Research Article

Unresolved endoplasmic reticulum stress engenders immune-resistant, latent pancreatic cancer metastases

See allHide authors and affiliations

Science  15 Jun 2018:
Vol. 360, Issue 6394, eaao4908
DOI: 10.1126/science.aao4908

Chronic stress as a survival tactic

Most patients with pancreatic ductal adenocarcinoma (PDA) develop liver metastases after surgical removal of their primary tumor. These metastases are thought to potentially arise from quiescent disseminated cancer cells, likely present at the time of surgery, which evade elimination by the immune system. Pommier et al. explored how these quiescent cells survive by analyzing mouse models and tissue samples from patients with PDA. They found that disseminated cancer cells do not express a cell surface molecule that triggers killing by T cells. This phenotypic feature is linked to their inability to resolve endoplasmic reticulum stress. When this stress is resolved, the disseminated cells begin proliferating and form metastases.

Science, this issue p. eaao4908

Structured Abstract

INTRODUCTION

Pancreatic ductal adenocarcinoma (PDA) is the fourth most common cause of death from cancer worldwide and has a 5-year survival rate of 6%. Patients who have had their primary PDA surgically resected often develop metastatic disease, despite intra-operative examination of the liver confirming the absence of macrometastatic lesions. These observations lead to the conclusion that latent metastases, detectable only microscopically, were present in these patients and were responsible for the postoperative development of metastatic disease.

RATIONALE

Latent metastases with the potential for outgrowth had been considered to be lesions in which cancer cell proliferation is balanced by immune-mediated cancer cell death, but a more recent explanation invokes quiescent, single disseminated cancer cells (DCCs). Single, nonreplicating DCCs have been observed in several cancer types, but whether quiescence is enforced by the microenvironment or is cancer cell–autonomous is not known. Immunity, both innate and adaptive, also is likely to have a role in the selection and/or maintenance of latent DCCs. This has long been suspected on the basis of the clinical observation of donor-derived cancer in immune-suppressed recipients of allografts. However, there is an unexplained paradox of immunity preventing the outgrowth of latent metastases but not eliminating them.

RESULTS

We studied the metastatic process in the context of an ongoing adaptive immune response because of the occurrence of cancer cell–specific immunity in human and mouse PDA. Livers from patients and mice with PDA contained single DCCs with an unusual phenotype of being negative for cytokeratin 19–negative (CK19)and major histocompatibility complex class I (MHCI). The absence of the expression of MHCI in DCCs and the occurrence of cancer-specific CD8+ T cells in the genetically engineered mouse model of PDA, and possibly in patients with PDA, suggested that DCCs may be selected by an anticancer immune response during the metastatic process.

To investigate this hypothesis, we created a mouse model that would allow us to determine how DCCs develop, their relationship to metastatic latency, and the role of immunity. Intraportal injection of immunogenic PDA cells into preimmunized mice seeded livers only with single, nonreplicating DCCs lacking MHCI and CK19, whereas naïve recipients of PDA cells had macrometastases. We found that a subpopulation of PDA cells with the phenotype of DCCs was present in vitro and that those cells are the precursors of DCCs in vivo. We found that T cells select DCCs by eliminating MHCI+ proliferating cancer cells.

To identify the cell-autonomous “switch” regulating the developmental state of the metastases, we preformed single-cell RNA sequencing of PDA cells with the DCC phenotype. This transcriptomic analysis demonstrated an endoplasmic reticulum (ER) stress response. Moreover, DCCs showed a lack of activation of the IRE1α (inositol-requiring enzyme 1α) pathway of the unfolded protein response, whereas the PERK (protein kinase RNA-like ER kinase) pathway was activated, suggesting that DCCs cannot resolve ER stress. Relieving ER stress pharmacologically with a chemical chaperone or genetically by overexpression of spliced XBP1, in combination with T cell depletion, stimulated outgrowth of macrometastatic lesions containing PDA cells expressing MHCI and CK19.

CONCLUSION

We find that a PDA-specific adaptive immune response selects DCCs, in which the ER stress response accounts for both quiescence and resistance to immune elimination. Accordingly, outgrowth of DCCs to macrometastases requires not only relief from the cancer cell–autonomous ER stress response, but also suppression of systemic immunity. Thus, the ER stress response is a cell-autonomous reaction that enables DCCs to escape immunity and establish latent metastases.

Unresolved ER stress allows disseminated cancer cells to escape the T cell response.

Quiescent cancer cells that exhibit unresolved ER stress lack expression of MHCI. They avoid killing by T cells and become latent disseminated cancer cells (DCCs). Resolution of ER stress allows DCCs to regain proliferative capacities and grow as overt metastases, only if the T cell response is disrupted, because they also regain the expression of MHCI. Ecad, E-cadherin.

Abstract

The majority of patients with pancreatic ductal adenocarcinoma (PDA) develop metastatic disease after resection of their primary tumor. We found that livers from patients and mice with PDA harbor single disseminated cancer cells (DCCs) lacking expression of cytokeratin 19 (CK19) and major histocompatibility complex class I (MHCI). We created a mouse model to determine how these DCCs develop. Intraportal injection of immunogenic PDA cells into preimmunized mice seeded livers only with single, nonreplicating DCCs that were CK19 and MHCI. The DCCs exhibited an endoplasmic reticulum (ER) stress response but paradoxically lacked both inositol-requiring enzyme 1α activation and expression of the spliced form of transcription factor XBP1 (XBP1s). Inducible expression of XBP1s in DCCs, in combination with T cell depletion, stimulated the outgrowth of macrometastatic lesions that expressed CK19 and MHCI. Thus, unresolved ER stress enables DCCs to escape immunity and establish latent metastases.

Pancreatic ductal adenocarcinoma (PDA) is the fourth most common cause of death from cancer worldwide (1) and has a 5-year survival rate of 6% (2). The majority of patients are diagnosed after the disease has spread beyond the primary tumor site. Patients who show no clinical evidence of local invasion or distant metastasis are treated by surgery, but about 75% of these patients develop metastatic disease within 2 years after resection of their primary tumors (3, 4), despite intra-operative examination of the liver confirming the absence of macrometastatic lesions (5). These observations lead to the conclusion that latent metastases, detectable only microscopically, were present in these patients at the time of surgery and were responsible for the postoperative development of metastatic disease.

