Two cancer stem cell‐targeted therapies in clinical trials as viewed from the standpoint of the cancer stem cell model

Abstract A key implication of the cancer stem cell model is that for a cancer therapy to be curative, it is imperative to eliminate the cancer stem cells (CSCs) that drive tumor progression. The California Institute for Regenerative Medicine is supporting two novel approaches that target CSCs, one an antibody‐mediated immunotherapy targeting CD47 and the other an antibody targeting ROR1. This article summarizes the evidence that CSCs are targeted and discusses the results of early clinical trials within the context of the CSC model.


| INTRODUCTION
The California Institute for Regenerative Medicine (CIRM), California's stem cell funding agency, selects highly meritorious projects using a rigorous peer review process and then partners with grantees to develop novel stem cell-based treatments, including therapies aimed at eradicating cancer stem cells (CSCs). This article highlights two such approaches that are now in clinical trials.

| CANCER STEM CELL MODEL
The CSC model asserts that tumors are comprised of heterogeneous hierarchies of cancer cells, not all of which are capable of sustaining tumor growth or initiating new tumors. Rather, there exists within most tumors a unique subset of cells, termed CSCs, that have stem cell-like properties such as relative quiescence as well as ability to self-renew and differentiate. It is these cells that drive both tumor progression and metastasis. 1 Evidence for the existence of CSCs was first provided in acute myelogenous leukemia (AML) 2 and has since been demonstrated in many other cancers. 1 CSCs from human tumors have largely been identified based on their ability to propagate tumors in murine xenograft models. Using such models, AML CSCs, also known as leukemic stem cells (LSC), have been defined based on their ability to (a) engraft and form tumors after primary transplantation into immunodeficient mice, (b) propagate and form tumors in serial transplants, and (c) recapitulate the heterogeneity of the tumors from which they were derived including giving rise to non-LSC progeny that do not engraft. 3 Phenotypic as well as functional overlap between normal tissue stem cells and CSCs has been described in a number of cancers. AML LSC exhibit cell surface marker expression patterns characteristic of normal hematopoietic stem or progenitor cells. 3 A gene expression signature specific for normal adult intestinal stem cells identified a population of colorectal CSCs that, as with AML LSC, robustly propagated tumors in immunodeficient mice and recapitulated the organization of the tumor of origin. 4 Expression of a stem cell-like signature in breast and colorectal tumors is predictive of more aggressive disease, 4,5 a finding consistent with the CSC model. Furthermore, a recent study integrating both gene expression and epigenetic features of multiple human cancers identified a "stemness index" to quantify stemness and found it to be most prominent in metastatic tumors. 6 The CSC model can explain why most cancers recur after remissions induced by standard anticancer therapies. Stem cell-like characteristics, such as relative quiescence as well as elevated levels of multidrug resistance transporters and DNA damage repair enzymes, enable CSCs to withstand chemotherapy or radiation and subsequently repopulate tumors and drive relapse. 7 Consistent with this notion, LSC were shown to persist in bone marrow of AML patients following chemotherapy, even in patients in morphologic remission 8 and there is evidence that the LSC population expands after relapse as predicted by the CSC model. 9 In the case of targeted therapies, expression of the target on more differentiated progeny cells within the tumor but not on CSCs would also spare CSCs and allow for relapse.
A clear corollary of the CSC model is that a cancer therapy will never be curative unless the CSC population is eliminated. Many efforts are therefore underway to develop CSC-targeted approaches. 10 In particular, several CSC signaling pathways and regulators of stemness have been identified and agents targeting those pathways are currently undergoing clinical evaluation (reviewed in Reference 11). This article discusses two novel approaches being advanced by CIRM and its grantees. The evidence for CSC targeting is summarized below and emerging clinical results are examined in light of the CSC model.

