Abstract

INTRODUCTION

Despite our increased understanding of pancreatic ductal adenocarcinoma (PDAC) from transgenic mouse models,¹ this human disease remains resistant to systemic therapies. Genomic and epigenetic alterations that are characteristic of all cancers provide a diverse set of antigens that immune cells can use to distinguish and selectively target malignant cells.^2-4 The importance of tumor-infiltrating lymphocytes, in particular anti-tumor CD8⁺ cytotoxic T-cell location, has been highlighted by their prognostic value in many human cancers.^2-7 However, similar immunohistological analysis investigating the correlation of T-cell infiltration with pancreatico-bilary cancer prognosis has been understudied to date. Increased infiltration of immunosuppressive cells such as tumor-associated macrophages (TAMs), myeloid derived suppressor cells (MDSCs) and regulatory T-cells (T_regs) has been detected in the inflammatory tumor microenvironment of genetic mouse models of PDAC (including ‘KPC’ mice: LSLKrasG12D/+; LSL-Trp53R172H/+; Pdx-1-Cre) during disease progression to invasive cancer.⁸ In human disease, increased T_regs infiltration was found to be increased in PDAC tissue when compared to non-malignant pancreas, whereas no differences were observed in peripheral blood mononuclear cells of healthy donors compared to cancer patients.⁹

PDAC is characterized by a pronounced desmoplastic reaction mediated by the activation of quiescent pancreatic stellate cells (PSC).¹⁰ We have previously demonstrated that this critical event is reversible.¹¹ Importantly, in situ hybridization studies from other groups have demonstrated differential gene expression signatures when comparing defined stromal compartments in PDAC that include the juxtatumoral stroma (immediately adjacent to the tumor epithelial cells) and the panstromal compartment (non-adjacent to tumor cells).^{12, 13} In this present study, we show for the first time that there is differential immune cell infiltrate unique to these PDAC stromal compartments and that enhanced juxtatumoral CD8+ T-cell infiltrate correlates with better patient survival. Our experimental assays, including the use of the KPC PDAC mouse model, show that CD8⁺ T-cells are prevented from targeting PDAC cells by activated PSC that reside in the panstromal compartment. Of translational relevance, rendering activated PSC quiescent reduced PDAC patient CD8⁺ T-cell in vitro migration and adhesion towards PSC and allowed enhanced in vivo juxtatumoral infiltration of CD8⁺ T-cells.

MATERIALS AND METHODS

RESULTS

Immune cell infiltration in pancreatico-biliary diseases (PBD)

We performed an unbiased, automated evaluation of immune cell infiltrate for pancreatico-biliary cancers such as PDAC, ampullary carcinoma (AC) and cholangiocarcinoma (CC) using dedicated tissue micro-arrays (TMA). These tumors were compared alongside other PBD including borderline malignant condition: mucinous cystic neoplasm (MCN); chronic inflammation: chronic pancreatitis (CP); and normal donor control TMAs (Supplementary Tables 1-2). Our immune cell density measurements were validated in whole tissue sections and by two distinct methods: as a percentage of all cells and by cell numbers per unit area (Supplementary Figures 1-4). Our results showed that immune cell densities were significantly increased in malignant (PDAC, CC, AC) tumors that are associated with an adverse prognosis,²¹ compared to other PBD and healthy controls (Figure 1). Increased T_reg (FoxP3⁺) infiltrate in human PDAC was confirmed as reported before.⁹ In contrast, the density of CD8⁺ T-cell infiltrate in PDAC was comparable to chronic pancreatitis but significantly reduced compared to AC and CC (P<0.01). These findings in human PDAC when compared with other PBD, reported for the first time, are consistent with reduced infiltration of CD8⁺ cytotoxic T-cells recently reported in the KPC mouse model of PDAC.^{8, 22, 23}

Open in a separate window

Figure 1

Immune cell infiltrate in human pancreatico-biliary diseases

Immune cells infiltrates were measured in pancreatico-biliary diseases such as chronic pancreatitis (CP, n=4), mucinous cystic neoplasms (MCN, n=6), ampullary carcinoma (AC, n=9), pancreatic ductal adenocarcinoma (PDAC, n=93-98, shaded box), cholangiocarcinoma (CC, n=21) and normal tissues (N, n=11-14) using Ariol (Genetix) software as described in Supplementary Figures 1-4. Box (median with interquartile ranges (25^th and 75^th)) and whisker (5^th and 95^th centiles) plots (outliers are represented by individual dots) demonstrate that all diseases demonstrate significant, yet varied, density and profile for CD3⁺(A), CD4⁺(B), CD8⁺(C), CD20⁺(D), FoxP3⁺(E), and CD68⁺(F) immune cells infiltrate as compared by Kruskal-Wallis test (p-value < 0.0001 for all immune cells). Further statistical comparisons between columns were done by Dunn’s post test and indicated by *** p< 0.001; ** p= 0.001 to 0.01; * p= 0.01 to 0.05.

