] (b) A complete data set from HEK293 cells expressing mutant or wild-type CFTR, and treated with CPX. The radial plots summarize on a global scale 588genomic drug vectors, abstracted from the data of Srivastava et al.[106] The angle in the radial plot is the angle that the drug causes the gene expression to move. Therelative magnitude of the change in position is given as the number of standard deviations (SDs) away from the average changes of all genes in the array. These arecolor-coded in the key. The radial plot on the left is at low resolution, and the majority of vectors move in the 45° to 225° axis, i.e. affected similarly in both cell types.The axes of the specific angles are subdivided into numbers of genes at any given angle. At high resolution, one can delineate the genes which not only move in the135° to 315° axis, but also whether they are members of the higher SD class of genes.[15,106] Genes that move in the 135° to 315° axis are those that are moving towardsequivalent expression in mutant and control cells. These gene expression changes could represent ‘repair’ of the deficit.



inal IB3 lines were treated with CPX. To mimic the clinical situ-ation in patients with CF, the cells were also exposed separatelyto a clinical isolate of Pseudomonas aeruginosa, the bacteriumthat chronically infects most CF patients. By measuring globalgene expression in each condition and analyzing the entire en-semble with the hierarchical clustering algorithm, we could testthe hypothesis that the mutant IB3 cell exposed to CPX wouldclosely resemble the wild-type CFTR-treated IB3-S9 cell line.We summarize the clustered conditions from the horizontal axisof the hierarchical clustering algorithm from Eidelman et al.[16]

in figure 5. The data clearly indicate that the most closely clus-tered conditions are IB3 + CPX and IB3-S9 cells. Thus the dataclearly indicate that CPX drug treatment and CFTR gene therapycause genomically similar consequences in the CF epithelial cellline. The extent to which this is also true for other CF cells andCF mutations will validate the choice of genes to be chosen fordrug discovery.

6.2.2 The GENESAVER Algorithm The gene space vector (GENESAVER) algorithm is a method

in which a physiological variable is used to identify genes whoseexpression is correlated with the variation in magnitude of thevariable. The details of how to apply this algorithm have beendescribed by Srivastava et al.[106] Briefly, the experimental sys-tem is subjected to a variety of conditions which are relevant tothe question at hand. One of the conditions is used as a control,and results in each of the other conditions is normalized to it.Thus, for CF, we chose exposure of CF cells to bacteria, as wellas gene or drug therapy, and used IL-8 expression as the relevantvariable. The control in each case is the untreated IB3-1 cell. Foreach condition, IL-8 secretion was measured, and the results (inlogarithmic form) were used to create a multidimensional vector.The outcome is a set of n-1 dimensional vectors (where n is thenumber of experimental conditions). The next step is to take eachgene in the array and calculate a similar multidimensional vector.The clustering step is done by calculating the angle in multidi-mensional space between the vector for each gene and the vectorfor the physiological parameter, e.g. IL-8 secretion. Small anglesbetween the vectors indicate a close correlation between the phys-iological variable and gene expression. By contrast, angles closeto 180° indicate a negative correlation with IL-8. Our CF studyresulted in the mining out of a small number of genes that turnedout mainly to be members of the TNFαR/NFκB pathway[16] (ta-ble I). The likelihood of that number of genes being identified bya random process is 1 in 10 billion (1/109), and, obviously, theprobability that these random processes would mostly identifygenes of a particular signal transduction pathway is far smaller.Many of the genes in table I have been described in a preceding

section. For example, in table I, NFκB-p105 expression shows aninverse correlation with that of IL-8 (vector angle = 128°). Thisis not surprising since, as we have reviewed above, the productof this gene is a precursor to the inhibitory subunit NFκB-p50 inthe NFκB-p65/NFκB-p50 complex. Thus, it makes sense that whenIL-8 expression increases, the message for the inhibitory subunitshould decline. We include for completeness a list of ‘not corre-lated’ genes whose magnitude of change is nonetheless ≥2SDs.

The main advantage of this algorithm is that it yields clustersof genes whose dynamic relationship to the disease or biologicalprocess is manifest by close correlation with a relevant physio-logical variable. On the one hand, GENESAVER is a discovery-driven process in that one need not know, a priori, what thosegenes might be. One has only to select a relevant physiologicalvariable. On the other hand, GENESAVER can be a hypothesis-driven process, if one has the foresight to pick the appropriatephysiological variable. For example, in the case of CF, in whichan intrinsic overproduction of proinflammatory IL-8 is an impor-tant characteristic of the CF airway, we did have reason to antici-pate, a priori, the likely involvement of the NFκB signaling path-way. Therefore, the fact that the GENESAVER algorithmselected this and the related TNFαR pathway is evidence to sup-port the utility of this new algorithm.

