Establishment and Characterization of 5-Fluorouracil-Resistant Human Colorectal Cancer Stem-Like Cells: Tumor Dynamics under Selection Pressure

5-Fluorouracil (5-FU) remains the gold standard of first-line treatment for colorectal cancer (CRC). Although it may initially debulk the tumor mass, relapses frequently occur, indicating the existence of cancer cells that are therapy-resistant and are capable of refueling tumor growth. To identify mechanisms of drug resistance, CRC stem-like cells were subjected to long-term 5-FU selection using either intermittent treatment regimen with the IC50 drug dose or continuous treatment regimen with escalating drug doses. Parental cancer cells were cultivated in parallel. Real-time PCR arrays and bioinformatic tools were used to investigate gene expression changes. We found the first method selected for cancer cells with more aggressive features. We therefore transplanted these cancer cells or parental cells in mice, and again, found that not only did the 5-FU-selected cancer cells generate more aggressive tumors with respect to their parental counterpart, but they also showed a different gene expression pattern as compared to what we had observed in vitro, with ID1 the top upregulated gene. We propose ID1 as a stemness marker pervasively expressed in secondary lesions emerging after completion of chemotherapy.


Introduction
Current first-line chemotherapy generally consists of cytotoxic agents, like 5-fluorouracil (5-FU), a uracil analogue that following intracellular conversion into three active metabolites (fluorouridine triphosphate, fluorodeoxyuridine triphosphate, and fluorodeoxyuridine monophosphate) causes alterations in RNA processing/function and severe DNA damage [1]. Evidence shows that DNA lesions are able to trigger apoptosis [2]. However, evasion of both death receptor and mitochondrial pathways is common in response to stress stimuli [3]. Antiapoptotic mechanisms together with cytoprotective autophagy [4], upregulated DNA repair pathways [5], and metabolic rewiring [6] allow tumor cells to thrive in conditions that otherwise would be lethal propagating the disease.
Despite intense studies, one important question still remains open: are resistant cells already present in the tumor at the time of first treatment or do tumor cells adapt to therapy? Several models have been proposed to address this important question. The longstanding cancer model, known as 'clonal evolution' postulates that tumors arise from a single mutated cell. This cell, in turn, generates an abnormal offspring that also mutates, forming a mass of genetically diverse tumor cells [7]. From this perspective, tumor progression proceeds via a process analogous to Darwinian evolution, where only cells with advantageous mutations survive and fuel tumor growth [8]. The opposing cancer stem cell (CSC) theory proposes instead that tumor growth is driven and sustained by a rare population of cells capable of self-renewal and differentiation [9,10]. While the CSC theory could explain the hierarchical heterogeneity observed in most tumors, and why some patients suffer relapses after initial successful therapy [11], clonal evolution driven by genomic instability is likely to play an important role in tumor progression by allowing CSCs to adapt to environmental, immunologic and pharmacologic cues [12]. Indeed, tumors are dynamic entities and understanding how their cells succeed in offsetting survival pressures imposed by therapy is of utmost importance for the development of more effective treatments.
Current drug design has been largely focused on initial efficacy, and the effects of drug selective pressures remain poorly studied. Panels of chemotherapy resistant cancer cell lines have been generated to investigate the molecular pathways that may be associated with resistance. However, the vast majority of these studies has been performed on commercially available cell lines and not on primary cells isolated from patients' biopsies. Both pulse and incremental drug treatment methods have been described. However, a comparison of gene expression in the respective treatment-derivatives remains incomplete. Moreover, to our knowledge, re-transplantation into an in vivo setting of such in vitro established resistant cell lines has been poorly addressed.
In this manuscript, we used stem cells that were identified, isolated, and characterized in our laboratory from colorectal cancer (CRC) patients' specimens [13]. We generated 5-FU resistant cells by using either intermittent treatment with the IC50 drug dose or continuous treatment with escalating drug doses. We then used real-time PCR arrays and ingenuity pathway analysis (IPA) to investigate gene expression changes. Our results indicated that the intermittent treatment with the IC50 of 5-FU could serve as a better model to investigate drug resistance. While biological functions like cell survival, proliferation, movement, invasion, and migration were predicted by IPA to be inhibited in cells undergoing continuous treatment with escalating 5-FU doses with respect to parental cells passaged in parallel, the same functions were predicted to be unaffected in cells undergoing intermittent treatment with the IC50 5-FU dose. Moreover, while 5-FU resistant cells under continuous treatment with escalating drug doses were predicted to be more prone to cell death with respect to parental cells, cell cycle progression was not predicted to be delayed in the other cell group despite DNA damage, suggesting a propensity to genomic instability, as previously described [13]. Tumor xenografts were generated from both parental and resistant cells undergoing intermittent treatment with the IC50 5-FU dose with the resistant cells generating more aggressive phenotypes. The gene expression changes observed in the in vivo xenograft setting were compared to those observed in the in vitro setting. We found in vivo and in vitro gene expression patterns to differ, indicating a capability for 5-FU resistant cells to dynamically adapt and evolve. We identified ID1 as the top upregulated gene in the tumor generated by 5-FU resistant cells.

