Epigenetically distinct sister chromatids and asymmetric generation of tumor initiating cells

ABSTRACT

Cancer stem cells (CSC) are thought to be an important source of cancer cells in tumors of different origins. Mounting evidence suggests they are generated reversibly from existing cancer cells, and supply new cancer cells during tumor progression and following therapy. Elegant lineage mapping stud(ies are identifying progenitors, and in some cases differentiated cells, as targets of transformation in a variety of tumors. Recent evidence suggests resulting tumor initiating cells (TIC) might be distinct from CSC. Molecular pathways leading from cells of tumor origin to precancerous lesions and cancer cells are only beginning to be unraveled. We review a pathway where asymmetric division of precancerous cells generates TIC in a K-Ras-initiated model of lung cancer. And, we compare unexpected steps in this asymmetric division to those evident in well-studied stem cell models.

Keywords:

• Cancer
• tumor initiating cells
• cancer stem cells
• asymmetric division

Introduction

Cancer stem cells (CSC) have been identified as a subpopulation in solid and hematologic cancers [1–5]. Functional identification of CSC has classically relied on xenographic transplant of cells isolated from tumors. Relative to the bulk of cancer cells in tumors, a small number of CSC can regenerate the original heterogeneous tumor following transplant. Like stem cells, it is widely accepted that CSC can divide asymmetrically to produce differentiated cancer cells. But, details of niches were CSC reside in vivo and interaction of CSC with these niche are just beginning to emerge. Surprisingly, recent studies demonstrate that existing cancer cells undergo reprogramming to generate CSC [1,2,5–8], and properties attributed to CSC appear plastic within tumors, implying transition between cancer cell and CSC is reversible, and this pathway might reflect a gradient of changes. Graduated progression back and forth between cancer cell and CSC might explain findings of distinct expression patterns in cells characterized as CSC in patients [2,5,6,9–11]. Similarly, it is though that unipotent adult tissue stem cells can revert back to a multipotent professional stem cell state in response to injury [12]. And, normal stem cells and progenitors have also been suggested to be sources of CSC [13]. As with normal stem cells, CSC display bivalent repressive and activating epigenetic marking of a set of genes linked to development and cancer initiation [8]. These poised bivalent genes can then be repressed or activated by environmental factors during cancer cell generation [13].

Reprogramming of cancer cells to CSC can be triggered by EMT driven by transcription factors including Zeb1 [2,14,15]. Notably, this EMT driven by Zeb1 in CSC appears distinct from that seen in normal tissue stem cells, which has been linked to Slug (Snai2) [15]. EMT in CSC has been shown to result in high expression of CD44, which marks CSC in tumors including breast and lung cancers [16–19], and a positive CD44/Zeb1 loop has been shown to drive EMT and reprogramming of existing cancer cells to a CSC phenotype [20,21]. Notably, this EMT distinguishes CSC from the bulk of carcinoma cells, which continue to express epithelial markers such as E-cadherin. This CD44/Zeb1 loop can be initiated by Tgf-β induction of Zeb1 in cell culture [2], but it is unclear if such a loop is present or functional in vivo. Mounting evidence suggests that EMT in CSC also causes these cells to act as an early-metastasizing population [22]. And, consistent with plasticity of the CSC phenotype, following metastasis, the cells undergo a differentiating MET to resemble cancer cells [5,23–27].

The notion that CSC are derived from a reversible dedifferentiation of existing cancer cells in expanding tumors raises the important question as to whether cells with CSC properties are also original tumor initiating cells (TIC). Lineage mapping studies are identifying progenitors and in some cases differentiated cells as cells of origin targeted for transformation [5,28,29]. But, whether these cells of origin undergo an intermediate transition to CSC in their pathway to cancer cell generation is unclear, highlighting the fact that the pathway leading from cell of origin to precancerous cells and cancer cells warrants further investigation.