Differential immune cell infiltrate in stromal sub-compartments

We next characterized the immune infiltrate within stromal sub-compartments of PDAC and identified that cytotoxic effector T-cells (CD8⁺), T_regs (FoxP3⁺), B-cells (CD20⁺) and natural killer or NK-cells (CD56⁺) were significantly reduced in juxtatumoral compartments compared to the panstromal areas. This was in contrast to neutrophil (myeloperoxidase⁺) and macrophage (CD68⁺) stromal infiltration (Figure 2, Supplementary Figure 5).¹³ This differential immune cell infiltrate, particularly for CD8⁺ T-cells, was confirmed in an independent TMA cohort of PDAC patients composed mainly of advanced cancers (Supplementary Figure 6). Interestingly, this differential immune cell infiltrate was not observed in AC (Supplementary Figure 7) or CC (Supplementary Figure 8). Our analysis showed that tissue immune cell infiltrate had no correlation with circulating immune cell counts (Supplementary Figures 9-11). The paucity of functional vasculature close to the PDAC tumor cells (juxtatumoral) is well established.²⁴ Taken together these observations strongly suggest a hitherto unreported PDAC-specific CD8⁺ T-cell trafficking defect within human PDAC tissues.

Open in a separate window

Figure 2

Stromal compartment specific immune cell infiltration in human PDAC

Proportion of immune cell infiltrate was determined as shown in Supplementary Figure 1-4. Juxtatumoral stromal (within 100μm of tumor cells (single or islands /ducts)) and panstromal (the rest of the tumor stroma) were defined. Each data point represents a single patient (median scores of all TMA cores (n=6)) and lines represent median for the cohort of patients. Comparison of immune cell infiltration in these two PDAC stromal compartments was made for CD4⁺ (A), CD8⁺ (B), FoxP3⁺ (C), CD20⁺ (D), CD68⁺ (E) and Myeloperoxidase⁺ (F) cells, and also for CD3⁺ and CD56⁺ cells (Supplementary Figure 5). Whilst PDAC tissues demonstrate a significantly higher density in the juxtatumoral stroma relative to the panstromal compartment for CD68⁺ cells, the reverse was true for CD8⁺, FoxP3⁺ and CD20⁺ cells with an equal density for CD4⁺ and Myeloperoxidase⁺ cells implying a differential immune cell infiltration defect. Mann Whitney U test; p-values are two-tailed.

Our data further demonstrates for the first time that a higher CD8⁺ T-cell juxtatumoral infiltrate is associated with improved survival post-surgical resection as analyzed by X-Tile software (Figure 3, Supplementary Table 5).^{17, 25} Of interest, a lower juxtatumoral T_reg (FoxP3⁺) infiltrate also demonstrated a pro-survival effect in PDAC patients. Taken together, this data suggests that reduced infiltration of anti-tumor CD8⁺ T-cells to target tumor epithelial cells in the PDAC microenvironment has important clinical impact.

Open in a separate window

Figure 3

Correlation of stromal compartment-specific immune cell infiltration with survival

Prognostic implications upon differential immune cell infiltration for CD8⁺ (A-C) and FoxP3⁺ (D-F) cells in whole PDAC tissue (A, D), the juxtatumoral compartment (B, E) and the panstromal compartment (C, F) was determined using optimal cut-off for proportion of immune cells using X-tile software (Yale).^15-17 Survival comparisons were made between patients with low (black) and high (grey) density of the immune cell infiltrate. Miller-Seigmund corrections of p-values were performed to account for multiple comparisons (Supplementary table 5). The various optimal cut-off for proportion of immune cells defining high versus low infiltrate are indicated. Patients with high densities of CD8⁺ in the whole PDAC tissue (A) or the juxtatumoral compartment (B) demonstrate a statistically significant survival benefit; however, this was not true for panstromal compartment (C). Lower FoxP3⁺ infiltration in all compartments resulted in better survival.