The general utility of this algorithm should also be empha-sized. Presuming that the investigator can pick an appropriatephysiological variable for the disease or process of interest, ap-propriate candidate genes are very likely to be mined out. Finally,from the perspective of disease or drug-dependent signaling path-way discovery, it is worth emphasizing that the GENESAVERalgorithm delivers the same kind of specific gene cluster prom-ised by discovery-driven studies in the classical time or frequencydomains.

7. Conclusions

During the process of drug discovery, the principal problemhas always been to define an experimental system which is notonly simple in operation, but also provides a faithful model of thedisease under investigation. The endpoint should indicate as un-ambiguously as possible whether a given compound has promiseas a possible drug or not. But no matter how shrewdly chosen, alarge fraction of the candidate lead compounds coming out ofsuch model systems have failed as they traverse the toxicologyprocess and into phase I studies. Development of surrogate end-points based on genomics, therefore, has the promise to increasethe probability of success in drug discovery, not only by focusingon human genes, but also by giving the investigator the opportu-nity to learn of possible adverse effects at the earliest stage of

234 Eidelman et al.


Fig. 5. Hierarchical clustering algorithm comparing pharmacotherapy with CPX and gene therapy with wild-type CFTR. Clustered gene expression data are shown forthe CF airway IB3 epithelial cells line IB3-1 and the wild-type CFTR-repaired IB3 cell line IB3-S9 (S9). The cell lines were either untreated, or treated with CPX, Ps.aeruginosa (P.a.), or both. Each condition is compared with all the others on the basis of global gene expression patterns. The most closely clustered conditions aredyadically joined. The IB3 cell treated with CPX and the IB3-S9 cells repaired by gene therapy are closely clustered (far right). These data support the hypothesis thatCPX and gene therapy have similar genomic consequences. Data are from Eidelman et al., 2001.[16]



development. For example, a drug could satisfy all the genomicrequirements needed to define it as a good candidate, yet thecompound could also induce elevation of proto-oncogenes or sup-pression of tumor suppressor genes. Either could signal a concernfor possible carcinogenic side effects of the compound in ques-tion. Alternatively, apoptotic genes might be induced in cellscultured from the CNS, signaling concern for possible neuro-degenerative potential. These and other approaches to preclinicalgenomics could be truly cost-effective preludes to the more costly‘go/no-go’ decision to advance to toxicology studies and beyond.

The use of genomics to define surrogate endpoints also has thepromise of developing drugs that are tailored to specific patients.With the recent conclusion of the Human Genome Project, weknow that each human substantially and significantly differs fromothers, even within the same family. The differences include vari-ations known as single nucleotide polymorphisms (SNPs), whichcan indicate variations in the susceptibility to various diseases.Conceivably, the complex pattern of SNPs in each human couldalso impact on the appropriate drug or dosage prescribed for thatindividual patient. Finally, the advent of a genomics-based sys-tem for drug discovery promises to modify the implicit businessmodel under which most pharmaceuticals are delivered to thepublic. Rather than the current one-size-fits-all approach, drugsmay become formulated in combinations and strengths suited tothe genomic constitution of a particular patient. The model hasalready proven to be effective for computers, in which the con-sumer goes to the web and assembles a computer with accessoriesas needed for his purpose and situation. A similar model mayprove to be appropriate in the genome-based pharmaceutical in-dustry of the future.

Acknowledgements

This work was supported in part by grants from the Cystic FibrosisFoundation and the National Institutes of Health (R01-DK53051).

