Cancer: When Normal Stem Cell Function Goes Awry
As we began to genetically dissect the underlying mechanisms by which skin stem cells transition from a non-tissue-generating to a tissue-generating mode, we realized that when we over-activated the pathway genetically in mice, they became more capable of repairing wounds but also more prone to skin cancers. Conversely, when we under-activated the pathway, mice repaired wounds more slowly but also became more resistant to cancers. These findings revealed that malignant progression derails the basic mechanisms that normal stem cells use to make and repair tissues.
The finding also led us to wonder what the relation might be between normal stem cells and their malignant counterparts. To address this question, we focused on squamous cell carcinomas (SCCs), since many of our mice, including those whose skin lacked the TGFβ receptor (TβRII) or adherens junction proteins α-catenin and E-cadherin, exhibited an increased propensity towards these tumors (Vasioukhin et al., 2001; Guasch et al., 2007). Additionally, SCCs of the skin are the second most common cancer world-wide, and if neglected, can metastasize to the lung and other less common sites. More broadly, squamous cell carcinomas can occur in most stratified epithelial tissues, including the cervix, anogenital tissues, lung, esophagus and oral tissues of the head and neck. Head and neck cancers alone are the 6th most deadly cancer world-wide with a devastating 50% mortality rate. Given their aggressiveness and frequency, surprisingly little is known about these cancers.
We began by using fluorescence activated cell sorting (FACS) to fractionate SCCs on four different genetic backgrounds. We tested the fractions for their ability to generate SCCs when engrafted into host recipient mice. On some backgrounds, at nearly the single cell level, a cell derived from one SCC was able to initiate an SCC resembling its parent when serially transplanted. Like normal stem cells, these tumor-initiating, so-called cancer stem cells (CSCs) exhibited high levels of integrin and focal adhesion kinase (FAK) activity, which was essential for maintaining the tumor (Schober and Fuchs, 2011). Like their normal counterparts, these CSCs also reside at the epithelial-mesenchymal border and express high levels of self-renewal factors such as Bmi1 and HMGA2 (Chen et al., 2012; Schober and Fuchs, 2011). With the exception of a few parallels, however, transcriptional profilings revealed hundreds of differences between CSCs and stem cells of the epidermis or the hair follicles (Schober and Fuchs, 2011). Many of these differences reflected expected distinctions, including not only excessive integrin activation, but also activation of epithelial-mesenchymal transition factors, VEGFα for angiogenesis, the autocrine growth factor TGFα, elevation of K-RAS and other growth-related genes, and induction of transcription factors such as SOX2, important for pluripotency and often upregulated in cancers, but not a normal factor in adult tissue stem cell biology. Notably, E-cadherin and α-catenin were downregulated, as expected given their tumor-suppressive roles in SCC. Beyond these known changes, however, were hundreds of changes whose significance remains unknown.
Initially, we thought we might be able to capitalize on the human cancer genome sequencing effort to narrow down our choice of putative tumor suppressor or oncogenic changes. The major focus of this exome and genome-wide sequencing of human SCCs is on SCCs of the head and neck (HN). However, there are literally hundreds of genetic differences in the genomes of head and neck SCCs versus those of normal patient tissue. Combining the human HNSCC data with our RNA-seq profiling, there are still >350 genes that are altered in HNSCC DNA and also ≥2X changed at the mRNA level in CSCs relative to normal skin SCs. Clearly, conventional conditional knockout technology as we’ve been doing for the past 25 years, is not a practical solution to the modern day genome era.
