Cell Culture and Cell Sorting
RNA yields from SP and nSP cells sorted directly from freshly trypsinized cell suspensions were only 0.1 to 0.4 pg/cell and their RNA quality was extremely poor (Wolosin JM, unpublished, 2006). This result may be due to the prolonged trypsinization needed to dissociate this epithelium, compounded by the inclusion of the anchoring-dependent epithelial cells in suspension. To surmount this intrinsic barrier to the molecular characterization of SP cells of solid tissues, we cultured the conjunctival cells at <100,000 cells/cm2 before cell sorting. Costaining for total keratin and keratin 14 after a 12-hour culture showed that approximately 97% of the adherent cells were bona fide epithelial (pan cytokeratins positive) cells and approximately 70% were positive for the basal cell keratin 14 (not shown). The effect of the time in culture on SP cell abundance was investigated using both rabbit and human conjunctivas (). Based on preliminary investigation, the analysis was performed using light-scattering gates that exclude the majority of cells of nonepithelial lineages.
FIGURE 1 Presence of SP cells in cultured conjunctival epithelium as a function of time in culture and medium. Rabbit cells after 16 (A), 40 (B, D), and 64 (C) hours of culture in SHEM or 40 hours in base medium (E). Human cells after 16 (F), 40 (G), and 64 ( (more ...)
In the rabbit, SP cells, ~1.6% of adherent cells at 16 to 24 hours (average for five independent experiments), decreased rapidly afterward and were essentially null by the third day (). The decreases appear to be linked to cell proliferation: when cells were seeded in serum and growth factor–free medium (D/F12 and 1% BSA) SP yields were preserved (). The SP yield for human cells cultured in SHEM, however, remained constant for the first 48 hours (). Cell cycle analysis of rabbit SP cells (the number of cells needed to complete this two-stage experiment preempted its completion with human cells), isolated from a pool of six rabbit conjunctivas, showed that at the time of collection, at least 99% of the SP cells (two experiments) were in G0/G1 (). Hence, the nSP was taken from the center of the G0/G1 Hoechst image as indicated in .
Cycling status of rabbit CNJE cells.
Adherent cell yields per human conjunctival pair (single donor) were typically 2 to 3 million cells. The light-scattering gate-restricted SP cohort accounted for approximately 0.3% to 0.5% of the human conjunctival epithelial adherent single cell population, or 6,000 to 12,000 SP cells. The average RNA yields for the sorted cells (n = 4) were 2.6 and 2.4 pmol/cells for the SP and nSP populations, respectively or between 15 to 30 ng/per experiment. These limited quantities of RNA hindered measurements of RNA purity or origami integrity. This problem was particularly important in this study because the need to trypsinize adherent cells and maintain them in suspension until and during sorting could cause some RNA degradation. Hence, we relied on the quality control probes included in the microarray (HG-U133A plus 2.0; Affymetrix) to generate a robust retrospective analysis of the effect of the cellular processing used on the final quality of the microarray results.
The microarray contains three distinct probes for glyceraldehyde-3 phosphodehydrogenase (GAPDH), each representing distinct sections of the gene in the 5′ to 3′ (polyA) direction. A fourfold decrease in the SI for the 5′ end probe (1000 bp away from the polyA) relative to the 3′ end probe () suggests that some degradation occurred in the course of cell sorting. Nevertheless, such a partial degradation is likely to have had minimal impact on our comparative analysis for two reasons: The degree of degradation was similar for both SP and nSP samples so that the SI ratios for probes that are as much as 1000 base pairs away from the polyA site is the same as for the 3′ end probes, and in fact, unlike the case for GAPDH, where probes spanning the whole length of the message were intentionally included in the microarray for quality control, for nearly all the genes, the microarray includes at least one probe derived from the polyA-proximal (3′ end) of the sequence. RNA equivalence is also supported by the similar percentage of present calls in the two populations ().