Establishment of 5-FU Resistant Human CRC Stem-Like Cells
CRC stem cells were derived from five human metastatic cancers (Tu11, Tu14, Tu27, Tu28, and Tu42) using a feeder-dependent cell culture system, previously described by our laboratory [13]. In order to establish chemotherapeutic resistant cells, we first tested the ability of tumor cells to grow in the absence of the feeder layer. Complete depletion of the feeder cells was obtained at the second passage of tumor cells on plastic (data not shown). At this passage, tumor cells often showed a high ratio of nucleus to cytoplasm, prominent nucleoli, and colony morphology similar to that of embryonic stem cells (data not shown). However, following repeated passages, colonies with these characteristics became less frequent. These passaged feeder-free cells will be hereinafter referred to as 'stem-like cells'. After one passage on the feeder layer, and two passages on plastic, all five cell lines were treated with serial dilutions of 5-FU to establish an IC50 dose. Low 5-FU doses (10-25 µM) surprisingly led to increased cell numbers in most of the cultures ( Figure 1A). We found the subpopulation of cells expressing the cancer-initiating cell marker EpCAM [14] to proliferate in response to low 5-FU doses. Indeed, a higher number of cells stained positive for EpCAM after treatment, and only EpCAM positive cells also stained positive for the proliferation cell marker Ki-67 (data not shown). Unfortunately, we were not able to further propagate Tu11 and Tu14 cells. In addition, Tu28 cells survived for only two more passages on plastic, while Tu27 and Tu42 cells were easily expanded (Figure 1 B-D). Along with passages, these cells became less resistant to 5-FU (IC50 for Tu27 = 250 µM; IC50 for Tu42 = 100 µM) ( Figure 1B,D).
Once the IC50 dose for 5-FU was established, we attempted the generation of 5-FU resistant cells by either intermittent treatment with the IC50 drug dose (hereinafter called R1 cells; two pulses of 100 µM 5-FU were given) or continuous treatment with escalating drug doses (hereinafter called R2 cells; six doses were given, starting with the 1/20 IC50 dose up to the IC50 dose). Unfortunately, we were not able to establish stable 5-FU resistant Tu27 cells: as early as 5 weeks after the start of treatment, all cells died. Conversely, we successfully generated resistant Tu42 cells.
We retrospectively reviewed microarray data from short-term feeder-expanded Tu11, Tu14, Tu27, Tu28, and Tu42 CRC stem cells to investigate why we failed to generate resistant cancer cell lines most of the time. We found that with respect to the other cell lines, Tu42 cells downregulated the tumor suppressor PTEN/MMAC1/TEP1 (PTEN, phosphatase deleted on chromosome ten; MMAC1, mutated in multiple advanced cancers; TEP1, tensin-like phosphatase) ( Figure 1E). Thus, PTEN/MMAC1/TEP1 status might be a crucial determinant of 5-FU resistance in CRC.
Low 5-FU doses (up to the 20 µM) resulted in accumulation of large foci in Tu42 cells ( Figure  2A, middle panels). We found the subpopulation of cells expressing the cancer-initiating cell marker EpCAM [14] to proliferate in response to low 5-FU doses. Indeed, a higher number of cells stained positive for EpCAM after treatment, and only EpCAM positive cells also stained positive for the proliferation cell marker Ki-67 (data not shown). Unfortunately, we were not able to further propagate Tu11 and Tu14 cells. In addition, Tu28 cells survived for only two more passages on plastic, while Tu27 and Tu42 cells were easily expanded ( Figure 1B-D). Along with passages, these cells became less resistant to 5-FU (IC50 for Tu27 = 250 µM; IC50 for Tu42 = 100 µM) ( Figure 1B,D).
Once the IC50 dose for 5-FU was established, we attempted the generation of 5-FU resistant cells by either intermittent treatment with the IC50 drug dose (hereinafter called R1 cells; two pulses of 100 µM 5-FU were given) or continuous treatment with escalating drug doses (hereinafter called R2 cells; six doses were given, starting with the 1/20 IC50 dose up to the IC50 dose). Unfortunately, we were not able to establish stable 5-FU resistant Tu27 cells: as early as 5 weeks after the start of treatment, all cells died. Conversely, we successfully generated resistant Tu42 cells.
We retrospectively reviewed microarray data from short-term feeder-expanded Tu11, Tu14, Tu27, Tu28, and Tu42 CRC stem cells to investigate why we failed to generate resistant cancer cell lines most of the time. We found that with respect to the other cell lines, Tu42 cells downregulated the tumor suppressor PTEN/MMAC1/TEP1 (PTEN, phosphatase deleted on chromosome ten; MMAC1, mutated in multiple advanced cancers; TEP1, tensin-like phosphatase) ( Figure 1E). Thus, PTEN/MMAC1/TEP1 status might be a crucial determinant of 5-FU resistance in CRC. Immunofluorescence staining for Wheat germ agglutinin (WGA), Laminin B1 (LMNB1), SF2, Fibrillarin (FBR), Nucleostemin (NS), Nucleophosmin (NPM), or PML suggested high nucleus-tocytoplasm ratio in these cells, as well as the presence of nuclear buds, and identified the previously observed foci as nucleolar structures (Figure 2A, bottom panels). Interestingly, the nucleolar structures were highly reactive for Ki-67 ( Figure 2B), indicating that the cells were likely in interphase and responding to DNA damage [15,16]. Moreover, with respect to parental cells, 20 µM 5FU-resistant cells upregulated the pluripotent stem cell marker TRA-2-49 [17] ( Figure 2B) and downregulated EpCAM, Mucin 2 (Muc2, the major secretory mucin synthesized and secreted by goblet cells [18]), and Villin (a cytoskeletal protein and a differentiation marker [19]) ( Figure 2C). Interestingly, resistant cells could reacquire EpCAM expression when directed to form organoids in Matrigel, although they failed to generate organoids with a central lumen ( Figure 2D). These observations suggest two possible outcomes: a) low 5-FU doses given under two-dimensional culture conditions reprogram CRC stem-like cells into less differentiated CSCs and this reprogramming is partially reversible when the cells are cultured in three-dimensional (3D) matrices; or b) the selective pressure under low 5-FU doses allows survival of less differentiated CSCs already present in the culture. Immunofluorescence staining for Wheat germ agglutinin (WGA), Laminin B1 (LMNB1), SF2, Fibrillarin (FBR), Nucleostemin (NS), Nucleophosmin (NPM), or PML suggested high nucleus-to-cytoplasm ratio in these cells, as well as the presence of nuclear buds, and identified the previously observed foci as nucleolar structures (Figure 2A, bottom panels). Interestingly, the nucleolar structures were highly reactive for Ki-67 ( Figure 2B), indicating that the cells were likely in interphase and responding to DNA damage [15,16]. Moreover, with respect to parental cells, 20 µM 5FU-resistant cells upregulated the pluripotent stem cell marker TRA-2-49 [17] ( Figure 2B) and downregulated EpCAM, Mucin 2 (Muc2, the major secretory mucin synthesized and secreted by goblet cells [18]), and Villin (a cytoskeletal protein and a differentiation marker [19]) ( Figure 2C). Interestingly, resistant cells could reacquire EpCAM expression when directed to form organoids in Matrigel, although they failed to generate organoids with a central lumen ( Figure 2D). These observations suggest two possible outcomes: a) low 5-FU doses given under two-dimensional culture conditions reprogram CRC stem-like cells into less differentiated CSCs and this reprogramming is partially reversible when the cells are cultured in three-dimensional (3D) matrices; or b) the selective pressure under low 5-FU doses allows survival of less differentiated CSCs already present in the culture.

