Interleukin-6 trans-signaling is a candidate mechanism to drive progression of human DCCs during periods of clinical latency

Although thousands of breast cancer cells disseminate and home to bone marrow until primary surgery, usually less than a handful will succeed in establishing manifest metastases months to years later. To identify signals that support survival or outgrowth in patients, we profiled rare bone marrow-derived disseminated cancer cells (DCCs) long before manifestation of metastasis and identified IL6/PI3K-signaling as candidate pathway for DCC activation. Surprisingly, and similar to mammary epithelial cells, DCCs lacked membranous IL6 receptor expression and mechanistic dissection revealed IL6 trans-signaling to regulate a stem-like state of mammary epithelial cells via gp130. Responsiveness to IL6 trans-signals was found to be niche-dependent as bone marrow stromal and endosteal cells down-regulated gp130 in premalignant mammary epithelial cells as opposed to vascular niche cells. PIK3CA activation rendered cells independent from IL6 trans-signaling. Consistent with a bottleneck function of microenvironmental DCC control, we found PIK3CA mutations highly associated with late-stage metastatic cells while being extremely rare in early DCCs. Our data suggest that the initial steps of metastasis formation are often not cancer cell-autonomous, but also depend on microenvironmental signals.


Introduction 49
In breast cancer, dissemination to distant sites precedes the clinical manifestation of 50 metastasis by six to eight years in median, ranging from less than one year to more 51 than 40 years 1-3 . These clinical data derived from breast cancer growth kinetics and 52 imaging studies are strongly supported by recent experimental evidence. Whereas 53 dissemination from the primary site occurs preferentially in early tumor stages 4-6 , 54 specific mechanisms reduce cancer cell dissemination in anatomically and molecularly 55 advanced stages 5 . Furthermore, analysis of cancer growth kinetics suggests that not 56 only cancer cells disseminate early, but also that micro-metastatic colony formation is 57 initiated early. However, its manifestation may take considerable time 1,3,7 . Such data 58 are consistent with the observation that early DCCs often lack critical genetic and 59 genomic alterations, which they need to acquire at the distant site in breast and other 60 cancers. This process could explain the much longer clinical latency periods observed 61 in humans as compared to mouse models 5,8,9 and be particularly relevant for cancers 62 displaying late relapses such as hormone receptor positive breast cancer. However, 63 early dissemination and prolonged clinical latency at distant sites raise questions about 64 the identity and nature of signals conferring survival, genomic progression and 65 outgrowth of DCCs over extended periods of time. 66 DCCs are extremely rare. They are detected at very low frequencies (1-2 DCCs 67 per 10 6 BM cells 10, 11 ) in bone marrow (BM) of about 30% of breast cancer patients 68 with no evidence of manifest metastasis. Besides genomic studies, the assessment of 69 the DCC phenotype has been limited to testing for selected antigens 12 and to 70 anecdotal transcriptomic studies 13, 14 . Moreover, spontaneous or transgenic mouse 71 models, such as the PyMT-or Her2-driven models, do not generate bone metastases. 72 Hence, there is no in vivo model available to study the spontaneous progression and 73 genomic evolution from early BM infiltration to manifestation of bone metastasis. To 74 unravel mechanisms operative during clinical latency periods, we interrogated 75 transcript-derived pathway information from DCCs isolated from BM of breast cancer 76 patients. Since early breast cancer DCCs often display close-to-normal genomes 5, 15 , 77 we used mammary epithelial cells isolated from reduction mammoplasties and 78 available immortalized pre-malignant breast cancer cell lines as cellular models for 79 functional testing of candidate mechanisms in vitro. We identified IL6 trans-signaling 80 as pathway that (i) activates normal and pre-malignant cells, (ii) induces a proliferative 81 stem/progenitor-like phenotype in mammary epithelial cells and (iii) whose activation 82 in DCCs depends on regulatory niche cells in bone marrow. These data shed light onto 83 the so far dark stage of early metastasis formation in patients. Moreover, it may inform 84 about future ways to delay or prevent metachronous metastasis in patients whose 85 breast cancer is diagnosed to be locally confined by standard clinical means. 86 87 Results 88