Latent metastases are thought to be lesions in which cancer cell proliferation is balanced by immune-mediated cancer cell death (68). A more recent hypothesis highlights the role of quiescent, single disseminated cancer cells (DCCs) (911). Single, nonreplicating DCCs have been observed in several cancer types, most often in the bone marrow (12, 13), but whether quiescence is enforced by the microenvironment or is cancer cell–autonomous is not known (14). Immunity, both innate (15) and adaptive (7, 16, 17), also is likely to have a role in the selection and/or maintenance of latent DCCs. This has long been suspected based on the clinical observation that immunosuppressed recipients of allografts occasionally present with donor-derived cancer (18, 19). However, there is an unexplained paradox of immunity preventing the outgrowth of latent metastases but not eliminating them.

We examined the nature of latent metastases in PDA by developing a mouse model that replicates the characteristics of hepatic DCCs that are found in human PDA and in spontaneously arising PDA in mice. We studied the metastatic process in the context of an ongoing adaptive immune response because of the occurrence of cancer cell–specific immunity in human and mouse PDA (2023).

Quiescent, single DCCs in the livers of humans and mice with PDA

To determine whether hepatic DCCs occur in human PDA, we microscopically examined tissue sections from the primary tumors and livers of five patients with PDA who had no clinically detectable hepatic metastases. The clinicopathologic characteristics of the patients are shown in table S1. The tumors were genotyped as having p53 loss of heterozygosity, which permitted staining for mutant p53 accumulation as an identifier of cancer cells (24). p53+ cancer cells were present in both the primary tumors and livers of all five patients. The p53+ cancer cells resided in the livers as single cells that were consistently negative for cytokeratin 19 (CK19), Ki67, and major histocompatibility complex class I (MHCI), in contrast to the cancer cells in the primary tumors, which exhibited all three markers (Fig. 1, A to C). We also examined the livers from mice bearing the autochthonous LSL-KrasG12D/+; LSL-Trp53R172H/+; Pdx-1-Cre; RosaYFP (KPCY) model of PDA (G, glycine; D, aspartic acid; R, arginine; H, histidine; YFP, yellow fluorescent protein) (2527), which recapitulates human PDA. In livers devoid of macrometastases, we found both YFP+ micrometastases and DCCs. Whereas the micrometastases always expressed CK19, Ki67, and MHCI, the single DCCs were mainly CK19 (32 of 40), Ki67 (22 of 22), and MHCI (28 of 28) (Fig. 1, D to F). Thus, the livers of patients and mice with PDA contain DCCs that share an unusual phenotype linking the loss of epithelial gene expression and quiescence with a potential for escape from T cell recognition.

Fig. 1 Single DCCs with a characteristic phenotype are present in the livers of humans and mice with PDA.

(A to C) Immunofluorescence (IF) of sections from the primary tumor and liver of a patient with PDA that were stained with anti-p53 to reveal cancer cells (red) and with (A) anti-CK19, (B) anti-Ki67, or (C) anti-MHCI (all green). Photomicrographs are representative of five patients. (D to F) IF of sections from a liver of a KPCY mouse with spontaneous PDA and no hepatic macrometastases. Sections were stained with anti-YFP to reveal cancer cells (green) and with (D) anti-CK19, (E) anti-Ki67, and (F) anti-MHCI (all red). Photomicrographs are representative of three mice. The ratios shown in the top right corners of the photomicrographs indicate the frequency of the observed DCC phenotype relative to the total number of DCCs that were assessed. All frequencies are compiled in table S2. White arrows designate DCCs and, in the sections of human livers, green arrows designate normal, liver-resident CK19+ or MHCI+ cells. Scale bars, 25 μm. Micromet, micrometastasis; DAPI, 4′,6-diamidino-2-phenylindole.

A mouse model of hepatic metastasis in the context of an adaptive immune response

The absence of MHCI expression in DCCs and the presence of cancer-specific CD8+ T cells in the genetically engineered mouse model of PDA (20), and possibly in patients with PDA (28), suggested that DCCs may be selected by an anticancer immune response during the metastatic process. Accordingly, we developed a mouse model of hepatic metastases that allowed us to assess the effect of a preexisting immune response. The mM1 (mouse metastasis 1) cell line was derived from a spontaneous liver metastasis of a mouse bearing an autochthonous PDA and was stably transfected with a transposon vector directing the expression of diphtheria toxin receptor (DTR), Herpes simplex thymidine kinase (HSV-TK), firefly luciferase, and mTagBFP2 to generate mM1DTLB cells. Syngeneic C57Bl/6 mice were injected subcutaneously with 106 mM1DTLB cells; tumors were grown for 14 days and then eliminated by treatment with diphtheria toxin (DTx) and ganciclovir (GcV). These “preimmunized” mice and naïve mice were challenged by intrasplenic injections of 106 mM1DTLB cells, followed by splenectomy, thereby seeding the liver via the portal vein, as in PDA (Fig. 2A).

Fig. 2 A mouse model for hepatic DCCs.

(A) Mice were preimmunized by subcutaneous injection of 106 mM1DTLB PDA cells derived from a hepatic metastasis of a KPC mouse. After 2 weeks, tumors were eliminated by treating mice with DTx and GcV. For hepatic metastases, 106 mM1DTLB PDA cells were injected intrasplenically into naïve and preimmunized mice, followed immediately by splenectomy. (B and C) Tumor growth was measured by whole-body bioluminescence imaging (p/s, photons per second; D, day). The dashed gray line represents the background luminescence in tumor-free mice. (D) Ex vivo photon flux of whole livers was measured at days 5, 10, 15, and 20 after cancer cell injection. Results are representative of three experiments with at least five mice per group. (E to J) IF of sections from the livers of a naïve mouse (left panels) and a preimmunized mouse (right panels) that were stained with anti-luciferase (green) to identify cancer cells and with (E) anti-CK19, (F) anti-Ecad, (G) anti-Ki67, (H) EdU, (I) anti-MHCI, and (J) CD3 (all red). For EdU staining, mice were injected every 12 hours with EdU for 3 days. Photographs are representative of 20 mice from three independent experiments. The ratios shown in the top right corners of the photomicrographs indicate the frequency of the observed DCC phenotype relative to the total number of DCCs that were assessed. All frequencies are compiled in table S2. White arrows designate DCCs. Scale bars, 25 μm.