References1. Welsh MJ, Ramsey BW, Accurso B, et al. Cystic fibrosis. In: Scriver CL, Beaudet

AL, Valle D, et al., editors. The metabolic and molecular bases of inheriteddiseases. 8th ed. New York: McGraw-Hill, 2001: 5121-88

2. Rommens JM, Lannuzzi MC, Karem B-S, et al. Identification of the cystic fibrosisgene: chromosome walking and jumping. Science 1989; 245 (1): 1059-65

3. Riordan JR, Rommens JM, Karem B-S, et al. Identification of the cystic fibrosisgene: cloning and characterization of complementary DNA. Science 1989; 245:1066-73

4. Kerem B-S, Rommens JM, Buchanan JA, et al. Identification of the cystic fibrosisgene: genetic analysis. Science 1989; 245: 1073-80

5. Shak S, Capon DJ, Hellmiss R, et al. Recombinant human DNAse I reduces vis-cosity of cystic fibrosis sputum. Proc Nat Acad Sci U S A 1990; 87: 9188-92

6. Konstan MW, Byard PJ, Hoppel CL et al. Effect of high dose ibuprofen in patientswith cystic fibrosis. N Eng J Med 1995; 332: 848-54

7. Bonfield TL, Konstan MW, Burfeind P, et al. Normal bronchial epithelial cellsconstitutively produce the anti-inflammatory cytokine interleukin 10 which isdownregulated in cystic fibrosis. Am J Respir Cell Mol Biol 1995; 13: 257-61

8. Bonfield TL, Panuska JR, Konstan MW, et al. Inflammatory cytokines in cysticfibrosis lungs. Am J Respir Crit Care Med 1995; 152: 2111-18

9. Dean TP, Dai Y, Shute JK, et al. Interleukin-8 concentrations are elevated inbronchoalveolar lavage sputum and sera of children with cystic fibrosis. PediatrRes 1993; 34: 159-61

10. Ruef C, Jefferson DM, Schlegel-Haueter SE, et al. Regulation of cytokine secretionby cystic fibrosis airway epithelial cells. Eur Resp J 1993; 6: 1429-36

11. Anderson MP, Rich DP, Gregory RJ, et al. Generation of cAMP activated chloridecurrents by expression of CFTR. Science 1991; 2516: 679-682

12. Collins FS. Cystic fibrosis: Molecular biology and therapeutic implications. Sci-ence 1992; 256 774-9

13. Pilewski JM, Frizzell RA. Role of CFTR in airway disease. Physiol Rev 1999; 79:S215-55

14. Griesenbach U, Scheid P, Hillery E, et al. Anti-inflammatory gene therapy directedat the airway epithelium. Gene Ther 2000; 7: 306-13

15. Pollard HB, Eidelman O, Jacobson KA, et al. Pharmacogenomics of cystic fibrosis.Mol Interventions 2001; 1: 54-63

16. Eidelman O, Srivastava M, Zhang J, et al. Control of the proinflammatory state incystic fibrosis lung epithelial cells by genes from the TNFαR/NFκB pathway.Mol Med 2001. In press

17. Bailey DS, Bondar A, Furness LM. Pharmacogenomics - it’s not just pharmacoge-netics. Curr Opin Biotechnol 1998; 6: 595-601

18. Ferrar P. Pharmacogenomics: a new approach to individual therapy of hyperten-sion? Curr Opin Nephrol Hypertens 1998; 7: 217-21

19. Graever G, Shoemaker DD, Jones TW, et al. Genomic profiling of drug sensitivitiesvia induced haploinsufficiency. Nat Genet 1999; 3: 278-83

20. Eidelman O, Guay-Broder C, van Galen PJM, et al. A1-adenosine-receptor antag-onists activate chloride efflux from cystic fibrosis cells. Proc Nat Acad Sci U SA 1992; 89: 5562-6

21. Guay-Broder C, Jacobson KA, Sarnoy S, et al. A-1 receptor antagonist‘cyclopentyl1,3-dipropylxanthine selectively activates chloride efflux from hu-man epithelial and mouse fibroblast cell lines expressing the cystic fibrosistransmembrane regulator ΔF508 mutation. Biochemistry 1995; 34: 9079-87

22. Jacobson KA, Guay-Broder C, van Galen PJM, et al. Stimulation by alkylxanthinesof chloride efflux in CFPAC-1 cells does not involve A1 adenosine receptors.Biochemistry 1995; 34: 9088-94

23. Kartner N, Hanrahan JW, Jensen TJ, et al. Expression of the cystic fibrosis gene innon-epithelial invertebrate cells produce a regulated anion conductance. Cell1991; 64: 681-91

24. Berger HA, Anderson MP, Gregory RJ, et al. Identification and regulation of thecystic fibrosis transmembrane conductance regulator-generated chloride chan-nel. J Clin Invest 1991; 88: 1422-31

