DKK1 drives immune suppressive phenotypes in intrahepatic cholangiocarcinoma and can be targeted with anti‐DKK1 therapeutic DKN‐01

Abstract Background and aims Dickkopf‐1 (DKK1) is associated with poor prognosis in intrahepatic cholangiocarcinoma (iCCA), but the mechanisms behind this are unclear. Here, we show that DKK1 plays an immune regulatory role in vivo and inhibition reduces tumour growth. Methods Various in vivo GEMM mouse models and patient samples were utilized to assess the effects of tumour specific DKK1 overexpression in iCCA. DKK1‐driven changes to the tumour immune microenvironment were characterized by immunostaining and gene expression analysis. DKK1 overexpressing and damage‐induced models of iCCA were used to demonstrate the therapeutic efficacy of DKK1 inhibition in these contexts using the anti‐DKK1 therapeutic, DKN‐01. Results DKK1 overexpression in mouse models of iCCA drives an increase in chemokine and cytokine signalling, the recruitment of regulatory macrophages, and promotes the formation of a tolerogenic niche with higher numbers of regulatory T cells. We show a similar association of DKK1 with FOXP3 and regulatory T cells in patient tissue and gene expression data, demonstrating these effects are relevant to human iCCA. Finally, we demonstrate that inhibition of DKK1 with the monoclonal antibody mDKN‐01 is effective at reducing tumour burden in two distinct mouse models of the disease. Conclusion DKK1 promotes tumour immune evasion in iCCA through the recruitment of immune suppressive macrophages. Targeting DKK1 with a neutralizing antibody is effective at reducing tumour growth in vivo. As such, DKK1 targeted and immune modulatory therapies may be an effective strategy in iCCA patients with high DKK1 tumour expression or tolerogenic immune phenotypes.


| INTRODUC TI ON
Cholangiocarcinoma (CCA) is a group of cancers of the biliary tree, a network of ducts that drain bile into the intestine. Typically described as either intrahepatic (iCCA) or extrahepatic (comprising of distal dCCA and perihilar pCCA), CCAs comprise a highly proliferative epithelium and dense, immune cell-rich stroma. [1][2][3] These malignancies constitute approximately 15% of primary liver cancers. 4 Following diagnosis, patient outcomes are dismal and treatment options are extremely limited, with less than one-third of patients being eligible for surgery. [4][5][6] Currently, the majority of patients receive palliative chemotherapy. 7 In recent years, research has focussed on molecular profiling of CCA and the development of targeted therapeutics such as mutant-IDH and FGFR inhibitors. These targeted approaches have shown promising results in clinical studies but their application is limited to a small proportion of patients with suitable genetic profiles. [8][9][10][11][12][13][14] Immune-directed therapies offer novel and widely applicable potential treatments for CCA, yet clinical trials are in their infancy 15,16 and to date have had variable success. 4,7 Importantly, results of the recently published TOPAZ-1 trial, 17 which evaluated the efficacy of Durvalumab (an anti-PD-L1 agent) and chemotherapy vs chemotherapy alone in advanced biliary tract cancers, demonstrated a promising improvement in survival for patients receiving the anti-PD-L1 antibody. Suggesting that immunotherapeutic approaches still hold promise for the disease.
CCA, therefore represents a cancer with a patient group who would substantially benefit from the identification of novel therapeutic strategies including immunomodulatory approaches.
Dickkopf-1 (DKK1) is a secreted WNT signalling modulator which has been shown to be highly expressed in around a third of iCCAs and is associated with worse prognosis. 18 Interestingly, iCCA is a cancer in which the canonical WNT signalling pathway is activated, 19 suggesting that DKK1 may not be fulfilling its classical role of preventing WNT receptor activation in these tumours. 20 Alternative DKK1 functions have been described 21,22 and increasing evidence suggests an immunological role for DKK1 in cancer. 3,[21][22][23][24] Notably, DKK1 drives the recruitment of myeloid-derived suppressor cells to the tumour microenvironment 25 and reduces activation of natural killer (NK) cells, 26 implicating DKK1 in the modulation of an immunosuppressive tumour microenvironment.
Secreted DKK1 can be specifically targeted with DKN-01 a neutralizing antibody that has been shown to reduce cancer growth in models of both melanoma and prostate cancer in an NK cell-dependent manner. 24,26 DKN-01 is being investigated as a monotherapy or combination therapeutic for various malignancies 27 and a retrospective analysis suggested that patients with elevated tumoral expression of DKK1 were the more likely to derive clinical benefit from a DKN-01 anti-PD-1 combination therapy. 28 As such, DKK1 tumoral expression is currently being investigated prospectively as a patient stratification strategy as part of a phase 2 clinical study (NCT04363801). In biliary tract cancers (including iCCA) DKN-01 used in combination with standard of care chemotherapy (Gemcitabine/Cisplatin) has been shown to be well tolerated but provides no improvement over Gemcitabine/Cisplatin alone (NCT02375880). 29 Circulating biomarker analysis in this study suggested that DKN-01 might be immune modulatory, transiently increasing inflammatory cytokines IFNγ, IL6 and IL8; however, DKK1 tumoral expression data were available for only a limited number of patients in this study so the potential benefit of stratification is unclear.
Further understanding of the mechanisms governing DKK1's immune modulatory and tumour-promoting activity in iCCA is required to provide a rationale for patient stratification or therapy combinations in which DKK1 inhibition can be levied in the most effective way. Using in vivo models of iCCA, we define the effects of high DKK1 expression on the tumour immune microenvironment and test the efficacy of DKK1 neutralization on tumour growth.

