© 2018 Nature America, Inc., part of Springer Nature. All rights reserved. Received Received 11 July 2016; accepted 12 January 2018; published online 12 March 2018; of patients by the adjuvant therapy options that are offered for group a particular reflected are subtype cancer breast each of characteristics predictive and and Pathology, Department of Clinical Sciences, Lund University, Lund, Sweden. subtype that do not express hormone receptors hormone express not do that subtype basal-like the oftumors carry women of 10–15% whereas (PgR), one receptor (estrogen receptorsfor estrogen mammary carcinomas and are characterized by expression of hormone Breast (A ortumors B) account subtypes offor luminal the of70% all Biochemistry Biochemistry and Biophysics, Karolinska Institutet, Stockholm, Sweden. Sophie Sophie Lehn paucity of therapeutic targets a to owing rate survival overall worst the and recurrence to time est short rate, the recurrence highest tumors have the basal-like nomas, Oncology, Juntendo University School of Medicine, Tokyo, Japan. 11 1 expression biomarker or analyses expression gene through tified can iden relevance, be which of subtypes clinical different molecular of range a as manifests that disease heterogeneous a is cancer Breast subtype is under control microenvironmental and is actionable. therapeutically sensitivity to endocrine therapy in previously resistant tumors. We conclude that of specification breast cancer to the basal-like mouse models of cancer resulted in conversion of basal-like breast cancers into a hormone receptor-positive state that enhanced cognate receptors in human basal-like mammary carcinomas. Genetic or intervention of pharmacological PDGF-CC activity in identified paracrine crosstalk between cancer cells growth expressing platelet-derived factor (PDGF)-CC and CAFs expressing the Here we delineate a previously role unappreciated for CAFs as determinants of the molecular subtype of breast cancer. We underlying stroma instigates cancer-associated fibroblasts (CAFs) to support most, if not all, hallmarks of cancer progression. of recurrence and poor survival in patients following treatment. Coevolution of the malignant mammary epithelium and its Breast tumors of the basal-like, hormone subtype receptor–negative remain an unmet clinical challenge, as there is high rate Pernilla Roswall CC signaling elicited through paracrine platelet-derived growth factor- Microenvironmental control of breast cancer subtype nature medicine nature Lund, Sweden. the malignant cells in the tumor. However, this tumor cell–centric view breast patients cancer arewith basal-like urgently required. Eliane Cortez of of Medical Research, Sydney, New South Wales, Australia. fulvestrant or aromatase inhibitors treatment with drugs that impinge on ER Jari Häkkinen Holger Moch Division Division of Translational Cancer Research, Department of Laboratory Medicine, Lund University, Lund, Sweden. These These authors contributed equally to this work. should Correspondence be addressed to K.P. ( Breast cancer subtypes are defined according to the characteristics of 4– 6 . . Patients luminal carrying tumors typically benefit from 6 Olivia Olivia Newton-John Cancer Research Institute and School of Cancer Medicine, La Trobe University, Melbourne, Victoria, Australia. 1 10 1 , , Jonas Sjölund 4 , Lao H Saal

, , Andrew M Scott

1

, Ingrid J G Burvenich advance online publication online advance , 2 , 11 , , Matteo Bocci 1 , 3 , 8 , 9 4 . Thus, new treatment approaches for

7 . . In contrast, out of all breast carci 1

, Christina Orsmark-Pietras , , Steven Reid α 6 signaling, such as tamoxifen, α , , Ulf Eriksson (ER 1 , 11 α 3 , , Michael Bartoschek 6 . The prognosis and prognosis The . )) and/or progesterand/or )) , , Elgene Lim 8 University University of New South Wales, Sydney, New South Wales, Australia. 1

, , Christer Larsson 10 Department Department of Pathology and Molecular Pathology, University Hospital Zürich, Zürich, Switzerland. 2

