The heterogeneous distribution of extracellular adenosine reveals a myeloid-dependent axis, 1 shaping the immunosuppressive microenvironment in pancreatic ductal adenocarcinoma

56 The prognosis for patients with pancreatic ductal adenocarcinoma (PDAC) remains extremely 57 poor. It has been suggested that the adenosine pathway contributes to the ability of PDAC to 58 evade the immune system and its resistance to immunotherapies (Immuno-Oncology Therapy, 59 IOT), by generating extracellular adenosine (eAdo). 60 Using syngeneic genetically engineered mouse allograft models of PDAC with differential 61 immune infiltration and response to IOT, we showed enrichment of the adenosine pathway in 62 tumour-infiltrating immune cells (in particular, myeloid populations). Extracellular adenosine 63 distribution is heterogeneous in tumours, with high concentrations in hypoxic margins that 64 surround necrotic areas, associated with a rich myeloid infiltration. Pro-tumorigenic M2 65 macrophages express high levels of the Adora2a receptor; particularly in the IOT resistant 66 model. Blocking the in vivo formation and function of eAdo (Adoi), using a combination of 67 anti-CD73 antibody and an Adora2a inhibitor slowed tumour growth and reduced metastatic 68 burden. In addition, blocking the adenosine pathway improved the efficacy of combinations 69 of cytotoxic agents or immunotherapy. Finally, Adoi remodelled the tumour microenvironment 70 (TME), as evidenced by reduced infiltration of M2 macrophages and Tregs. RNAseq analysis 71 showed that genes related to immune modulation, hypoxia and tumour stroma were 72 downregulated following Adoi and a specific adenosine signature derived from this is 73 associated with a poorer prognosis in PDAC patients. The formation of eAdo appears to promote the development of the immunosuppressive TME 75 in PDAC, contributing to its resistance to conventional and novel therapies. Therefore, inhibition of the adenosine pathway may represent a strategy to modulate the stroma and 77 improve therapy response in patients with PDAC.


81
Survival for patients with pancreatic ductal adenocarcinoma (PDAC) has not changed 82 significantly in the last 50 years and remains poor (https://www.cancerresearchuk.org/health-83 professional/cancer-statistics-for-the-uk). There is a need for new treatments, given that 84 current standard of care for patients with metastatic disease is associated with low survival, 85 with less than 10% living more than 2 years (1). In addition to relative resistance to 86 conventional therapies, cancer immunotherapy (Immuno-Oncology Therapy, IOT) is also 87 ineffective in this disease, except in the small group of patients (1-2%) with microsatellite 88 instability/mismatch repair deficient (MSI-H/dMMR) tumours (2). Several authors consider 89 that the reason for this resistance can be ascribed to the low mutational burden of this neoplasm, 90 which leads to lymphocyte exclusion and anergy (3,4). However, the tumour 91 microenvironment in PDAC has been shown to be populated by a rich variety of immune cells, 92 but most of them bear strong immune suppressive features, which contribute to the resistance expressed on infiltrating macrophages (13,14). The adenosine signature recently published by 106 Sidders and colleagues (15) shows that this pathway correlates with resistance to 107 immunotherapies and is associated with other genetic features of tumour aggressiveness, such  Some recent publications link the adenosine pathway to the biology and aggressiveness 127 of PDAC, which has been shown to have an increased adenosine pathway RNA signature 128 resulting in a worse prognosis (15), and receptors for eAdo as well as CD73 have been found pancreatic tissue (33). Further, pancreatic cancer cells are known to be strongly CD73 positive 131 (34). However, little is known about the complex mechanism generated by the adenosine 132 pathway resulting in the immunosuppressive characteristics of pancreatic cancer 133 microenvironment and stroma, in particular the role that the adenosine pathway has in shaping 134 the immune infiltration of this disease. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022 5% FBS (Gibco, #10270106). All the cell lines were analysed for STR fingerprinting and 155 mycoplasma testing routinely. Editing Core team. Tumours were detected by palpation followed by ultrasound imaging by 164 the Genome Editing Core. Tissues from KPC mice were provided once tumour dimensions or 165 health status rendered them unsuitable for therapeutic studies. KPC mice were killed when 166 showing clinical signs of the disease. 167 Tumour allograft studies were performed with technical assistance from CRUK-CI BRU staff. 168 Mice were subcutaneously injected in the right flank with 1x10 6 KPCY-derived cells in 50% 169 PBS and 50% Matrigel basement membrane matrix (#354234, Corning). In the interventional 170 experiments, mice were treated as indicated, starting 12-14 days from tumour cells 171 implantation, to allow the microenvironment to establish. Tumour volume was calculated 172 using the formula; (π/6)*(width) 2 *length. Tumour response was defined based on the % of 173 change of the longest diameter from start of therapy (stable disease < 20% increase and < 30% 174 decrease of target lesion RECIST v. 1.1). Mice were then killed at specific endpoints (e.g. 14 175 days from start of treatment) or when the tumour reached 2000 mm 3 (or before in case of 176 appearance of clinical signs). When indicated the following drugs were used: AZD6738 177 (ATRi; 25 mg/kgdaily for 4 days), AZD4635 (Adora2ai; 50 mg/kg bid), 2c5mIgG1 (anti-CD73; 10 mg/kg twice weekly), AB740080 D265A (anti-PD-L1: 10 mg/kg twice weekly), 179 NIP228 mouse IgG1 control kappa (isotype; 10 mg/kg twice weekly) and NIP228 muIgG1 180 D265A (isotype; 10 mg/kg twice weekly) were provided by Astrazeneca; gemcitabine 181 hydrochloride (Tocris, 3259) was used at 100 mg/kg twice weekly; inVivoPlus anti-CD40 182 (clone FGK4.5/FGK45; bioxcell BE0016-2) and InVivoPlus rat IgG2a isotype control, anti-183 trinitrophenol (clone 2A3; bioxcell BE0089) were used as a single injection of 100 µg.