Differential Morphology, Gene Expression Profiles, and Biological Functions in 5-FU Resistant Tu42 CRC Stem-Like Cells as Compared to Parental Cells
While parental cells passaged in parallel progressively acquired a differentiated morphology comprising crypt-villus structures ( Figure 3A, top image), R1 cells organized in round colonies with defined edges following 8 months of 5-FU-induced selection pressure ( Figure 3A, middle image). Conversely, R2 cells treated with increasing drug doses (up to the 20 µM 5-FU dose) were initially organized in round colonies (data not shown) and eventually showed a more differentiated morphology that was similar to the parental cells ( Figure 3A, bottom image).

Differential Morphology, Gene Expression Profiles, and Biological Functions in 5-FU Resistant Tu42 CRC Stem-Like Cells as Compared to Parental Cells
While parental cells passaged in parallel progressively acquired a differentiated morphology comprising crypt-villus structures ( Figure 3A, top image), R1 cells organized in round colonies with defined edges following 8 months of 5-FU-induced selection pressure ( Figure 3A, middle image). Conversely, R2 cells treated with increasing drug doses (up to the 20 µM 5-FU dose) were initially organized in round colonies (data not shown) and eventually showed a more differentiated morphology that was similar to the parental cells ( Figure 3A, bottom image).  We isolated RNA from all three-cell groups (parental, R1, and R2) derived from Tu42, and analyzed 588 genes using several RT 2 Profiler PCR arrays in 96-well format. Each 96-well plate included 84 pathway-or disease-focused genes, as well as five housekeeping genes. Moreover, each plate included control elements for genomic DNA contamination detection, RNA sample quality, and three replicate controls to check general PCR performance. We found the system to be very reliable and reproducible, with Pearson correlation of gene expression between two technical replicates >0.99 (Supplementary Figure S1). Figure 3B shows the list of top up/downregulated genes in R1 and R2 cells with respect to parental cells. Both resistant cell groups upregulated CA9, MUC1, and CD24, while they downregulated HEY1, NANOG, and IL8. Biological functions including cell cycle progression, DNA replication, G1/S phase transition, metabolism of DNA, mitosis, and S phase were predicted by IPA to be activated in R1 cells with respect to parental cells ( Figure 3C). Conversely, biological functions including cell movement, cell proliferation, cell survival, cell viability, invasion of cells, metastatic solid tumor, migration of cells, repair of DNA, and self-renewal of cells were predicted to be inhibited in R2 cells with respect to parental cells ( Figure 3C). R2 cells were also predicted to activate biological functions like apoptosis and cell death ( Figure 3C). Thus, the intermittent treatment with the IC50 drug dose selected for cells with more aggressive features than the continuous treatment with escalating drug doses. Genes expressed differentially between parental and R1 cells or parental and R2 cells are shown in Figure 3D. Official full names for each reported gene are listed in Table 1. Table 1. Official full names for genes differentially expressed in vitro.

ABCC1
ATP binding cassette subfamily C member 1 ABL1 ABL proto-oncogene 1, non-receptor tyrosine kinase ACSL4 Acyl-CoA synthetase long chain family member 4 AKT2 AKT serine/threonine kinase 2 ATM ATM serine/threonine kinase ATR ATR serine/threonine kinase ATRX ATRX, chromatin remodeler Bone morphogenetic protein 4 CA9 Carbonic anhydrase 9 CBX1 Chromobox Cell division cycle 42 CDK1 Cyclin dependent kinase 1 CDK2 Cyclin dependent kinase 2 CDK4 Cyclin dependent kinase 4 CDK5 Cyclin dependent kinase 5 CEBPD CCAAT enhancer binding protein delta CHD3 Chromodomain helicase DNA binding protein 3 CHEK1 Eukaryotic translation initiation factor 4B EIF4E Eukaryotic translation initiation factor 4E ERBB2 Erb-b2 receptor tyrosine kinase 2 ESCO2 Establishment of sister chromatid cohesion N-acetyltransferase 2 FABP1 Fatty acid binding protein 1 FANCA FA complementation group A FANCG FA complementation group G FAS Fas (TNF receptor superfamily member 6) FEN1 Flap structure-specific endonuclease 1 FGFR2 Fibroblast growth factor receptor 2 FOSL1 FOS like 1, AP-1 transcription factor subunit FOXP1 Forkhead box P1 FZD7 Frizzled class receptor 7 GADD45A Growth arrest and DNA damage inducible alpha GCLC Glutamate-cysteine ligase catalytic subunit GCLM Glutamate-cysteine ligase, modifier subunit GSR Glutathione-disulfide reductase HAT1 Histone acetyltransferase 1 HDAC1 Histone deacetylase 1 HDAC11 Histone deacetylase 11 HDAC4 Histone deacetylase 4 HDAC6 Histone deacetylase 6 HDAC7 Histone deacetylase 7 HDAC9 Histone deacetylase 9 HES1 Hes family bHLH transcription factor 1 HEY1 Hairy/enhancer-of-split related with YRPW motif 1 Whereas normal cells or drug-responsive cancer cells repair DNA damage and go into senescence or apoptosis, loss of cell cycle checkpoints allows cancer cells to proliferate and perpetuate harmful mutations [10]. We analyzed multiple DNA damage and cell cycle-related genes in resistant cell derivatives. The top upregulated genes of these categories in R1 cells included CDK1, CDC25A, BLM, and AURKB. CDK1 and CDC25A force proliferation despite drug-induced DNA damage, due to their ability to completely disable cell-cycle checkpoints [20,21], and are therefore associated with tumor progression and drug resistance [22]. The BLM product is a helicase of the RECQ family with roles in DNA replication and repair [23]. AURKB overexpression is associated with multinucleation and polyploidy and therefore has a high correlation with genomic instability [24]. CDK1 and BLM were also upregulated in R2 cells, although to a lower extent. Finally, both R1 and R2 cells downregulated DDIT3, which has been implicated in stress responses leading to apoptosis [25].