Ras-pathway-initiated lung tumors

Ras pathway mutations, including K-Ras itself and EGFR, which signals Ras activation, have been widely utilized in mouse models of human lung cancer [30]. These mutations are mutually exclusive in human lung cancer, suggesting they are redundant and thus equivalent in Ras pathway activation in the lung [31]. Mutations such as Pten or p53 affect tumor progression in this K-Ras model, and they have been widely utilized with K-Ras to evaluate their roles in tumors. Notably, Pten is not mutated in K-Ras initiated tumors such as lung and pancreatic cancer, but, instead, its expression is somehow repressed as these tumors progress, accounting for Pten mutation accelerating tumor progression in these mice [32–34]. Compound mutation of p53 does not affect cancer cell generation or their expansion into tumors [35,36]. Instead, its mutation allows K-Ras-initiated tumors to transition to metastasis, implying p53 is acting later to promote cancer cell metastasis in this model.

As opposed to compound mutations generated simultaneous in mouse models, mutations are though to arise sequentially over a long period in patients. In this regard, it is of note K-Ras mutation alone initiates a pathway leading to lung cancer in mice, but with this single mutation the process is highlighted by a protracted period of precancerous lesion expansion [37–39]. In these mice, precancerous subpleural adenomas form around bronchial airways. Cancer cells appear later as foci within these expanding adenomas, and they expand into large tumors that invade airways. Elegant studies in the lung have shown SPC+ ATII cells are the targets of K-Ras transformation responsible for precancerous proliferating adenoma generation in this model [29,40]. Cancer cells are also derived from ATII cells, but it is still unclear if they are derived from adenoma cells, or if they are generated directly from ATII cells in a pathway parallel to adenoma. We recently focused studies on sites of initial cancer cell generation within expanding adenomas in the lungs of mice expressing mutant K-Ras with the aim of characterizing a precancerous-to-cancer transition [39].

A zeb1-dependent pathway in lung precancerous-to-malignant transition

We demonstrated previously that Zeb1 is important for tumor progression in K-Ras-initiated lung cancer [38]. In these earlier studies, we attributed the role for Zeb1 to a classic function for EMT in cancer cell behavior (e.g., migration and invasion). However, our more recent studies suggest that Zeb1 mutation had no effect on lung adenoma formation, but cancer cell generation within these expanding adenomas was blocked, demonstrating a critical role for Zeb1 earlier, in transition of precancerous adenoma cells to cancer cells [39]. We identified a Zeb1-driven pathway critical for this transition containing the Zeb1 family member Zeb2, CD44, Akt, Pten, Yap1, Bmi1, and Numb. Indeed, we found that expression of components of this pathway, e.g., Pten and Zeb2, is a remarkable predictor of patient outcome.

Zeb1-dependent TIC in expanding adenomas are not CSC

Previous studies have found that Zeb1 expression can drive cancer cells toward a CSC phenotype [2,14,15]. The Zeb1 promoter under these conditions was shown to be in a poised chromatin state [2,8]. Our results identifying a Zeb1-dependent pathway important for cancer cell generation in vivo might then provide new details linked to how Zeb1 expression is regulated and what genes it targets in CSC formation [39]. But, in these studies we found that TIC generated in the lung of K-Ras mutants do not have properties of CSC. Thus, this Zeb1 pathway seems aimed at TIC as opposed to CSC per se. Despite a high level of Zeb1 in invading cancer cells in the K-Ras model, the cells remained CD44^lo and we failed to detect cells with properties of CSC within the tumors. We interpreted these findings to mean that cells with properties of CSC are not being generated from existing cancer cells in this model. Notably, migrating cells with properties of CSC are also linked to metastasis, and despite formation of large invasive tumors, no metastasis was evident in this K-Ras model [30,37,39]. Following isolation of CD44^hi TIC from expanding adenomas, the cells rapidly lost CD44 expression, failed to expand in culture and did not produce xenographic tumors, indicating they are not CSC [39]. In contrast to the K-Ras model, a small subset of cells could be cultured from K-Ras, p53 compound mutant tumors, and these cells were highly tumorigenic when delivered back into the lung [41]. These cells correspond to a CD44^hi subset of the tumors. p53 represses CD44 [21], and we concluded that p53 induction in culture, which is classically responsible for senescence in primary rodent cell cultures [42], is likely causing loss of CD44 expression and failure of expansion of the cells in culture. Evidence suggests CSC in tumors are derived from existing cancer cells that undergo reversible p53-regulated reprogramming, they are important for ongoing tumor expansion and tumor renewal following therapy, and p53 mutation is required for a CSC pool in mouse breast cancer [43]. p53 triggers stable repressive epigenetic changes [44], and we propose p53 in expanding cancer cells in the K-Ras model is responsible for maintaining CD44 silencing [39]. We suggest inability to induce CD44 in the cancer cells prevents their reprogramming to CSC until p53 is mutated. Although it remains to be seen how the gene signature of CD44^hi, Zeb1^hi TIC we identified might be related to CSC, we propose CSC generation from existing cancer cells reflects reprogramming aimed at restoring the cancer cell-generating phenotype of TIC.