We found that in naïve mice, whole-body bioluminescence increased from day 1 after injection, consistent with the growth of hepatic metastases. In preimmunized mice, however, whole-body bioluminescence decreased after day 1, and by day 7, luminescence was at background levels (Fig. 2, B and C). In additional cohorts of naïve and preimmunized mice, livers were removed at intervals after the intrasplenic injection of mM1DTLB cells and assessed for bioluminescence. Both photon flux (Fig. 2B) and visually detectable metastases increased in the livers of naïve mice between days 5 and 20, whereas metastatic foci were barely detectable in preimmunized mice at day 5 and were absent at later time points (Fig. 2D). A potential role for T cells in the elimination of mM1DTLB cancer cells was suggested by the finding that in the livers of preimmunized mice, tumor cells were frequently surrounded by both CD8+ and CD8 CD3+ T cells by 24 hours (fig. S2A). This possibility was confirmed by treating preimmunized mice with depleting antibodies against CD4 and CD8, alone or together, or with isotype control antibody (fig. S2B). Depleting either CD4+ or CD8+ T cells abrogated the ability of preimmunization to suppress the development of macrometastases (fig. S2B).

Microscopic examination of livers from naïve and preimmunized mice revealed the presence of macrometastatic lesions in the former, but only single DCCs in the latter. The DCCs differed from cancer cells in the macrometastases in the following ways: They did not express the epithelial markers CK19 and E-cadherin (Ecad), they did not express Ki67, they did not incorporate 5-ethynyl-2′-deoxyuridine (EdU), and they did not express MHCI (Fig. 2, E to I). Importantly, the phenotype of the DCCs in this metastasis model is similar to that of DCCs in human PDA and the KPCY mouse. T cells in the vicinity of DCCs were infrequent, whereas they surrounded macrometastatic lesions in naïve mice (Fig. 2J). The absence of CK19 and Ecad expression did not indicate an epithelial-mesenchymal transition (EMT) because DCCs did not express the EMT markers desmin, αSMA, Snail1, or Slug (fig. S3).

When naïve and preimmunized mice were challenged with mM1DTLB cells via the tail vein, lung macrometastases developed only in the naïve mice, whereas CK19 DCCs were observed in the lungs of preimmunized mice (fig. S4). We also assessed the 1242 cell line derived from a primary PDA tumor of a KPC mouse (29), which had been modified with the same transposon vector, for its ability to generate DCCs. When naïve and preimmunized mice were challenged by intrasplenic injection of 1242DTLB cells, hepatic macrometastases developed in the naïve mice, whereas only CK19 DCCs were observed in the livers of preimmunized mice (fig. S5). To assess more stringently the proliferation of DCCs, we labeled the mM1DTLB cells with carboxyfluorescein diacetate succinimidyl ester (CFSE) before intrasplenic injection into naïve or preimmunized mice. Whereas all cancer cells in the macrometastases of naïve mice became CFSE, DCCs in pre-immunized mice retained CFSE (fig. S6A). Mice that had received intrasplenic injections of mM1DTLB cells were also given EdU in drinking water for 20 days. Whereas cancer cells in macrometastases in naïve mice incorporated EdU, almost all DCCs in preimmunized mice were EdU (fig. S6B), indicating a lack of proliferation. Therefore, the occurrence of quiescent, MHCI DCCs in the absence of macrometastases is a consequence of an ongoing cancer-specific immune response.

A latent capacity for outgrowth of DCCs is controlled by T cells

We next examined whether a latent capacity of DCCs for outgrowth into macrometastatic lesions might be revealed by T cell depletion. When depleting antibodies against CD4 and CD8 were administered to preimmunized mice 3 weeks after the establishment of DCCs, macrometastases appeared in 10 of 15 mice. When T cells were depleted at 9 weeks, macrometastases appeared in 2 of 15 mice (Fig. 3). The metastases were composed of cancer cells that had reexpressed CK19 and MHCI, suggesting that DCCs revert to an epithelial phenotype to initiate the formation of macrometastases. The lower frequency of macrometastases in mice in which T cells were depleted at 9 weeks suggests that DCCs with a capacity for reversion decreased between 3 and 9 weeks. Indeed, we found that livers of preimmunized mice had fewer DCCs at 9 weeks than at 3 weeks (fig. S7A). This loss of DCCs may reflect the killing by T cells of spontaneous revertants, as suggested by the occasional occurrence of a CK19+ DCC surrounded by T cells in the preimmunized mice (fig. S7B). The reexpression of MHCI by these growing metastases also suggests the means by which T cells control the outgrowth of spontaneously reverting DCCs. To demonstrate that T cells alone are both necessary and sufficient for controlling the growth of MHCI DCCs, we depleted natural killer (NK) cells by administering antibody against NK1.1 at the time of mM1DTLB cell challenge. Preimmunized mice lacking NK cells were indistinguishable, with respect to the occurrence of DCCs and absence of macrometastases, from control antibody–treated mice (fig. S8). This evidence for a dominant role of the T cell in controlling DCCs is supported by the finding that DCCs were never seen to be in contact with CD45+ (fig. S9A), F4/80+ (fig. S9B), CD19+ (fig. S9C), CD31+ (fig. S9D), αSMA+ (fig. S9E), or Ly-6G+ (fig. S9F) cells; the distribution of these cell types was similar to that in the normal liver (fig. S9, G to N).

Fig. 3 T cells control outgrowth of latent DCCs.

(A) Using bioluminescence imaging, we assessed the growth of hepatic metastases in preimmunized mice that had been depleted of T cells by administration of antibodies against CD4 and CD8 beginning 3 or 9 weeks after splenic injection of mM1DTLB PDA cells. One group of mice was also treated with isotype control antibody. (B and C) IF of sections containing macrometastases from the liver of a preimmunized mouse that had been depleted of T cells 3 weeks after splenic injection of mM1DTLB PDA cells. Anti-luciferase identifies cancer cells. Scale bar, 25 μm.

