Altered Dynamics of Intestinal Cell Maturation in Apc^1638N/+ Mice

Reprogramming of intestinal epithelial cells as they leave the progenitor cell compartment at the base of the crypt and mature along the villi toward the lumen generates the multiple cell lineages necessary for normal functioning of the tissue. Homeostasis is established by the correct allocation of cells to these different lineages. The reprogramming during this cell maturation involves altered expression of genes that drive proliferation and markers of differentiation lineages (1, 2). Disruptions in this reprogramming—either persistent expression of underlying drivers of proliferation or failure of proper differentiation—cause tumor development.

In Apc^Min/+, Apc^Δ716, and Apc^1638N/+ mice, intestinal tumors develop when the inherited mutant Apc allele is complemented by somatic focal loss, mutation, or silencing of the WT allele (3–7), conforming to the hypothesis that for tumorigenesis, both alleles of a tumor suppressor gene must be inactivated (8, 9). Unclear, however, is how the inherited mutation affects the intestinal mucosa, and the probability of tumor formation, before reduction of Apc mutation to homozygosity. For example, an ∼85% decrease of APC protein is necessary for the generation of ∼1 tumor per mouse (10), but mice that inherit one mutant Apc allele, or at tumor risk for other reasons, exhibit an expanded proliferative compartment (11).

We therefore compared the intestinal mucosa of C57Bl6 wild-type (WT) mice to histologically normal mucosa of congenic Apc^1638N/+ littermates, before tumors develop due to focal loss of the WT allele. Unlike Apc^Min/+ or Apc^Δ716 mice, in which large numbers of tumors develop within months of birth, only approximately three tumors develop from 6 to 9 months in Apc^1638N/+ mice (6), thus permitting the analysis of effects of the inherited mutant allele before loss of the WT allele and development of mucosal pathology.

A novel method of active transcription site imaging in single cells in situ revealed that during normal maturation of intestinal cells along the crypt-villus axis (CVA), regulation of genes associated with Wnt and Notch signaling was much more dynamic than apparent from the analysis of steady-state RNA or protein levels. Moreover, there was significant displacement in the Apc^1638N/+ mice of oscillating patterns of these active transcriptional units responsible for cell reprogramming. These pathways cooperate to maintain crypt cells in a progenitor cell phenotype and in lineage-specific allocation (12). In addition, we also found that the inherited mutation dampened cell reprogramming and perturbed the expression pattern of lineage-specific markers in the villus. Moreover, crypt cells exhibited altered expression of genes that encode enzymes of the tricarboxylic acid (TCA) cycle, and perturbed the expression of Hif1α and its targets. Thus, the single WT Apc allele in the Apc^1638N/+ mouse is insufficient to maintain normal pathways and patterns of cell maturation along the CVA.

Generation, maintenance, genotyping, and pattern of tumor formation of Apc^1638N/+ mice are described (4, 6, 13). Experiments were approved by the Institutional Animal Care and Use Committee of Montefiore Medical Center and the Albert Einstein College of Medicine. Apc^1638N/+ mice and Apc^+/+ littermates were fed a completely defined diet (AIN76A; ref. 13). Upon sacrifice, the intestine was rapidly dissected, portions of each region fixed in formalin and then embedded in paraffin, or were used for isolation of cells from along the CVA.

