Biomarkers of the Hedgehog/Smoothened pathway in healthy volunteers

Abstract

The Hedgehog (Hh) pathway is involved in oncogenic transformation and tumor maintenance. The primary objective of this study was to select surrogate tissue to measure messenger ribonucleic acid (mRNA) levels of Hh pathway genes for measurement of pharmacodynamic effect. Expression of Hh pathway specific genes was measured by quantitative real time polymerase chain reaction (qRT-PCR) and global gene expression using Affymetrix U133 microarrays. Correlations were made between the expression of specific genes determined by qRT-PCR and normalized microarray data. Gene ontology analysis using microarray data for a broader set of Hh pathway genes was performed to identify additional Hh pathway-related markers in the surrogate tissue. RNA extracted from blood, hair follicle, and skin obtained from healthy subjects was analyzed by qRT-PCR for 31 genes, whereas 8 samples were analyzed for a 7-gene subset. Twelve sample sets, each with ≤500 ng total RNA derived from hair, skin, and blood, were analyzed using Affymetrix U133 microarrays. Transcripts for several Hh pathway genes were undetectable in blood using qRT-PCR. Skin was the most desirable matrix, followed by hair follicle. Whether processed by robust multiarray average or microarray suite 5 (MAS5), expression patterns of individual samples showed co-clustered signals; both normalization methods were equally effective for unsupervised analysis. The MAS5- normalized probe sets appeared better suited for supervised analysis. This work provides the basis for selection of a surrogate tissue and an expression analysis-based approach to evaluate pathway-related genes as markers of pharmacodynamic effect with novel inhibitors of the Hh pathway.

Introduction

The Hedgehog (Hh) proteins are a family of secreted proteins that control cell differentiation, survival, and growth [1,2]. The receptor for Hh is Patched (PTCH), which is a 12-span transmembrane protein. Smoothened (SMO), which is a 7-span transmembrane receptor, serves as an obligatory signal transducing protein for the system. In the absence of the Hh ligand, PTCH catalytically inhibits SMO. Upon Hh binding, the PTCH-mediated suppression of SMO is released. Activation of SMO then triggers a series of intracellular events, ultimately leading to the stabilization of GLI zinc-finger transcription factors (GLI1, GLI2, and GLI3) with the subsequent expression of GLI-dependent genes, including GLI and PTCH.

The understanding of the Hh pathway and its aberrations in cancer is important for the rationale design of potential anticancer therapies. Inappropriate signaling of the Hh pathway has been implicated in the development of several cancers, including breast, lung, prostate, and pancreas [3]. Cyclopamine, which was the first described Hh inhibitor, binds to and inactivates SMO [4,5]. Because cyclopamine has poor oral availability, suboptimal pharmacokinetics, and low affinity, other inhibitors have been sought [3]. Several synthetic, small molecule inhibitors targeting various components of the Hh pathway are being tested in preclinical and early phase trials in humans [3,6]. Vismodegib, a Smo antagonist was recently approved by Food and Drug Administration for the treatment of advanced metastatic basal cell carcinoma [7]. Additionally, monoclonal antibodies, which prevent binding of Hh to PTCH, are under development [3,6]

During testing of molecules that target components of signal transduction, pharmacodynamic markers are essential for confirming that the study drug is affecting the desired pathway. In addition to defining markers, the best tissue and analytical conditions for measuring these markers must be determined. The primary objectives of this study were to measure levels of mRNA markers of the Hh pathway in skin biopsies, hair follicles, and peripheral blood mononuclear cells (PMBCs), and to identify markers that are consistently expressed in normal tissues at levels such that a measurable change (decrease) produced by a pathway specific inhibitor would indicate a pharmacodynamic effect.