A rare subpopulation of PDA cells in vitro with the phenotype of DCCs

The absence of MHCI expression by DCCs raised the possibility that these cells were present in the injected PDA population and were negatively selected by T cells. Indeed, ~1% of the mM1DTLB cells in tissue culture were Ecad and CK19, and all Ecad cells were MHCI (Fig. 4, A and B). The Ecad cells resided in a nonproliferating subpopulation of cells, as indicated by the resistance of these HSV-TK–expressing cells to GcV (Fig. 4C). A phenotypic plasticity of the mM1DTLB cells was shown when we cultured FACS (fluorescence-activated cell sorting)–purified Ecad+ and Ecad cells for 3 days and found that they generated Ecad and Ecad+ cells, respectively (Fig. 4D). The Ecad MHCI phenotype shared by PDA cells in vitro and DCCs in vivo suggested that the former may be the precursors of the latter. We assessed this possibility by intrasplenically injecting 106 Ecad+ or 104 Ecad cells into naïve and preimmunized mice. The growth of Ecad+ macrometastases in the livers of the naïve mice receiving Ecad+ cells was similar to that in naïve mice receiving unsorted mM1DTLB cells (Fig. 4E). Microscopic examination of the livers of these mice, however, revealed no Ecad DCCs (Fig. 4F). Injection of Ecad+ cells into preimmunized mice resulted in a rapid decline in photon flux, with no specific signal by day 7 (Fig. 4E), and microscopic examination of these livers also demonstrated the absence of DCCs (Fig. 4F). Injection of Ecad cells into naïve mice resulted in delayed development of hepatic macrometastases (Fig. 4E), and microscopy revealed the presence of DCCs (Fig. 4F). Injection of Ecad cells into preimmunized mice led to DCCs but no macrometastases, confirming that an ongoing T cell response controls DCC outgrowth (Fig. 4, E and F). The macrometastases found in naïve mice injected with Ecad cells were CK19+ (Fig. 4G) and MHCI+ (Fig. 4H), indicating reversion to an epithelial phenotype.

Fig. 4 A subpopulation of PDA cells in vitro shares phenotypic features with DCCs.

(A) Flow cytometry analysis of mM1DTLB PDA cells that were stained with anti-CK19 or anti-Ecad. Results are representative of five independent experiments. FSC, forward scatter. (B) Flow cytometry measurement of anti-MHCI staining of Ecad+ and Ecad mM1DTLB PDA cells and of lymph node cells as a comparator. Results are representative of five independent experiments. (C) mM1DTLB PDA cells were treated in vitro for 48 hours with increasing doses of GcV to kill proliferating cells, and the proportion of viable cells that were Ecad and MHCI was measured by flow cytometry. Results are representative of two independent experiments. **P < 0.01; ***P < 0.001. (D) FACS analysis of purified Ecad+ and Ecad mM1DTLB PDA cells that were cultured for 3 days. Dot plots (left) and histogram (right) are representative of three independent experiments. (E) Growth of hepatic metastases after intrasplenic injection of 106 Ecad+ or 104 Ecad mM1DTLB PDA cells into naïve and preimmunized mice was assessed by whole-body bioluminescence imaging (n = 5 mice per group). The dashed gray line represents the luminescence background in tumor-free mice. (F) Table summarizing the occurrence of DCCs and/or metastases in each group of mice that had been injected with Ecad+ or Ecad mM1DTLB cells. N, naïve; PI, preimmunized. (G and H) IF of sections from the liver of a naïve mouse that had received an intrasplenic injection of Ecad mM1DTLB cells. Anti-luciferase (green) identifies cancer cells. Photomicrographs are representative of five mice. Scale bars, 25 μm. Error bars, mean ± SEM.

In summary, the origin of the DCC is the Ecad cell, because DCCs were not present in naïve or preimmunized mice after the injection of Ecad+ cells. Reversion from the quiescent, Ecad state to the proliferating, Ecad+ phenotype was observed in naïve mice but was masked by the ongoing T cell response in preimmunized mice because reversion is associated with reexpression of MHCI. Thus, the two states of mM1DTLB cells that were observed in vivo—the proliferating MHCI+, Ecad+, CK19+ macrometastasis and the quiescent MHCI, Ecad, CK19 DCC—occur in vitro, reflecting a developmental plasticity that may be controlled by a cell-autonomous process.

Unresolved endoplasmic reticulum (ER) stress characterizes PDA cells with the DCC phenotype

To identify the cell-autonomous “switch” regulating the developmental state of the metastases, we performed single-cell RNA sequencing (scRNA-seq) of in vitro sorted Ecad+ and Ecad cells. The most up-regulated pathway in Ecad cells relative to Ecad+ cells was “response to ER stress” (Fig. 5A and fig. S10C), and the most down-regulated pathway was “cell division” (Fig. 5B and fig. S10D). Network analysis of other up-regulated pathways that distinguish the Ecad and Ecad+ PDA populations, such as “autophagy,” showed that they were linked to “response to ER stress”. Similarly, network analysis of other down-regulated pathways demonstrated linkage to “cell division” (Fig. 5B). These two populations of the mM1DTLB PDA cells were also distinct in principal components analysis, in which Ecad cells appeared to be more heterogeneous than Ecad+ cells (fig. S10A). Indeed, analysis at the single-cell level identified four subpopulations of Ecad cells, in three of which the dominant up-regulated pathway was related to “response to ER stress,” and the major down-regulated pathway was “cell division” (fig. S11). An EMT signature was not present among the 1639 genes that were differentially expressed between Ecad+ and Ecad cells (fig. S10B), confirming the immunofluorescence analysis of Ecad cells (fig. S6E). In addition, neither Ecad+ nor Ecad cells expressed three of the four major NKG2D ligands—Ulbp1, H60b, and H60c—and both strongly expressed the inhibitory ligand, Qa1, providing an explanation for the absence of a role for NK cells in controlling outgrowth of macrometastases (fig. S7C) (30). Genes that are involved in the processing of MHCI were not differentially expressed, which is consistent with reports that the ER stress response suppresses the expression of MHCI by a posttranscriptional mechanism (31, 32). Last, the conclusion that the ER stress response alters the expression of MHCI and Ecad is supported by the finding that treatment of mM1DTLB PDA cells with tunicamycin, an inducer of ER stress, increased the proportion of cells that were MHCI and Ecad (fig. S12).

Fig. 5 The ER stress response in PDA cells that share phenotypic features of DCCs.

scRNA-seq was performed on 104 Ecad+ and 98 Ecad mM1DTLB PDA cells. (A and B) Network analysis, following pathway enrichment analysis, shows ontology relationships between the pathways (left panels). Their relative representation is depicted as a pie chart (right panels) for (A) up-regulated pathways and (B) down-regulated pathways in Ecad cells relative to Ecad+ cells. Pathways are significant with an adjusted P < 0.01 after Benjamini-Hochberg false discovery rate correction. (C) Liver sections from a naïve mouse and a preimmunized mouse were stained with anti-luciferase (green) to identify PDA cells and with anti-CHOP (red) to identify cells exhibiting an ER stress response. (D) Liver sections from a KPCY mouse were stained with anti-YFP (green) to identify PDA cells and with anti-CHOP (red). (E) Sections from the primary tumor and liver of a patient with PDA were stained with anti-p53 to identify PDA cells (red) and with anti-CHOP (green). Photomicrographs are representative of five patients who had no detectable liver metastases. The ratios shown in the top right corners of the photomicrographs indicate the frequency of the observed DCC phenotype relative to the total number of DCCs that were assessed. All frequencies are compiled in table S2. White arrows designate DCCs. Scale bars, 25μm.