| Animal work
All animal work was performed under the UK Home Office project licence held by Dr Luke Boulter (PFD31D3D4) or Prof Jeffrey Pollard (P9C3F6964). Animals were maintained in colonies in 12 h light-dark cycles and were allowed access to food and water ad libitum.

| Hydrodynamic tail vein injection
Female, FVB/N mice were purchased from Charles River, UK and were used at 4-6 weeks of age. Animals were injected with a  Jwp.-Porcn tmros (Csf1r-iCre;Porcn flox ) mice have been previously described. 31 Mice were bred with equivalent Cre-negative controls (WT). Gene-targeted mice display myeloid-specific loss of either Ctnnb1 or Porcn genes. Hydrodynamic injections were performed in these animals using PGK-SB13, pT3-EF1a-Nicd and pT3-myr-Akt-HA plasmids as described above.
Cells were stained for 30 min at 4°C with antibodies for immune markers CD45, CD11b, CD115, LY6C, LY6G, CD45R/B220, CD49b and CD3 (antibodies are described in detail in Table S1). Cells were sorted using a BD FACS Aria II Flow Cytometer with the addition of DAPI as a live/dead cell marker. Gating for CD115+ monocytes was performed as described in Figure S4.

| Histology and Immunohistochemistry
Livers were flushed with saline and lobes were dissected into 10% neutral buffered formalin for 24 hrs. Fixed tissue was processed into paraffin blocks. For immunohistochemistry 4 μm sections were dewaxed in xylene and rehydrated. Following antigen retrieval, samples were incubated with 3% hydrogen peroxide and blocked for avidin and biotin (Abcam, ab64212) followed by a pan-species protein block (Abcam, ab64226). Primary antibodies were incubated overnight and detected using species-specific biotinylated secondary antibody and HRP-DAB detection. Slides were either counterstained with Harris haematoxylin in the case of DAB staining or stained in haematoxylin and eosin (H&E) for assessing tissue histology. Antibodies are listed in Table S1.

| Tissue microarray
Slides containing samples from tissue microarray (TMA) LVC1261 (Pantomics) were used for immunohistochemical analysis following the standard IHC protocol described above. The TMA contained 126 cores comprised of two tumours cores and one normal adjacent tissue core from 42 patients with iCCA.

| Tumour burden and immune quantification in tissue sections
Tissue slides were scanned using the Nanozoomer slide scanner

| Reverse transcription and quantitative PCR
Reverse transcription of 1 μg of RNA was performed using Quantitect Reverse Transcription Kit (Qiagen) per the manufacturer's instructions. cDNA was diluted 1:10 and qPCR was performed using the