Methods

Patient eligibility

Overtly healthy men and women ≥18 years of age were eligible for this study. Patients with stable medical problems were also eligible if, in the opinion of the study investigator, those conditions would not place the patients at increased risk or interfere with data interpretation. Subjects must have had healthy skin that was amenable to punch biopsies. Subjects were ineligible to participate if they met any of the following exclusion criteria: history of bleeding diathesis or inflammatory disorders of the skin; a requirement for long-term anticoagulation or antiplatelet therapy; use of an aspirin-containing product within the past 14 days or a nonsteroidal anti-inflammatory within the previous 72 hours. Subjects were also excluded if, within the last 30 days, they had enrolled in or discontinued from a clinical trial involving an investigational drug or device or off-label use of a drug or device.

The center’s institutional review board approved the protocol in accordance with an assurance filed with and approved by the US Department of Health and Human Services. This study was conducted in accordance with good clinical practices and the Declaration of Helsinki. Healthy volunteers provided written informed consent prior to undergoing study procedures. This research was conducted at Dawes Fretzin Clinical Research (Indianapolis, IN).

Study objectives and statistical considerations

The planned enrollment was approximately 30 subjects to ensure that approximately 25 high-quality samples would be obtained. This sample size was not based on statistical considerations.

The primary objective was to obtain samples suitable for measurement of mRNA markers of the Hedgehog-Smoothened pathway in skin biopsies, hair follicles, and PBMCs. Secondary objectives included the following: 1) compare the quantitative real-time polymerase chain reaction (qRT-PCR) data for selected genes across the 3 tissue types and select the appropriate sample type for further studies; 2) query microarray data from matched sample sets for correlation with selected gene qRT-PCR data for trend verification; 3) evaluate extracted RNA for measurement of expression using qRT-PCR and for measurement of signal intensity using custom microarrays with robust multiarray average (RMA) and microarray suite 5 (MAS5) data normalization; and 4) evaluate microarray data using signal intensity levels and coefficients of variation to find other Hh-pathway-related genes not included in the candidate gene set.

Tissue collection and handling

A maximum of 20 mL of blood was collected by venipuncture, and 10-16 hair follicles were extracted from each subject. Two adjacent (at least 1 cm apart) skin biopsies were taken from the iliac crest using a standard skin punch of approximately 3 mm. For the biopsies, a standard skin preparation was used, and the investigator or a designee provided follow-up care of the biopsy site as needed. All samples were collected on the same day prior to 12 PM.

RNA extraction

Whole-blood samples were collected in PAXgene tubes (PreAnalytX, Hombrechtikon, Switzerland). Skin punches and hair follicle specimens (10 to 16 hairs with a visible bulb) were collected and preserved in RNAlater® (Applied Biosystems/Ambion, Austin, TX). Ribonucleic acid (RNA) was extracted from all collected samples according to procedures recommended by the reagent manufacturer. Samples were prioritized on the basis of RNA quantity and quality for qRT-PCR; remaining samples with adequate mass and quality were processed for microarray analysis.

Quantitative RT-PCR assays

The TaqMan^® (Life Technologies/Applied Biosystems, Carlsbad, CA) assay was performed on RNA derived from skin, blood, and hair using standard procedures and commercially available primers and probes. Thirty one (26 target and 5 potential normalizing) genes were analyzed using the ABI 7900 RT-PCR instrument (Life Technologies/Applied Biosystems, Carlsbad, CA). Data were reported as cycle threshold (Ct), which corresponds to the number of cycles required to observe a predefined signal intensity, and delta Ct (ΔCt) values, calculated as the difference between target gene Ct and a single normalizing gene Ct. Analysis was performed with duplicate reverse transcriptase reactions and triplicate PCRs. This required a minimum of 500 ng of total RNA per sample. For samples with less than 500 ng total RNA yield, a core set of transcripts, consisting of 5 Hh pathway-related gene transcripts and 2 normalizing gene transcripts, were selected for qRT-PCR-based analysis.

Further, for each of the potential normalizing genes, the stability score for that gene was obtained by averaging all pairwise variations between that gene and other potential normalizing genes. The genes were then ranked according to their stability scores in increasing order in Table 1 (genes with low stability scores were more stably expressed) [8].

Table 1.