The most differentially expressed gene was Ddit3/CHOP, the mRNA level of which was 18-fold as high in Ecad cells as in Ecad+ cells. This gene encodes a transcription factor, C/EBP homologous protein (CHOP), that is part of the ER stress response. We examined the expression of CHOP protein by immunofluorescence in the Ecad DCCs in human and mouse PDA. Anti-CHOP staining was demonstrated in hepatic DCCs in preimmunized mice (Fig. 5C) and in KPCY mice (Fig. 5D), but not in PDA cells of macrometastases from naïve mice or of micrometastases from KPCY mice (Fig. 5E). Importantly, DCCs in the livers of three of five patients with PDA were also stained by antibody against CHOP (Fig. 5E).

The higher expression level of CHOP in DCCs versus in growing PDA cells with an epithelial phenotype suggested that the response of DCCs to ER stress was not as effective as the response of growing CK19+/Ecad+/MHCI+ PDA cells. To restore protein homeostasis, cells undergoing ER stress activate the unfolded protein response (UPR) (33, 34). CHOP is induced by the protein kinase RNA-like ER kinase (PERK) pathway of the UPR (33, 34), and, indeed, this pathway was activated in both Ecad+ and Ecad cells, as shown by phosphorylation of PERK and EIF2α (fig. S14A). In contrast, the inositol-requiring enzyme 1 (IRE1α) pathway was activated in Ecad+ cells but not in Ecad cells, as shown by phosphorylation of IRE1α and splicing of the Xbox binding protein 1 (XBP1) transcription factor mRNA (fig. S14A). Impaired activation of the IRE1α pathway in Ecad cells was supported by scRNA-seq results showing decreased expression of XBP1 target genes relative to their expression levels in Ecad+ cells (fig. S14B). The cancer cells in primary PDA tumors in patients and hepatic macrometastases in mice also demonstrated phosphorylation of both EIF2α and IRE1α, but hepatic DCCs in patients and mice were stained only by anti-pEIF2α (Fig. 6, A to D). Thus, DCCs may not be able to resolve the ER stress response because they do not activate the IRE1α pathway, which is required to generate the spliced, active form of XBP1 (XBP1s) that regulates the transcription of multiple proteins that promote protein folding in the ER.

Fig. 6 Absence of IRE1α pathway activation in DCCs.

(A) Liver sections from naïve and pre-immunized mice were stained with anti-luciferase (green) to identify PDA cells and with anti-pEIF2α (red) to assess activation of the PERK pathway. (B) Sections from the primary tumor and liver of a patient with PDA were stained with anti-p53 to identify PDA cells (green) and with anti-pEIF2α (red). (C) Liver sections from naïve and preimmunized mice were stained with anti-luciferase (green) to identify PDA cells and with anti-pIRE1α (red). (D) Sections from the primary tumor and liver of a patient with PDA were stained with anti-p53 to identify PDA cells (green) and with anti-pIRE1α (red). Photomicrographs are representative of five patients who had no detectable liver metastases. The ratios shown in the top right corners of the photomicrographs indicate the frequency of the observed DCC phenotype relative to the total number of DCCs that were assessed. All frequencies are compiled in table S2. White arrows designate DCCs. Scale bars, 25 μm. (E) Preimmunized mice (n = 5) that had received splenic injections of mM1TetXBP1s PDA cells were treated with Dox starting on the day of injection. The number of hepatic DCCs was determined 3 weeks later and compared with that of mice not receiving Dox (n = 5). (F) Preimmunized mice that had received splenic injections of mM1TetXBP1s PDA cells were treated (n = 10) or not treated (n = 8) with Dox beginning 3 weeks later. T cells were depleted by administering antibodies against CD4 and CD8. Growth of hepatic metastases was assessed by whole-body bioluminescence imaging. The dashed gray line represents the background luminescence in tumor-free mice. (G) The mean number of bioluminescent metastases formed with or without Dox treatment was determined in resected livers. **P < 0.01; ***P < 0.001. Error bars, mean ± SEM.

To determine whether this unresolved ER stress of DCCs contributes to their phenotype, we treated mM1DTLB PDA cells with the chemical chaperone 4-phenylbutyrate (4-PBA), which binds to solvent-exposed hydrophobic segments of unfolded or improperly folded proteins, thereby “protecting” them from aggregation and relieving ER stress (35). Treatment with 4-PBA decreased the proportion of cells lacking expression of Ecad (fig. S15A) and increased MHCI expression in both Ecad+ and Ecad cells (fig. S15B). We also assessed the effect of 4-PBA on the proliferative capability of mM1DTLB PDA cells by first eliminating proliferating mM1DTLB PDA cells through treatment with GcV and then pulsing the residual quiescent cells with EdU overnight in the presence or absence of 4-PBA. Relieving ER stress with 4-PBA increased the proportion of cells incorporating EdU by a factor of 10 (fig. S15C). We extended the analysis to DCCs by administering 4-PBA to preimmunized mice, beginning on the day of the mM1DTLB cell injection and continuing for 3 weeks. This treatment decreased the number of hepatic DCCs by a factor of 4 relative to the number in control mice (fig. S15D), consistent with the possibility that promoting the conversion of DCCs to replicating MHCI+ cells leads to their T cell–mediated immune elimination. To confirm this interpretation, we began 4-PBA treatment of preimmunized mice 3 weeks after the intrasplenic injection of mM1DTLB cells and depleted the mice of T cells. Within 2 weeks of 4-PBA treatment, all mice had developed MHCI+ macrometastases (fig. S15, E and F), and the number of macrometastatic lesions in the livers of 4-PBA–treated, T cell–depleted mice was six times the number in the livers of mice subjected only to depletion of T cells (fig. S15E).