Early DCCs do not engraft in immunodeficient mice 89
For decades attempts to culture early DCCs, i.e. DCCs from non-metastasized M0-90 stage patients, have failed. Only anecdotal reports have been published that were not 91 reproduced since then [16][17][18] . We recently observed in melanoma that early DCCs failing 92 to generate xenografts differ genomically from DCCs that successfully engrafted, 93 indicating a causative role of genomic "maturation" for metastasis and xenograft 94 formation. Specific alterations were identified that are closely linked to colonization 95 success in mice and patients 9 . We repeated these experiments for BM-derived breast 96 and prostate cancer DCCs from early (M0) and advanced (M1) stages. Given the very 97

Identification of EpCAM-positive DCCs in bone marrow 146
In order to test whether any of these pathways was enriched in DCCs isolated from 147 BM of breast cancer patients, we aimed to isolate DCCs with confirmed malignant 148 origin 14 . We followed the reasoning that epithelial cell identity in bone marrow plus the 149 presence of genetic alterations is sufficient to claim malignant epithelial origin or 150 malignant potential of a cell. DCCs were detected by screening diagnostic BM 151 aspirates from 246 M0-stage and 18 M1-stage patients for cells that stained positively 152 for the epithelial marker EpCAM ( Supplementary Fig. 1a). Forty percent of M0-stage 153 and 72% of M1-stage patients harbored EpCAM + cells. However, EpCAM is a surface 154 marker that is not as specific for DCCs as the diagnostically used cytokeratins 14 in 155 bone marrow because it is expressed by cells from the B cell lineage (own unpublished 156 data and 23 ). Therefore, we sought to differentiate between EpCAM-positive cells of 157 breast cancer patients and non-cancer patients. 158 Copy number alterations (CNAs) are found in less than 5% of non-malignant cells with 159 a median of 1.8% [24][25][26] and are diagnostically used to differentiate normal and 160 malignant cells 27 . We therefore performed combined genome and transcriptome 161 analysis and isolated genomic DNA and mRNA from the same single cell 5,14 . Although 162 this approach fails in 10-50% 13, 14, 28 and a CNA profile cannot be obtained for every 163 cell, we found that 50% and 80% of successfully analyzed EpCAM-positive cells from 164 M0 and M1 patients, respectively, harbored CNAs (Fig. 3a, Supplementary Fig. 1a). 165 We selected these DCCs for single cell RNA-Seq analysis (M0: n=30 DCCs, 21 166 patients; M1: n=11 DCCs, 5 patients). To provide additional evidence that the aberrant 167 EpCAM-positive cells are derived from a non-hematopoietic cell lineage, we compared 168 them with autochthonous EpCAM-positive bone marrow cells. The latter were isolated 169 from patients without known malignant disease undergoing hip replacement surgery. 170 Of note, cancer patient-derived and non-cancer patient-derived EpCAM-positive cells 171 could be clearly separated using the overall gene expression as well as epithelial and 172 B-cell annotated genes -with the exception of one M0 cell, which was therefore 173 excluded from further analysis (Fig. 3b). Moreover, many cells from M0-stage breast 174 cancer patients strongly expressed the marker gene KIT (Fig. 3b), characteristic for 175 mammary luminal progenitor cells 29 (see below). Together, copy number alterations 176 and the epithelial, non-hematopoietic phenotype of cells isolated from patients with 177 breast cancer provided compelling evidence that the selected cells were true DCCs. 178

IL6 pathway activation in DCCs 179
We then tested whether any of the pathways enriched in mammary stem cells (LRCs,180 Fig. 2d) were also enriched in DCCs using pathway membership enrichment analysis. 181 We found four out of the twelve pathways to be significantly enriched in DCCs (Fig. 3c, 182 d, Supplementary Table 4) including the pathway "IL6-mediated signaling events", the 183 "TCPTP" pathway, the "VEGF-VEGFR3" and "Angiopoietin-Tie2 receptor" pathways. 184 We decided to experimentally follow-up on the pathway "IL6-mediated signaling 185 events" for several reasons: (i) IL6 signaling was previously found to be relevant for 186 stemness maintenance, i.e. mammosphere-formation of ductal breast carcinoma and 187 normal mammary gland 30 ; (ii) the TCPTP pathway, a negative regulator of IL6 188 signaling 31 , was also enriched and (iii) assessment of individual genes expressed in 189 these pathways ( Supplementary Fig. 1c