Reversal of the juxtatumoral exclusion of cytotoxic T-cells in murine PDAC

We next identified similar juxtatumoral exclusion of CD8⁺ T-cells and CD45R⁺ B-cells, but not F4/80⁺ macrophages in the PDAC microenvironment of KPC mice (Figure 4, Supplementary Figure 12). Previously we have demonstrated that treating these PDAC mice with ATRA renders activated pancreatic stellate cells (PSC) quiescent leading to stromal collapse and a detrimental effect on resident tumor cells through a multitude of signaling cascades.¹¹ In the present study, we found that ATRA treatment enhanced juxtatumoral CD8⁺ T-cell infiltration but not CD45R⁺ B-cells or F4/80⁺ macrophages. Recent studies suggest that MDSCs have an important role in the PDAC-associated CD8⁺ T-cell infiltration defect.^{22, 23} Thus, we examined this population of immune cells in our ATRA-treated KPC mice experiments but did not observe any changes in MDSC (CD11b⁺) infiltrate. Thus, it is conceivable that activated PSC in PDAC sequester anti-tumor CD8⁺ T-cell migratory and adhesive function preventing access to the tumoral cells. Since it is not technically feasible to study the dynamic spatial interactions between PSC and immune cells by current in vivo models, we designed novel in vitro chemotaxis assays examining the possible immunomodulatory role of PSC.

Open in a separate window

Figure 4

Juxtatumoral exclusion of CD8⁺ T-cells in KPC mice is reversed upon stromal collapse engineered by targeting stellate cells with ATRA

KPC mice with fully developed tumours (day 60-200) were enrolled into a pre-clinical trial where they were either given vehicle (n=4) or ATRA (n=6). ATRA, selectively targets and restores stellate cells quiescence.¹¹ Proportion of CD8⁺ (A,B), CD4⁺ (C), CD45R⁺ (D), F4/80⁺ (E) and CD11b⁺ (F) cells in the panstromal and juxtatumoral compartments were measured as shown in Supplementary Figure 12. All data are presented as box (median with interquartile ranges (25^th and 75^th)) and whisker (5^th and 95^th centiles) plots. The juxtatumoral exclusion of CD8⁺ (cytotoxic T-cells) and CD45R⁺ (B-cells) cells was demonstrated in vehicle-treated mice. There was a significant increase in the CD8⁺ T-cell infiltrate (A,B) into the juxtatumoral compartment upon treatment of mice with ATRA which was not seen for CD4⁺ (helper T-cells), CD45R⁺ (B-cells), F4/80⁺ (macrophages) or CD11b⁺ (MDSC) cells.

Scale bar: 100μm

Mann-Whitney U-test, all p-values are two-tailed. ** p= 0.001 to 0.01; n.s.= not significant

CD8⁺ T-cells migrate preferentially towards activated PSC

Quiescent (normal pancreas) and activated (PDAC) PSCs have distinct phenotypes.¹¹ Our chemotaxis assays showed that healthy donor CD3⁺ T-cells adhered and migrated preferentially to activated PSC (aPSC) conditioned medium compared to quiescent PSC (qPSC) or cancer cell (Capan1 and AsPc1) conditioned medium that contained serum (Figure 5, Supplementary Figure 13). Notably, quiescent PSC induced chemotactic responses were remarkably similar to baseline levels.

Open in a separate window

Figure 5

In vitro immune-stellate cells interactions

Immuno-magnetic bead separated immune cells from un-diseased human donors (n=4) were used. Pancreatic stellate cells (PSC) were rendered quiescent using ATRA.¹¹ Proportion of adherent T-cells to aPSCs or qPSCs were quantified after incubation for one hour and fixation and staining. (A; paired t-test; p-value is two-tailed). Transwell (5 μm) migration assays of different immune cells towards conditioned media (CM) from activated (aPSC), quiescent (qPSC) PSC and cancer cells (Capan1 and AsPc1) were carried out over 4-8 hours. Background migration of T-cells towards serum-free RPMI was subtracted to give ‘specific’ migration and normalized to migration towards RPMI supplemented with 10% FBS (‘basal’ serum directed migration) as shown in Supplementary Figure 13. These conditions served as internal controls for comparison across biological replicates.