| CD47 AS A CSC TARGET
CD47 is expressed widely on both cancer and normal cells throughout the body. 12,13 It was initially identified as a potential CSC target by the observations that CD47 expression is elevated on AML LSC compared with normal bone marrow stem cells and high CD47 expression at diagnosis predicts worse overall survival (OS) in AML patients. 14 The importance of CD47 for LSC-driven tumor formation was demonstrated in a series of xenograft transplantation experiments in which precoating of human AML tumor cells with an anti-CD47 antibody prevented leukemic engraftment in immunodeficient mice. 14 Furthermore, when mice carrying established human AML tumors were treated with the anti-CD47 antibody, there was almost complete elimination of circulating human AML LSC as well as a significant decrease in LSC remaining in the bone marrow. Secondary transplants from anti-CD47-treated mice resulted in no leukemic engraftment, further indicating that AML LSC had been eliminated. 14 Parallel experiments using cells from human acute lymphoblastic leukemia (ALL) patients showed similar inhibition of leukemia formation by the anti-CD47 antibody. 15 Collectively, these experiments verified CD47 as an LSC target and provided preclinical proof-of-concept for CD47 blockade as a strategy to target LSC. This strategy has been extended to other cancers. Overexpression of CD47 is correlated with poor prognosis in non-Hodgkin's lymphoma (NHL), ovarian cancer, gastric cancer, and lung cancer. In xenograft models of multiple patient-derived solid tumors, treatment with an anti-CD47 antibody inhibited tumor growth and prevented metastasis, consistent with an effect on CSCs. 16

| CD47 BLOCKADE MECHANISM OF ACTION
CD47 functions as a ligand for signal regulatory protein-α (SIRPα) on phagocytic macrophages, transmitting a "don't eat me" signal that inhibits phagocytosis. 12,13 Macrophage phagocytosis is determined by the balance between various prophagocytic and antiphagocytic signals. Overexpression of CD47 increases the net antiphagocytic signal and appears to be a general mechanism used by cancer cells to evade phagocytosis. 16 Blocking CD47 presumably tips the balance in favor of phagocytosis and as predicted, disruption of the CD47-SIRPα interaction with anti-CD47 antibodies has been demonstrated to enable phagocytosis of AML, ALL, and solid tumor cancer cells by human macrophages in vitro. [14][15][16] These findings support the premise that the observed effects of CD47 blockade in xenograft models in vivo, that is depletion of LSC from blood and bone marrow of engrafted mice and inhibition of secondary transplants, occur via macrophagemediated phagocytosis of CSCs. [14][15][16] Additionally, anti-CD47 antibody-mediated phagocytosis of cancer cells has been shown to induce an antitumor T-cell response via cross-presentation of cancer cell antigens to the adaptive immune system, 17 providing a second potential antitumor mechanism of action (MOA).

| ANTI-CD47 COMBINATION THERAPY MOA
The anticancer effects of the anti-CD47 antibody in mouse models were synergistically enhanced by combining with another anticancer drug such as rituximab in a NHL model 18 or azacytidine in an AML model, 19 furnishing a rationale for combination clinical trials. In both cases, it was hypothesized that the observed synergy is due to augmentation of prophagocytic signals on tumor cells by the second drug as described below.
Rituximab is an anti-CD20 monoclonal antibody that binds to nor- Azacytidine is an anticancer chemotherapeutic indicated for the treatment of AML and myelodysplastic syndromes (MDS). Its mechanisms of action include inhibition of DNA methylation and cytotoxicity due to incorporation into DNA and RNA. 21 Evidence suggests that CD47 blockade alone is insufficient to induce macrophage phagocytosis and that target cells must also express a strong prophagocytic signal in order to trigger phagocytosis. 22 Azacytidine has been shown to induce upregulation of calreticulin 23 which has been identified as a dominant prophagocytic signal expressed on many human cancers. 22 In preclinical studies using an AML model, combining azacytidine with CD47 blockade resulted in enhanced macrophage-mediated phagocytosis in vitro and enhanced antitumor activity in vivo. 19 Analogous to the anti-CD47-rituximab combination, the anti-CD47-azacytidine synergy is postulated to be due to the dual mechanisms of blocking of the CD47 antiphagocytic signal while enhancing phagocytosis via upregulation of a prophagocytic signal, in this case, calreticulin.