Over the past 5 years, we’ve overcome this problem and devised a method that allows us to rapidly assess the functional relevance of a particular gene mutation in a timeframe that is several orders of magnitude quicker than conventional gene targeting in mice. The procedure uses lentiviral delivery of genes to living mouse embryos (Beronja et al., 2010). It takes advantage of the fact that lentivirus infects the first epithelial cell layer it encounters. By injecting high titer lentivirus into the amniotic sac of living mouse E9.5 embryos, the virus is taken up selectively by the single layer of unspecified surface epithelial progenitor cells. Twenty four hours later, the DNA has stably integrated into the mouse genome, and thereafter is stably propagated to the progenitors and their progeny, including the epidermis, hair follicles, mammary glands and oral epithelia. The procedure doesn’t introduce an immune response and expression of integrated genes can be monitored into adulthood (Beronja et al., 2010). When coupled with short hairpin RNA (shRNA) technology, it becomes feasible to knockdown expression of specific mRNAs in a matter of days, and with our recent incorporation of CRISPR/CAS technology, we can now use lentivirus to deliver guide RNAs and conditionally knock out genes as expeditiously. The technology permits knocking down single genes but also carrying out complex genetics that would otherwise take years, if at all possible.
Returning to the issue of sifting through the complexity of the hundreds of gene differences revealed by CSCs and the human cancer sequencing project, we’ve also developed a large scale screening method to identify which of these myriads of gene changes are likely to reflect alterations in tumor-suppressor gene expression (Schramek et al., 2014).
In our recent tumor suppressor screen, we pooled ~2000 lentiviruses corresponding to ~350 genes which were either mutated in The Cancer Genome Atlas (TCGA) of human HNSCCs or altered by ≥2X in our CSCs versus normal skin SCs. By transducing embryos that were either wild-type or defective for TGFβ signaling, we then simply waited for the adult mice to develop tumors, and then through high throughput sequencing, identified the hairpin(s) that was selected for in the tumor (Schramek et al., 2014). In contrast to wild-type mice, mice conditionally null for TβRII develop skin and HNSCCs if they receive an additional mutation in a tumor-suppressor (loss of function) or oncogene (gain of function).
Surprisingly, in >30 different SCCs, one of three different Myh9 shRNAs were present in abundance. Myh9, encoding myosin IIA, has been ignored as a potential tumor suppressor, and yet on retesting, each of the three Myh9 shRNAs enhanced SCC formation on both the TβRII-deficient background and also in mice harboring an HRasG12V knockin allele, typically mutated in human SCCs. Moreover, when we engineered mice conditionally null for both TβRII and Myh9, these mice exhibited a marked increase in tumor susceptibility relative to TβRII alone. This was even true for mice harboring only a single mutated Myh9 allele (Schramek et al., 2014). These findings establish Myh9 as a bona fide tumor suppressor.
Sifting through the complexities of how this cytoskeletal protein functions in cancer will be a major challenge for the future. Of note, however, it is intriguing that reductions in α-catenin, which also promote an SCC-like phenotype, are associated with reductions in RhoA, which would be expected to lead to a corresponding reduction in myosin II activity (Vaezi et la., 2002). Conversely, reductions in FAK, which enhance RhoA function in vitro, also lead to a decreased susceptibility for SCCs in vivo (Schober et al., 2007; 2011). Indeed it has been more than 30 years since changes in the actin cytoskeleton have been associated with cancers, and in addition to myosin IIA, there are many putative cytoskeletal regulators that have surfaced both from our comparisons of CSCs and normal stem cells and also from our screens for potential tumor suppressors and oncogenes. Given my lab’s long-standing interest and expertise on cytoskeletal biology, we are poised to delve further into the significance of these genes and their actions in the future. These research efforts could have important implications for designing future therapies to treat this critical and life-threatening cancer.
On the flip side of this coin, we’ve also recently carried out screens for oncogenic regulators of epithelial growth. In this case, we carried out a genome-wide RNAi screen, involving 78,000 lentiviruses against the entire mouse genome (Beronja et al., 2013). For this screen, we used wild-type embryos and embryos harboring the H-RasG12V mutation, which already exhibit a hyperproliferative phenotype. By transducing hundreds of embryos such that on average, at least 30 different cells received the exact same virus and no cells received more than one virus, we were able to carry out the screen in triplicate and obtain highly reproducible results. We used high throughput sequencing of E10.5 epidermis to determine the overall representation of each shRNA in the skin, and then we sequenced newborn transduced embryos to identify shRNAs that were disproportionately over- or under-represented in newborn versus E10.5 epidermis. By comparing the data from H-RasG12V versus wild-type embryos, we could weed out “housekeeping” genes required for key cellular processes, and instead focus on those shRNAs which differentially affected oncogenic versus wild-type growth. Beta-catenin, broadly implicated in human cancers, was at the top of our list of oncogenes (Beronja et al., 2013).