Active transcription sites were detected based on methods described (14, 15). Formalin-fixed, paraffin-embedded sections (4 μm) were heated in a 55°C dry oven for 1 hour, placed in decloaking buffer for deparaffinization, cooled, and treated with ammonia/70% ethanol (20 min) and sodium borohydride (50 min, 4°C) to reduce autofluorescence. Prehybridization with 50% formamide/2XSSC was at room temperature, 30 minutes; slides were hybridized overnight with 20 ng of probe at 37°C, protected from light in a humidified chamber. Probes for fluorescence in situ hybridization were designed using the OLIGO-6.0 software, with specificity verified using the National Cancer Institute (NCI) GeneBank BLAST program. For each target nascent transcript, three to eight 50-mer DNA probes were synthesized with four to five modified thymidine bases conjugated to succinimidyl ester fluorescent Cy3 or Cy5 dyes (GE Healthcare). Fluorescence of the multiple probes for each target mRNA localized to a specific site. Following hybridization, slides were washed in prewarmed buffers on a shaker protected from light: 50% formamide/2XSSC, 20 minutes; 2XSSC, 1XSSC, and 0.5XSSC for 15 minutes each; then in PBSM and nuclei stained with 4′,6-diamidino-2-phenylindole (DAPI) before mounting with Prolong gold antifade (Molecular Probes). In transcription site detection, fluorescence intensity of hybridized probes is amplified by the many nascent transcripts within a small volume. Fluorescent signals were detected with an Olympus AX70 microscope, UApo 40X, 1.35NA and PlanApo60X, 1.4NA objectives, and a CoolSNAP-HQ CCD camera (Photometrics) using filters for DAPI (#SP100), FITC (#SP101), Cy3 (#SP-102v2), and Cy5 (#SP104v2; Chroma Technology). Three-dimensional images acquired with a 200-nm Z step size were analyzed using the IPLab software version 3.61 (BD Biosciences). Fluorescent spots were identified as transcription sites on the basis of location within the DAPI-stained nuclei, fluorescence intensity, volume, shape, and absence of autofluorescence detected in the FITC channel. Active transcription sites for each locus were counted in a minimum of 50 well-oriented crypt-villi from three mice per genotype. The percentage of transcription sites for each gene was calculated from the total number of transcription sites and nuclei detected.

Epithelial cells were isolated progressively from the top of the villus (F1) to the bottom of the crypt (F10) as described (e.g., refs. 1, 2). Expression profiling with RNA isolated from the F1 and F10 cells for each genotype was done on GeneChip Mouse Genome 430 2.0 Arrays (Affymetrix) for four mice for each cell position and genotype. Mean values by genotype and position for each sequence were calculated; the F1/F10 ratios were calculated for each genotype; and then the absolute value of the ratio of these ratios were calculated.

Lysates of cells from fractions F10 (crypt bottom) to F2 (near villus top) were fractionated by electrophoresis, blotted, and specific peptides detected as described (1). Antibodies were as follows: Hif1α (Cayman Chemical), vascular endothelial growth factor (VEGF; Santa Cruz), and Hk2 (Cell Signaling). Cells from fraction 1 (villus tip) were not analyzed due to low yield. The experiment was repeated three times using cells isolated from three different mice of each genotype.

Analysis of intestinal cell reprogramming during maturation along the CVA used a validated method that detects active transcription of specific loci in single cells in situ (14–18). Because the highest concentration of an RNA sequence is at its site of transcription, fluorescent oligonucleotide probes hybridizing to these transcripts produce a bright image at this site, indicating active transcription of the locus queried (Fig. 1A; additional examples in refs. 14–20). Assay of activation of gene transcription in single cells in situ, unlike measurement of steady-state levels of RNA or protein, assesses the state of the “rheostats” (i.e., functional state of the transcriptional machinery) that respond rapidly to internal and external signals. In WT mice, the frequency of cyclin D1 and c-myc active transcription sites was higher in cells near the bottom of the crypt (Fig. 1B), consistent with function of these genes in driving cell cycling in this compartment, and with reports that they are direct targets of Wnt signaling, more active at the bottom of the crypt (21–23). Similarly, active Notch1 or Hes1 transcription sites, the latter a direct target of activated Notch signaling, were also more frequent in cells in the lower half of the CVA, although localization was not as striking as for cyclin D1 and c-myc. Importantly, reprogramming of cells as they mature along the CVA was highly dynamic: for each gene, there was an oscillating pattern of cells with locus activation along the CVA, which may reflect compensatory regulation that overshoots at each position of stimulation and repression, or biological compartmentalization of activation resulting from lateral inhibition during differentiation. These oscillations contrast with monotonic decreases and increases in steady-state expression levels of genes that characterize proliferation or differentiated functions, respectively, along the CVA axis that we reported (1, 2, 24, 25).

Figure 1B (inset) also shows that oscillations of cells with active Hes1 or Math1 transcription sites, both components of Notch signaling, had different periodicities along the CVA, especially at cell positions 18 and 36. Consistent with repression of Math1 by Hes1, there was weak pairwise correlation of Hes1 and Math1 active transcription sites (Spearmann coefficient of correlation, r = 0.116, P = 0.625), showing that these genes were not transcriptionally active in the same cells at the same time. Indeed, when assayed simultaneously, no cell was identified in which Hes1 and Math1 were simultaneously active, regardless of genotype. In contrast, pairwise correlation of Notch1 and c-Myc was significant (r = 0.618, P = 0.035), showing concordant transcriptional activation of these genes that drive proliferation.