Apoptosis
DNA-damaging agents induce apoptosis through both the mitochondrial and the death receptor pathway. Surprisingly, both R1 and R2 cells upregulated FAS and TNFSF10. While the overwhelming majority of studies on FAS and TNFSF10 have explored their role as initiators of apoptosis, sporadic reports suggest that they could actually be drivers of cell proliferation, migration, and invasion, and therefore contributors of tumor progression [26]. The tumorigenic activity of FAS is likely mediated by a pathway involving JNK and c-Jun. [27]. On the other hand, TNFSF10 can antagonize apoptosis induction and enhance proliferation through activation of the NF-κB pathway [28]. R1 cells also upregulated BIRC5. Mostly known as an anti-apoptotic factor, BIRC5 additionally controls cell division by regulating the G2/M phase [29]. The expression levels of the anti-apoptotic genes BBC3 and MCL1 were reduced in both resistant cell groups, especially in R2 cells. Moreover, R2 cells also downregulated an additional anti-apoptotic gene, BCL2L1.

Stemness
Both R1 and R2 cells upregulated the putative CSC marker CD24 and PROM1, to a lower extent. Both markers correlate with invasiveness in CRC [30]. KITLG, encoding the ligand for the c-kit proto-oncogene [31], was highly upregulated in R1 cells, but not in R2 cells. Both cell groups downregulated the pluripotent stem cell marker NANOG [32], especially R2 cells. Finally, ABCC1, which encodes a multidrug resistance-associated protein [33], was downregulated in R2 cells but not in R1 cells.

mTOR Pathway and Other Pathways
Several genes implicated in the mTOR pathway [50,51] were downregulated in R2 cells and unchanged or slightly changed in R1 cells. With the exception of HEY1, several genes encoding members of the Notch family (JAG1, LFNG, MAML1, NOTCH2 [52]) showed a more enhanced downregulation in R2 cells than R1 cells. Notch pathway activation has been suggested to function as a protective mechanism in response to chemotherapy [53]. Finally, STAT3 was slightly upregulated in R1 cells.

Metabolism
Metabolic alterations are common in tumors. These include increased uptake of glucose, amino acids, and lipids. Under normal physiological conditions, glucose is catabolized to pyruvate. Most of the pyruvate undergoes several changes to become acetyl-CoA and enter the Kreb's cycle. Only a small amount of pyruvate is converted to alanine or lactate. In contrast, tumor cells prefer lactic acid fermentation as a source of energy; that is, they convert pyruvate to lactate even under normoxic conditions. This phenomenon, referred to as the Warburg effect [54], is possible thanks to the upregulation of LDHA, a key glycolytic enzyme [55]. We found LDHA to be upregulated in both cell groups, especially in R2 cells. Carbon dioxide is released during aerobic glycolysis, leading to a decrease in the intracellular and extracellular pH. Studies have shown that cancer cells adapt to this acidic milieu thanks to the activity of carbonic anhydrases including CA9, which catalyzes the conversion of carbon dioxide into bicarbonate [56]. CA9 was highly upregulated in both groups, especially in R2 cells. Lipid metabolism is often elevated in cancer cells and plays an important role in their growth and malignancy. FABP1, which encodes a protein critical for fatty acid uptake and intracellular transport [57] was upregulated in both cell groups, especially in R2 cells. Other genes whose products are lysosomal proteases involved in lipid metabolism (CTSB, CTSD, and CTSL [58]) were also downregulated in R2 cells. Finally, resistant cells also showed alterations of genes encoding proteins involved in fatty acid synthesis (ACSL4 [59]) or oxidation (CPT2 [60]), as well as glutathione metabolism (GCLC, GCLM, GSR, and NQO1 [61]) and oxidative stress (HMOX1 [62]).

