Phenotype screens of murine pancreatic cancer identify a Tgf-α-Ccl2-paxillin axis driving human-like neural invasion

Solid cancers like pancreatic ductal adenocarcinoma (PDAC), a type of pancreatic cancer, frequently exploit nerves for rapid dissemination. This neural invasion (NI) is an independent prognostic factor in PDAC, but insufficiently modeled in genetically engineered mouse models (GEMM) of PDAC. Here, we systematically screened for human-like NI in Europe’s largest repository of GEMM of PDAC, comprising 295 different genotypes. This phenotype screen uncovered 2 GEMMs of PDAC with human-like NI, which are both characterized by pancreas-specific overexpression of transforming growth factor α (TGF-α) and conditional depletion of p53. Mechanistically, cancer-cell-derived TGF-α upregulated CCL2 secretion from sensory neurons, which induced hyperphosphorylation of the cytoskeletal protein paxillin via CCR4 on cancer cells. This activated the cancer migration machinery and filopodia formation toward neurons. Disrupting CCR4 or paxillin activity limited NI and dampened tumor size and tumor innervation. In human PDAC, phospho-paxillin and TGF-α–expression constituted strong prognostic factors. Therefore, we believe that the TGF-α-CCL2-CCR4-p-paxillin axis is a clinically actionable target for constraining NI and tumor progression in PDAC.


Introduction 81
Pancreatic ductal adenocarcinoma (PDAC) is currently the third leading cause of cancer-associated death 82 worldwide, and is projected to become the second by 2030 (1). Remarkable progress has been made in the 83 last five years in disentangling the complex genetic and molecular drivers, and subtypes of PDAC. Oncogenic

84
Kras-driven, genetically engineered mouse models (GEMMs) of PDAC have uncovered several aspects of 85 the co-evolution of cancer lesions and tumor microenvironment during carcinogenesis (2). As faithful models, 86 they are powerful in modelling the molecular events that lead to metastasis, intra-tumoral heterogeneity, and 87 several defining features of the tumor microenvironment such as local immunosuppression (3).

88
Human PDAC also exhibits yet an unparalleled high frequency of neural invasion (NI) and neuroplastic 89 alterations in the pancreas and in the central nervous system (4-6). NI was shown to be present in up to 90 100% of classical ductal adenocarcinomas of the pancreas (7). NI in human PDAC typically manifests as 91 "perineural" invasion, which implies the circular alignment of cancer cells along the epineural sheaths (8,9) 92 ( Figure 1A). Importantly, in PDAC, the severity of NI, i.e., the penetration depth of cancer cells into the 93 intrapancreatic nerves, is an independent prognostic factor for overall and disease-free survival, as well as 94 local recurrence (10). Indeed, nerve-invading cancer cells use these as highways for rapid spread, which 95 results in massive local tumor invasion, surgical irresectability, and severe pain (4, 11-13). In human PDAC, 96 NI, starts, however, at the earliest stages of cancer, e.g. in T1 tumors (14). In fact, Schwann cells of peripheral 97 nerves were shown to emerge already around the precursor lesions of PDAC, i.e. PanINs (pancreatic 98 intraepithelial neoplasia), which suggests that it is nerves, and not cancer cells, that first migrate to initiate NI 99 (15). Neural and extrapancreatic tumor invasion toward the spinal cord have been previously reported to be 100 present in the oncogenic Kras-driven p48-Cre;LSL-Kras G12D ;p53 lox/-(KPC) GEMM of PDAC (16, 17). 8 phosphate pathway ( Figure S3). This suggests that metabolic alterations can further aggravate the human-165 like perineural invasive phenotype of murine pancreatic cancer.