Finally, an examination of SI values and P/A calls for nonepithelial markers (), demonstrated that the collected SP and nSP populations were essentially free of the three nonepithelial cells most commonly found within the conjunctival epithelial strata, hematopoietic (CD45), antigen presenting dendritic (CD1), and melanin-producing (tyrosinase) cells. Conversely, the average levels of expression for all 24 intermediate filament genes expressed in the tissue were very similar in the two populations.
Global Differential Gene Expression Analysis
The AE gene list contained 10,265 unique genes or approximately 40% of all the unique annotated genes represented in the array. Of these 10,265 AE genes, 1,617, or approximately 16%, were differentially expressed. This high percentage indicates that the SP cells are markedly different from the generic G0/G1 basal cells represented by the nSP cohort. Of the 1617 genes, 1254 were SP overexpressed and 363 underexpressed (i.e., a more than 3:1 ratio). This large asymmetry can be visualized in a plot of SI values (average for the SP and nSP) versus log2 of the ratio between the SP and nSP SIs for the whole AE list (, main frame). The SP cells seem to overexpress a very extensive set of genes of low to moderate SI. Since the microarray results were normalized to identical total expression (sum of SIs for SP = sum of SI for nSP), the excess expression in the SP decreased the relative weight of high-expression genes, mostly invariant housekeeping genes. As a result, these genes appeared to have SP/nSP ratios below 1 (e.g., GAPDH, see ). The equivalent plot for the nonpresent transcripts (i.e., mostly random false-positives) was essentially symmetric around the y-axis (, inset), indicating that the expression asymmetry did not represent an artificially generated bias.
FIGURE 3 Plot of log2 of SP/nSP ratios versus signal intensity (SI). Main frame. Expressed transcript set. Note that there are many more transcripts in the >+1 (ratio, 2) than in the >−1 range. Left side inset: nonexpressed transcript set. (more ...)
The microarray expression ratios for 10 genes were subjected to independent validation by qPCR (). The small amount of RNA collectable per experiment made it necessary to preamplify the cDNA. Given the implicit risk of nonlinear amplification, in particular since the amount of RNA imputed was well below the amount suggested by the manufacturer, the PCR results cannot be viewed as truly quantitative. Nevertheless, the qualitative concordance obtained provides general validation of the microarray ratios.
Comparison of SP/G1 Expression Ratios Calculated from a qPCR Experiment and from the Microarray Study
displays the most prominent SP overexpressed and underexpressed genes. The lists include genes involved in multiple unrelated cellular activities. No single central or dominant cellular function is readily identifiable, but there are many intriguing genes for future analysis and more detailed delineation of the SP cell phenotype.
Genes for Which SP/nSP or nSP/SP SI Ratios Are Larger Than 10
Cell Surface Markers
One of the most valuable results from a differential global gene expression study is identification of exclusive cell surface markers, as such markers facilitate minimally intrusive tracking and isolation of the cell of interest in the viable state. The gene with the highest SP/nSP SI ratio represented CD62E/E-selectin (). The presence of this gene in the CNJE was surprising. CD62E, a cell surface protein involved in leukocyte adhesion, is thought to be expressed only in activated vascular endothelia.17
Nevertheless, consistent with the microarray results, staining of cytospun SP and nSP CNJE with anti-CD62E antibodies identified a highly expressed epitope in the cell surface of most SP cells, but not in the nSP population () and costaining of 12-hour cultures with anti-CD62E and keratin 14 Abs proved that the expression occurred in basal epithelial cells (). After 72 hours of culture, when the cells had spread, migrated, and proliferated, cytokeratin-rich CD62E cells with classic epithelial morphology could be seen intercalated between CD62E-negative cells (). Most of these CD62E+
cells were isolated. In two rare cases, though, we observed adjacent CD62E+
cells of identical stain intensity suggestive of derivation from a proliferation event. In tissue sections, we infrequently identified CD62E staining in well-defined intracellular granules (). Unlike the case for CD62E, three other endothelial surface proteins CD62P/P-selectin, CD31/P-CAM and CD54/ ICAM-1, were not identified in the microarrays or, at the protein level, by immunostaining (not shown).