Chromatin Remodeling
The chromatin remodeling genes which showed the greatest changes in R1 cells were ESCO2, CBX1, and SUV39H1, which were all upregulated. ESCO2 is critical for the cohesion of sister chromatids during DNA replication and DNA damage repair [63]. Besides its role in DNA repair, CBX1 has opposite effects in differentiated cells and pluripotent stem cells: it is required for the differentiation of differentiated cells while it maintains pluripotency in pluripotent stem cells [64]. Conversely, SUV39H1 regulates cell migration and tumor growth [65]. While ESCO2 was also upregulated in R2 cells (although 2.5-fold less than in R1 cells), R2 cells highly downregulated the chromatin remodeling genes HDAC4 and DOT1L. HDAC4 promotes growth of colon cancer cells via repression of p21 [66]. DOT1L deficiency leads to the dysregulation of cell cycle regulators and G1 cell cycle arrest [67].
Collectively, our findings suggested that R1 cells might have had a more resistant phenotype than R2 cells. This hypothesis was further supported by the finding that R2 cells lost the expression of EIF4B ( Figure 3E), a substrate of the mTOR pathway, required for cell proliferation and survival of cancer cells [68]. Our analysis also identified EGF and MYCN as two de novo expressed genes in the population of resistant cells ( Figure 3E). EGF has been reported to induce MYCN expression and to favor tumor growth [69].
Overexpression of CD24 and Muc1 (CD227) in R1 cells was confirmed using flow cytometry ( Figure 3F). Moreover, we confirmed unaltered and decreased mTOR activity in R1 and R2 cells, respectively, by staining for the downstream mTOR target S6RP. We found that most of the phospho-S6RP (pS6RP-S235/236 and pS6RP-S240/244) expression localized at the cell colonies' edges ( Figure 3G). Mathematical models suggest that the growth in size of a cellular colony is due to a ring of growing and proliferating cells at the colony edge [70]. The mTOR signaling plays a major role in promoting cell-cycle progression, and downregulation of phospho-S6RP is a feature of slow-cycling cancer cells [71]. Thus, we hypothesize that parental and R1 cell-derived colonies contained proliferating cells at their edges expressing pS6RP. Proliferation is reduced in R2 cells, and this might be an effect of chemotherapy-induced mTOR pathway downregulation. Importantly, 5-FU chemoresistance in R1 cells was maintained over time, as indicated by high 5-FU IC50 value two months after their establishment (Supplementary Figure S2).

Differential Histology, Gene Expression Profiles, and Biological Functions of 5-FU Resistant Tu42 CRC Stem-Like Cells as Compared to Parental Cells In Vitro and In Vivo
We next established tumor xenografts using both parental and R1 cells. Morphologically, the tumors were different ( Figure 4A).   Characteristics of R1 cell-derived tumor xenograft included a pseudostratified glandular epithelium with marked nuclear atypia and cellular overlapping, and few mucinous goblet cells. Conversely, parental cell-derived tumor xenograft showed a lesser degree of architectural abnormalities including glands with low-grade cytologic dysplasia and abundant goblet cells ( Figure 4A). We therefore ran the same PCR arrays as before in the two tumor xenografts and compared their expression profiles. Table 2 lists the official full names for each gene found to be differentially regulated. Table 2. Official full names for genes differentially expressed in vivo.

ABCC1
ATP binding cassette subfamily C member 1 BCL2L1 BCL2 like 1, Bcl-xl BMP2 Bone morphogenetic protein 2 BMP4 Bone morphogenetic protein 4 CA9 Carbonic anhydrase 9 CEBPD CCAAT enhancer binding protein delta CDC25A Cell division cycle 25A CHD3 Chromodomain helicase DNA binding protein 3 DNMT3A DNA methyltransferase 3 alpha DOT1L DOT1 like histone lysine methyltransferase ERBB2 Erb-b2 receptor tyrosine kinase 2 EIF4B Eukaryotic translation initiation factor 4B FAS Fas (TNF receptor superfamily member 6) FABP1 Fatty acid binding protein 1 FGFR2 Fibroblast growth factor receptor 2 FOXP1 Forkhead box P1 FZD7 Frizzled class receptor 7 HDAC11 Histone deacetylase 11 HDAC9 Histone deacetylase 9 ID1 Inhibitor of DNA binding 1, HLH protein IGF1 Insulin like growth factor 1 IGF1R Insulin like growth factor 1 receptor IGFBP3 insulin like growth factor binding protein 3 IRS1 Insulin Solute carrier family 2 member 1, GLUT-1 TWIST1 Twist family bHLH transcription factor 1 TWIST2 Twist family bHLH transcription factor 2 YAP1 Yes associated protein 1 ZEB2 Zinc finger E-box binding homeobox 2 While we could not detect any mRNA transcript for IGF1, IGFBP3, KIT, PRKCB, PTEN, PTGS2, SERPINE, TWIST1, TWIST2, and ZEB2 in vitro, all these genes were expressed in vivo ( Figure 4B), possibly indicating different promoters' methylation status or different microRNA regulation in the two conditions. Importantly, all the above-mentioned genes were upregulated in R1 cell-derived tumor xenograft with respect to parental cell-derived tumor xenograft ( Figure 4B) and some of them (KIT, SERPINE, TWIST1, TWIST2, and ZEB2) could even be included in the top upregulated gene list of R1 cell-derived tumor xenograft versus parental cell-derived tumor xenograft ( Figure 4C). These genes encode for proteins with well-known roles in cancer progression and normal stem cell maintenance (KIT [72]), angiogenesis (SERPINE [73]), and epithelial-mesenchymal transition (TWIST1, TWIST2, and ZEB2 [74,75]). The list of top upregulated genes in R1 cell-derived tumor xenograft versus parental cell-derived tumor xenograft also included ID1 and BMP4 ( Figure 4C). Moreover, FAS and CDKN1A were identified as the top downregulated genes in R1 cell-derived tumor xenograft versus parental cell-derived tumor xenograft ( Figure 4C). Biological functions associated with cancer were predicted by IPA to be activated in R1 cell-derived tumor xenograft versus parental cell-derived tumor xenograft, while biological functions associated with death were predicted to be inactivated ( Figure 4D).
Interestingly, we found that only a few of the genes which were upregulated in vitro by R1 cells were also upregulated in vivo. These included CDC25A, PARP1, PROM1, ERBB2, CA9, and HDAC9 ( Figure 4E). None of the genes which were downregulated in vitro were also downregulated in vivo. Opposing trends were observed for FAS, NANOG, CCL5, CEBPD, FGFR2, FOXP1, JAG1, and FABP1. As expected, in vivo, fewer of the genes associated with DNA damage and cell cycle, metabolism, or chromatin remodeling were changed. The greatest changes affected the invasion, metastasis and angiogenesis gene category, with ID1 and BMP4 being the most two upregulated genes. Furthermore, many genes in the mTOR pathway category were upregulated in vivo by R1 cells. Most of these genes (EIF4B, PRAS40, RAPTOR, RPS6KA2, RPS6KB2, and RRAGA) are directly or indirectly involved in the mTORC1 pathway [50]. This agrees with our previous observation that mTORC1 pathway activation is a feature of colorectal tumors [76].

