1. ¹Cancer Biology Group, Laboratory of Molecular Toxicology, National Institute of Environmental Health Sciences, Research Triangle Park, North Carolina, USA
2. ²Alpha-Gamma Technologies, Inc., Raleigh, North Carolina, USA
3. ³Laboratory of Cellular Carcinogenesis and Tumor Promotion, Center for Cancer Research, National Cancer Institute, Bethesda, Maryland, USA
4. ⁴Columbia University Medical Center, Department of Dermatology, New York, New York, USA
5. ⁵Laboratory of Signal Transduction, National Institute of Environmental Health Sciences, Research Triangle Park, North Carolina, USA
6. ⁶University of Pennsylvania School of Medicine, Department of Dermatology, Philadelphia, Pennsylvania, USA.
   E-mail: trempus@niehs.nih.gov

⁷Current address: GE Global Research, Niskayuna, New York, USA

TO THE EDITOR

Recent developments in animal models (Morris et al., 2004; Tumbar et al., 2004), as well as the discovery of cell surface markers (Jones and Watt, 1993; Tani et al., 2000; Trempus et al., 2003; Nijhof et al., 2006), have made it possible to isolate living epidermal hair follicle stem cells (HFSCs) from mouse skin, facilitating the study of the biological and molecular features inherent to HFSCs. A complexity of stem and progenitor cell populations within the hair follicle has been revealed. For example, the presence of discrete basal and suprabasal stem cells within the hair follicle bulge region was described recently (Blanpain et al., 2004), as well as a grouping of progenitor cells located above the bulge region distinguished by expression of the cell surface marker MTS24 (Nijhof et al., 2006). The cell surface protein CD34 has been shown to be uniquely expressed on stem and progenitor cells in the mouse hair follicle bulge region (Blanpain et al., 2004; Trempus et al., 2003), and the cell surface properties of CD34 has facilitated isolation of HFSCs for gene expression analysis (Blanpain et al., 2004). In addition, gene expression profiles have been generated from animal models developed to enrich for HFSCs (Morris et al., 2004; Tumbar et al., 2004). Although these datasets provided detailed insight into the complex biology of HFSCs, the scope was somewhat limited by the sensitivity of the microarray platform used to generate the profiles. Here, we report comprehensive profiling of mouse CD34-expressing HFSCs using the Agilent mouse oligo microarray platform to extend and enrich the existing HFSC databases. The Agilent-derived dataset was validated both by comparison to the Affymetrix-based expression profile of HFSCs isolated from the K15-eGFP transgenic mouse (Morris et al., 2004), and Gene Ontology biological process enrichment analysis. Additionally, to correlate expression patterns of genes with functional categories, HFSC datasets were clustered with datasets from hematopoeitic and neural stem cells.

Keratinocyte isolations were made as described previously (Trempus et al., 2003; Wu and Morris, 2005). All animal procedures were conducted in accordance with institutional guidelines governing the care and use of experimental animals. Gating of the sorted populations excluded the majority of the alpha6^lowCD34^high population (Blanpain et al., 2004). Total RNA was prepared from CD34(+) and CD34(-) keratinocytes obtained from three biological replicates, labeled with two different fluorescent dyes, and hybridized to the Agilent oligo microarrays containing 22,000 mouse genes and expression sequence tag probes (Supplementary Methods). The relative abundance of a particular mRNA was derived from the differential fluorescent signals, which identified 1,981 differentially expressed (DE) genes with statistical significance of P0.001 from all replicate experiments (Table S1; data is accessible at NCBI Gene Expression Omnibus, http:www.ncbi.nlm.nih.gov/geo/, accession number GSE7690). Confirmation of the microarray measurements was made of selected genes by real-time PCR (Supplementary Methods), as plotted in log 10 ratios (Figure 1, red circles; Tables S2 and S4).

Figure 1.

Validation of microarray measurements from CD34-expressing mouse hair follicle bulge stem cells. Log10 ratios of DE genes in CD34(+) versus CD34(-) keratinocytes are plotted against similarly transformed log 10 ratios determined by the Taqman assays (red circles), and log 10 ratios derived from the Affymetrix dataset by Morris et al. (2004) (blue diamonds). Linear regressions through (0, 0) point are shown as the dotted lines, with the indicated coefficients in red or blue fonts, respectively.