Asymmetric division: what is the initiating event?

Asymmetric divisions of embryonic stem cell populations are the cornerstone of tissue formation during development, and tissue-specific stem cells are similarly responsible for ongoing tissue maintenance in adults [45,46]. Surprisingly, mounting evidence suggests that unipotent tissue stem cells can convert to multipotential professional stem cells in response to tissue damage [12]. Polarization of stem cells within niches via interaction with extracellular matrix and/or cell-cell contacts leads to a stromal/macrophage-induced gradient of Notch and Wnt signaling, causing a Numb-dependent reorientation of mother and daughter centrosomes and thereby establishing an asymmetric plane of division in cells, resulting in unequal inheritance cytoplasmic content in the two daughters [45,47–49]. One target of this signaling is miR-205, which in turn regulates both Numb and Zeb1 [49], and in a negative feedback loop, Zeb1 represses miR-205 [14]. It is presumed that resulting asymmetric segregation of cytoplasmic content in turn drives differential gene expression in the two daughters that is responsible for maintaining a stem cell phenotype in one daughter and driving a differentiated phenotype in the other daughter [45].

Key to establishing gene expression patterns that distinguish the differentiated daughter from the stem cell daughter are epigenetic modifications of histones [50–53]. Differential histone modifications in daughter cells might be driven by unequal distribution of histone modifying enzymes, nevertheless such modifications must be transmitted to new histones that bind to DNA strands following replication. Notably, histones are removed from template DNA in front of the replication fork [54–58]. In symmetrically dividing cells new and old histones appear to be reassigned similarly to the new DNA strands, but in some asymmetrically dividing Drosophila stem cells new histone H3 was shown to selectively associate before mitosis with sister chromatids destined for transfer to the differentiated daughter, whereas old histones remained associated with sister chromatids to be transferred to the stem cell daughter [56–58]. Importantly, phosphorylation of new synthesized H3 was delayed, and it was hypothesized that differences in mother and daughter centrosome maturation caused the two centrosomes to distinguish H3 phosphorylation, allowing asymmetric sister chromatid segregation to the two daughter cells [56]. As opposed to H3, H3.3, which is deposited onto DNA later after replication, was equally distributed on the sister chromatids [56], consistent with old vs. new H3 incorporation being replication-dependent. Somewhat complementary studies in human cells have examined the timing of reconstituting histone epigenetic marks on newly synthesized histones during symmetric mitosis [59]. Notably, acetylation of newly synthesized histones is required for their transfer to DNA [54]. Thus, newly synthesized histones are enriched for acetylation marks [59]. However, these marks were replaced by other key nucleosome regulatory marks such as trimethylation, which regulates both transcriptional repression and activation, in a fashion matching surrounding old histones. But, this mark replacement is largely delayed until the following cell cycle [59]. Thus, in the case of asymmetric division, the DNA strand repopulated primarily by new histones would remain relatively free of epigenetic remarking for some time. Further, new histones under these conditions might lack the instructive information for epigenetic remarking provided by surrounding old histone complexes in symmetrically dividing cells. Taken together, these results imply that in some cases asymmetric division can initiate with directed segregation of sister chromatids based on differences in histone epigenetic marks to the two daughter cells [57,60]. But, it is less clear if aspects of this replication-initiated model of asymmetric division apply to mammalian stem cells, CSC or TIC.