166
Both genotypes were associated with extensive desmoplasia / fibrosis, more pronounced than in KPC mice.

167
The TPC and TPAC tumors displayed a ductal-like phenotype and, accordingly, strongly expressed the ductal 168 cancer cell marker cytokeratin 19 (CK-19) ( Figure S4A), yet the TPAC genotype with loss of RelA/p65 and 169 p53 had an even more ductal-like appearance ( Figure S4A&C), more desmoplasia ( Figure S4C), a lower 170 cancer cell proliferation rate ( Figure S4B) than the TPC mice. TGFa is a known driver of acinar-to-ductal 171 metaplasia (ADM). Importantly, the rate of ADM in the TPAC genotype was more pronounced both in vivo 172 and in vitro than in TPC mice (not shown). Overall, these data suggested a more ductal, fibrotic, and slowly 173 growing tumor phenotype in TPAC mice than in TPC or KPC mice. Accordingly, the overall survival of 174 analyzed TPAC mice was significantly longer than that of TPC mice. TPAC mice had a median survival of 175 370 days compared to 297 days in TPC cohort (P < 0.0001, Figure S1B). To further study the biological 176 consequences of RelA loss in TPC mouse model, 46 TPC or TPAC tumors were analyzed. A remarkably 177 high incidence of pancreatic ductal adenocarcinoma (PDAC) was observed in TPAC compared with TPC 178 mice (75% vs 9.1%). In addition, RelA deficiency in TPC model resulted in a significant decrease in 179 metastasis rate (31.8% in TPC vs 8.3% in TPAC, Figure S4D).

180
These survival times clearly surpassed the lifespan of KPC mice (median 61 days), but were comparable to 181 the lifespan of KC mice, which do not harbor a priori loss of p53 in carcinogenesis ( Figure S1A). However,

182
KC mice, despite their slower tumorigenesis, did not exhibit human-like perineural invasion. Interestingly, 183 neural hypertrophy was present in all analysed genotypes, independent of NI ( Figure S1F). To exclude the 184 presence of mutant Kras in the TPAC mouse model, we performed targeted sequencing of the Kras locus in 185 the spleen of the TPAC mice, as well as in cancer cells isolated from the TPAC mice. As predicted, the 186 tissues and cells from TPAC mice did not harbor Kras mutations and were thus Kras-wildtype (Table S3).

187
Despite the absence of the oncogenic Kras mutation, we explored whether the TPAC cancer cells also 188 exhibited elevated Ras activity. In Ras activity assays, we detected the Ras activity to be even higher in the 189 isolated TPAC cancer cells than in KPC cancer cells, which confirms the strong Ras-activating capacity of 190 TGFa hypersignalling. (Fig S4E). Altogether, these data strongly suggested that TGFa signaling, without the 9 need for mutant Kras, gives rise to slowly growing, highly fibrotic, ductal pancreatic cancer, which seem to 192 be necessary for the emergence of human-like perineural invasion in murine PDAC.

201
P3 C57Bl/6J mice and analysed cell migration through time-lapse microscopy ( Figure 1D). We did not identify 202 any differences in the velocity of migrating cancer cells of all genotypes toward neurons ( Figure S1G).

203
However, the forward migration index (FMI), which indicates directional chemotaxis, was much higher in

216
(22), were the two most prominently upregulated neurotrophic genes ( Figure 1H). In addition, TPAC cancer 217 cells overexpressed Fgf9, which promotes proliferation of neuronal precursors (23), and Nrg1, which has 218 10 been shown to drive proliferation and/or induce myelin differentiation (24); and Nrg4, which is involved in the 219 establishment of early sensory innervation in the skin (25) ( Figure 1H).

220
Another recently discovered aspect of nerve-cancer interactions is the appearance of Schwann  Figure 1J).

233
To further explore the temporal relationship between TGFa overexpression and NI, we also compared the  Figure S4F).

11
To confirm the relevance of cancer-derived TGFa in human PDAC, we analysed three different publicly 247 available single cell RNA sequencing datasets of human PDAC and searched for the prime source of TGFa 248 in the human PDAC tissue. In line with our expectations, we detected cancer cells as the main source of 249 TGFa in human PDAC ( Figure S5), expressing 19.7 times higher amounts of TGFa when compared to other 250 cells in the tumor microenvironment, which had negligible amounts of TGFa expression (Fig S5).

251
In the next step, we analyzed the transcriptome of pancreatic cancer cells from KPC and TPAC mutants 252 using the Affymetrix Mouse Gene ST1.0 array with subsequent bioinformatics analysis at the GSEA platform.