25. Rommens JM, Dho S, Bear CE, et al. cAMP-inducible chloride conductance inmouse fibroblast lines stably expressing the human cystic fibrosis transmem-brane conductance regulator. Proc Natl Acad Sci U S A 1991; 88: 7500-4

26. Bear CE, Reyes EF. cAMP-activated chloride conductance in the colonic cell lineCaco-2. Am J Physiol 1992; 262: C251-6

27. Denning GM, Ostergaard LS, Cheng SH, et al. Localization of cystic fibrosis trans-membrane conductance regulator in chloride secretory epithelia. J Clin Inv1992; 89: 339-49

28. Tsui L-C. Mutations and sequence variations detected in the cystic fibrosis trans-membrane conductance regulator (CFTR) gene: a report from the Cystic Fibro-sis Genetic Analysis Consortium. Hum Mutat 1992; 1: 197-203

29. Tsui L-C. The spectrum of cystic fibrosis mutations. Trends Genet 1992; 8: 392-8

30. Hung LW, Wang IX, Nikaido K, et al. Crystal structure of the ATP-binding subunitof an ABC transporter. Nature 1998; 396: 703-7

31. Diederichs K, Diez J, Greller G, et al. Crystal structure of MalK the ATPase subunitof the trehalose/maltose ABC transporter of the archaeon ThermococcusLitoralis. EMBO J 2000; 19: 5951-61

32. Drumm ML, Wilkinson DJ, Smit LS, et al. Chloride conductance expressed byDF508 and other mutant CFTRs in xenopus oocytes. Science 1991; 254: 1797-9

33. Arispe N, Rojas F, Hartman J, et al. Intrinsic anion channel activity of the recom-binant first nucleotide binding fold domain to cystic fibrosis transmembraneconductance regulator protein. Proc Nat Acad Sci U S A 1992; 89: 1539-43

236 Eidelman et al.


34. Cheng SH, Gregory RJ, Marshall J, et al. Defective intracellular transport andprocessing of CFTR is the molecular basis of most cystic fibrosis. Cell 1990;63: 827-34

35. Yang Y, Janich S, Cohn JA, et al. The common variant of cystic fibrosis transmem-brane conductance regulator is recognized by hsp70 and degraded in a pre-golginon-lysosomal compartment. Proc Nat Acad Sci U S A 1993; 90: 9480-4

36. Lukacs GL, Mohamed A, Kartner N, et al. Conformational maturation of CFTRbut not its mutant counterpart (ΔF508) occurs in the endoplasmic reticulum andrequires ATP. EMBO J 1994; 13: 6076-86

37. Ward CL, Kopito RR. Intracellular turnover of cystic fibrosis transmembrane con-ductance regulator inefficient processing and rapid degradation of wildtype andmutant protein. J Biol Chem 1994; 269: 25710-18

38. Denning GM, Anderson MP, Amara JF, et al. Processing of mutant cystic fibrosistransmembrane conductance regulator is temperature-sensitive. Nature 1992;358 (6389): 761-4

39. Denning GM, Ostedgaard LS, Welsh MJ. Abnormal localization of cystic fibrosistransmembrane conductance regulator in primary cultures of cystic fibrosis air-way epithelia. J Cell Biol 1992; 118 (3): 551-9

40. Engelhardt JF, Yankaskas JR, Ernst SA, et al. Submucosal glands are the predom-inant site of CFTR expression in the human bronchus. Nat Genet 1992; 2: 240-7

41. Kartner N, Augustinas O, Jensen TJ, et al. Mislocalization of ΔF508 CFTR in cysticfibrosis sweat gland. Nat Genet 1992; 1: 321-7

42. Srivastava M, Eidelman O, Pollard HB. Pharmacogenomics of the cystic fibrosistransmembrane conductance regulator (CFTR) and the cystic fibrosis drug CPXusing genome microarray analysis. Mol Med 1999; 5: 753-67

43. Abman SH, Ogle JW, Harbeck RJ, et al. Early bacteriologic immunologic andclinical courses of young infants with cystic fibrosis identified by neonatalscreening. J Pediatr 1991; 119: 211-17

44. Khan TZ, Wagene JS, Bost T, et al. Early pulmonary inflammation in infants withcystic fbrosis. Am J Respir Crit Care Med 1995; 151: 1075-82

45. Cruse JM, Lewis RE. Illustrated dictionary of immunology. Boca Raton: CRCPress Boca, 1995: Appendix 3

46. Richman-Eisenstat JB, Jorens PG, Hebert CA, et al.Interleukin-8: an importantchemoattractant in sputum of patients with chronic inflammatory airway dis-eases. Am J Physiol 1993; 264: L413-8