Heterochromatin changes and transcription factor binding asymmetrically mark sister chromatids prior to cytokinesis in TIC generation

We identified avascular niches in the interior of expanding lung adenomas in K-Ras mutant mice, which were hypoxic [39]. And, precancerous adenoma cells in these niches divided asymmetrically to produce TIC daughters. Interestingly, this asymmetric division was highlighted by changes in heterochromatin in sister chromatids at opposite poles of the cells during mitosis. Chromatin decondensation marked sister chromatids at the pole destined to give rise to a TIC, whereas the opposite pole retained the densely heterochromatic sister chromatids of adenoma cells. Zeb1 was not evident in quiescent or symmetrically dividing adenoma cells, but it was induced during mitosis in these asymmetrically dividing cells [39]. Notably, Zeb1 selectively associated with decondensed chromatin at the forming TIC pole of the cell. As opposed to Zeb1, Zeb2 is expressed in quiescent and symmetrically dividing adenoma cells, but it is repressed by Zeb1 in TIC. Consequentially, Zeb2 remained associated selectively with sister chromatids in the adenoma pole of the asymmetrically dividing cells. It is unclear if diminished heterochromatin is responsible for induction of Zeb1 during mitosis. But, it appears that Zeb1 bound to sister chromatids at the TIC pole during mitosis is functional because its downstream targets, Bmi1 and nuclear Yap1, are likewise expressed in this pole of the mitotic cells. Further, these transcription factors regulate expression of Numb, Pten, activated Akt and E-cadherin, whose expression pattern distinguishes adenoma and TIC daughter cells, and these factors were already expressed in a polarized fashion in the cells before cytokinesis. Activation of oncogenic K-Ras triggers inflammation in the lung tumor microenvironment, and this inflammation is important for K-Ras driven EMT [39,61–63]. Our results show this inflammation is closely linked to deposition of Tgf-β, which is a known driver of EMT [39]. miR-34a is regulated by inflammation and it controls EMT by targeting Zeb1, but not Zeb2 [64]. But, in an apparent negative feedback loop, Zeb1 can repress miR-34a [41]. Notably, inflammatory regulation of miR-34a controls expression of Numb to drive stem cell asymmetric division [65]. These findings then link inflammation, Zeb1, miR-34a, Numb and asymmetric division.

The above findings point to a role for unequal sister chromatid segregation during mitosis based on epigenetic differences as an initiating event in asymmetric generation of a TIC from a precancerous adenoma cell. And, they provide the first evidence that differential gene expression patterns responsible for the phenotype of adenoma and TIC daughters is already established at poles of asymmetrically dividing cells prior to cytokinesis.

Conclusions and discussion

Beyond Drosophila stem cells, unequal distribution of old and new DNA strands has been observed in daughter cells following asymmetric division of muscle satellite stem cells [66]. And, segregation of parental vs. new DNA strands has been reported in asymmetric division of some non-small cell lung cancer cell lines in culture [67]. But, the basis for this segregation is still unclear. As in Drosophila, in C. elegans asymmetric generation of cells driving bilateral asymmetry of the nervous system has been linked to differential heterochromatin assembly on sister chromatids, in a process termed replication-coupled chromatin assembly [55]. In this process, sister chromatids were differentially sorted to daughter cells based on heterochromatin content (via an unknown mechanism), driving distinct daughter phenotypes. PCNA promotes continuous DNA synthesis on the leading strand, and a single PCNA is required per replication origin. But, a PCNA clamp is required for each Okazaki fragment of 100–200 nucleotides on the lagging strand. Because PCNA in complex with Caf1 drives repopulation of histones onto new DNA strands following replication fork passage, it was suggested that differences in PCNA content on leading vs. lagging strands was responsible for differences in the number of histones loaded onto the two strands, in turn causing distinct heterochromatin assembly on the two sister chromatids. But, this mechanism would not seem to fully explain why differences in sister chromatid heterochromatin content are restricted to asymmetric division. Histone H3-H4 complexes with existing nucleosome-regulatory marks dissociate in front of the replication fork, and in symmetrically dividing cells these complexes are re-distributed equally on the other side of the fork to the two DNA strands, as noted above, in a PCNA-dependent fashion [54,57,58]. Newly synthesized H3-H4 then equally fill in gaps left by the 2-fold dilution of old histone complexes. These newly synthesized histones have pre-deposition marks of acetylation important for their deposition [54]. And, these marks are replaced during the next cell cycle with new nucleosome regulatory marks that reflect the surrounding parental pattern on old histones [59]. As noted above, evidence in Drosophila suggests that old and new H3 is deposited selectively onto leading vs. lagging strands during asymmetric division, and that phosphorylation retained on old H3 is differentially recognized by mother and daughter centrosomes, causing epigenetically distinct sister chromatids to be segregated to the two daughters [56]. It was hypothesized such differential recognition of phospho-H3 might be based on differences in mother vs. daughter centrosome maturation [56,68,69]. But, neither the molecular basis for selective deposition of old vs. new histone complexes on leading vs. lagging strands in asymmetric division nor how phosphorylation marks on old H3 are recognized by one centrosome but not the other is clear.