We demonstrated reduced migration of CD3⁺ (B), CD4⁺ (C) and CD8⁺ (D), CD56⁺ (E) but not CD19⁺ qPSC CM compared with aPSC CM. The absolute CD4⁺ (B) cell number migration was minimal with little difference over the basal migration. Migration of all subsets of T-cells towards aPSC CM was higher than migration towards cancer cell CM. The most dramatic, and perhaps clinically relevant, fold change in migration of T-cells was seen with CD8⁺ T-cells (D) and NK cells (E).

Bar chart represents mean ± SEM. *** p< 0.001; ** p= 0.001 to 0.01; * p= 0.01 to 0.05, Comparisons were conducted with ANOVA with comparisons between columns using Bonferroni’s Multiple Comparison Test.

In line with our in vivo histological observations, the differential migration of CD4⁺ helper T-cells towards cancer cells or PSC was not as significant when compared to CD8⁺ T-cell experiments (Figure 5C,D). CD56⁺ NK-cells showed preferential migration to aPSC compared to qPSC, while CD19⁺ B-cells did not (Figure 5E,F). Together with our earlier KPC mouse study findings, we believe that these results suggest that panstromal-associated aPSC attract and adhere to CD8⁺ T-cells preventing their effective juxtatumoral infiltration.

CXCL12 mediates the increased T-cell migration towards aPSC

Re-analysis of our previous Affymetrix gene expression data of PSC in PDAC revealed that aPSC alter the expression of a number of cytokines/chemokines and adhesion-mediating molecules compared to qPSC (Figure 6A, Supplementary Figure 14).¹¹ The transcriptomic changes for a number of these molecules was validated at the protein level both in vitro and in vivo in human PDAC by others²⁶ and us (Figure 6B-E).^{11, 20} Notably, fibronectin expression was markedly increased in human PDAC (especially in the panstromal compartment compared to normal pancreas, Figure 6B). Fibronectin expression in murine PDAC was markedly reduced following ATRA treatment (Figure 6C).

Open in a separate window

Figure 6

Adhesion molecule and cytokine changes upon activation of PSC

(A) Re-analysis of gene expression microarrays of qPSC compared to aPSC¹¹ for significantly differentially expressed genes involved in cell adhesion (Gene Ontology GO:0007155) gene set enrichment test = 0.007, performed using DAVID (Benjamini-Hochberg test).¹⁹ Hierarchy clustering of all samples (replicates) from four groups, aPSC or control (black) and qPSC under different treatment conditions such as when plated on Matrigel and treated with 1 μM ATRA (blue), when plated on Plastic and treated with 1 μM ATRA (burlywood) and plastic with 10 μM ATRA (dark brown), was performed based on the expression profiles of these differentially expressed probes using the Euclidean metric. Also see Supplementary Figure 14 for the significant changes in gene expression derived from the time-course experiment as well as changes in cytokine-cytokine receptor interaction pathway.

(B) Fibronectin was further assessed in normal human pancreas and human PDAC demonstrating upregulation of fibronectin expression in cancer, particularly in the panstromal compartment where they sequester immune cells (black arrow). Scale bar, 100 μm. Box (median with interquartile ranges (25^th and 75^th)) and whisker (5^th and 95^th centiles) plots of fibronectin immunostain is demonstrated. Mann-Whitney U-test.

(C) Upon treatment with ATRA, KPC mice demonstrated a significant reduction of fibronectin deposition. Scale bar, 100 μm. Box (median with interquartile ranges (25^th and 75^th)) and whisker (5^th and 95^th centiles) plots of fibronectin immunostain is demonstrated. Mann-Whitney U-test.

(D) qRT-PCR demonstrated significant alteration of some of the transcripts identified by gene-expression microarray analysis. Other transcripts have been studied before.¹¹ Bar chart represents mean ± SEM. Paired t-test; p values are two-tailed.

(E) Concentrations of the proteins in conditioned media collected from qPSC and aPSC using ELISA. Surprisingly at protein level, there was no/little difference in secretion between qPSC and aPSC for sICAM1/ CD54 and MCP1/ CCL2. However, confirming the gene expression data, aPSC secreted less IL8 than qPSC and the reverse was true for SDF-1α/ CXCL12. Summary data represents mean ± SEM. Paired t-test; p values are two-tailed.