Evaluation of the expression stability score for each tested housekeeping gene

Rank	Whole Blood	Stability Score	Hair Follicles	Stability Score	Skin	Stability Score	Combined	Stability Score
1	PPIA	0.351	HPRT	2.269	HPRT	0.543	HPRT	1.661
2	RPLP0	0.364	PGK1	2.605	PPIA	0.571	PGK1	1.952
3	GAPDH	0.371	PPIA	2.925	GAPDH	0.597	PPIA	2.034
4	HPRT	0.420	RPLP0	3.104	PGK1	0.606	GAPDH	2.470
5	PGK1	0.434	GAPDH	3.825	RPLP0	1.146	RPLP0	2.556

Note: For each sample, the most stably expressing gene is rank ordered with its corresponding stability score. The combined stability score was calculated for all three sample types and rank order from the most stable to the least. Scores were calculated using SAS 9.2 (SAS Institute Inc, Cary, NC). Abbreviations: GAPDH = glyceraldehydes-3-phosphate-dehydrogenase; HPRT = hypoxanthine guanine phosphoribosyl transferase; PGK1 = phosphoglycerate kinase 1; PPIA = peptidyprolyl isomerase A (cyclophilin A); RPLP0 = ribosomal protein, large, PO, ribosomal protein PO-like.

Global gene expression analysis using RNA microarray

Samples with a minimum of 100 ng of total RNA with acceptable quality (RNA integrity number >6 on a scale of 10) remaining after qRT-PCR analysis were analyzed for global gene expression. Initially, multiplexed microarray analysis of a subset of individuals with matched skin, hair, and whole-blood RNA was used to compare the relative presence of markers of important signaling pathways identified in cancer, including the Hedgehog/Smoothened pathway. Samples were analyzed using a Affymetrix Human U133 microarray (part #510681, Affymetrix, Santa Clara, CA).

Microarray data analysis and comparison with qRT-PCR data

Microarray data were used to determine the preferred normalization algorithm, either RMA or MAS5, for future studies. After the microarray data were normalized using RMA- and MAS5-processed algorithms, the data were log₂-transformed and the mean value was subtracted from each gene across samples. Probe set signal intensities for the candidate genes that were determined either from RMA or from MAS5 normalized array data were then correlated with qRT-PCR data. Where more than 1 probe set corresponded to a gene, each signal value was used for correlative analysis.

In correlating with microarray data, the TaqMan qRT-PCR ΔCt values were normalized to a single gene, GAPDH as it is a commonly used “housekeeping” (normalizing) gene for data normalization across tissue types [9]. The TaqMan and microarray data were compared by computing probe set signal correlations across samples between the corresponding transcript values from TaqMan measurement. For each TaqMan gene, corresponding probe sets were located on the microarray; typically, multiple probe sets on the microarray corresponded to the same gene in the TaqMan assay. The Pearson correlation and corresponding P-value of TaqMan data were computed for each individual probe set/transcript corresponding to the same gene on the microarray. The data were fitted to a linear regression model to identify the slope, which elucidated the dynamic range of transcript expression. Individual correlation plots having a P<0.01 in either TaqMan/RMA or TaqMan/MAS5 correlations were displayed. Tab-delimited text files were then generated to summarize all correlation P-values as well as the presence or absence statistic for a given gene on the array. After the initial multiplexed microarray analysis, expression and variability of expression of selected genes in skin and hair follicles were determined. Ninety-one genes were identified on the areas, selected on the basis of published association with Hh pathway signaling. Data normalized by MAS5 and RMA algorithm were used to process the microarray values for skin and hair follicles. These data were log₁₀ transformed, and for each gene, the average log₁₀ (intensity), corresponding standard deviation (SD) across subject profiles, and percent (%) coefficient of variation (CV) were computed. To identify high-expressing genes with low variability, the data were ordered by the CV in increasing fashion.

The Hh pathway-related genes were subjected to functional enrichment analysis using molecular function gene ontologies. Open source bioinformatics tools BiNGO [10] and Cytoscape [11] were respectively used for determining gene set enrichment and data visualization. Hypergeometric distribution method for P-value calculation was used to find over represented gene-ontology terms. The P-value significance threshold was set at 0.01 in order to ensure that genes found to be functionally enriched in their corresponding molecular function class are not picked purely by random chance. The P-value of the significantly enriched molecular function categories was further corrected by adjusting for multiple testing based on Benjamini and Hochberg false discovery rate correction method [12].