ID1-Expressing Cell Enrichment Is a Feature of 5-FU Resistant Tu42 Cell-Derived Tumor Xenograft
This study has identified ID1 as the top upregulated gene in R1 cell-derived tumor xenograft. Using immunohistochemistry, we confirmed ID1 overexpression at the protein level ( Figure 5A). Interestingly, both ID1 positive and negative cells could be observed, leading to the speculation that ID1 could mark a subpopulation of cells with stemness features in the tumor. To establish ID1 as a stemness marker, the presence of ID1 in normal intestinal stem cells and its overexpression in cancerous intestinal stem cells need first to be demonstrated. To this purpose, we retrospectively compared the levels of ID1 in short-term feeder-expanded primary epithelial cells isolated from human fetal small or large intestine to short-term feeder-expanded primary epithelial cells isolated from CRC biopsies. All three-cell types were cultured under conditions promoting the enrichment of cells with stemness features and are herein indicated as SiSCs, LiSCs, and CoCSCs, respectively [13,77].
We previously found intrinsic differences between SiSCs and LiSCs, with a tendency for cells expanded from the large intestine to be more closely related to cancer than their small intestine counterparts, possibly explaining why cancer is 20 times more prevalent in the large intestine than the small intestine in humans [77]. Accordingly, while ID1 levels were statistically significant lower in SiSCs with respect to CoCSCs, there was no significant difference between LiSCs and CoCSCs, although ID1 levels tended to be higher in CoCSCs ( Figure 5B). These observations seem to suggest that ID1/ID1 could indeed mark cells with stemness features in our system. ID1/ID1 might be enriched in the cell-of-origin of intestinal cancer besides being expressed by cancer-propagating cells possessing stemness properties. Not only ID1, but also additional ID family members might identify cells with these characteristics. Indeed, we found ID2-4 to also be expressed ( Figure 5A). While ID1 was detected in both cytoplasm and nucleus, ID2 and ID3 showed a clear nuclear Interestingly, both ID1 positive and negative cells could be observed, leading to the speculation that ID1 could mark a subpopulation of cells with stemness features in the tumor. To establish ID1 as a stemness marker, the presence of ID1 in normal intestinal stem cells and its overexpression in cancerous intestinal stem cells need first to be demonstrated. To this purpose, we retrospectively compared the levels of ID1 in short-term feeder-expanded primary epithelial cells isolated from human fetal small or large intestine to short-term feeder-expanded primary epithelial cells isolated from CRC biopsies. All three-cell types were cultured under conditions promoting the enrichment of cells with stemness features and are herein indicated as SiSCs, LiSCs, and CoCSCs, respectively [13,77].
We previously found intrinsic differences between SiSCs and LiSCs, with a tendency for cells expanded from the large intestine to be more closely related to cancer than their small intestine counterparts, possibly explaining why cancer is 20 times more prevalent in the large intestine than the small intestine in humans [77]. Accordingly, while ID1 levels were statistically significant lower in SiSCs with respect to CoCSCs, there was no significant difference between LiSCs and CoCSCs, although ID1 levels tended to be higher in CoCSCs ( Figure 5B). These observations seem to suggest that ID1/ID1 could indeed mark cells with stemness features in our system. ID1/ID1 might be enriched in the cell-of-origin of intestinal cancer besides being expressed by cancer-propagating cells possessing stemness properties. Not only ID1, but also additional ID family members might identify cells with these characteristics. Indeed, we found ID2-4 to also be expressed ( Figure 5A). While ID1 was detected in both cytoplasm and nucleus, ID2 and ID3 showed a clear nuclear expression, again confined to a subset of cells. Conversely, ID4 was weakly expressed ( Figure 5A). IDs have been described as BMP4 targets in different cell types including embryonic stem cells [78]. In agreement with real-time PCR arrays, BMP4 was highly expressed in tumor sections ( Figure 5A). Potential activated pathways and targets in R1-cell derived tumor xenograft are shown in Figure 6. expression, again confined to a subset of cells. Conversely, ID4 was weakly expressed ( Figure 5A). IDs have been described as BMP4 targets in different cell types including embryonic stem cells [78]. In agreement with real-time PCR arrays, BMP4 was highly expressed in tumor sections ( Figure 5A). Potential activated pathways and targets in R1-cell derived tumor xenograft are shown in Figure 6.