Of the multitude of signaling cascades/adhesion molecules deregulated with an activated PSC phenotype, we chose to study upregulated CXCL12 in aPSC because of its well-established chemotactic properties for T-cells.²⁷ CXCL12 knock-down in both in vitro aPSC and primary PDAC patient PSC resulted in a reduction of healthy donor and PDAC patient CD8⁺ and CD4⁺ T-cells T-cell migration to levels comparable to qPSC (Figure 7, Supplementary Figure 17). Our flow cytometric analysis showed that whilst the circulating effector/memory T-cell ratio^{28, 29} was similar in PDAC patients compared to age-matched healthy donors (Supplementary Figures 15-16), the CXCR4 expression was markedly upregulated in PDAC patient T-cells (Figure 7F, Supplementary Figure 17A). Taken together, this data suggests an important role for the CXCL12/CXCR4 axis in regulating T-cell trafficking.

Open in a separate window

Figure 7

T-cell migration is mediated by CXCL12

CXCL12 was knocked down in PSC cell line (PS1: A) and primary PSCs (B) with two distinct targeting siRNAs and appropriate controls. CXCL12 secretion was measured in conditioned media. PSC treated with siRNA for CXCL12 demonstrated significant reduction of CXCL12 secretion (A,B) which was equivalent to quiescent PSC levels. Migration of normal human donor CD8⁺ T-cells to conditioned media of CXCL12 knockdown PSC (C: PS1 cell line and D: primary PSC) demonstrated significant reduction, which was equivalent to qPSC levels. Similar results were found with PDAC patient CD8⁺ T-cells (E). PDAC patient CD8⁺ T-cells demonstrated significant upregulation of CXCR4 as measured by Flow-cytometry (F).

*** p< 0.001; ** p= 0.001 to 0.01; * p= 0.01 to 0.05. Comparisons were conducted with ANOVA with comparisons between columns using Bonferroni’s Multiple Comparison Test.

DISCUSSION

PDAC has a 5-year overall patient survival of 3-5% making it the PBD with the worst prognosis.²¹ In this report, we performed a comprehensive global as well as stromal compartment-specific comparison of immune infiltrate in a variety of PBD. We show an increasing immune cell infiltrate from normal (healthy pancreas tissue) to inflammatory (CP), borderline (MCN) and to malignant (AC, PDAC and CC) diseases. To our knowledge, we identify for the first time that higher CD8⁺ T-cell infiltration in PDAC tumors correlates with better patient survival, a finding that that provides evidence for anti-tumor CD8⁺ T-cell cancer immunoediting in human PDAC.²

The PDAC stromal microenvironment is composed of activated myo-fibroblasts including aPSC, extracellular matrix, blood vessels and various immune cells, whose individual contribution to tumor progression remains elusive.²⁶ The PDAC stroma has been classified into two compartments based on differential gene expression:^{13, 30} the juxtatumoral and panstromal compartments. Our immunohistochemistry analysis revealed that whilst some immune cells (CD68⁺ macrophages, myeloperoxidase⁺ neutrophils) infiltrate the juxtatumoral compartment, others (CD8⁺ T-cells, FoxP3⁺ T_regs CD56⁺ NK-cells, CD20⁺ B-cells) are sequestrated in the panstromal compartment. The use of the KPC mouse model of PDAC that exhibits an intact in situ cellular desmoplastic component allowed us to confirm panstromal sequestration of CD8⁺ T-cells and CD45R⁺ B-cells, but not CD4⁺ T-cells or F4/80⁺ macrophages, in agreement with the human data.