We also specifically assessed the role of the IRE1α pathway in generating PDA cells with the DCC phenotype. In a loss-of-function experiment, mM1DTLB cells were treated with the IRE1α inhibitor Kira6 (36), which converted the mM1DTLB cells from an Ecad+ MHCI+ state to an Ecad MHCI phenotype in a dose-responsive manner (fig. S14C). We then performed gain-of-function experiments with the mM1TetXBP1s cell line expressing a doxycycline (Dox)–inducible form of XBP1s. This allowed us to test the hypothesis that circumventing the defect in IRE1α activation by expression of the spliced form of XBP1 would resolve ER stress in a more specific and physiological manner than was achieved with 4-PBA (fig. S13). Overnight treatment of mM1TetXBP1s cells with Dox induced the expression of XBP1s and almost eliminated the Ecad MHCI subpopulation (fig. S14D). To examine the role of the IRE1α pathway in DCCs in vivo, we continuously treated preimmunized mice with Dox beginning on the day that mM1TetXBP1s cells were injected intrasplenically. After 3 weeks, the frequency of hepatic DCCs was significantly decreased relative to their frequency in control mice not treated with Dox (Fig. 6E). The possibility that this decrease in XBP1s-expressing hepatic DCCs indicated killing by immune T cells was confirmed by repeating the experiment with the additional intervention of T cell depletion. Within 3 weeks of Dox-induced XBP1s expression in the DCCs, 9 of 10 T cell–deficient mice had developed hepatic macrometastases, whereas only 1 of 8 of the T cell–depleted mice without Dox treatment developed these lesions (Fig. 6F). Furthermore, the number of macrometastatic lesions in the livers of Dox-treated, T cell–depleted mice was seven times that in the livers of mice subjected only to depletion of T cells (Fig. 6G). These results lead us to conclude that unresolved ER stress has a nonredundant, cell-autonomous role in the maintenance of quiescent, immune-resistant DCCs.

Discussion

The clinical observation that PDA metastases develop in the majority of patients after the surgical removal of their primary tumors, despite no evidence of metastases at the time of surgery, indicates that these patients had harbored latent metastatic lesions. The nature of these latent metastases was suggested by our finding of single DCCs in the livers of patients and KPCY mice with PDA that have a distinctive phenotype of absent CK19, Ecad, and MHCI. The absence of two typical markers of epithelial ductal adenocarcinoma cells, without the occurrence of characteristic markers of EMT, indicates that these DCCs are distinct from the PDA cells that make up growing macrometastases (37). Moreover, the absence of MHCI implies an unusual relationship of DCCs to the adaptive immune system (38), as previously suggested in studies of breast, stomach, and colon carcinoma (12, 39). These descriptive findings provided the rationale for developing a mouse model that replicates DCCs with this distinctive phenotype and allows mechanistic studies defining both the cell-autonomous response responsible for the phenotype and the role of the immune system.

We hypothesized that metastases in patients with PDA may occur in the context of a cancer-specific adaptive immune response (2023). The absence of MHCI expression at the surface of DCCs in human and mouse PDA, and the absent or relatively infrequent occurrence of hepatic macrometastases, raised the possibility that immunity prevents the outgrowth of macrometastases while ignoring MHCI DCCs. This prediction was also based on the observation that DCCs in human and mouse PDA, in contrast to cancer cells in macrometastases, were quiescent. The finding that preexisting immunity prevented the occurrence of hepatic macrometastases while permitting the seeding of nonreplicating, MHCI hepatic DCCs verified this prediction and provided a potential explanation of how quiescent metastases can persist in the presence of an adaptive immune response that is capable of suppressing the growth of macrometastases.

This selective effect of adaptive immunity on macrometastases requires that the cell-autonomous mechanism that is responsible for the phenotype of DCCs invariably links the expression of MHCI to a capacity for cellular replication. This association not only was demonstrated in additional cell lines from primary and metastatic PDA tumors from KPC mice, but also was supported by the finding that ER stress, which inhibits MHCI expression (31, 32), and cell division (40) were the major up-regulated and down-regulated transcriptional signatures that distinguished Ecad from Ecad+ PDA cells. Thus, MHCI expression is “off” in quiescent cells and “on” in replicating cells. In addition, if the markedly increased expression in Ecad PDA cells of CHOP, a transcription factor that is induced by the PERK pathway of the UPR (33, 34), is taken as an indicator of the ER stress response, then the finding that in human and mouse PDA, hepatic DCCs are MHCI, Ki67, and CHOP+ links quiescence and immune concealment to an ongoing, unresolved ER stress in vivo. Our results confirm the previous suggestion that sustained activation of the UPR allows persistence of DCCs in the bone marrow of breast cancer patients (41) and may be related to the recent observation that knockdown of IRE1α inhibits tumor growth (42). We showed that resolution of ER stress by a chemical chaperone or, more definitively, by the induced expression of XBP1s causes reversion of DCCs to an epithelial, proliferating, MHCI+ phenotype. Thus, the DCC phenotype is caused by unresolved ER stress secondary to a block in IRE1α activation. Our study does not address how PDA cells avoid activating IRE1α, but other studies suggest that under high ER stress, IRE1α acquires endonucleolytic activity against RNA targets, in addition to the mRNA encoding XBP1, in a reaction termed regulated IRE1-dependent decay of mRNA (RIDD) (43). These additional endonucleolytic events may be associated with apoptosis, so that the ability of the PDA cell to adopt the DCC phenotype may protect it not only from death caused by immunity, but also from death caused by the RIDD reaction.

A latent capacity of hepatic DCCs for reversion to growing cancer cells was revealed by the outgrowth of macrometastases after the depletion of T cells. This finding also implied that T cells eliminated those DCCs that spontaneously reverted to a replicating MHCI+ epithelial phenotype. T cell killing of reverting DCCs must be efficient and occur before an immune-suppressive microenvironment is established, given that no macrometastases were observed during a 12-month period in T cell–replete mice with hepatic DCCs. The causal link between reversal of ER stress in the DCC and a replicating MHCI+ epithelial phenotype was established by the use of the chemical chaperone 4-PBA and by the induced expression of XBP1s in DCCs. Relief of ER stress by each intervention enhanced the expression of Ecad and MHCI and the proliferation of DCCs in vitro and, more importantly, caused the outgrowth of macrometastases from DCCs in T cell–depleted mice. This observation, coupled with the capacity of 4-PBA and XBP1s to decrease hepatic DCCs in T cell–replete mice, supports the essential role of the unresolved ER stress response in maintaining the DCC phenotype. Other factors, such as CXCR2 (C-X-C chemokine receptor type 2)–expressing neutrophils, may also have a role in this response (44).

The implications for therapy to prevent the occurrence of metastatic disease in patients after the surgical removal of their primary PDA may be twofold. First, outgrowth of latent DCCs in the mouse model required suppression of T cell immunity. Elevations of plasma cortisol after pancreatectomy in patients with PDA (45) are in the range that was found to be immune-suppressive in cachectic mice with PDA (46). This stimulation of the hypothalamic-pituitary-adrenal axis may also occur with the caloric deprivation that commonly occurs in patients after this surgical procedure (47). Therefore, postoperative parenteral hyperalimentation may be an effective means to decrease the occurrence of metastatic disease. Second, and more speculative, the administration of a chemical chaperone such as 4-PBA preoperatively, when tumor immunity is intact, might purge organs of latent DCCs, thereby decreasing the likelihood of postoperative metastatic disease.