202
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022 Confluence was calculated as the average of the 3 fields using the Incucyte algorithm.

203
Experiments were repeated at least 3 times.

204
Single cell suspension preparation 205 For experiments in s.c. allografts, tumours were weighed and placed in RPMI and finely 206 minced with a scissor in a 2 ml tube which was then washed with up to 2.5 ml of digestion 207 buffer (Tumour dissociation kit, Miltenyi, 130-096-730) plus Deoxyribonuclease I (300 µg/ml, 208 Sigma, DN25-1G). Dissociation was performed using the protocol suggested by Miltenyi. For 209 KPC tumours, a trypsin inhibitor (250 µg/ml, Sigma, T6522) was added to the digestion buffer.

212
Spleens, inguinal and mesenteric lymph nodes were mashed on a 100 µM filter (Grainer 213 Bio-one, 542-000) over a 50 ml tube, using a syringe plug and the filter was washed with 214 MACS buffer and centrifuged (at 4° C as for all the following centrifugation

251
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022 Tumours were snap frozen in liquid nitrogen immediately after resection and the tissues 252 were embedded in a HPMC/PVP hydrogel as previously described (36  CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022 data processing was performed in SCiLS Lab (version 2021b, Bruker Daltonik, Bremen, 276 Germany).

277
MALDI-MSI analysis was performed on a RapifleX Tissuetyper instrument (Bruker Daltonik,278 Bremen, Germany) operated in negative detection mode. 9-Aminoacridine (9-AA) prepared in 279 80:20 methanol:water was used as a MALDI matrix and spray deposited using an automated  (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is

324
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ; https://doi.org/10.1101/2022.05.24.493238 doi: bioRxiv preprint 3.3.2541.405 (Indica Labs). Cell density was calculated as the number of positive cells x mm2 325 of tumour tissue analysed. Sections of mouse spleen were used on each slide as internal control.

326
For the analysis of the lung metastatic burden of any individual mouse, the 4 right lobes and 327 the left lobe were cut in multiple pieces and together fixed and then embedded, then treated as 328 above. A p53 staining was used for helping the detection of smaller lesions (min. of 5 cells).

329
Analysis was performed using Halo software and expressed as % of metastatic areas/total lung 330 area analysed. Mice with intra-abdominal/thoracic organs direct infiltration were excluded 331 from the analysis.

333
RNA was extracted from tumour tissues weighing up to 30 mg. Tissues were firstly 334 disrupted and homogenised using TissueLyser II and then RNA was extracted using Qiagen 335 RNA kit, according to manufacturer instructions. RNA was then quantified using Qubit 3.0 336 (life technologies) and purity and quality were assessed using Agilent 4150 (G2992AA) 337 TapeStation system (Agilent). Library construction was followed by paired-end 50 bp 338 sequencing on Novaseq 6000 sequencer.   (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ; https://doi.org/10. 1101 In order to the evaluate the correlation of the adenosine-related gene expression in 373 correlation to human PDAC we analysed 712 genes which had at least a 50% decrease (Log2FC 374 <-0.58) following adenosine inhibition treatment, of which 561 had a human ortholog (suppl.  Using PDAC specific data from TCGA (40) available in https://www.cbioportal.org/, we 391 derived the z-score of these 52 genes for each patient with known disease-specific-survival 392 (DSS), Progression free survival (PFS) and disease-free-survival (DFS). The z-scores for all 393 genes were summed up for each patient and was deemed high adenosine score if >0 or low 394 adenosine score if <0, as previously shown (15).