More importantly, we could demonstrate that in vivo ATRA treatment, that converts aPSC to a qPSC phenotype¹¹, specifically enhanced CD8⁺ T-cell infiltration towards tumor cells but no change was demonstrable for the CD4⁺ T-cells, CD45R⁺ B-cells, F4/80⁺ macrophages and MDSC (CD11b⁺) infiltrate. In the absence of appropriate tools we were unable to define MDSC infiltrate in stromal tissue compartments in human PDAC.³¹ MDSCs have been postulated to drive CD8⁺ T-cells exclusion recently in murine KPC-derived subcutaneous PDAC models devoid of in situ cellular desmoplastic component, particularly stellate cells.^{22, 23} Of interest, in vitro but not in vivo studies demonstrate the PSC may modulate differentiation of MDSC in an IL6-STAT3 dependent manner.³² Depletion of FAPα⁺ stromal cells in a subcutaneous mouse model of PDAC resulted in immunological control (CD8⁺ T-cell-mediated) of tumor growth, a phenomenon not seen in Rag-2 deficient mice.³³ The relationship between FAPα cells and PSC remains to be established. Furthermore, in a cohort of surgically incurable PDAC patients and in KPC mice, combining gemcitabine therapy with an agonist CD40 antibody led to significant tumor regression. CD40-activated macrophages were found to infiltrate the tumors to facilitate the depletion of the stroma. This depletion of tumor stroma was associated with tumor immunoclearance.³⁴ Our data suggest CD8⁺ T-cells may be involved in tumor immunoclearance if their trafficking is unimpeded by the ‘PDAC stromal barrier’, a finding substantiated by murine PDAC investigations.^{8, 34}

Our ex vivo migratory bioassays using both healthy donor and PDAC patient cytotoxic CD8⁺, CD4⁺ helper T-cells and NK-cells showed that aPSC conditioned media attracted increased numbers of these lymphocytes compared to tumor cells or qPSC. Taken together, these data provide evidence that aPSC are key regulators of the immune cell infiltrate in the PDAC stromal microenvironment. Understanding the complex temporal (disease progression) and spatial (stromal compartment) relationship between aPSC, FAPα⁺ stromal cells, CD8⁺ T-cells, MDSC, T_regs and NK-cells in the PDAC disease process will be important for identifying novel cellular and molecular targets for therapy.³⁵

Recently, we demonstrated that aPSC modulate a range of cellular process in tumor cells in a dose-dependent manner. These findings suggest a vital role for aPSC in PDAC disease progression mediated through perturbation of multitude of signaling cascades.²⁰ Our analysis of the transcriptome of PDAC-associated aPSC revealed deregulated adhesion and chemokine/cytokine genes.¹¹ We hypothesized that these molecules likely contribute to stromal microenvironment-specific regulation of immune cell trafficking and infiltration into the tumor.^35,36 In this present study, we focused on CXCL12 that was upregulated in aPSC and is known to be an important chemoattractant for T-cells.²⁷ Our chemotaxis assays showed that both donor and PDAC patient CD8⁺ and CD4⁺ T-cells showed elevated migration towards aPSC compared to qPSC that was CXCL12-dependent in CD8⁺ T-cells. CXCL12-siRNA treated aPSC cells or conversion of aPSC to qPSC reduced CXCL12 expression and T-cell chemotaxis. Notably, we show that PDAC patient T-cells express higher levels of the CXCL12 receptor (CXCR4) compared to healthy donor T-cells, but with no change in the proportion of effector or memory T-cells. Taken together, this data suggests an important role for stromal PSC:T-cell interaction and the CXCL12/CXCR4 axis in PDAC. Moreover, CXCL12 has been implicated in the maintenance of PDAC cancer stem cells in a PSC niche by regulating the expression of sonic hedgehog.^{37, 38}

Further work remains to be done to delineate the relative contribution of NK-cells and T_reg cells, both of which are also be sequestered in the panstromal compartment in human PDAC. Of note, it has been suggested that this intra-stromal cross-talk (tumor immuno-suppressive network) can further impede CD8⁺ T-cell trafficking and immuno-editing.³⁵ The relative contribution of CXCR4/CXCL12 and other cytokines/chemokines/adhesion molecules such as IL-8, fibronectin amongst others will need to be precisely defined to refine therapeutic strategies.

Taken together, these data suggest that the complex tumor microenvironment, in particular, the increasingly recognized PSCs^{11, 20, 26} regulate the tumor immune response. We demonstrate that PDAC-associated aPSC orchestrate many such interactions by sequestration of CD8⁺, CD4⁺ T-cells, NK-cells and T_regs in the panstromal compartment. Our data provides the basis for further studies to explore this clinically significant, yet unrecognized, stromal-immune cell cross-talk in PDAC. Successful targeting of desmoplastic and immune stroma cells may provide an interesting approach in tackling human PDAC in an attempt to improve survival in this incurable disease.

Supplementary Material

Acknowledgements

Abbreviations

Footnotes

REFERENCES