Full figure and legend (77K)

Because our analysis generated significantly more genes than other datasets, we were interested in determining the degree of individual gene and differential expression overlap to validate the quality of our data. For this comparison, the dataset generated by Morris et al. (2004), using the Krt1-15.eGFP mouse model (http://www.ncbi.nlm.nih.gov/geo/gds/gds_browse.cgi?gds=840), was re-analyzed using the Significance Analysis of Microarrays (SAM) software (Tusher et al., 2001). The SAM software calculates a set of gene-specific differential expression score using t-tests and estimates the false discovery rate of the gene set through permutations, which identified 752 Affymetrix probe sets from the Morris dataset at false discovery rate <10% (Supplementary Methods, Table S3). As shown in Figure 1, all but 4 of the 332 mapped genes through common Unigene clusters are concordant between the two datasets, with an R²=0.827. The high concordance of over 300 genes between the two studies (Table S4) lends confidence that the remaining genes on our list are of high quality and of potential biological significance, including the 800 genes (Table S5) absent from the Affymetrix arrays used in this comparison. Finally, 31 upregulated genes from the dataset published by Blanpain et al. (2004) are upregulated as well in our data (Table S1), providing additional confidence in our analysis.

To define further expression patterns of HFSC genes and their functional correlations, we compared differentially expressed genes in the HFSCs to microarray datasets from other adult stem cells, including neural stem cells and hematopoietic stem cells (Ramalho-Santos et al., 2002; Blanpain et al., 2004; Venezia et al., 2004) (Supplementary Methods, Table S6); and segregated the genes into six K-mean clusters (Eisen et al., 1998) (Supplementary Methods, Table S7). Enrichments of particular biological process annotations among the six K-mean clusters of genes were analyzed against the current Gene Ontology annotations (Supplementary Methods, Table S8, Figure S1) with the HT-GOMiner software (Zeeberg et al., 2005). The enrichment analysis compares the fraction of genes annotated to a biological process from the DE gene list to that on the whole chip, and calculates the random chance (i.e., false discovery rate) of the fraction in DE genes being derived from the whole chip, through permutations. Three gene expression clusters (Figure 2, clusters labeled nos. 1, 4, and 6) showed significant enrichments at false discovery rate <10% with 200 permutations, which are consistent with our current knowledge about HFSCs. The quiescent nature of adult HFSC and hematopoietic stem cell (Ramalho-Santos et al., 2002; Cotsarelis, 2006) is reflected by downregulated gene expression (Figure 2, cluster no. 1, 1P), which is highly enriched with cell-cycle-related processes. Upregulated genes in HFSCs are enriched mainly for development-related processes (Figure 2, cluster no. 4, 4P), including "negative regulation of Wnt receptor signaling", reported to play crucial roles in HFSC maintenance and epidermal differentiation (Alonso and Fuchs, 2003; Honeycutt and Roop, 2004) (Table S9). Cluster no. 6 contains genes that are downregulated in HFSC, unchanged in neural stem cell, but upregulated in hematopoietic stem cells. This expression pattern is highly enriched with the biological process of "antigen processing and presentation" (Figure 2, 6P), emphasizing the notion that the HFSCs are immune privileged (Christoph et al., 2000; Morris et al., 2004). Combining microarray datasets from diverse stem cell populations has therefore allowed us not only to derive gene expression patterns with significant enrichment of biological processes that are consistent with our current understanding of HFSCs, but also has identified three patterns (nos. 2, 3, and 5), which have not been well studied and annotated. Understanding the biological processes associated with these genes may advance our knowledge of HFSCs.

Figure 2.

Biological processes associated with stem cell gene expression patterns. The compiled log 10 ratios of 1,200 genes from various stem cell datasets, as described in Supplementary methods, were grouped into six gene expression patterns by K-means clustering (upper panel, Table S7). The lists of genes in each cluster were used for Gene Ontology Biological Process enrichment analyses using HT-GOMiner (Zeeberg et al., 2005) (Table S8). Three clusters (1, 4, and 6) showed significant enrichments (or over representation) of biological processes (1P, 4P, and 6P), whose statistical significances are represented by the intensities of the red-gray heatmaps.

Full figure and legend (529K)

In summary, we have shown that different isolation strategies and microarray platforms targeting the same biological population can yield highly concordant global gene expression profiles, as demonstrated by the remarkable similarity between the HFSC profiles described here. In addition, comparison between different adult HFSC populations has revealed expression patterns that cluster to characteristic biological processes that are similar between mouse (Figure 2) and human (Ohyama et al., 2007). Our data contribute 800 novel DE genes that were not present in the Affymetrix microarray used in other studies (Morris et al., 2004; Tumbar et al., 2004), and comparative analysis between stem cells of different tissue origin reveal novel functional patterns, opening up exciting possibilities for further molecular and functional analysis of HFSCs in mice.