Roche Lightcycler 480 II instrument and Lightcycler 480 Sybr Green
Master Mix (Roche) following the manufacturer's instructions with target specific primers used at 10 μM. A full list of primers used can be found in Table S2.

| Nanostring gene expression analysis
RNA was isolated as described above. RNA quality was assessed using the Agilent 2100 Bioanalyzer. Nanostring gene expression analysis was performed using the PanCancer IO 360 panel. Data were quality controlled and analysed using the nCounter Advanced Analysis 2.0 software. Gene lists used for Nanostring pathway analysis are shown in Table S3.

| Statistics
All statistical analysis was performed in Graphpad Prism 9 unless otherwise stated. In the case of normally distributed data (determined by Shapiro-Wilk testing) unpaired Student's t-test was used.
With non-normally distributed data, the non-parametric Mann-Whitney test was applied. Pearson's Rank test and associated p-values were calculated in R.

| DKK1 overexpression modulates chemokine and cytokine signalling in intrahepatic cholangiocarcinoma
Previous human studies have shown that high DKK1 expression correlates to significantly poorer survival in patients with iCCA. 18 To investigate the role of high DKK1 expression on tumour progression, we modified a hydrodynamic tail vein injection model (HTVI) to overexpress DKK1 in tumour cells. HTVI utilizes high-volume injections of naked DNA plasmids and the SB13 retrotransposase to promote DNA integration into hepatocytes. 32 Here, we drive the formation of iCCA by expression of constitutively active myristoylated Akt (myrAkt) and the Notch intracellular domain (Nicd) as previously described. 33,34 Tumour-specific overexpression of DKK1 was achieved by co-injecting a plasmid expressing both HA-tagged DKK1 and RFP ( Figure 1A) 35  This analysis identified alterations in immune modulatory pathways as a key differential between DKK1 overexpressing and control t umours in this model.

| DKK1 recruits immune suppressive myeloid cells to the tumours
Due to the association of DKK1 with both immune modulation in our in vivo model and worse outcomes in patients with iCCA, 18,[24][25][26] we sought to explore whether DKK1 expression promotes immune

Sb13/Nicd/Akt
Control  ** macrophages may be naïve or poorly polarized. To better understand the nature of these cells we performed gene expression analysis using markers previously shown to be specifically upregulated in MHCII low immunosuppressive TAM2 macrophages. 36 These genes demonstrated an overall increase when DKK1 is overexpressed in our models ( Figure 2E and F) compared to control tumours. These genes include markers of alternative macrophage polarization (Arg1, Cd163), monocyte chemoattractant Ccl6, and ligands for CCR2 (Ccl2, Ccl7, Ccl12) and CCR5/1 (Ccl3, CCl4, Ccl9) suggesting that F4/80+ cells in these tumours represent an increase in TAM2-like myeloid cells. Interestingly, we were able to demonstrate that this is not a consequence of DKK1mediated WNT signalling inhibition in myeloid cells themselves.
Using transgenic mouse models for the myeloid-specific deletion of β-catenin (Csf1r-iCre/Ctnnb1 flox/flox ) required for canonical WNT signalling or Porcupine (an O-acetyltransferase required for WNT ligand processing and therefore production) (Csf1r-iCre/Porcn flox/flox ) we induced iCCA formation over 6 weeks using hydrodynamic injection of Nicd/Akt ( Figure S4). Using these mouse models, we were able to validate effective loss of Porcn or Ctnnb1 in monocytic cells, and this did not affect the circulating numbers of CD115+ monocytes in the blood ( Figure S4A,B) of either Csf1r-iCre/Ctnnb1 flox/flox or Csf1r-iCre/ Porcn flox/flox mice. Furthermore, myeloid cell ablation of either Porcn or Ctnnb1 had no effect on tumour formation in the Nicd/Akt iCCA model ( Figure S4C). Consistent with these data on tumour growth, loss of Porcn or Ctnnb1 did not result in a change in the recruitment of F4/80positive macrophages to tumour regions ( Figure S4D) nor of FOXP3+ regulatory T recruitment ( Figure S4E). These data collectively show that neither loss of macrophage WNT production nor reception are sufficient to recapitulate immune phenotypes produced by DKK1 overexpression.