47. Francoeur C, Denis M. Nitric oxide and interleukin-8 as inflammatory componentsof cystic fibrosis. Inflammation 1995; 19: 587-98

48. Armstrong DS, Grimwood K, Carlin JB, et al. Lower airway inflammation ininfants and young children with cystic fibrosis. Am J Respir Crit Care Med 1997;156: 1197-204

49. Briars GL, Dean TP, Murphy JL, et al. Interleukin-8 and tumour necrosis factor-alpha concentrations in cystic fibrosis. Arch Dis Child 1995; 73: 74-6

50. Tabary O, Zahm JM, Hinnrasky J, et al. Selective upregulation of the chemokineIL8 expression in cystic fibrosis bronchial gland cells in vivo and in vitro. AmJ Pathol 1998; 153: 921-30

51. Bedard M, McClure CD, Schiller NL, et al. Release of interleukin-8 interleukin-6and colony-stimulating factors by upper airways epithelial cells: implicationsfor cystic fibrosis. Am J Resp Cell Mol Biol 1993; 9: 455-62

52. DiMango E, Ratner AJ, Bryan R, et al. Activation of NF-kappaB by adherentPseudomonas aeruginosa in normal and cystic fibrosis respiratory epithelialcells. J Clin Invest 1998; 101: 2598-605

53. Black HR, Yankaskas JR, Johnson LG, et al. Interleukin-8 production by cysticfibrosis nasal epithelial cells after tumor necrosis factor-alpha and respiratorysyncytial virus stimulation. Am J Respir Cell Mol Biol 1998; 19: 210-15

54. Massion PP, Inoue H, Richman-Eisenstat J, et al. Novel Pseudomonas productstimulates interleukin-8 production in airway epithelial cells in vitro. J ClinInvest 1994; 93: 26-32

55. Schwiebert LM, Estell K, Probst SM. Chemokine expression in CF epithelia: im-plications for the role of CFTR in RANTES expression. Am J Physiol 1999,276: C700-10

56. Griesenbach U, Scheid P, Hillery E, et al. Towards anti-inflammatory lung genetherapy. Pediatr Pulmonol 1999; S19: 237

57. Arispe N, Ma J, Jacobson KA, et al. Direct activation of cystic fibrosis transmem-brane conductance regulator by 8-cyclopentyl-1,3-dipropylxanthine and 1,3-diallyl-8-cyclohexylxanthine. J Biol Chem 1998; 273: 5724-34

58. Pollard HB. Role of CPX in promoting trafficking and chloride channel activity ofwildtype and mutant CFTR. Pediatr Pulmonol 1997; S14: 128-31

59. Cohen BE, Lee G, Jacobson KA, et al. CPX (1,3-dipropyl-8-cyclopentyl xanthine)and other alkyl-xanthines differentially bind to the wild type and ΔF508 mutantfirst nucleotide binding fold (NBF-1) domains of the cystic fibrosis transmem-brane conductance regulator (CFTR). Biochemistry 1997; 36: 6455-61

60. Schultz BD, Frizzel RA, Bridges RJ. IBMX stabilizes the ATP-bound state ofΔF508CFTR. J Gen Physiol 1994; 104: 35a

61. Stutts MJ, Canessa CM, Olsen JC, et al. CFTR as a cAMP-dependent regulator ofsodium channels. Science 1995; 269: 847-50

62. Ismailov II, Awayda MS, Jovov B. Regulation of epithelial sodium channel by thecystic fibrosis transmembrane conductance regulator. J Biol Chem 1996; 271:4725-32

63. Egan M, Flotte T, Affione S, et al. Defective regulation of outwardly rectifying Clchannels by protein kinase A corrected by insertion of CFTR. Nature 1992; 358:581-4

64. Gabriel SE, Clarke LL, Boucher FCC, et al. CFTR and outward rectifying chloridechannel are distinct proteins with a regulatory relationship. Nature 1993; 363:263-8

65. Schwiebert EM, Flotte T, Cutting GR, et al. Both CFTR and outwardly rectifyingchloride channels contribute to cAMP-stimulated whole cell chloride currents.Am J Physiol 1994; 266: C1464-7