396
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The data generated in this study are available within the article and its supplementary data files 404 or from the corresponding authors upon reasonable request. Code for differential expression  The adenosine pathway is enriched in immune cells infiltrating PDAC models 456 We hypothesized that the adenosine pathway might have a more impactful role in the 457 TME, as opposed to a cell autonomous effect. For this reason, we investigated the expression 458 of the adenosine pathway components on tumour-infiltrating immune cells which represent 459 significant proportion cells seen in PDAC lesions. We showed a highly significant enrichment 460 in both the IOT-resistant and IOT-responsive models for CD39 + CD73 + double-expressing 461 immune cells, when compared to secondary lymphoid organs (spleen and nodes). In particular, 462 the majority (65-91% in tumour vs 36-56% in spleen) of tumour-infiltrating CD11b + myeloid 463 cells express the two receptors, due to an increase in expression of CD73, given that those cells 464 are normally CD39 + ( fig. 2A-B). Similar results were shown for Tregs and CD8 + T-cells, 465 which are normally CD73 + and displayed an increase in expression of CD39 in tumour, 466 compared to the secondary lymphoid organs counterparts (suppl. fig. 4A; p<0.05). There was 467 no significant difference in these findings when comparing the two models, despite their 468 differential response to IOT. We then confirmed these findings in KPC autochthonous tumours 469 . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is  Having shown in these models that in the PDAC microenvironment, immunosuppressive 501 adenosine is present abundantly, we then investigated which cells within the microenvironment 502 might be responsive to this. We investigated the expression of the adenosine A2a receptor 503 (Adora2a, the receptor with the highest affinity for adenosine) that has been found frequently 504 overexpressed in human tumours. We found that Adora2a was highly expressed by tumour-505 infiltrating myeloid population when compared to the spleen (suppl. fig. 3D-E) and this  fig. 4F).

519
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ; https://doi.org/10. 1101 In addition to pro-tumorigenic macrophages, Adora2a expression was found enriched in 520 other myeloid immune populations infiltrating the tumours. In particular CD11bdendritic 521 cells, CD11b + dendritic cells (suppl. fig. 4G), M1 macrophages (suppl. fig. 4H) and mo-  The anti-CD73 was extremely effective in reducing the expression of CD73 on the surface of 539 all live cells (suppl. fig. 5A). Again, MSI data showed that the treatment was very effective in 540 reducing adenosine formation in the TME (Fig. 4 B-C). In particular, adenosine was 541 completely abolished in the viable tumour areas, while a small amount remained in the necrotic 542 margins, although massively decreased ( fig. 4C), highlighting the importance to inhibit not 543 only the formation of adenosine through CD73 inhibition, but also targeting adenosine 544 . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ; https://doi.org/10.1101/2022.05.24.493238 doi: bioRxiv preprint receptors. The efficiency of the treatment on the extracellular purinergic pathway was also 545 supported by the decrease of molecules downstream adenosine (adenine and inosine in viable 546 tumour and necrotic margin areas), and the increase of upstream and alternative pathway 547 molecules as AMP (in the necrotic margin) and xanthine (in both viable tumour and necrotic 548 margin) respectively. There was no change in the distribution of ATP, ADP and hypoxanthine 549 (suppl. fig 5B).

550
The treatment led to a significant reduction of growth rate and tumour weight ( fig 4D and   551 suppl. fig. 6A). The adenosine pathway has been shown to control the metastatic process and The TME of PDAC is an intricated structure that relies on the presence of multiple non-616 malignant cells. This TME is well recapitulated in pre-clinical models of PDAC (35). In order 617 to investigate the broader effect of Adoi in 6419c5 PDAC model, given the effect we have seen 618 . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022 in the infiltrating immune suppressive populations, we performed a bulk RNAseq analysis of 619 the 6419c5 model treated with Adoi or control (5 vs 6 mice, see methods). Following the 14-620 day treatment with Adora2a inhibitor and anti-CD73 a total of 712 genes had at least a 50% 621 decrease (logfold <-0.58; fig. 6A). KEGG and GO Biological process pathway analysis  . 6D), that has been originally associated with a poorer prognosis when compared to others.

643
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022 The score remained significantly higher even when comparing the squamous subtypes with the 644 grouped non-squamous ( fig. 6E).

645
In order to create a PDAC-specific adenosine signature and to evaluate its performance in 646 PDAC-specific outcomes, we created a signature starting from the 561 human ortholog genes 647 with a 50% decreased following Adoi treatment. Of these, 52 genes were selected for the 648 signature ( fig. 6F), that according to our RNAseq dataset were clearly associated with the 649 pathway areas dependent on adenosine (hypoxia response, immunity, tumour stroma), were  Notably, the presence of a high adenosine signature is also associated with shorter disease-free-661 survival (mDFS high ado 23.54 vs low Ado 49.68, p=ns; suppl. fig. 7L).