Results

RNA extraction and comparative qRT-PCR

A total of 30 healthy subjects, 7 male and 23 female, between the ages of 21 and 59 years participated in this study. Matched skin, hair follicles, and whole-blood samples were acceptable for analysis for 98 specimens (22 hair follicles, 48 skin punches and 28 blood specimens); of these, 10 specimens (8 hair follicle, 1 skin and 1 blood) did not yield enough RNA or failed quality for subsequent analysis of all genes using qRT-PCR and microarray. Samples with greater than 1.5 ug total RNA yield were evaluated by qRT-PCR for the expression of 26 Hh pathway-related genes derived from preclinical animal experiments and known Hh pathway genes (manuscript in preparation) and 5 normalizing genes. Specimens with less than 100 ng of total RNA were analyzed for expression of a core set of 5 Hh pathway-related genes and 2 normalizing genes. Twelve sample sets analyzed by qRT-PCR were also analyzed for global gene expression using the Affymetrix microarray platform (Affymetrix, Santa Clara, CA).

The expression of the 5 normalizing genes was initially evaluated to find the most stably expressed gene across all sample types. Table 1 shows the evaluation of the expression stability score for each potential normalizing gene. The genes are rank ordered for each sample type with PPIA the most stable in blood and HPRT the most stable in corresponding hair follicles and skin. The combined stability score across sample types determined that HPRT was the most stably expressed gene (Table 1). The relative expression of each selected Hh pathway gene to the corresponding most stable gene was subsequently evaluated for each sample type and HPRT, the most stable gene across all three sample types.

Data summarized in Table 2 shows that, relative to the normalizing gene PPIA, the 3 most abundant target gene transcripts in blood were MYC (0.9±0.55), HSPB1 (2.5±0.64), and BCL2 (3.0±0.38). Relative to the normalizing gene HPRT, the three most abundant target gene transcripts in hair were HSPB1 (-8.5±0.40), MYC (-3.0±0.41) and SFRP1 (-2.8±1.36), whereas in skin they were HSPB1 (-7.5±0.42), MYC (-3.2±0.40) and SFRP2 (-4.7±0.85) Skin samples had the highest level of expression of the tested transcripts as depicted by the most negative numbers when normalized to HPRT expression. The known Hh pathway-associated genes, PTCH1, PTCH2, GLI1, GLI2, and SMO, were also notably present in skin relative to hair follicles and blood (except for GLI2 in blood). GLI1 and GLI2 were only clearly present in skin, as indicated by lower mean Ct values. These data supported the choice of skin as the most appropriate normal tissue surrogate for assessment of Hh pathway gene expression in human studies.

Table 2.

Relative expression of selected hedgehog pathway candidate genes in corresponding normal blood, skin, and hair follicles