The histologically normal mucosa of Apc^1638N/+ and Apc^+/+ littermates were compared, with all mice on a congenic C57Bl6 background and fed a chemically defined control diet (AIN76A). This defined, invariant diet minimizes environmental variables (e.g., phytochemical content) that can markedly alter Apc-initiated tumorigenesis and thus underlying mechanisms (e.g., refs. 26–29). Differences between the two genotypes were clear: preferential transcription of the cyclin D1 and c-myc genes at the bottom of the crypt of WT mice was shifted to higher positions along the CVA of the Apc^1638N/+ mice (Fig. 1C). Moreover, a major peak of active transcription of Math1 in the WT mice at cell positions 30 to 36 and a smaller peak at 36 to 42 were absent in the Apc^1638N/+ mice (P < 0.002; Fig. 1D), with additional repression of active Math1 transcription sites at cell position 9 that approached significance (P = 0.06).

Because canonical Wnt and Notch signaling cooperate in regulating intestinal cell maturation, we determined whether alterations in transcription site patterns of Apc^1638N/+ mice perturbed maturation by assaying expression profiles of villus and crypt cells. We have documented that the cell fractions isolated by sequential elution exhibit gradients of proliferation and differentiation markers that decrease and increase monotonically along the CVA (1, 2, 24, 25). Here, we determined gene expression profiles of the extremes of these gradients: top of the villus (F1 fraction) and bottom of the crypt (F10 fraction) of each of 4 Apc^+/+ and 4 Apc^1638N/+ mice.

As expected from different functions of cells in crypts and villi, unsupervised clustering using all 31,213 probe sets unambiguously distinguished cells from these two compartments, regardless of genotype (branch 1 and 2, Supplementary Fig. S1). These data also suggested genotypes were distinguishable, with three of four mice of each genotype separating between branches 1a and 1b, and 2c and 2d (Supplementary Fig. S1), leading to more in-depth analyses.

Four sets of differentially regulated sequences were distinguished as follows: by crypt or villus position and by genotype (Supplementary Table S1–S4; Supplementary Fig. S2). The 3457 sequences differentially expressed between villus (F1) and crypt (F10) for WT mice were consistent with those we previously reported for intestinal epithelial maturation along the CVA (e.g., enriched in genes contributing to cell cycle regulation, enterocyte differentiation, cytoskeleton assembly, and lipid metabolism; ref. 1). Genotype comparisons identified 73 sequences differentially expressed between the WT and mutant genotypes in the villi (criteria: change >2-fold, and P < 0.05) and 259 in crypt cells. Thus, changes were ∼4-fold more frequent between genotypes in crypts, in which Wnt signaling is normally active, than in the villi, in which Wnt signaling may be shutdown and therefore not readily compromised.

To address whether the WT Apc allele in the Apc^1638N+ mice was haploinsufficient for regulating normal reprogramming along the CVA, the villus/crypt ratios (F1/F10) for the mean of each sequence were compared. Using combined criteria (>2-fold change, P < 0.05), there was no sequence for which F1/F10 was elevated or repressed for either WT or Apc^1638N/+ mice, but showed reciprocal change in the other genotype (Supplementary Fig. S3), consistent with the apparent normal histology and functioning of the intestinal mucosa in Apc^1638N/+ mice. However, although there was overlap between sequence subsets for which F1/F10 was altered in WT or mutant mice (Supplementary Fig. S3), there were also distinct differences in extent of change, as illustrated by the following: Sequences were identified differentially expressed between the crypt and villi in the WT mice (F1/F10 >2-fold change, P < 0.05), and also altered in the same direction in the mutant mice, but with no criteria imposed for the magnitude of change or P value. For the 1,720 sequences downregulated, 82% were downregulated less in the Apc^1638N/+ mice (Fig. 2A, area under the red dotted line); for the 1,187 sequences upregulated, 68% were upregulated less in Apc^1638N/+ mice (Fig. 2B). A Z test for binomial proportions showed that the probability that these results were due to chance was very small (P < 0.0001). Inserts (Fig. 2A and B) show P values for differences in expression in villus(F1)/crypt(F10) for each sequence in the Apc^1638N/+ mice, with sequences in the same order on the abscissas of both the figures and the inserts. As predicted, sequences that changed less in Apc^1638N/+ than in WT mice (i.e., those to the left of the X-axis; Fig. 2A and B) show higher P values in mutant mice and, hence, less significant change. Importantly, for the mutant mice (inserts), most P values remain above the significance criteria (i.e., P < 0.05); thus, general patterns of cell reprogramming along the CVA in Apc^1638N/+ mice are maintained, consistent with the normal histology of the tissue, but are dampened compared with reprogramming in the WT mice.