Material and Methods

Animals

Male C57Bl/6 mice 10 to 12 weeks of age purchased from The Jackson Laboratory were used. All procedures were approved by the Cold Spring Harbor Laboratory Institutional Animal Care and Use Committee (IACUC) and were conducted in accordance with the NIH “Guide for the Care and Use of Laboratory Animals.”

Cell culture

Cell lines used in that study were a gift of D. Tuveson (Cold Spring Harbor Laboratory). The metastatic cell line mM1 and the primary cells line 1242 were derived from a KPC mice liver metastasis and a primary tumor, respectively. Cells were cultured in DMEM medium (#10-013-CV, Cellgro) supplemented with 10% fetal calf serum (FCS) (#1500-500, Seradigm), 100 units/ml penicillin, and 100 μg/ml streptomycin.

Human samples

Postmortem tissues were obtained following the Iacobuzio-Donahue laboratory rapid autopsy program previously described in detail (48). Patients were selected based on two criteria: p53 loss-of-heterozygosity, which allowed the staining for accumulation of mutant p53 protein, and absence of clinically detectable liver metastases.

Cell culture reagents

Ganciclovir (GcV, Sigma-Aldrich #G2536), was used at concentration ranging from 1-100mM. Tunicamycin (#11445, Cayman chemicals) was used at 5μg/ml. Sodium 4-phenylbutyrate (4-PBA, #11323, Cayman chemicals) was used at 5 mM. 5-Ethynyl-2'-deoxyuridine (EdU, #sc-284628, Santa Cruz biotechnology) was used at 10 mM.

EdU labeling

For short term EdU pulse, mice were injected every 12 hours with 1.4 mg of EdU over a 3-day period. For long term experiments, EdU was given in the drinking water at 0.82 mg/ml + 2.5% sucrose. Revelation of EdU containing cells was performed with a kit (Click-iT Plus EdU Imaging Kit, Molecular probes, #C10640) following the manufacturer protocol.

CFSE labeling

Cells were harvested from the cell culture and CFSE (CellTrace CFSE Cell Proliferation Kit, Molecular probes, #C34554) was used according to the manufacturer protocol.

Immune cell depletion

200μg of anti-CD4 (clone GK1.5, BioXcell) and/or anti-CD8 (clone 53.6-7, BioXcell) or anti-NK1.1 (clone PK136, BioXcell) were injected intraperitoneally 2 times per week.

Fluorochrome conjugation

For some applications (Ki67 staining in Fig. 1 and CHOP staining in Fig. 5), the anti-p53 antibody was conjugated using Zenon labeling reagent 555 (Life Sciences, # Z25005) following the manufacturer’s protocol.

Chemical chaperone treatment

4-PBA was given to mice in drinking water at 1 g/kg/day.

Doxycycline treatment

Dox was given to mice in drinking water at 2mg/ml + 5% sucrose

Plasmid preparation and transfection

For the generation of mM1DTLB PDA cells: Sequences of the genes of diphtheria toxin receptor (D), HSV-TK (T), firefly luciferase (L) and mTagBFP2 (B) were generated by gene synthesis (Invitrogen GeneArt) with appropriate restriction sites. For HSV-TK and firefly luciferase, sequences were digested and migrated with agarose gel electrophoresis. Appropriate size bands were purified with gel extraction kit (Qiagen, #28706) and cloned into the piggyback transposon vector (PB-EF1-MCS-IRES-Neo cDNA Cloning and Expression Vector, System Bioscience #PB533A-2). For mTagBFP2, the sequence was amplified by PCR and cloned into the linearized piggyback vector already containing HSV-TK and firefly luciferase with Gibson assembly master mix (New England Biolabs #E2611L). Plasmids were purified using Zippy plasmid miniprep kit (Zymo research, #D4037). Cells were transfected with the resulting vector and the plasmid of the super piggyback transposase (System Bioscience, #PB210PA-1) with Lipofectamine 3000 (Thermofisher, #L3000015). Cells were selected for stable transfectant with G418 (Sigma-Aldrich #A1720) and sorted for the highest expression of mTagBFP2 (fig. S1A). The functionality of Luciferase, DTR and HSV-TK was verified in vitro (fig. S1, B and C).

For the generation of mM1TetXBP1s PDA cells: The sequence of spliced XBP1 (XBP1s) and mCherry linked by a 2A peptide sequence was generated by gene synthesis (Genscript) with appropriate restriction sites. The sequence was digested and migrated with agarose gel electrophoresis. Appropriate size bands were purified with gel extraction kit (Qiagen, #28706) and cloned into an inducible TetON plasmid (gift of S. Lyons). Plasmids were purified using Zippy plasmid miniprep kit (Zymo research, #D4037). mM1DTLB PDA cells were transfected with the resulting vector with Lipofectamine 3000 (Thermofisher, #L3000015). Cells were selected for stable transfectant with puromycin (Sigma-Aldrich #P8833) and sorted first for no mCherry expression without doxycycline (Dox) to remove cells with potentially leaking expression and then for the highest expression of mCherry upon Dox treatment to select inducible cells (fig. S13A). Tight inducible overexpression of XBP1s was confirmed in vitro by Western blot analysis (fig. S13B).

Generation of preimmunized mice

Mice were anesthetized by isoflurane and dorsal fur was removed with a clipper. A subcutaneous injection of 106 PDA cells was performed. Tumors were grown for 2 weeks. In order to eliminate the tumors, diphtheria toxin (DTx, List Biologicals, #150) was injected intraperitoneally on 2 consecutive days at a dose of 25 ng/g. GcV was injected intraperitoneally on 2 consecutive days at a dose of 5 mg/kg. In approximately 20% of the cases, one round of DTx and GcV injection was not sufficient to eliminate the tumors, and a second round of injection was performed 1 week later. On rare instances two rounds of injections were insufficient, and these animals were not used. The absence of remaining cancer cells was confirmed by bioluminescence (fig. S1D) and mice were used for experiments at least 3 weeks after the last injection of DTx and GcV.