265
Dcn and Ccdc80 genes, when we compared TPC with KPC cancer cells ( Figure S6A). In addition, the 266 expression of genes that support neuritogenesis, namely Npy and Fos, was also increased in TPC as 267 compared to KPC derived cancer cells ( Figure S6B). We also found an enrichment of genes encoding the 268 extracellular matrix, including ECM glycoproteins, collagens and proteoglycans involved in biological 269 oxidation, nuclear receptors, and proteins encoding drug metabolism via cytochrome P450 and retinoic acid 270 signaling ( Figure 2B). Overall, these analyses showed that the

13
In addition, we sorted out murine neurons out of DRG of TPAC mice and other cell types (cancer cells, 301 myeloid cells, lymphoid cells) from the primary tumors of TPAC mice and KPC and cross-compared the 302 expression levels of CCL2 ( Figure S7E). In line with above, we again detected the highest expression in the 303 DRG neurons, which was much higher in TPAC-derived DRG neurons (610.9±157.8% of KPC-derived 304 neurons, p=0.03), as compared to KPC-derived DRG neurons ( Figure S7E).

305
We also quantified the mRNA content of the Ccl2 gene in granulocytes, monocytes, endothelial cells,

322
Cre;LSL-KrasG12D; 100±15.8%), which might explain the receptivity of TPAC-derived DRG fro TGFa from 323 cancer cells (Fig S9A-B). We complemented these analyses on murine DRG with the spatial transcriptome 324 analysis of nerves within human pancreatic cancer specimens to explore whether we can detect the EGFR 325 also in human nerves at the periphery that would bind TGFa (Fig S9C-D). In the Nanostring GeoMx® DSP 326 spatial transcriptome-based analysis of EGFR expression, we found that the levels of EGFR expression in 327 nerves with versus no neural invasion did not vary, but EGFR was detectable in the nerves in all analysed 328 14 human PDAC specimens (Fig S9C-D). We confirmed the expression of EGFR in FACS-sorted DRG of TPAC,

329
KPC and wildtype mice (Fig S9E), in nerves of human PDAC specimens ( Fig S9F) and in the DRG of TPAC 330 mice ( Fig S9G)

344
To elucidate the cytoskeletal adaptation in cancer cells that are confronted with neurons, we investigated the 345 morphology of filopodia and lamellipodia of cancer cells in the 3D migration assays with neurons. We 346 quantified filopodia formation during migration toward DRGs using the FiloQuant® software ( Figure 3A, B).

347
Interestingly, cancer cells at the migration front showed a significant increase in filopodia numbers and length

367
To test a functional effect of the CCL2-CCR4 pathway on migration, we pre-treated cancer cells with rCCL2 368 or C021 for 15 minutes and used them in the migration assay in the presence of DRG neurons ( Figure 3J).

369
Vehicle pre-treated cancer cells served as a negative control. Here, treatment with rCCL2 increased cancer   pan-CK (for labeling cancer cells) and p-paxillin. We found that cancer cells located close to nerves (within 382 200µm diameter) expressed more p-paxillin (15.57% of total tissue area) than cancer cells located in the 383 pancreas distant from nerves (4.23% of total tissue area) (Figure4A, B). As NI is a strong and independent 384 prognostic factor for overall survival in PDAC (39), we then analyzed whether p-paxillin content also relates 385 to patient survival. First, we divided the patient samples into low p-paxillin (n=31) and high p-paxillin (n=23) 386 groups based on a cut-off value of 3.77%, which corresponded to the median of the p-paxilin/paxillin ratio

392
4E). We also found that CCL2 immunoreactivity was increased in the high p-paxillin expression group ( Figure   393 4F), suggesting that CCL2 overexpression is indeed linked to paxillin phosphorylation in human PDAC tissue.