66. Jovov B, Ismailov II, Benos DJ. Cystic fibrosis transmembrane conductance reg-ulator is required for protein kinase A activation of an outwardly rectified anionchannel purified from bovine tracheal epithelia. J Biol Chem 1995; 270: 1521-8

67. Fulmer SB, Schwiebert EM, Morales MM, et al. Two cystic fibrosis transmem-brane conductance regulator mutations have different effects on both pulmonaryphenotype and regulation of outwardly rectified chloride currents. Proc NatAcad Sci USA 1995; 92: 6832-6

68. Naren AP, Nelson DJ, Xie W, et al. Regulation of CFTR chloride channels bysyntaxin and Munc18 isoforms. Nature 1997; 390: 302-5

69. Naren AP, Quick MW, Collawn JF, et al. Syntaxin 1A inhibits CFTR chloridechannels by means of domain-specific protein-protein interactions. Proc NatAcad Sci U S A 1998; 95: 10972-7

70. Moyer BD, Denton J, Karlson K, et al. Interacting domain in CFTR is required forapical polarization and export from the endoplasmic reticulum. PediatrPulmonol 1999a; S19: 164

71. Moyer BD, Denton J, Karlson K, et al. A PDZ interacting domain in CFTR is anapical membrane polarization signal. J Clin Invest 1999b; 104: 1353-61

72. Raghuram V, Hallows KR, Foskett JK. Multiple PDZ domains involved in theassociation of NHERF with CFTR. Pediatr Pulmonol 1999; S19: 178

73. Casavola V, Turner RJ, Guay-Boder C, et al. CPX: a selective A1-adenosine-re-ceptor antagonist regulates intracellular pH in cystic fibrosis cells. Am J Physiol1995; 269: C226-33

74. Sun F, Hug MJ, Yun CH, et al. The PDZ domain-containing protein E3KARPcouples EZRIN to cystic fibrosis transmembrane conductance regulator(CFTR). Pediatr Pulmonol 1999; S19: 164

75. Short DB, Trotter KW, Reczek D, et al. An apical PDZ protein anchors the cysticfibrosis transmembrane conductance regulator to the cytoskeleton. J Biol Chem1998; 273: 19797-801

76. Bretscher A. Regulation of cortical structure by the Ezrin-Radaxin-Moesin proteinfamily. Curr Opin Cell Biol 1999; 11: 109-16

77. Pearson MA, Reczek D, Bretscher A, et al. Structure of the ERM protein Moesinreveals the FERM domain fold masked by an extended actin binding tail domain.Cell 2000; 101: 259-70

78. Lamprecht G, Weinman EJ, Yun CH. The role of NHERF and E3KARP in thecAMP-mediated inhibition of NHE3. J Biol Chem 1998; 273: 29972-8

79. DiMango E, Zar HJ, Bryan R et al. Diverse Pseudomonas aeruginosa gene productsstimulate respiratory epithelial cells to produce interleukin-8. J Clin Invest 1995;96: 2204-10

80. Steffan NM, Bren GD, Frantz B, et al. Regulation of IκB alpha phosphorylation byPKC-and Ca2+ dependent signal transduction pathways. J Immunol 1995; 155:4685-91

81. Sun L, Carpenter G. Epidermal growth factor activation of NF-κB is mediatedthrough IκBa degradation and intracellular free calcium. Oncogene 1998; 16:2095-105



82. Trushin SA, Pennington KN, Algeciras-Schimnich A, et al. Protein kinase C andcalcineurin synergize to activate IκB kinase and NF-κB in T lymphocytes. J BiolChem 1999; 274: 22923-31

83. Bren GD, Pennington KN, Paya CV. PKC-zeta-associated CK2 participates in theturnover of free IκBa. J Mol Biol 2000; 297: 1245-58

84. Taylor JA, Bren GD, Pennington KN, et al. Serine 32 and serine 36 of IκBa aredirectly phosphorylated by protein kinase CKII in vitro. J Mol Biol 1999; 290:839-50

85. Han Y, Meng T, Murray NR, et al. Interleukin-1-induced nuclear factor-κB-IκBaautoregulatory feedback loop in hepatocytes. A role for protein kinase Ca inpost-transcriptional regulation of IκBa resynthesis. J Biol Chem 1999; 274:939-47

86. Vertegaal AC, Kuiperj HB, Yamaoka S, et al. Protein kinase C-alpha is an upstreamactivator of the IκBa complex in the TPA signal transduction pathway to NF-κBin U2OS cells. Cell Signal 2000; 12: 759-68