Of the 2,217 sequences attenuated in change along the CVA in mutant compared with WT mice (i.e., below the lines in Fig. 2A and B), 255 overlapped with a c-myc target data base of 1,160 sequences (30), a highly significant enrichment of c-myc target genes among sequences modulated in alteration in the Apc^1638N/+ mice (P < 0.0001; Fig. 2C). This is consistent with the displaced distribution of c-myc active transcription sites in the mutant mice (Fig. 1) and is important in that c-myc expression drives normal maturation of intestinal epithelial cells along the CVA (1), and its derepression is necessary for Apc-initiated tumor formation (31, 32).

Wnt and Notch signaling are determinants of lineage-specific allocation of intestinal progenitor cells (33): Math1 expression drives secretory cell differentiation (34, 35), whereas in its absence, the default is enterocyte differentiation (35). We therefore determined whether alterations in transcription site distribution in the mutant mice, and especially repression of Math1 transcription site activation (Fig. 1), was reflected in the expression pattern of lineage-specific markers.

Because sample size was limited and expected differences in expression modest, we present the individual data points for each gene in each of four Apc^1638N/+ mice compared with four WT mice, along with the median value for each gene, which minimizes the contribution of outliers. Although changes by genotype for each gene were not statistically significant, the overall differences in pattern of expression in the mutant compared with the WT mice were striking (note log scale). The median of 9 of 10 secretory cell markers was lower in the Apc^1638N/+ mice than in WT (Fig. 3A). This included two markers of the secretion of mucus by goblet cells (Muc2 and Tff3) and seven of eight markers of the heterogenous population of enteroendocrine cells, which secrete a variety of hormones, growth, and neuroendocrine factors. The exception was glucagon, normally a product of pancreatic α cells, but processed by posttranslational cleavage to produce GLP-1 in intestinal enteroendocrine cells. In contrast to this decreased pattern of secretory cell markers, all six enterocyte cell markers were elevated in villus cells of mutant mice compared with WT (Fig. 3A). Thus, although perturbation of expression of each gene is modest in the histologically normal intestinal mucosa of Apc^1638N/+ mice, the pattern of decreases of secretory and increases in enterocyte markers is consistent with the decrease in active Math1 transcription sites along the CVA.

We reported that dietary risk for colon and intestinal tumors in mice was characterized by decreased expression in the histologically normal intestinal and colonic mucosa of sequences encoding enzymes of the TCA cycle, and suggested that elevated Wnt signaling was associated with shift toward glycolytic metabolism—a metabolic state well documented in colon tumors but not known to exist in the normal tissue at risk (27). We therefore analyzed sequences encoding TCA cycle enzymes in relationship to genetic risk. Expression of the sequence encoding at least one enzyme of every step in the TCA cycle, as well as of 3-ketoacyl-CoA thiolase, which generates acetyl-CoA from metabolism of lipids, was decreased by 20% to 50% in the crypts of Apc^1638N/+ mice compared with WT (Fig. 4A), similar to changes in the mucosa for dietary-induced risk (27). In contrast, these genotype-associated changes were not seen in villus cells (Fig. 4B). Thus, the overall pattern of expression suggests a functional shift in metabolism specifically localized to the bottom of the crypt.