FIGURE 4 Expression of CD62E epitopes in the CNJE. (A) Cytospin-sorted SP and nSP cells (SP cells aggregated before centrifugation). (B) Double immunostaining of a 12-hour CNJE culture on glass. Green: FITC-conjugated anti-keratin14 Ab; red: polyclonal anti-CD62E (more ...)
The third overexpressed gene in , CD93
also represents a cell surface protein present in the myeloid lineage with no previous association with epithelial cells.18
It has been identified as a marker for myeloid and hepatic-yielding stem cells.19
The fifth gene in , CXCR4
, is a widely expressed, pleiotropic cytokine receptor that may have critical roles in stem cell function throughout mammalian lineages.20
Detailed results on the expression of these and other surface proteins in the CNJE cells and the relationship between them and the SP phenotype will be presented elsewhere (Wolosin et al., manuscript in preparation).
Topical Analysis and Cell Cycle Pathways
To identify potentially significant features of the SP cells that may depend on a few pivotal genes and hence would not be highlighted by the overrepresentation analysis, we scanned the DE list for a few specific gene categories, including (1) homeodomain genes that have been implicated in either stem cell renewal or in development, (2) genes that may contribute to slow or infrequent cycling, and (3) genes that may help with cell survival. Selected results of these studies are shown in . The first gene in this table, MSX1
, is involved in regulating cellular plasticity. Its ectopic expression inhibits the differentiation of a variety of mesenchymal and epithelial progenitor cell types and reverses myotubal differentiation.21,22
Another overexpressed gene, MEIS1
, has been shown to enhance expansion of the pools of hematopoietic precursors.23
Two other homeodomain morphogenes that are critical for eye development, PAX6
and sina oculis1 (SIX1
) were also highly overexpressed. PAX6
is highly expressed in the ocular surface epithelia and may contribute to the stem cell phenotype as a proliferation moderator.24
Substantially Over- or Underexpressed Homeodomain and Development-Related Genes
The highly overexpressed HES1
genes act as dominant negative blockers for the differentiation-inducing helix-loop-helix proteins. HES-1
expression preserves purified hematopoietic cells ex vivo and augments the hematopoietic SP cell population in vivo.25 ID1
is critical for long-term repopulating hematopoietic stem-cell maintenance.26
In fact HES1
expression may be interrelated. During development, HES1
expression is enhanced by ID
Within the stem-cell–related canonical WNT/β catenin/ TCF pathway,28,29
the DE gene list included only two genes, Wnt9
, and only the latter had a very moderate twofold increase expression in the SP. In addition, SP-nSP expression differences were not observed for the frizzled (FZD) receptors for these ligands. However, SP cells strongly underexpressed DKK1 and DKK3, main antagonists of the canonical WNT/β catenin pathway.30
Low levels of DKKs may facilitate WNT signaling in the SP respective to the signal strength in the DKK-high nSP cell. SP cells have augmented levels of the noncanonical WNT5A
gene and of its receptor, FZD5. The pathways associated with, and effects of noncanonical WNTs are poorly understood.31
When acting as an inhibitor, WNT5A
has been shown to foster a slow-cycling status that stabilizes the hematopoietic stem cell phenotype.32
Furthermore, WNT5a may act as an activator or inhibitor of the (canonical) β catenin/TCF signaling depending on receptor context.33
Hence, the noncanonical classification of Wnt5a does not preclude a relevant role for this gene in these epithelial stem cells.