Finally, in the past year, we’ve carried out a screen for oncogenic microRNAs. MicroRNAs play a key role in normal stem cell behavior and show marked alterations in SCCs (Yi and Fuchs, 2008; Zhang et al., 2013). To distinguish which of these changes are drivers in malignancy, we made pooled lentiviral library of microRNA expression vectors and then transduced them according to the strategy described above. The two top hits from our screen were miR21, implicated broadly in human cancers, and miR21*, the passenger strand, whose function was previously unexplored (Ge et al., 2016). Probing deeper, we showed that not only is this miR functionally important in SCC, but in addition, it is associated with poor prognosis in human HNSCCs.
Because our strategies and our findings are relatively new, it is too early to say how successful the approach will be in identifying new potential targets that might impair oncogenic stem cell growth without affecting normal stem cell growth. That said, embryonic growth is rapid, thereby mimicking oncogenic growth. In addition, our strategy tests the effects of knocking down genes in vivo, and in progenitors that are exposed to their normal constituency of heterologous, systemic and cell-autonomous signals. Moreover, in contrast to in vitro culture, which invariably induces a stress response, the approach we use is non-invasive. Collectively, these features offer a compelling argument for the potential value of this functional approach over conventional methods for identifying driver mutations in cancers. As we continue our efforts in probing the basic biology underlying normal tissue-initiating stem cells and their tumor-initiating counterparts, we hope to be able to make significant inroads towards the interface between the basic science and clinical applications for cancers.
In addition to genetic screens, the lab is taking other approaches to make inroads into our studies on skin cancers. On one avenue, we’ve carried out ChIP-seq and super-enhancer analyses to unravel new regulators in tumor-initiating CSCs that distinguish them from both epidermal stem cells and hair follicle stem cells, two established origins of SCCs. Our studies led to our uncovering a major role for pETS2 in SCCs, which is a transcription factor phosphorylated and super-activated by the RAS/MAPK pathway at the root of many SCCs (Yang et al., 2015).
On another route, we were intrigued by the fact that two distinct populations of tumor-initiating cells could be purified from SCCs. They differed by their response to TGFβ, but we were unable to identify a hierarchical relation between the two populations (Schober et al., 2011). Probing deeper into mechanism, we devised a lentiviral vector that harbored two genes: 1) a constitutively active tetracycline-sensitive transactivator and 2) a TGFβ-sensitive gene expressing mCherry and a tamoxifen inducible CreER. When transduced into a tetracycline-inducible oncogenic HRas mutant X Rosa26fl-STOP-fl-YFP mouse embryo at low titer, we could add tetracycline to activate mutant Ras, but only in transduced skin clones, thereby controlling tumor burden. Additionally, as tumors developed and progressed to SCCs, we detected mCherry exclusively in the TGFβ-responding CSCs, enabling us to show that these stem cells were ones adjacent to blood vessels. Moreover, when we added tamoxifen, we could lineage trace these CSCs, showing that this population of stem cells responded to TGFβ by becoming slower-cycling but invasive, undergoing an EMT-like transition. Most intriguingly, these TGFβ-responding CSCs resisted chemotherapy, and regrew the tumor (Oshimori et al., 2015). Thus, as cancer develops, heterogeneity in the tumor microenvironment leads to striking changes in malignant stem cell behavior. As a blood vessel comes close to the tumor, the stem cells become slow-cycling, invasive and resistant to chemotherapy. More distant from the vessel, the stem cells are fast-cycling, giving rise to the bulk of the tumor. These CSCs self-renew and generate many more tumor-initiating cells, and since blood vessels come and go, the process is dynamic, yielding a double edge sword to cancer, and providing an explanation for why cancers are so difficult to treat.