| DKK1 overexpression promotes a tolerogenic immune microenvironment
Having established a DKK1-driven increase in TAM2-like macrophages in our hydrodynamic models, we next sought to explore whether DKK1 expression can promote tolerogenicity in the iCCA tumour microenvironment. We looked at the number of tumour associated FOXP3+ regulatory T cells (T reg ) located within DKK1 overexpressing and control tumours ( Figure 3A-D). T reg are highly immunosuppressive and their abundance has been associated with worse prognosis in a number of cancers. [37][38][39] When DKK1 was overexpressed in our Nicd/Akt model, we found a significant increase in FOXP3+ cells within tumour boundaries (p = .0076).
On average 2.07% of cells were FOXP3 positive when DKK1 was highly expressed (n = 130), compared to 0.933% in control tumours (n = 57) ( Figure 3B). A similar increase in FOXP3+ cells was found in our Kras G12D /gTrp53 model ( Figure 2C and D). When fold change = 9.49 and p = .0107, fold change = 16.41, respectively, Figure 3E). Finally, the fact that DKK1 is inducing a tolerogenic immune microenvironment in these tumours was supported by Nanostring gene expression data, which confirmed that Nicd/Akt/DKK1 tumours have reduced gene signatures associated with antigen presentation and costimulatory signalling compared to Nicd/Akt alone (p = .0014 and p = .0093 respectively) ( Figure 3F).
To support evidence from our in vivo models, we assessed whether DKK1 might recruit FOXP3 regulatory T cells in human iCCA by analysing transcriptomic Illumina beadchip array data from 104 cholangiocarcinomas (GSE26566) (Figure 4A-C). These data were interrogated for expression of T reg transcriptional signatures. 40,41 When stratified by mean DKK1 expression (DKK1 high n = 23, DKK1 low n = 81) ( Figure 4B), genes associated with T reg were significantly higher in DKK1-high patients when compared to the DKK1-low group

| Anti-DKK1 therapeutic mDKN-01 reduces tumour burden in pre-clinical models of iCCA
Having established a role for DKK1 in promoting tumour immune modulation and defined the relationship between DKK1 levels and FOXP3+ cells in a cohort of human samples, we next wanted to test whether DKK1 inhibition was effective at reducing iCCA growth.
Vehicle in the biliary epithelium (by Keratin19-CreERT, herein referred to as the KPP model) following the administration of Tamoxifen ( Figure 6A). 30 When combined with thioacetamide (TAA) induced liver damage, mice develop diffusely distributed, well-differentiated iCCA within 8 weeks. KPP mice were dosed with mDKN-01 or vehicle control twice weekly for the final 2 weeks before the end of the experiment ( Figure 6A). We found that mDKN-01 was able to reduce tumour burden in the liver from a mean of 24.5% in the vehicle control (n = 8) down to 4.7% in mDKN-01-treated mice (n = 9) (p = .0037) ( Figure 6B). (Representative images of tumour burden is shown by CK19 immunohistochemistry, Figure 6C). This suggests that the inhibition of DKK1 could be an effective strategy at reducing tumour growth in a more physiologically and clinically relevant model with an active inflammatory driver, and without the artificial overexpression of DKK1.

| DISCUSS ION
DKK1 has been shown to be overexpressed and associated with worse outcomes in a number of cancers 42 18,52 This suggests that the association between DKK1 levels and Treg abundance is unlikely to be due to a role for DKK1 expression in a common underlying inflammatory pathology, and that mechanisms of DKK1-driven immune suppression may be relevant to a subset of DKK1 high patient iCCAs.
Finally, DKK1 appears to drive an immunosuppressive microenvironment associated with intrinsic or acquired PD-1 resistance

CO N FLI C T O F I NTE R E S T
This study was partially funded by Leap Therapeutics. M.K. and W.N. are employees and stockholders and/or stock option holders of Leap Therapeutics Inc. JWP is a founder, shareholder in and on the board of Macomics Ltd an immuno-oncology company. These studies, however, do not conflict with those of the company. All other authors have no further conflicts to disclose.

PATI ENT CO N S ENT S TATEM ENT
All patient data used in this study was fully anonymised.