| CD47 BLOCKADE IN THE CLINIC
A humanized anti-CD47 antibody (magrolimab, formerly known as Hu5F9-G4 or 5F9) has been tested in early clinical trials. Despite widespread expression of CD47 on normal cells, CD47 blockade selectively targets cancer cells and not normal cells (except for aging red blood cells), presumably because cancer cells express prophagocytic signals that are absent from normal cells. 24 In agreement with preclinical findings, magrolimab has been well tolerated in humans. 25,26 The efficacy outcomes of four phase 1 trials in AML, NHL, and solid tumors, using magrolimab either as monotherapy or in combination with rituximab or azacytidine, are summarized in ROR1 is a receptor for Wnt5a and ROR1-dependent Wnt5a signaling has been implicated in CSC maintenance and self-renewal and also in metastasis. 36 A number of ROR1-dependent, Wnt5a-mediated signaling pathways have been uncovered in CLL cells, including activation of Rac1/2, which is important for CLL cell proliferation. 36,37 A humanized anti-ROR1 antibody (cirmtuzumab) was developed that inhibits ROR1-dependent Wnt5a signaling by binding with high affinity to an epitope in the extracellular domain of ROR1.
Cirmtuzumab blocked Wnt5a-enhanced proliferation of CLL cells and inhibited leukemic engraftment in a mouse model. 35 In an ovarian cancer study, cirmtuzumab inhibited the ability of ovarian cancer CSCs to migrate, form spheroids, or engraft and form tumors in immunodeficient mice. 33 These studies validated inhibition of ROR1 signaling by cirmtuzumab as a strategy for targeting CSCs.  (Table 2). These early results, in particular the encouraging CR rate, which is higher than would be expected with ibrutinib alone, are suggestive of synergy between the two drugs as predicted by the preclinical data.

| DISCUSSION
While acknowledging the extreme heterogeneity and complexity of human tumors, a simplified interpretation of the CSC model predicts the following: Therapies that target bulk tumor but not the CSC subpopulation may cause tumor shrinkage, inducing objective responses and even complete responses, but the responses will not be durable.
In contrast, a therapy that targets primarily the CSC subpopulation but not bulk tumor cells might arrest tumor progression without inducing tumor shrinkage, resulting in SD without objective responses.
A key prediction of the CSC model is that durable objective/complete responses will only be achieved by combined targeting of both bulk tumor cells and the CSC subpopulation.
The two CSC-targeted therapeutic strategies described above work by completely different mechanisms. Yet it is interesting to note a number of parallels. First, in both cases, high-level expression of the target in human tumors is a negative prognostic marker, predicting shorter OS. Second, clinical evaluation of these strategies as monotherapy resulted primarily in SD, a result consistent with the predictions of the CSC model. Third, there is strong preclinical rationale for combination therapy with other anticancer drugs and although it is still too early to tell what the clinical combination trials will show, initial results appear promising and are consistent with additive or synergistic effects. Importantly, both approaches are well tolerated in humans and there appear to be no limiting toxicities.
While overall response rate (ORR) is an accepted endpoint for evaluating efficacy of anticancer therapies, it does not test the basic premise of the CSC model. More meaningful endpoints will be duration of response, progression-free survival, relapse-free survival, and ultimately, OS. It remains to be seen whether the approaches described in this article will lead to long-term, durable responses when combined with another anticancer therapy, as predicted by the CSC model.

ACKNOWLEDGMENT
We thank Dr. Mark Chao, Dr. Thomas Kipps, and the members of the CIRM Therapeutics team for helpful editorial feedback.