662
Overall these data highlight the role of the innate immune system in shaping a pro-tumorigenic, 663 immune suppressive microenvironment in PDAC, dependent on the formation of extracellular 664 adenosine in the context of a hypoxic milieu. We can speculate that this unfavourable 665 environment may create the condition for a more aggressive PDAC phenotype which would 666 then translate in the ability to escape the immune system, resist to cytotoxic treatment and 667 easily metastasize.

668
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022.  The extracellular adenosine pathway has also been shown to influence the TME 691 fostering the immune suppression provided by some innate immune subpopulations (as 692 . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ; https://doi.org/10.1101/2022.05.24.493238 doi: bioRxiv preprint myeloid and NKs) and inhibiting the function of the adaptive immune system, in particular, T-693 cells (6,33,44). By activating its receptors, adenosine is able to increase the intracellular 694 concentration of cAMP which leads to the induction of a M2 phenotype of macrophages and 695 block the secretion of IL1β increasing the release of CXCL1, IL-6, IL-10 and IL-8 among 696 others from myeloid population which are known to orchestrate immune exclusion (8,54). 697 eAdo also favours the formation and maintenance of Treg cells (8), which are known to favour 698 cancer progression and IOT resistance .

699
Emerging data correlate the adenosine pathway to the promotion and aggressiveness of PDAC.

700
CD73 is also overexpressed on the surface of cancer cells of PDAC (34,55), but its role on 701 cancer cells is controversial, as is the cancer cell direct role of extracellular adenosine. A recent 702 publication, shows that genomic targeting on mouse PDAC cells of CD73 leads to a reduced 703 in vivo tumour formation and change in the circulating and infiltrating immune system (57).

704
Data from bulk RNA-sequencing have determined that CD73 along with Adora2a and Adora2b 705 are strongly overexpressed in tumours when compared to normal pancreatic tissue (33).

706
Recently an adenosine signature has been published which clearly identifies enrichment in 707 PDAC and a correlation with progression free survival as well as overall survival (15).

708
To date, little was known about the expression of the adenosine pathway in the context 709 of the innate and adaptive immune system in PDAC, how the extracellular adenosine is 710 generated and what are the targets of adenosine also in regard to its spatial distribution and 711 formation of adenosine.

712
Our results show for the first time, that the mechanism of generation of extracellular 713 adenosine in pancreatic cancer TME is finely orchestrated by tumour infiltrating myeloid cells 714 and tumour cells, due to the expression of high level of CD39 in infiltrating myeloid cells and 715 CD73 on both cell types. We have also demonstrated that the pathway can be overexpressed 716 . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022 in T-cells infiltrating the tumours, regardless of their activation status ( fig. 1,2 and suppl. fig.   717 2,4). The distribution of extracellular adenosine is spatially heterogeneous and a high level of 718 extracellular adenosine correlates with the presence of a hypoxic environment and is favoured 719 by the presence of necrosis, where the myeloid population is enriched (fig.3). Necrosis is 720 common in human PDAC and related to poor prognosis for all stages (56). The enrichment of 721 a CD39 + CD73 + double population, potentially able to independently produce adenosine, does 722 not seem to correlate with IOT-resistant or responsive tumour models, but there is a difference 723 when the target of adenosine (Adora2a receptor), is considered. Adora2a on myeloid 724 populations, in particular in pro-tumorigenic M2 macrophages but also in antigen presenting 725 cells, is differentially expressed in regard of IOT response phenotype, with the resistant 726 tumours abundantly overexpressing the receptor in these populations ( fig.3 and suppl. fig. 4).

727
The bulk RNAseq analysis of tumour treated with adenosine inhibition revealed indeed a 728 broader role for adenosine in PDAC TME ( fig. 6). Genes related to immunosuppression and (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022 ( fig. 5 and suppl. fig. 7). Notably, targeting the pathway can reduce tumour growth in an IOT- would not only need to be designed to target the generation of adenosine through CD73 765 inhibition but also adenosine specific receptors.

766
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022 In summary, we have shown that tumour-infiltrating myeloid immune cells contribute (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022    CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is      CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022 was dosed twice per week intraperitoneally at 10 mg/kg. Adora2a inhibitor (AZD4635) was 865 given by oral gavage twice daily at 50 mg/kg.

889
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted May 25, 2022. ;https://doi.org/10.1101https://doi.org/10. /2022 (J) Percentage change in the long diameter length following 14 days of treatment per group.