Discussion
Chemotherapeutic resistance represents a major treatment obstacle as most cancer patients ultimately relapse, becoming refractory to additional chemotherapeutic drugs.
Panels of chemotherapy resistant cancer cell lines have been generated in order to establish phenotypic signatures that might help predict chemotherapeutic sensitivity. However, despite the number of reported studies, there is no standardized protocol for reproducible generation of chemotherapeutic resistant cells. Both pulse and incremental methods have been described. However, a comparison of gene expression in the respective treatment-derivatives remains incomplete. Moreover, to our knowledge, re-transplantation into an in vivo setting of such in vitroestablished resistant cell lines has been poorly studied.
In this study, we aimed at generating 5-FU resistant cells lines from five patient-derived CRC stem cell lines previously isolated and characterized in our lab [13]. Our data suggested that PTEN/MMAC1/TEP1 status could be a key determinant for successful generation of drug resistant cells. Indeed, only Tu42 cells, which downregulated PTEN/MMAC1/TEP1, could survive long-term 5-FU treatment. PTEN/MMAC1/TEP1 is the most frequently inactivated tumor suppressor gene in sporadic cancer [79] and acts by inhibiting cancer cell proliferation and invasiveness and promoting apoptosis through its antagonism of PI3K. Tu42 cells did not express PTEN/MMAC1/TEP1 in vitro; however, PTEN/MMAC1/TEP1 expression could be detected in Tu42 cell-derived tumor xenograft, indicating reversible PTEN/MMAC1/TEP1 loss in these cells.
We used two treatment regimens to generate resistant cells, and we compared the transcriptional profiles of the resistant cell derivatives by focusing on almost 600 genes with roles in cell cycle, DNA repair, apoptosis, stemness, metastasis, metabolism, and epigenetics. We identified similarities and differences of 5-FU resistant cells generated by intermittent treatment with the IC50 drug dose or continuous treatment with escalating drug doses. Similarities included upregulation of CA9, MUC1, and CD24, and downregulation of HEY1, NANOG, and IL8. Similarities also included de novo acquisition of two potent oncogenes with crucial roles in cell growth and proliferation, EGF and MYC [80]. Despite these similarities, 5-FU resistant cells generated by continuous treatment with escalating drug doses appeared to be less resistant than the other set of treated cells, showing downregulation of numerous genes involved in invasion, metastasis, and angiogenesis, as well as mTOR pathway downregulation. From a therapeutic standpoint, this data suggests that escalating the dose of chemotherapy might enhance efficiency. However, at the beginning of the treatment, cells that are being treated with escalating drug doses might pose a

Discussion
Chemotherapeutic resistance represents a major treatment obstacle as most cancer patients ultimately relapse, becoming refractory to additional chemotherapeutic drugs.
Panels of chemotherapy resistant cancer cell lines have been generated in order to establish phenotypic signatures that might help predict chemotherapeutic sensitivity. However, despite the number of reported studies, there is no standardized protocol for reproducible generation of chemotherapeutic resistant cells. Both pulse and incremental methods have been described. However, a comparison of gene expression in the respective treatment-derivatives remains incomplete. Moreover, to our knowledge, re-transplantation into an in vivo setting of such in vitro-established resistant cell lines has been poorly studied.
In this study, we aimed at generating 5-FU resistant cells lines from five patient-derived CRC stem cell lines previously isolated and characterized in our lab [13]. Our data suggested that PTEN/MMAC1/TEP1 status could be a key determinant for successful generation of drug resistant cells. Indeed, only Tu42 cells, which downregulated PTEN/MMAC1/TEP1, could survive long-term 5-FU treatment. PTEN/MMAC1/TEP1 is the most frequently inactivated tumor suppressor gene in sporadic cancer [79] and acts by inhibiting cancer cell proliferation and invasiveness and promoting apoptosis through its antagonism of PI3K. Tu42 cells did not express PTEN/MMAC1/TEP1 in vitro; however, PTEN/MMAC1/TEP1 expression could be detected in Tu42 cell-derived tumor xenograft, indicating reversible PTEN/MMAC1/TEP1 loss in these cells.
We used two treatment regimens to generate resistant cells, and we compared the transcriptional profiles of the resistant cell derivatives by focusing on almost 600 genes with roles in cell cycle, DNA repair, apoptosis, stemness, metastasis, metabolism, and epigenetics. We identified similarities and differences of 5-FU resistant cells generated by intermittent treatment with the IC50 drug dose or continuous treatment with escalating drug doses. Similarities included upregulation of CA9, MUC1, and CD24, and downregulation of HEY1, NANOG, and IL8. Similarities also included de novo acquisition of two potent oncogenes with crucial roles in cell growth and proliferation, EGF and MYC [80]. Despite these similarities, 5-FU resistant cells generated by continuous treatment with escalating drug doses appeared to be less resistant than the other set of treated cells, showing downregulation of numerous genes involved in invasion, metastasis, and angiogenesis, as well as mTOR pathway downregulation. From a therapeutic standpoint, this data suggests that escalating the dose of chemotherapy might enhance efficiency. However, at the beginning of the treatment, cells that are being treated with escalating drug doses might pose a threat. Indeed, our data of TRA-2-49 upregulation together with EpCAM, Muc2, and Villin downregulation in 20 µM 5-FU resistant cells seems to suggest mechanisms of epithelial-mesenchymal transition, oncogenic dedifferentiation, and acquisition of stemness features [81,82]. It is worth spending a few words on EpCAM because it might help track phenotypic changes during chemotherapeutic treatment: EpCAM-expressing cells proliferated in response to 10 µM 5-FU; later doubling of the 5-FU concentration resulted in the downregulation of EpCAM, however EpCAM expression could be restored when the cells were cultured in 3D matrices, which represents an assay of cell differentiation. This data highlights the dynamic nature of CRC stem-like cells which undergo constant dynamic changes to adapt and survive in adverse conditions. By engrafting 5-FU resistant cells generated by intermittent treatment with the IC50 drug dose in the mouse host, we confirmed the capacity of such cells to regenerate a tumor in vivo. The observation that several genes implicated in stem cell maintenance and cancer progression were de novo expressed, in vivo indicated that resistant cells continued to adapt in the new environment, even in the absence of further chemotherapeutic insult.
Interestingly, while resistant cells generated by intermittent treatment with the IC50 5-FU dose and control parental cells passaged in parallel did not differ for the expression of ID1 in vitro, ID1 was the top upregulated gene in the resistant cell-derived tumor xenograft as compared to the control tumor xenograft. Using immunohistochemistry, we confirmed ID1 overexpression in the resistant cell-derived tumor xenograft at the protein level. Both ID1 positive and negative cells could be observed. The mechanisms driving this intratumoral heterogeneity remain to be fully investigated. A tumor is a complex ecosystem of relatively differentiated cancer cells and CSCs as well as other cell types. In the tumor, individual cells display a diverse set of characteristics and function together to support the growth and maintenance of the tumor as a whole. ID1-expressing cells within the drug resistant cancer cell pool might have preferentially proliferated following transplantation into the mouse host, while the ID1-expressing cells within the untreated cancer cell pool might have had a decreased proliferation potential. Emerging technologies including single-cell transcriptomics could allow us to test our hypothesis and determine whether ID1 is restricted to a CSC subpopulation in the primary colorectal tumor and is pervasively expressed in secondary lesions emerging after completion of chemotherapy. Undoubtedly, our retrospective analysis of microarray ID1-expression indicated that ID1 is already expressed in normal intestinal stem cells in accordance with a previous study [83], and therefore, the ID1-expressing population might play a role in tumor initiation besides propagation.
In conclusion, our study connects the capacity for adaptation and propagation of CRC stem-like cells after chemotherapeutic pressures, knowledge that could be exploited in future therapeutic strategies.