Pancreatic cancer metastasis model

For the liver metastasis model, animals were anesthetized by isoflurane, a 1 cm incision was made in the left subcostal region and the spleen was exposed. A suspension of 106 pancreatic cancer cells in 50 μl of PBS was injected into the body of the spleen. Immediately following injection, mice were splenectomized to prevent growing of extra-hepatic tumors. The peritoneum was closed with a 5-0 absorbable surgical suture (Vicryl, Ethicon) and the skin with wound clips (Roboz surgical instruments). For the lung metastasis model 106 cells in 100 μl were injected through the tail vein.

Bioluminescence imaging

Mice were anesthetized by isoflurane and ventral fur was removed with a clipper. 150 μl of a 30mg/ml D-Luciferin K+ salt (Perkin-Elmer, #122799) solution was injected intraperitoneally. Mice were imaged with an IVIS Spectrum in Vivo Imaging System (Perkin-Elmer) 14 min after the injection.

Immunofluorescence

Organs were harvested and fixed with PLP buffer (1% PFA, 80mM L-lysin, 10mM NaIO4); through portal vein perfusion in the case of livers. Organs were embedded in Tissue-Tek OCT compound (Sakura, #4583) and section of 10 μm were cut on a Leica cryostat. Between each of the following steps, sections were washed three times for five min with PBS. Tissue sections were post-fixed for 15 min at room temperature with PLP buffer and permeabilized for 15 min at room temperature with 0.1% Triton X100 in PBS for mouse tissues or 30 min at room temperature with 0.2% Triton X100 in PBS for human tissues. Sections were surrounded with hydrophobic barrier pen (ImmEdge, Vector labs, #H-4000) and unspecific antibody labeling was blocked for 1 hour at room temperature with 10% donkey serum (Jackson immunoresearch, #017-000-121) in PBS for mouse tissues or 10% goat serum (Thermo Fisher, #16210-064) in PBS for human tissues. Primary antibodies (table S3) were incubated overnight at 4°C in the dark. Secondary antibodies (table S3) were incubated for 2 hours at room temperature in the dark. Nuclei were counterstain with DAPI (Molecular probes, #R37606) for 10 min at room temperature in the dark and sections were mounted with ProLong diamond antifade mountant (Molecular probes, #P36965) for mouse tissues or ProLong gold antifade mountant (Molecular probes, #P36934). Images were acquired using a spinning disk confocal microscope (Perkin-Elmer UltraVIEW VoX, High speed spinning disk Yokogawa CSU-X1) with 20X or 60X objectives. Images were analyzed with ImageJ. CFSE (CellTrace CFSE Cell Proliferation Kit, Molecular probes, #C34554) and EdU (Click-iT Plus EdU Imaging Kit, Molecular probes, #C10640) staining were performed following the manufacturer protocol.

Flow cytometry analysis and sorting

Cells were harvested with TrypLE (Gibco, #12605036), distributed in 96-well round-bottom plate and centrifuged at 2000 rpm for 1 min at 4°C. Supernatant was removed and the cells incubated in ice cold FACS Buffer (1% FBS and 0.02% Sodium Azide in PBS) with Fc Block (Biolegend clone 93) at 4°C for 15 min. After a wash in FACS Buffer, the cells were incubated with primary and secondary antibodies (table S3) for 15 min at 4°C in the dark. The cells were then washed twice in FACS Buffer and resuspended in 450 ml PBS before analysis using a LSR Fortessa (BD Biosciences) operated by a FACSDIVA (BD Biosciences) software. Data analysis was performed on FlowJo 10 (FlowJo, LLC). For sorting, cells were submitted to the same procedure as for flow cytometry analysis and processed using BD FACSAria (BD Biosciences).

Disseminated cancer cell count

Sections were taken every 100 μm through the entire thickness of the liver. The thickness of the liver was measured at the same time. DCCs were identified by luciferase expression and counted in every section. The number of DCCs per whole liver was then calculated.

scRNA-seq

scRNA libraries were generated from viable Ecad+ and Ecad tumor cells obtained by flow cytometry sorting. PDA cells were adjusted to 200 cells/μl and applied to the C1 system for single-cell capture with a 10-17μm IFC (Fluidigm). In the C1, whole-transcriptome amplification was performed with the SMARTer kit (Clontech), and the product was converted to Illumina sequencing libraries using Nextera XT (Illumina). RNA-seq was performed on a NextSeq instrument (Illumina) single read 75. Quality control was performed using FastQC and cells with low read numbers were eliminated. Reads were aligned against reference mouse genome (EnsMart72) and transcript expression values were determined after transcript normalization (transcript per million; TPM) with AltAnalyze. Cells with aberrant expression for Gapdh, Actb and Hprt were eliminated leaving 104 Ecad+ and 98 Ecad cells for further analysis. Transcripts were considered significantly expressed if TPM ≥ 1. Differential expression analysis was performed considering fold-change between Ecad+ and Ecad cells ≥ 2 fold as a cut-off. Pathway enrichment analysis and network maps were performed with Cytoscape v3.5.1 and the ClueGo v2.3.3 plugin. Pathways with P < 0.01 after Benjamini-Hochberg procedure for false discovery rate were considered significant.

Statistical analysis

Two-column comparisons were performed with an unpaired t test. Comparison of three or more columns was performed using a one-way ANOVA followed by Tukey’s procedure. A P value less than 0.05 was considered significant: ***P < 0.001, **P < 0.01, and *P < 0.05. All statistical analyses were performed using GraphPad Prism software version 6.

Supplementary Materials

References and Notes

Acknowledgments: We thank D. Tuveson for providing mM1 and 1242 cells and the KPCY liver tissues. We thank R. Kappagantula for her help in processing patients’ information. Funding: This work was supported by a Distinguished Scholar Award to D.T.F. from the Lustgarten Foundation, an award from the Cedar Hill Foundation, and 5P30CA45508-29 from NIH-NCI. A.P. was supported by the Philippe Foundation. Author contributions: Conceptualization: A.P. and D.T.F. Acquisition of data: A.P., N.A., N.M., Z.L.K., A.G., R.Y., C.A., J.A., M.E., C.A.I.-D., and S.K.L. Analysis of data: A.P., N.A., Z.L.K., A.G., R.Y., J.A., M.E., and D.T.F. Writing (original draft and review and editing): A.P. and D.T.F. Competing interests: D.T.F. is a cofounder of Myosotis (a company developing cancer immunotherapies) and is on the Scientific Advisory Boards of iTEOS Therapeutics (a company developing immuno-oncology drugs), IFM Therapeutics (a company developing therapies targeting the innate immune system), and Kymab (a company developing therapeutic antibodies). Data and materials availability: scRNA-seq data were deposited in GEO with the accession number GSE108811.

Stay Connected to Science

Navigate This Article