Systems that may regulate premature apoptosis or determine cell survival under challenging conditions were also examined. Annexin A1, one of the most SP-overexpressed genes (; SP/nSP ratio, 12.55) is a pleiotropic blocker of inflammatory mediator release.34
Members of the type 2C ser/threo protein phosphatase family, a set of enzymes that seem to negatively regulate cellular stress signaling, also displayed substantial overexpression. The most overexpressed family members, PPM1D () and PPM1H (ratio, 6.94) has been shown to attenuate oxidative stress-induced phosphorylation of p38 MAP kinase in epithelial cells.35
But, seemingly more significant, PPM1D
, also called Wip1
, for wild type 53-induced phosphate-1, has been shown to increase the threshold for an apoptotic response to DNA damaged through its role in the p53-Mdm2 autoregulatory loop.36,37
SP cell protection from oxidative damage to DNA may be afforded by the high overexpression of CYP1A1 (), a dioxin-inducible enzyme that degrades aromatic genotoxic compounds.38
Polymorphisms in this gene correlate with increased incidence of cancers.39
Within the complement cascade, substantially overexpressed genes were limited to complement factors CFB (SI, 3534; ratio, 5.64) and CFH (SI, 2122; ratio, 4.57). These two factors ensure that the complement system is directed toward pathogens and does not damage host tissue. Polymorphisms of these genes have a high correlation with age-related macular degenerations.40,41
The nature of the genes with the highest SP underexpression in (nSP SI>>SP-SI) is consistent with the results of DAVID analysis indicating paucity of cell components related to cell motility and migration in the SP cells.
In addition to the topical analysis, GenMapp was used to generate maps of a large number of intracellular pathways and gene clusters (http://cgap.ncbi.nih.gov/pathways
; provided in the public domain by the National Center for Biotechnology Information). While the interpretation of the significance of individual gene expression differences in many of these maps was not apparent or given to unambiguous interpretation, all the pathways associated or affecting activity at the cell cycle restriction (R) point that control transition from late G1
to S, namely (1) the mitogen activated kinase (MAPK) cascades (GenMapp’s file Hs_MAPK_signaling_pathway_KEGG); (2) the cyclin D associated kinases (GenMapp’s Hs_Cell_cycle_KEGG); and (3) TGFβ (GenMapp’s Hs_TGFβ_KEGG) yielded readily interpretable images on unique features of the CNJE SP cell ().
FIGURE 5 Over- and underexpression of genes in cellular pathways affecting activity at the G1 to S cell cycle restriction (R) point. Gen-Mapp pathways have been modified from the originals by removing genes not included in the AE gene file and highlighting genes (more ...)
To build , it was necessary to ascribe MAPK (i.e., ERK1/2, p38 or JNK) specificity to MAPK phosphatases. This attribution was made on the basis of recently published data on MAPK-MAPK-phosphatase interplay showing that (1) DUSP5 and DUSP6/MKP3 are, respectively, nuclear and cytosolic phosphatases wholly selective for ERK enzymes42–44
; (2) DUSP1 and DUSP4 are nuclear phosphatases that can deactivate enzymes from all three branches of the MAPK cascades45,46
and can block ERK-dependent proliferation, but, due to differences of affinities for the kinases representing each of the three branches of the cascade,47
DUSP1 and/or DUSP4 may not be effective in many instances as ERK function blockers. 48–50
For the MAPK pathways (, left side) the most significant facts are that (1) there are only nominal changes in the aggregate expression levels of all MAPK isoforms comprising each of the three terminals kinase nodules: ERK (~25% lower in the SP), p38 (20% higher in the SP), and JNK (25% lower in the SP); (2) the dual specificity phosphatases (DUSPs) the only enzymes that are available to dephosphorylate and thereby inactivate the ERK, p38, and JNK enzymes, in particular the ERK-specific DUSP5, are substantially overexpressed in the SP cells. Our qPCR measurements validated the microarray results for DUSP 4, -5, and -6 at the gene level (). In addition, DUSP4 immunostaining of sorted cytospun cells validated the result for this gene at the protein level (). The stain was acutely localized to the nuclei and the fluorescence intensity in the SP nuclei was 3.4 times higher than that of nSP cells (average value for 15 cells of each). Remarkably, a similar multi-DUSP overexpression has been observed in keratinocyte SP cells.51
DUSP4 and ERK expression and phosphorylation in SP and nSP cells. DUSP4 immunostaining of cytospins of HCNJE nSP (A) and SP (B) cells. (C) Western blot of sorted rabbit CNJE cells. Right: (a) β-actin, (b) ERK1/2, and (c) pERK1/2.