Assay ID	Gene ID	Gene Name	Type	AMP Length	Blood Mean Ct (SD)	Blood (Mean-PPIA) (SD)	Blood (Mean-HPRT) (SD)	Hair Mean Ct (SD)	Hair (Mean-HPRT) (SD)	Skin Mean Ct (SD)	Skin (Mean-HPRT) (SD)
Hs99999018_m1	BCL2	Oncogene B-cell leukemia 2	Target	76	29.3 (0.65)	3.0 (0.38)	−0.9 (0.6)	35.0 (1.02)	3.2 (0.84)	30.9 (0.67)	−0.1 (0.47)
Hs99999004_m1	CCND1	Cyclin D1	Target	79	36.0 (0.59)	9.9 (0.46)	5.9 (0.77)	30.9 (2.27)	−1.1 (1.43)	28.7 (0.94)	−2.3 (0.69)
Hs00176469_m1	CDC2	Cell division cycle 2, G1 to S and G2 to M	Target	101	35.1 (0.72)	8.8 (0.67)	4.8 (0.68)	33.3 (1.76)	1.2 (0.71)	33.0 (0.99)	1.9 (0.37)
Hs00178327_m1	EPHA3	EPH receptor A3	Target	66	nd	nd	nd	nd	nd	33.7 (0.96)	2.6 (0.64)
Hs99999905_m1	GAPDH	Glyceraldehydes-3-pho sphate-dehydrogenase	Control	122	24.2 (0.52)	−2.0 (0.39)	−6 (0.28)	25.5 (1.17)	−6.7 (0.62)	24.7 (0.96)	−6.3 (0.28)
Hs00171790_m1	GLI1	GLI family zinc finger 1	Target	80	34.8 (0.62)	8.5 (0.60)	4.6 (0.67)	33.0 (1.25)	10.8 (15.67)	31.5 (0.99)	0.5 (0.52)
Hs01119974_m1	GLI2	GLI family zinc finger 2	Target	61	nd	nd	nd	32.8 (1.96)	10.0 (15.59)	30.8 (0.70)	−0.2 (0.42)
Hs01112674_m1	GPR153	G protein-coupled receptor 153	Target	77	33.6 (0.80)	7.3 (0.80)	3.3 (0.78)	32.5 (1.83)	0.3 (0.69)	30.0 (0.92)	−1.0 (0.46)
Hs01114113_m1	HEY1	Hairy/enhancer-of-split related with YRPW motif 1	Target	82	33.9 (1.08)	7.6 (1.17)	3.6 (1.04)	34.2 (1.6)	2.2 (0.59)	32.1 (0.92)	1.1 (0.42)
Hs01011009_m1	HHIP	Hedgehog interacting protein	Target	91	nd	nd	nd	31.7 (1.31)	10.0 (15.60)	35.4 (0.88)	4.6 (0.82)
Hs01003267_m1	HPRT	Hypoxanthine guanine phosphoribosyl transferase	Control	72	30.3 (0.60)	4.0 (0.43)	0 (0.00)	32.3 (1.53)	0.0 (0.00)	31.1 (0.86)	0.0 (0.00)
Hs00356629_g1	HSPB1	Heat shock 27kDa protein 1	Target	66	28.8 (0.46)	2.5 (0.64)	−1.4 (0.50)	23.8 (1.51)	−8.5 (0.40)	23.6 (0.82)	−7.5 (0.42)
Hs00426289_m1	IGFBP3	Insulin-like growth factor binding protein 3	Target	84	33.0 (0.97)	6.7 (0.70)	2.7 (0.99)	34.3 (1.00)	2.9 (0.95)	28.8 (0.93)	−2.2 (0.65)
Hs01081800_m1	IHH	Indian hedgehog homolog	Target	70	nd	nd	nd	nd	nd	nd	nd
Hs99999003_m1	MYC	v-myc myelocytomatosis viral oncogene	Target	65	27.2 (0.61)	0.9 (0.55)	−3.1 (0.31)	29.2 (1.29)	−3.0 (0.41)	27.9 (0.85)	−3.2 (0.40)