394
Analyzing the clinical parameters, we found that the high p-paxillin group had a higher percentage of patients 395 with large tumors (T4>6 cm) than the low p-paxillin group (chi-square test: p=0.01, Figure S10A, C). In 396 addition, the percentage of moderately differentiated tumors (G2: 76%) was increased, whereas differentiated 397 G1 tumors were not detected in the high p-paxillin group compared with the low p-paxillin group (G2: 63%, 398 G1: 9%) (chi-square test: p=0.006, Figure S10A, C). We also observed a marked, but not significant, increase 399 in the proportion of affected lymph nodes with tumour cells ( Figure 4G). These data point out that the tumor 400 size and higher tumor grading were indeed more prevalent in the high-paxillin group, which suggest that the 401 survival difference of high vs low-paxillin groups might be indirect due to other clinico-pathological 402 parameters. Overall, these results also suggested that p-paxillin is a prognostic factor and may predict poor 403 prognosis and aggressive tumor biology in human PDAC.

405
To understand the role of the CCL2-CCR4 axis in an organismal context, we modulated CCL2-CCR4 406 signalling in vivo by intraperitoneal administration of recombinant CCL2 protein (50ug/kg i.p.) or of the CCR4 407 17 receptor antagonist C021 (1mg/kg), starting with KPC mice that exhibit hyperinnervation ( Figure S1F), yet no 408 perineural invasion ( Figure 5A). No metastases were detected in the brain, lung, heart, liver, kidney, jejunum, 409 and colon in the studied groups ( Figure S11). Here, the amounts of nerves as detected through intratumoral   In a next step, we investigated whether inhibition of the paxillin-Src-Erk signalosome has effects on cancer 436 cell migration toward neurons and tumor severity. We used the small molecule inhibitor of paxillin protein 437 disruptor 6-B345TTQ, which inhibits the binding of paxillin to integrin4-alpha and regulates cell migration (40-438 42). As expected, we found a significant decrease in paxillin phosphorylation at Y118 in SU.86.86 and T3M4 439 cells treated with 6-B345TTQ by Western blots (Figure S12A, B). Ex vivo, we found that cells pre-treated with

446
Next, we tested the effects of paxillin inhibition with 6-B345TTQ on pancreatic cancer progression in vivo. To 447 this end, we treated 5-month-old TPAC animals with 6-B345TTQ (1mg/kg) 5 days per week for 4 weeks and 448 analyzed the pancreas for tumor growth ( Figure 6C). All animals developed tumors with comparable 449 pancreatic weights in both groups ( Figure S13A). Remarkably, the tumor innervation as measured through 450 the PGP9.5 content was decreased in the tumors of TPAC mice treated with 6-B345TTQ (mean 0.31% of 451 total area) compared with control TPAC mice treated with DMSO (mean 0.64% of total area) ( Figure 6D, E).

452
Moreover, a reduction in p-paxillin was detected in the pancreas of the treated TPAC mice (mean 4.72% of 453 total area) compared with the DMSO controls (mean 9.79% of total area) ( Figure 6D, F). These results 454 strongly suggested that blockade of paxillin phosphorylation leads to a rapid decrease in tumor innervation.

455
In addition, we tested the activity of highly proliferating preselected in vitro clones from TPAC tumors by 456 implanting them orthotopically into the pancreas of 8-10 weeks old wild type mice (129xC57BL/6) ( Figure   457 6G). We treated transplanted mice for 4 weeks with 6-B345TTQ (1mg/kg) by applying it for 5 days weekly 458 and treated the control animals with DMSO ( Figure 6G). We did not detect any changes in body weight of 459 recipient animals transplanted with TPAC cancer cells ( Figure S13B). Interestingly, pancreatic weight of 460 recipients implanted with TPAC cells decreased after treatment with 6B345TTQ (0.138 g) compared with 461 19 DMSO (0.168 g) ( Figure S13C) due to strong reduction of the primary tumor area in mice treated with 462 6B345TTQ (0.95% of total area) compared with DMSO-treated controls (3.43% of total area) ( Figure 6H, I).

463
Furthermore, we identified a strong reduction of paxillin phosphorylation in cancer cells of recipients treated 464 with 6B345TTQ (4.72% of total area) compared with control mice treated with DMSO (9.79% of total area) 465 ( Figure 6H, J). These results confirmed paxillin phosphorylation as a key actor in tumor progression and a 466 promising target for therapeutic approaches.

468
Finally, we interrogated whether TGFa, which is the prime driver of carcinogenesis in the neuro-invasive