IL6 trans-signaling converts progenitor into stem-like cells 256
We noted that IL6-and HIL6-stimulated mammosphere-cultures showed an increase 257 in the relative abundance of CD44 high /CD24 low cells (Fig. 5a, b), a phenotype that has 258 been ascribed to neoplastic and non-tumorigenic mammary cells enriched in tumor-259 initiating and sphere-forming cells, respectively 33, 40 . Here, HIL6-stimulated cultures 260 displayed the highest increase ( Fig. 5a, b; one-way ANOVA/Dunnett's test ctrl. vs. IL6 261 +/-sgp130-Fc, P<0.01; ctrl. vs. HIL6, P<0.0001). The increase in CD44 high /CD24 low 262 cells was not the result of increased proliferation of any CD24/CD44 subpopulation, 263 but seemed to be caused by conversion of non-stem-like CD44 high /CD24 high/int into 264 CD44 high /CD24 low stem-like cells ( Supplementary Fig. 3a-c). To corroborate these 265 findings, we compared IL6/HIL6-induced differential gene expression in MCF 10A cells 266   HIL6-group were significantly larger than in the control group (Fig. 5h). This was not 295 caused by increased proliferation or decreased apoptosis as the percentage of Ki-67-296 positive tumor cells did not differ significantly between the groups (Fig. 5i), and 297 caspase-3 positive cells were not detected in any of the tumors. 298 299 Bone marrow niche cells regulate responsiveness to IL6 trans-signaling 300 As gp130 expression is essential for IL6 signaling, we tested whether BM stromal cells 301 modulate the ability of mammary epithelial cells to receive IL6 signals. We isolated 302 but not with HUVECs (Fig. 6b). Separation of MCF 10 A and MSCs by a transwell or 310 using MSC-conditioned medium (CM) showed gp130 cell surface down-regulation to 311 be independent from cell-cell contact ( Supplementary Fig. 4b). Moreover, down-312 regulation was not immediate but observed between 6 and 14 hours after initiation of 313 the co-culture with MSCs, OBs or MSC-conditioned medium from healthy donors or 314 breast cancer patients (Fig. 6c, Supplementary Fig. 4c ). This kinetic is consistent with 315 the known independency of gp130-internalization from ligand binding 43 and points 316 towards a transcriptional regulation of gp130 surface expression. To test this, we 317 determined gp130 gene expression levels in single cells isolated from MCF 10A/MSCs 318 and MCF-7/MSC co-cultures. Interestingly, both cell lines decreased their gp130 gene 319 expression in response to MSCs (Fig. 6d), which is consistent with transcriptional 320 regulation and demonstrates that early DCCs as well as more advanced cancer cells 321 can respond to signals from neighboring cells. 322