87. Khoshnan A, Bae D, Tindell CA, et al. The physical association of protein kinaseC-theta with a lipid raft-associated inhibitor of κB factor kinase (IKK) complexplays a role in the activation of the NF-κB cascade by TCR and CD28. J Im-munol 2000; 165: 6933-40

88. Wang G, Siow YL, O K. Homocystein stimulates nuclear factor κB activity andmonocyte chemoattractant protein-1 expression in vascular smooth-musclecells: a possible role for protein kinase C. Biochem J 2000; 352: 817-26

89. Maiuri L, Raia V, De Marco G, et al. DNA fragmentation is a feature of cysticfibrosis epithelial cells: a disease with inappropriate apoptosis? FEBS Lett 1997;408: 225-31

90. Durieu I, Amsellem C, Paulin C, et al. Fas and Fas ligand expression in cysticfibrosis airway epithelium. Thorax 1999; 54: 1093-8

91. Gottlieb RA, Dasanjh A. Mutant cystic fibrosis transmembrane conductance reg-ulator inhibits acidification and apoptosis in C127 cell: possible relevance tocystic fibrosis. Proc Nat Acad Sci U S A 1996; 93: 3587-91

92. DeFranco AL. Macrophage signaling in response to bacterial lipopolysaccharides.Pediatr Pulmonol 1999; S19: 124

93. Prince A. Pulmonary pathogens activate Ca2+-dependent MAPK expression inrespiratory epithelial cells. Pediatr Pulmonol 1999; S19: 100-1

94. Laub MT, McAdams HH, Feldblyum T, et al. Global analysis of the genetic net-work controlling a bacterial cell cycle. Science 2000; 290: 2144-8

95. Chu S, DeRisi J, Eisen M, Mulholland J, et al. The transcriptional program ofsporulation in budding yeast. Science 1998; 282: 699-705

96. Wen X, Fuhrman S, Michaels GS, et al. Large-scale temporal gene expressionmapping of central nervous system development. Proc Nat Acad Sci U S A 1998;95: 334-9

97. van Someren EP, Wessels LF, Reinders MJ. Linear modeling of genetic networksfrom experimental data. Proc Int Conf Intell Syst Mol Biol 2000; 8: 355-66

98. Akutsu T, Miyano S, Kuhara S. Inferring qualitative relations in genetic networksand metabolic pathways. Bioinformatics 2000; 16: 727-34

99. Szallasi Z. Genetic network analysis in light of massively parallel biological dataacquisition. Pac Symp Biocomput 1999; 16 (2 Suppl.): 5-16

100. Wahde M, Hertz J. Coarse-grained reverse engineering of genetic regulatory net-works. Biosystems 2000; 55: 129-36

101. D’haeseleer P, Liang S, Somogyi R. Genetic network inference: from co-expres-sion clustering to reverse engineering. Bioinformatics 2000; 16: 707-26

102. Eisen MB, Spellman PT, Brown PO, et al. Cluster analysis and display of genome-wide expression patterns. Proc Nat Acad Sci USA 1998; 95: 14863-8

103. Alizaheh AA, Eisen MB, Davis RE, et al. Distinct types of diffuse large B-celllymphoma identified by gene expression profiling. Nature 2000; 403: 503-11

104. Perous CM, Sorlie T, Eisen MB, et al. Molecular portraits of human breast tumours.Nature 2000; 406: 747-52

105. Ross DT, Scherf U, Eisen MB, et al. Systematic variation in gene expressionpatterns in human cancer cell lines. Nat Gen 2000; 24: 227-35

106. Srivastava M, Eidelman O, Pollard HB. cDNA microarrays for pharmacogenomicanalysis of cystic fibrosis. Methods in Molecular Medicine 2001. In press

107. Zeitlin PL, Lu L, Hwang T-C, et al. A cystic fibrosis bronchial epithelial cell line:immortalization by adeno12-SV40 infection. Am J Respir Cell Mol Biol 1991;4: 313-9

Correspondence and offprints: Prof. Harvey B. Pollard, Department of Anat-omy, Physiology and Genetics, USU School of Medicine, USUHS, 4301 JonesBridge Road, Bethesda, MD, 20814, USA.E-mail: [email protected]

238 Eidelman et al.


Documents

Cystic Fibrosis and the Use of Pharmacogenomics to Determine Surrogate Endpoints for Drug Discovery