Hs00232074_m1	MYCN	v-myc myelocytomatosis viral related oncogene	Target	81	35.6 (1.00)	9.9 (0.38)	5.2 (1.20)	32.5 (1.33)	0.2 (1.38)	34.2 (0.89)	3.1 (0.57)
Hs00234994_m1	PDGFA	Platelet-derived growth factor alpha polypeptide	Target	93	31.5 (0.76)	5.2 (0.98)	1.3 (1.11)	32.4 (1.33)	0.2 (0.68)	30.5 (0.80)	−0.6 (0.33)
Hs00998018_m1	PDGFRA	Platelet-derived growth factor receptor, alpha	Target	84	35.5 (1.04)	9.2 (1.00)	5.2 (1.30)	35.1 (0.68)	3.8 (1.14)	28.5 (0.77)	−2.6 (0.55)
Hs99999906_m1	PGK1	Phosphoglycerate kinase 1	Control	75	26.6 (0.63)	0.3 (0.34)	−3.6 (0.58)	29.1 (1.3)	−3.2 (0.88)	28.7 (0.92)	−2.4 (0.41)
Hs99999904_m1	PPIA	Peptidyprolyl isomerase A (cyclophilin A)	Control	98	26.3 (0.59)	0.0 (0.00)	−4 (0.43)	27.3 (1.61)	5.7 (16.29)	26.0 (0.84)	−5.1 (0.37)
Hs00181117_m1	PTCH1	Patched homolog 1	Target	72	31.6 (0.57)	5.3 (0.52)	1.4 (0.65)	31.2 (0.66)	9.5 (16.07)	31.6 (0.73)	0.6 (0.41)
Hs00184804_m1	PTCH2	Patched homolog 2	Target	76	35.8 (0.51)	9.5 (0.29)	5.6 (0.44)	33.1 (1.44)	11.5 (15.96)	32.9 (0.93)	1.8 (0.55)
Hs00420895_gH	RPLP0	Ribosomal protein, large, PO, ribosomal protein PO-like	Control	76	24.9 (0.60)	−1.4 (0.25)	−5.4 (0.39)	24 (0.88)	2.3 (15.94)	24.3 (1.80)	−6.1 (6.21)
Hs03928965_s1	SFRP1	Secreted frizzled-related protein 1	Target	99	32.0 (2.78)	5.6 (2.83)	1.6 (2.63)	29.3 (2.14)	−2.8 (1.36)	28.8 (1.05)	−2.2 (0.53)
Hs00293258_m1	SFRP2	Secreted frizzled-related protein 2	Target	129	31.4 (1.28)	5.1 (1.47)	1.2 (1.65)	34.5 (1.41)	2.6 (1.80	26.3 (1.02)	−4.7 (0.85)
Hs00179843_m1	SHH	Sonic hedgehog homolog	Target	70	nd	nd	nd	33.1 (0.67)	2.2 (0.98)	36.1 (0.66)	5.8 (0.48)
Hs00170665_m1	SMO	Smoothened homolog	Target	96	36.4 (0.13)	10.7 (0.73)	5.4 (0.43)	33.4 (1.88)	1.4 (1.40)	31.4 (1.04)	0.3 (0.62)
Hs00195591_m1	SNAI1	Snail homolog 1	Target	66	35.7 (0.48)	9.6 (0.44)	5.6 (0.63)	36.2 (0.32)	4.2 (1.76)	33.9 (0.81)	2.9 (0.54)
Hs01057642_s1	SOX1	SRY (sex-determining region Y) – box 1	Target	96	33 (1.30)	6.6 (1.49)	2.7 (1.24)	nd	nd	nd	nd
Hs00165814_m1	SOX9	SRY (sex-determining region Y) – box 9	Target	102	nd	nd	nd	30.8 (2.92)	−1.4 (1.5)	31.5 (1.27)	0.4 (0.65)
Hs00183662_m1	WIF1	wnt inhibitory factor 1	Target	72	nd	nd	nd	32.4 (1.91)	0.3 (1.90)	31.2 (1.03)	0.1 (0.95)