PCNA dissociates differently from leading and lagging strands at slowly progressing and stalled replication forks, and this can lead to selective deposition of old H3-H4 onto one strand and new H3-H4 onto the other strand [54]. Because nucleosome regulatory marks are not established on new histones until the following cell cycle, such asymmetric deposition of new vs. old H3-H4 with inhibition of fork progression would lead to period of epigenetic difference between the two DNA strands. Hypoxia, which is a hallmark of stem cell niches, classically inhibits replication fork progression [54,70]. Hypoxic ATM activation prevents DNA damage at the stalled or slowing progressing replication forks, and this fork stalling is linked to increased repressive H3K7me3 marks [70]. Thus, hypoxia is a candidate for initiating epigenetic differences on sister chromatids during asymmetric division (Figure 1). But, these sister chromatids still have to be recognized differentially by mother and daughter centrosomes based on epigenetic modification, causing their selective sorting to the two daughter cells (Figure 1). As noted above, asymmetrically dividing stem cells are polarized within the niche via their interaction with extracellular matrix and/or other cells, and this polarization is necessary for assembly of distinct mother and daughter centrosomes that 1) initiate an altered plane of division leading to unequal transfer of cytoplasmic content to the two daughter cells, and 2) distinguish and segregate sister chromatids based on histone modification. A key “chicken vs. egg” question is then what are the relative contributions of segregating cytoplasmic content vs. segregating sister chromatids to the phenotype of the two emerging daughter cells.

Figure 1. Model suggesting hypoxia-induced replication-coupled chromatin assembly is driving TIC generation. TIC are derived from asymmetric division of precancerous adenoma cells (AD) in hypoxic, avascular niches in expanding lesions in K-Ras mutant lungs [39]. Red circles depict existing histones with nucleosome regulatory marks, and white circles are histones lacking these marks. Hypoxia in the niche is postulated to slow replication fork progression, leading to asymmetric distribution of modified (old) histones and unmodified (new) histones on sister chromatids. These sister chromatids are then sorted to opposite poles during mitosis based on differences in mother and daughter centrosomes. The TIC in turn give rise to cancer cells.

We found TIC in hypoxic, avascular niches within the interior of expanding lung adenomas [39]. Following their asymmetric generation from adenoma cells, we found TIC in these sites could divide symmetrically to expand their number, and Zeb1 remained high in these symmetrically dividing TIC, or they undergo a second asymmetric division to produce a cancer cell, where Zeb1 expression was diminished. These niches displayed a distinct morphology, but it remains to be determined if symmetrically and asymmetrically dividing cells in these niches show distinct orientations, display polarized differences in association with extracellular matrix and/or other cells or are linked to gradient of Wnt signaling, as reported with normal stem cell niches. And, as with experiments in Drosophila and C. elegans, the full extent and nature of histone mark differences on sister chromatids is still unknown for asymmetric division producing TIC. Nevertheless, our findings suggest that a form of replication-coupled chromatin assembly is at play in generating heterochromatin differences in sister chromatids, and that sorting of these chromatids based on epigenetic differences is driving asymmetric production of TIC from precancerous cells in the lung.

Acknowledgments

The authors acknowledge grants from from the NIH to D.C.D., an Institutional Development Award (IDeA) from the National Institute of General Medical Sciences (P20GM103453) to Y.L., NIH Core Grant EY015636, Research to Prevent Blindness, and by La Marató de TV3, AVON-SAU, MINECO (SAF2014-52874-R and SAF2017-84918-R), and AGAUR (2017-SGR-1174) to A.P.

Disclosure statement

No potential conflict of interest was reported by the authors.