Oncogenic mutations in enzymes of the TCA cycle can generate a pseudohypoxic response and a shift of metabolism toward glycolysis, characterized by increased expression of Hif1α, a transcription factor regulated posttranslationally that coordinates hypoxic response (36, 37). Hif1α protein levels from each of the fractions along the CVA axis of Apc^1638N/+ mice were elevated in a representative experiment (Fig. 5A), and the mean Hif1α protein levels were elevated in each fraction in the mutant mice (three experiments, three different mice of each genotype; Fig. 5B). The elevation increased in fractions 6 to 2 that encompass villus cells expressing differentiated functions (1), although this was not statistically significant for individual cell positions or when the curves were modeled by a repeated measure analysis of patterns of expression across CVA levels and genotypes (Fig. 5). However, VEGF protein, an angiogenic factor regulated by Hif1α, was progressively increased in expression in the same fractions six to two in the villus in Apc^1638N/+ mice relative to WT (Fig. 5A and B). A quadratic model showed this to be a significant genotype by CVA level interaction (P < 0.015), suggesting a curvilinear increasing trend from crypt bottom to villus top (Fig. 5). Finally, hexokinase 2, another Hif1α target, was at background levels in every fraction in WT mice but was elevated in every fraction in Apc^1638N/+ mice (Fig. 5A and B). Here, a linear model (Fig. 5) revealed a significant genotype by CVA level interaction (P < 0.029), suggesting a decreasing trend from crypt bottom to villus top.

These data show that although the intestinal mucosa of the Apc^1638N/+ mouse exhibits normal histology and function, the complex reprogramming of intestinal epithelial cells as they migrate from the progenitor cell compartment in the crypt is perturbed. This was determined by altered patterns of transcriptional activation along the CVA of key genes in the Notch pathway (Hes1 and Math1) and two genes (cyclin D1 and c-myc) that drive proliferation in the progenitor cell compartment regulated by Wnt and Notch signaling, fundamental developmental pathways that cooperate in maintaining the crypt progenitor cell compartment. Expression profiling then showed that although general patterns of reprogramming were maintained in villus compared with crypt cells in the mucosa of Apc^1638N/+ mice, there was a significant compromise in the reprogramming in the mutant mice compared with their WT littermates. These changes in the mutant mice encompassed perturbed profiles of lineage-specific markers and altered expression of sequences that govern metabolic patterns. Changes are modest, consistent with the fact that the tissue continues to appear morphologically and functionally normal, but overall patterns of change indicate that it is significantly perturbed in comparison with the mucosa in WT mice. It has been reported that enterocyte migration along the CVA is decreased in Apc^Min+ mice (38), consistent with the dampened maturation of cells in the villi. Although this was not detected in the Apc^1638N/+ mouse (39), this may be related to the much more modest tumor phenotype in Apc^1638N/+ compared with Apc^Min/+.

In the Apc^1638N/+ mouse, loss of the mutant allele is not detected until the development of frank tumors (40), an observation we confirmed (data not shown; ref. 41). Moreover, we determined that the expression of the WT Apc allele is reduced in the histologically normal mucosa by 40% to 60%, coupled with a 25% reduction in APC protein of 25% (data not shown). In addition, the Apc^1638N/+ mice do not accumulate significant levels of a truncated APC protein encoded by the mutant allele. Therefore, we conclude that the changes in the underlying molecular biology of the mucosa are due to haploinsufficiency of the WT allele.

The shift of cells exhibiting active cyclinD1 and c-myc transcription, and Notch signaling, along the CVA in the mucosa is important because these genes and pathways likely drive the expanded proliferative compartment that characterizes the mucosa at genetic and/or nutritional risk for tumor development (11). Continued proliferation of cells with a progenitor cell phenotype into zones where cell cycling is normally repressed can contribute to hyperplastic growth and expand the stem-like cell compartment, which must be targeted by the loss of the second Apc allele for tumors to form (42). Effects on apoptosis may be less important: rates in the intestinal mucosa are very low and a mutation affecting short-chain fatty acid metabolism that reduce this >90% does not cause tumor development (43). Further, targeted inactivation of Tcf4 in the mouse, which, in complex with β-catenin, is a major effector of intestinal Wnt signaling regulated by Apc, leads to postpartum lethality as the mucosa deteriorates and cannot be regenerated due to premature differentiation, but not apoptosis, of intestinal progenitor cells (44).

Math1 drives secretory cell differentiation of intestinal epithelial cells, and in its absence, there is default to the enterocyte lineage (35). Thus, the decrease in 9 of 10 secretory cell markers and the complementary increase in 6 of 6 enterocyte markers were consistent with that predicted by the repression of Math1 transcription sites in the mutant mice. However, altered expression of these markers likely reflects perturbed coordination of differentiation programs, rather than significant shifts in overall lineage allocation, because the mucosa appears normal until focal loss of the second Apc allele and tumor initiation.