Note: Data were compiled from replicate qRT-PCR expression values and mean Ct and ΔCt values relative to the most stable housekeeping control gene for each sample type is shown. Normalization was performed using R v2.14.2 (http://www.r-project.org). Abbreviations: AMP = amplicon; Ct = cycle threshold; HPRT = Hypoxanthine guanine phosphoribosyl transferase; nd = not detected above detection threshold; PPIA; peptidyprolyl isomerase A (cyclophilin A); qRT-PCR = quantitative real time polymerase chain reaction; SD=standard deviation.

Global gene expression microarray analysis

Twelve sample sets, each with adequate RNA derived from hair, skin, and whole blood, were analyzed for global gene expression using Affymetrix U133 microarrays (Affymetrix, Santa Clara, CA). Hierarchical clustering of the expression signals of processed gene expression values for each sample type is shown in Figure 1. Individual samples were processed using either RMA or MAS5 algorithms to estimate potential variability associated in overall expression pattern in divergent samples. This comparative analysis showed that, generally for each sample type, the RMA and corresponding MAS5 signals co-clustered (see left side of maps), indicating that the microarray analysis accurately represented a wide range of signals corresponding to both low- and high-abundance transcripts. However, MAS5-processed signals overall had a broader dynamic range, as indicated by the range of color in the heat map pattern. There appear to be several clusters for each sample type on the vertical axis in Figure 1; for example, blood (A) has 2 main clusters, whereas hair follicles (B) and skin (C) have multiple small clusters perhaps indicating biological variability from many contributing genes.

Figure 1.

Hierarchical clustering of MAS 5.0 and RMA normalized global expression signals from corresponding blood (A), hair follicles (B) and skin (C). Each row represents a color coded range of normalized signal values corresponding to probe sets for a gene. The number of represented probe sets and their relative proximity for each sample type is represented on the x-axis. The y-axis represents the sorted nearest neighbor based on the clustering algorithm such that closely related samples are in proximity and further removed from distantly related samples. Abbreviations: MAS 5.0 = Microarray Suite 5.0; RMA = Robust Multiarray Average.

Association between TaqMan qRT-PCR data and corresponding RMA- and MAS5-Normalized probe set and gene level data

The correlation was determined between candidate gene quantitative expression measurements and corresponding global expression array signal at the probe set level. For each array probe set corresponding to each gene and real-time qRT-PCR expression level (TaqMan), correlations between MAS5 and RMA signal intensity were estimated using the Pearson correlation coefficient for probe set level correlation (Figure 2). The probe set correlations for MAS5 microarray/TaqMan versus RMA microarray/TaqMan for whole blood (Figure 2A), hair follicle (Figure 2B), and skin (Figure 2C) show overall acceptable association (R-value range from 0.61-0.73 for blood, 0.60-0.77 for hair and 0.56-0.69 for skin). However, the number of data points in each sample type and its proximity to the simulated 45 degree line is indicative of presence of an acceptable signal value from the array data and the presence of a Ct value below the set threshold for qRT-PCR measurement. On the basis of these considerations, skin had the most valuable data points and was also the tissue type in which the qRT-PCR and corresponding array signals were most highly correlated. By the same measures, hair follicle sample type was the least correlated (Figure 2B).

Figure 2.

Correlation of the TaqMan derived expression Ct values of individual genes with corresponding probe set signals normalized with MAS 5.0 and RMA for blood (A), hair follicles (B) and skin (C). Each point represents an association with one or more probe sets depending upon the gene compared for each sample set. Abbreviations: Ct = cycle threshold; MAS 5.0 = Microarray Suite 5.0; RMA = Robust Multiarray Average.

Gene annotation for additional Hh pathway markers from global gene expression analysis

To assess other potential markers with high baseline levels and low variability but not included in the “candidate gene” set, a set of 91 genes was selected from literature data mining on the basis of a previously reported association with the Hh pathway as well as their inclusion in commercially available pathway mapping expression cards (Origene). Plots of probe set signal intensity versus %CV were generated for corresponding RMA- and MAS5-normalized probe set data for the 91-gene set. Since expression measurement in skin was the most correlated for qRT-PCR and microarray analysis of the 26 Hh pathway-related genes, the search for additional markers was restricted to the data obtained from skin samples (Figure 3). These plots indicated that MAS5-normalized probe sets were better correlated than RMA-normalized probe sets for selecting possible additional markers, primarily because the data sets have lower variability associated with higher signal intensity as seen in the top left corner of each plot. It was rationalized that the most appropriate candidates would likely represent those probe sets that had the highest signal values and the lowest %CV. The Gene Ontology tool (GO) was used to evaluate the probe set signals representing the 91-gene set for molecular function enrichment. Data shown in Table 3 shows these additional markers that were selected through data mining. Details of the gene lists and P-values are provided in Supplemental Tables.

Figure 3.