Specimens
Tumor tissues were obtained by David Geller (surgeon, University of Pittsburgh Medical Center, Pittsburgh, PA, USA) after informed consent in accordance with institutional review board (IRB) protocols from patients with colorectal adenocarcinoma (IRB:PRO08010372).

Cytotoxicity Assay
Tu27, Tu28, and Tu42 cells were seeded onto 96-well plates (10,000 cells/well) in 100 µL of culture medium overnight, and then treated in triplicate with serial dilutions of 5-FU (10-500 µM) for 72 h. The number of viable cells was measured as above described.

Establishment of 5-FU Resistant Cell Lines
Tu42 cells were treated with 100 µM 5-FU for 72 h. Surviving cells were allowed to recover 21 weeks in fresh media before a second pulse of the drug was given. Surviving cells were allowed to recover additional 14 weeks before further applications (R1 cells). Alternatively, 5-FU resistant cell lines (R2 cells) were established after six sequential treatments with 5-FU during an 8-month period (5 µM and 100 µM, initial and final 5-FU concentration, respectively). Control parental cells were passaged in parallel.

Organoid Preparation
Tu42 cells were suspended in 50µL Growth Factor Reduced (GFR) Matrigel (Corning, cat.no. 354230) and plated in tissue culture dishes at low density (50,000 cells/well of 48-well plate). After solidification at 37 • C for 30 min, the Matrigel was overlaid with culture media. Whole intact Tu42 organoids were fixed 30 min in 4% PFA, and embedded in optimal cutting temperature compound (OCT, Fisher Scientific, cat. no. 4853) for cryostat sectioning.

RNA Extraction, cDNA Synthesis, PCR Arrays
Total RNA was isolated from cells or tissues stored in RNAlater reagent (QIAGEN, cat. no. 76104; Valencia, CA, USA) using the RNeasy Mini kit (QIAGEN, cat. no. 74104), according to the manufacturer's instructions. Potentially contaminating genomic DNA was digested using DNase (QIAGEN, cat. no. 79254). Purity of isolated RNA was determined by measuring ratio of the optical density of the samples at 260 and 280 nm. The OD 260 / 280 ratio was ranging from 2.06 to 2.12 for all samples. cDNAs were synthesized using the RT 2 First Strand Kit (QIAGEN, cat. no. 330401). QIAGEN's RT 2 PCR Array Human RNA QC (cat. no. PAHS-999Z) was used to test the quality of the RNA samples before proceeding with the PCR arrays. A total of 588 genes were analyzed in duplicate using the following QIAGEN's RT 2 Profiler PCR arrays: Human mTOR Signaling (cat. no. PAHS-098A), Human Cancer Stem Cells (cat. no. PAHS-176A), Human Signal Transduction Pathway Finder (cat. no. PAHS-014ZA), Human Damage Signaling Pathway (cat. no. PAHS-029ZA), Human Cancer Targets (cat. no. PAHS-507ZA), Human Epigenetic Chromatin Modification Enzymes (cat. no. PAHS-085ZA), and Human Epigenetic Chromatin Remodeling Factors (cat. no. PAHS-086ZA). 0.5 µg total RNA was used for each plate. A StepOnePlus™ Real-Time PCR System (Applied Biosystems, Foster City, CA, USA) was used for gene amplification. Dissociation (melting) curve analysis was performed to verify PCR specificity. The analysis was performed automatically according to the SABiosciences company web portal (QIAGEN), and further confirmed manually using the 2 −∆∆Ct method. Changes in gene expression obtained from two technical replicates were illustrated as fold regulations.

Ingenuity Pathway Analysis (IPA)
To identify biological functions perturbed in response to chemotherapy-induced selective pressure, the dataset representing the log2 (fold change) of the expression level of 439 genes analyzed under different culture conditions was imported into the QIAGEN's IPA tool. Both core and comparison analyses were performed. The following settings were used: the Ingenuity Knowledge Base (genes only) database was used as a reference set; direct and indirect relationships were included, as well as molecule relationships with endogenous chemicals, with a maximum of 35 focus molecules per network and a maximum of 25 networks per analysis. All node types and data sources were included. The confidence level was set to include experimentally observed relationships. Only human genes were