The large overexpression of DUSPs imply dramatic changes in absolute kinase/kinase phosphatase gene ratios. The (MAPK1/ERK)/(DUSP5+DUSP4+DUSP6) SI quotient decreased 15 times, from 5469/716 to 4413/9163 (SIs for MAPK3/ERK1 measured in the microarray are very low). Including DUSP1 in the analysis does not radically modify this calculation. For the other two branches of the MAPK cascades, p38 and JNK. Similar calculation using MAPK13+MAPK14 for p38 and MAPK8+MAPK9+MAPK10 for JNK and dividing by DUSP1+DUSP4 (DUSP5 and -6, are ERK specific) yielded SI quotient decreases of six- and fivefold, respectively. Such large changes could be expected to prevent establishment of a phosphorylated MAPK state in SP cells and hence, attenuate or completely block SP cell responses to mitogenic and/or migration-inducing and inflammatory inputs.
To test this hypothesis, we cultured a large batch of rabbit CNJE cells in growth factor-free (base) medium for 24 hours, sorted equal amounts (80,000) of SP and nSP cells into the same base medium and determined EGF induction of ERK phosphorylation by Western blot analysis. A representative result from three independent trials is shown in ; ERK phosphorylation was never detected in the SP cells. Not directly predictable from the human microarray results, in the rabbit total ERK protein was markedly reduced in the SP cells. For p38, SP phosphorylation was detectable but at a much lower level than that for the nSP cell (not shown).
also incorporates the expression of CXCL12/SDF-dependent CXCR4 (). The SDF-CXCR4 interaction enhances skin wound healing and clonogenic growth of keratinocyte precursors in vitro.52,53 CXCR4
downstream signals include activation of PI-3 kinase and RAS
. The latter is one of the early genes in the mitogen activated ERK cascade (). Hence, the very large CXCR4
overexpression in the SP may provide a SP cellular activation that is synergetic with the EGF-initiated ERK activation.
The right side of focuses on the TGFβ/TGFβ-like (activin) receptor pathway. The primary effect of TGFβ in ectodermally derived epithelial cells is cell cycling arrest. The arrest is secondary to phosphorylative activation of the nuclear p-type kinase inhibitors of cyclin D (). In the extracellular/cell membrane domain of the pathway the microarray analysis reveals only moderate changes in gene expression of TGFβs and TGFβ receptors (). However, SP cells showed a marked underexpression of the follistatin (FST
) gene. Follistatin is a secreted protein that sequesters and thereby neutralizes activins.54,55
The low FST expression (nSP/SP ratio, 0.11) could render the SP cell to be highly sensitivity to TGFβs.
Within the nucleus, there were further differences in SP-nSP gene expression patterns that are highly meaningful for cyclin D system function. First, SP cells underexpressed the cyclin D kinases (CDK4 and 6) responsible on association with cyclin D isoforms for activating retinoblastoma, the final stage in releasing cells from the G1 to the S phase of the cell cycle. Second, this negative effect of cyclin kinase deficiency on cell cycle progression is compounded by the overexpression of several cyclin-dependent kinase inhibitors, which specifically act on the cyclin D/cyclin D kinase complexes. It is noteworthy that, in contrast to the marked differences in CDK4 and -6, other kinases associated with other regulation points of the cell cycle showed no major expression differences. For instance, cyclin A-associated p21 shows minimal differences () and of 17 cell division cycle (CDC) kinases identified in the microarray 15 are unchanged, 1 was overexpressed 2.6 times and 1 was underexpressed 2.5-times.
Finally, two cytokeratins that have been linked to the stem/precursor cell phenotype in epidermis, K15 and -19,56
were also part of the SP-overexpressed gene set (ratios of 3.78- and 3.38-fold, respectively). Yet, at the same time, expression of the basal cytokeratin K14 was substantially underexpressed in SP cells. This is a seemingly surprising finding although it has also been observed in epidermal SP cells.15