Distribution of probe set intensities and percent signal coefficient of variance (%CV) for 91 Hedgehog pathway genes using MAS 5.0 and RMA normalized data for skin. For each sample type and microarray data normalization method the top left corner represents the region with highest expression signal and lowest %CV for selecting potential Hedgehog pathway activation markers. Panel A, RMA; Panel B, MAS 5.0. Abbreviations: CV = coefficient of variation; MAS 5.0=Microarray Suite 5.0; RMA= Robust Multiarray Average.

Table 3.

Additional Pharmacodynamic Markers from Array Data Mining

Interactor Gene	mas5_mean_log intensity	mas5_std	mas5_cv	Np
PTCH1	3.061	0.06791	2.219	100
NPC1	2.855	0.06223	2.18	100
DISP1	2.42	0.0681	2.814	100
PTCH2	1.403	0.2193	15.64	21.74
C6ORF138	0.7896	0.2946	37.32	8.696
NPC1L1	0.7867	0.2967	37.71	13.04
PTCHD3	0.7235	0.4579	63.29	0

Abbreviations: C6ORF138 = chromosome 6 open reading frame; DISP1=dispatched homolog 1; mas5_cv = % coefficient variation for Microarray Suite5 (mas5), defined as 100 * mas5_std / mas5_mean_logintensity; mas5_mean_log intensity = average log10 (intensity) for a transcript across samples for data processed with Microarray Suite5; mas5_std = standard deviation of log10 (intensity) of transcript across samples for data processed with Microarray Suite5; Np = percent value indicating how many times the transcript was present with p < 0.01 across samples; NPC1L = Flotillin 1; PTCH1 = Patched homolog 1; PTCH2 = Patched homolog 2; PTCHD3 = patched domain containing 3.

Safety

No deaths or other serious adverse events occurred during this study.

Discussion

This study sought to evaluate methods for collection, handling, and extraction of hair follicle, skin, and blood sample RNA from healthy volunteers to quantitatively measure Hh/Smo pathway-related gene transcripts and to create a database of global transcripts using gene expression microarrays. The specific sample collection procedures, matched to sample type, were adequate for providing acceptable RNA for analysis. In the case of hair follicle samples, the procedures produced less-than-optimal yield, and several samples lacked material to conduct both proposed RNA analyses (qRT-PCR and cDNA-probe hybridization).

Tissue samples collected in RNA stabilization reagents were of high quality; however, without mechanical disruption, hair follicle specimens resulted in lower-than-expected yields. Quantitative RT-PCR measurement of pathway-related transcripts provided the data necessary, based on the levels and variability of SHH, GLI1/2, HHIP, and PTCH1/2, to recommend skin and potentially hair sampling and to exclude blood sampling in future studies (Ct range, 28 to 34).

Treatment-associated reduction in relative copy number was determined to be a reliable measure of a pharmacodynamic effect using these genes, particularly from skin. Treatment is expected to decrease Hh pathway-related signaling and corresponding expression of pathway-associated proteins; therefore, the selected pretreatment Ct range should allow reliable measurement of transcript reduction such that each unit increase in Ct represents a 2-fold reduction in transcript copy number when the efficiency of the reaction is 100%. In this regard, gene transcripts with low CTs and low variability, as expressed by CV, are preferred as potential pharmacodynamic markers.

The collection of microarray data was particularly informative for assessment of data normalization methods and demonstrated that both RMA and MAS5 are equally effective for unsupervised analysis, but supervised analysis (in which only a subset of the data for selected genes was compared) produced better correlation with the MAS5 subset. These correlation patterns were particularly evident for skin and hair samples; in both cases, the MAS5-normalized data is better aligned than the RMA-normalized data.

In conclusion, skin was the most desirable matrix for expression assessment followed by hair follicle, whereas whole blood lacked several key Hh pathway-related genes (Table 2). Microarray analysis provided the basis for evaluation of data analysis tools of global transcript data and the best potential for new biomarker discovery. Finally, no safety issues emerged in this non-drug interventional study in healthy volunteers. This work provides the necessary tools needed to assess the pharmacodynamic effect of novel anticancer agents that inhibit the Hh pathway.