323
To explore the functional impact of gp130 downregulation induced by MSCs, we tested 324 MCF 10A cells pre-treated for 14 hrs with MSC-CM for their sphere-forming ability. 325 Pre-treated MCF 10A showed a significant decrease in sphere-number and an 326 increase in single, non-sphere forming cells in the presence of both, endogenously 327 produced IL6/sIL6RA or exogenously added HIL6 (Fig. 6e,  IL6/sIL6RA complexes. Therefore, in both niches pathway activation by mutation 343 would provide a selection advantage for DCCs that otherwise might depend on 344 microenvironmental signals. We consequently sought for corroborating evidence that 345 genetically variant DCCs may evade the need for IL6 trans-signaling and become 346 selected. We considered the PIK3CA pathway a strong candidate for such a selected 347 oncogenically activated pathway 49 as (i) IL6 signaling activates not only the JAK/STAT 348 pathway, but also the PI3K/AKT pathway 50 , (ii) early DCCs expressed PIK3CA as core 349 element of the four identified stemness-associated pathways ( In this study we provide evidence for a role of the niche microenvironment to enable 388 and drive the earliest stages of human metastasis formation. We identified IL6 trans-389 signaling as an activator of stem-like and progenitor traits underlying epithelial colony 390 formation, a mechanism that is characterized by dependence on both IL6 and sIL6RA, 391 in contrast to IL6 alone, when a cell is equipped with mIL6RA. Our finding questions 392 the concept of fully-malignant and autonomous cancer cells as founders of metastasis. 393 However, during subsequent malignant evolution DCCs may evade 394 microenvironmental control by acquiring IL6-independence. Our data indicate that this 395 could occur via mutational activation of the PI3K pathway. 396 Several observations characterize early metastatic bone marrow colonization. 397 First, breast cancer dissemination in humans often starts from lesions often measuring 398 less than 1-4 mm in diameter 2, 5 . Second, initially DCCs often do not display the typical 399 karyotypic changes of breast cancer tumors or metastases, however are detected long 400 before manifestation of clinical metastasis 5,8,15 . Third, metastasis formation in BM 401 usually takes years to decades 53 indicating that evolutionary mechanisms and slow 402 growth kinetics dictate the pace. This framework precludes the use of transgenic 403 mouse models that are too short-lived to mimic the process in patients and rarely form 404 bone metastases. The same applies to in vitro and in vivo studies involving commonly 405 used and genomically highly aberrant breast cancer cell lines derived from manifest 406 metastases or primary tumors. All of these do not represent the biology under 407 investigation. 408 We therefore aimed at expanding early DCCs in immune-deficient mice, but 409 failed. Reasons for this can be manifold: DCCs are extremely rare and we estimate 410 that often less than 10 cells/mouse were injected. Also, species barriers may preclude 411 engraftment of cancer cells that may critically depend on certain microenvironmental 412 signals, among them the species specific IL6 signaling 54 . Our failure is also consistent 413 with data from melanoma where we had identified that engraftment of DCCs requires 414 activation of specific oncogenic pathways 55 -which in case of PIK3CA mutations early 415 (M0-stage) breast cancer DCCs commonly lack. Consistently, M1-stage DCCs formed 416 xenografts in two out of four cases. We therefore based our approach on two starting 417 points: First, we analyzed transcriptomic data of early human breast cancer DCCs to 418 identify candidate pathways and second, we used genomically close-to-normal and 419 normal mammary epithelial cells as models to functionally interrogate the 420 transcriptome data. 421 We focused our analysis on EpCAM-positive cells from bone marrow. Currently, 422 EpCAM is the best marker to identify and isolate viable epithelial cells from bone 423 marrow. However, it is neither fully specific 14 as cells of the B cell lineage may express 424 EpCAM 23 nor does it identify all breast cancer initiating cells 56 . However, EpCAM-425 positive mammary progenitor cells give raise to chromosomally instable cancers, such 426 as triple negative basal like / post-EMT and hormone receptor positive cancers 57 . 427 Since we used chromosomal instability as inclusion criterion to identify DCCs our 428 findings do not apply to claudin-low cancers which are derived from EpCAM-negative, 429 chromosomally stable cells. Interestingly, early M0-stage DCCs isolated from bone 430 marrow highly expressed KIT, which is characteristic for luminal progenitor cells. This reasoning is fully consistent with our observed effects of IL6 on normal 437 mammary stem and progenitor cells. We identified an activated IL6 pathway in DCCs 438 and our in vitro models revealed that IL6 trans-, but not classical signaling, induces de-439 differentiation of mammary epithelial cells and endows them with stemness and 440

Patient material 734
Human non-cancerous mammary tissue was obtained from female patients