A novel finding was decreased expression of genes that encode enzymes of every step of the TCA cycle in crypt cells in Apc^1638N/+ mice compared with Apc^+/+ mice, in contrast to the lack of such changes in the villus cells of the same mice. This is similar to alterations we reported in the mucosa of both the small and large intestine of mice at nutritional risk for tumor formation (27), which we recently found is also enriched in the crypt (data not shown). Here, we have shown association of these changes with perturbed expression of Hif1α and its targets VEGF and hexokinase 2. We hypothesize that these data reflect a shift in the tissue toward glycolytic metabolism and generation of a (pseudo)hypoxic state that promotes tumorigenesis. It has been suggested that a shift toward glycolysis favors proliferation, higher in the crypt, by providing biochemical intermediates for synthesis of macromolecules and increase in biomass (45). In regard to the potentially greater shift in the crypts of mutant compared with WT mice, it is important that hypoxia and Hif1α expression are characteristics of stem cell niches (46–50), and that genes encoding enzymes of the TCA cycle are bona fide proto-oncogenes that, when mutated, lead to the accumulation of TCA cycle intermediates (51–53) that can trigger Hif1α expression either by succinate inhibition of prolyl hydroxylase activity and/or by generating increases in reactive oxygen species (54). Moreover, elevated Hif1α expression is a direct cause of intestinal polyp formation in Peutz-Jaeghers syndrome, mediating a metabolic shift that drives tumorigenesis (55). Downregulation of the TCA cycle, a shift toward glycolytic metabolism, and a hypoxic response contributing to higher probability of tumor development in the intestinal mucosa by either genetic or environmental influences can be important in both screening strategies for early detection and as targets for chemoprevention.

Wnt signaling may contribute to intestinal tumorigenesis as a continuum of effects related to extent of altered signaling (56), increases in Wnt signaling beyond those sufficient for initiation are necessary for intestinal tumor progression (57), and embryonic stem cell differentiation is modulated as a function of extent of β-catenin signaling levels (58). Thus, it is tempting to speculate that the inherited Apc¹⁶³⁸ mutation causes modest changes in Wnt signaling that drives the altered transcriptional and expression patterns. Although steady-state levels of expression of several Wnt target genes (c-myc, cyclin D1, Sox9, Lgr5, and jagged1) were, as expected, higher at the bottom of the crypt of both normal and mutant mice, these steady-state levels were not significantly different at any position along the CVA of Apc^1638N/+ compared with WT mice (data not shown). However, whether Wnt signaling is functionally altered in the mucosa of Apc^1638N/+ mice is not easily resolved. For example, if inactivation of one Apc allele decreases Apc expression, modestly increasing Wnt activity, this would greatly increase the probability that stochastic variations in expression of the WT allele could transiently exceed a threshold sufficient to significantly alter steady-state levels of direct Wnt targets (59). This also applies to the variation in VEGF and Hk2 levels that are seen (Fig. 5), although these changes in Apc^1638N/+ compared with Apc^+/+ mice reach statistical significance. Although such focal and transient changes might not be detected in cell populations isolated from the mucosa, the important effect of these stochastic variations in tumor suppressor gene expression has been discussed in detail (59). Alternatively, the alterations in the mucosa of Apc^1638N/+ mice may depend on perturbation of one of the many other functions that have been reported for APC, rather than changes in Wnt signaling.

In summary, in Apc^1638N/+ mice, few tumors develop over an extended period and the histologically normal intestinal mucosa can be readily investigated. We have found that in this histologically normal mucosa, there are significant alterations in the dynamics of cell reprogramming along the CVA and of markers of normal cell maturation. We have previously shown that some of these changes are present in the mucosa at dietary risk (27). Thus, just as alterations at distant tissue sites contribute to tumor metastasis by generating receptive environments, probability of tumor development at the primary site may be modulated by alterations that precede the reduction of the inherited mutation to homozygosity, or the generation of initiating mutations. Therefore, these findings have important implications for understanding the mechanism of risk and tumor formation in this tissue, and for clinical approaches to early detection and prevention.

No potential conflicts of interest were disclosed.

Grant Support: Grants U54CA100926, RO1CA114265, RO1CA135561, R33CA-083208, EB2060, GM80247, and Cancer Center Support Grant P013330 from the NCI.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.