Isolation of human primary mammary epithelial cells 843
Primary human non-cancerous mammary tissue was dissociated as previously 844 described 20 . Briefly, upon mechanical digestion the tissue was subjected to enzymatic 845 digestion overnight at 37°C in DMEM/F12 (Pan-Biotech, Germany) supplemented with 846 10mM HEPES (Sigma-Aldrich, Germany), 2% bovine serum albumin (Sigma-Aldrich, 847 Germany), 5 μg/ml insulin, 0.5 μg/ml hydrocortisone, 10 ng/ml cholera toxin (Sigma-848 Aldrich, Germany), 300 Units/ml collagenase and 100 Units/ml hyaluronidase (all from 849 Sigma-Aldrich, Germany). After removal of organoids and adipocytes by centrifugation 850 at 210 g for 2 min, the cell suspension was passed over a 100 μm and 40 μm cell 851 strainer to obtain a single cell suspension. Separation of fibroblasts from epithelial cells 852 was accomplished by centrifugation at 350 g for 4 min and epithelial cells from the cell 853 pellet were cultured as mammospheres. The number of spheres with a diameter ≥ 50 µm was determined by manually counting 884 of a complete plate/dish at day 7 using an inverted microscope (Olympus, 10xair 885 objective). Alternatively, spheres were counted using the Operetta CLS high-content 886 imaging system (PerkinElmer, Hamburg, Germany) by adding CyTRAK Orange 887 (BioStatus Ltd, United Kingdom) at day 4 to the wells at a final concentration of 10 µM. 888 After 60 min incubation, fluorescence imaging of the plates was performed using a 5x 889 air objective and imaging of nine regions per well that were stitched to cover the entire 890 well surface. Harmony high content analysis software was used to analyze the images 891 and to count formation of spheres with diameter ≥ 50 µm (Version 4.8;PerkinElmer,892 Hamburg,Germany). MSCs and HUVECs were plated at a density of 4x10 5 cells/well of a 6 well plate 948 (Corning, Germany) in their respective growth medium. The next day, medium was 949 exchanged to MCF 10A growth medium and 1x10 5 MCF 10A-GFP cells were added 950 to each well. In case of co-cultures with OBs, 4x10 5 MSCs per 9.6 cm 2 surface of a 6 951 well plate (Corning, Germany)  EpCAM. The other half of the cell suspension was transplanted without ex vivo 964 expansion into NSG-mice using one to two injection routes (non-metastasized 965 patients) or three to four different injection routes (metastasized patients). In some 966 cases mononuclear cells were cultured as mammospheres in 6 cm culture plates 967 coated with polyhydroxyethylmethacrylate (12 mg/ml, Sigma), under hypoxic 968 conditions (7% O2) at 37°C and in mammosphere medium containing 10 nM HEPES 969 (Sigma-Aldrich, Germany), 10 µg/ml insulin (all from PAN-Biotech, Germany), 5 ng/ml 970 GRO-α (R&D Systems, Germany), 20 ng/ml hyper interleukin-6 (kindly provided by S. 971 Rose-John) and 0.2% Methylcellulose (Sigma-Aldrich, Germany). Cultures were 972 monitored weekly for sphere growth. 973 To transplant spheres or EpCAM-enriched or CD45/CD11b/erythrocyte depleted bone 974 marrow, cells/spheres were collected in a microwell (volume 10-15 µl, Terasaki, 975 Greiner Bio-One, Germany) pre-coated with polyhydroxyethylmethacrylate (12 mg/mL, 976 Sigma-Aldrich, Germany). Cells or spheres were transplanted in a final volume of 30 977 µl and 25% high-concentration matrigel (BD Biosciences, Germany) as published 978 before 9 . Cells were injected with an insulin syringe (Microfine, 29G, U-50, BD 979 Biosciences, Germany) sub-cutaneously, intra-venously, intra-femorally or sub-renally 980 in 4-8 weeks old male or female NSG or NOD.Cg-Prkdc scid mice. Mammary fat pad 981 injections were performed in the 4 th pre-cleared mammary fat pad of 3 weeks old 982 female mice in 50% matrigel (BD Biosciences, Germany). Breast or prostate cancer-983 origin of xenografts were verified by a pathologist. 984 To assess the differentiation ability of HMEC-spheres in vivo, secondary 985 mammospheres were dissociated and 200,000 cells were mixed with 225,000 pre-986 irradiated (15 Gy) C3H10T1/2 mouse fibroblasts. The cell suspension was then mixed 987 1:1 with matrigel (growth factor reduced without phenolred, BD Bioscience, Germany) 988 and injected in the 4 th pre-cleared mammary fat pad of 3 weeks old female NSG mice.