3

Institute Institute of Pathology, University Hospital Bonn, Bonn, Germany. & Kristian Pietras 7 1 , 8 , 2 , Akira , Orimo Akira - - - - . 5 Division Division of Clinical Genetics, Department of Laboratory Medicine, Lund University, 5 , Eugenia Cordero vasive factors tion of growth-stimulatory, angiogenic, immune-modulatory andmany proindifferent aspects of tumor growth and progression through secre major constituent cell type of tumor stroma and are known to support spective,tumors must regarded be multicellularas organs. CAFsare a the basis of morphological criteria and marker expression histological subtypes of the fibroblastic stroma of breast carcinomas on such as melanoma and cervical carcinoma and cervical as melanoma such of in CAFs and in PDGFs the tumors, recruitment solid functionality with diverse and opposing functions possibilityrepresentthethat ingCAFs heterogeneous a groupcellsof or as prognostic and/or predictive biomarkers is still under debate, open been reported to express both PDGF receptors, i.e., PDGFR- i.e., receptors, PDGF both express to reported been malignant initiation and progression munication, thus creating a dynamic signaling circuitry that promotes com continuousparacrine through state activated an into coevolves microenvironmentsurrounding the progresses, lesion cancerous the does not take into account the context in which cancer cells subsist. As 1 doi:10.1038/nm.449 We have previously demonstrated a role for signaling through through signaling for role a demonstrated previously have We , 11 1 , , Pontus Eriksson , , Hong Li [email protected] 11 1 9 , , , Mattias Höglund 1 2 . However, the potential for CAFs as therapeutic targets 2 , 11 4 , , Glen Kristiansen 1 , Bengt Kristian Haller 2 Division Division of Vascular Biology, Department of Medical 4 , , Charlotte Anderberg ). 13 1 0 , 9 1 4 . Hence, from a therapeutic per Department Department of Pathology and 4 , Lisa , RydénLisa . Indeed, a recent study defined 16 3 , 1 , , Sara Jansson 7 4 . Although CAFs have CAFs . Although Division Division of Oncology 2 s e l c i t r a , 7

Garvan Garvan Institute 4 , 1

1 , 5

. 4 α ,

and and  - - - - - © 2018 Nature America, Inc., part of Springer Nature. All rights reserved. vival in breast cancer ( cancer breast in vival sur poor for factor prognostic independent an as served PDGF-CC of expression epithelial that demonstrated others, among expression receptor hormone and status node lymph grade, stage, diagnosis, at as such age factors, risk clinical for established adjusted analysis sion homodimers as well as PDGFR- as well as homodimers with with breast cancer (the Lund cohort poor survival ( for factor prognostic significant a to statistically be found was to 3+) of 1+ (score cells malignant in of PDGF-CC expression high to erate sion analysis ( regres Cox univariate in survival poor for factor risk a as identified nor analysis survival Kaplan–Meier in outcome patient with ciated Table 2 ( status node lymph or stage the to not but tumor, the of index proliferative and grade the to correlated positively was of expression PDGF-CC and epithelial stromal Both malignant eters. Fig. Fig. 1a gest that PDGF-CC may activate PDGF receptor alpha (PDGFR- alpha receptor PDGF activate may PDGF-CC that gest PDGF-CC, remains undetermined, although although undetermined, remains PDGF-CC, The PDGF-DD). and PDGF-CC PDGF-BB, PDGF-AB, (PDGF-AA, isoforms ligand PDGF different five the via receptors these of activation preferential the staining intensity for PDGF-CC in the independent epithelial epithelial independent and stromal compartments (for scoring see schemes, the in PDGF-CC for intensity staining the ( and fibroblasts stromal capillaries tumoral intra cells, by malignant expressed was PDGF-CC In tumors, breast and uniformly expressed by stromal fibroblasts in tumors from the the from tumors in fibroblasts stromal by expressed uniformly and PDGFR- A  cohort ing 890 tumor as specimens well as healthy breast (The tissues Zurich cancer, we performed immunostaining of a tissue microarray contain breast human in PDGF-CC of pattern expression the Toinvestigate in breast cancer PDGF-CC is an independent prognostic factor for poor survival RESULTS tumors. breast basal-like with patients for implications therapeutic significant has and subtype define cancer that breast mechanisms unknown previously on light sheds work Our tumors. resistant previously in therapy endocrine to sensitivity ER an into tumors breast basal-like of conversion in resulted models mouse experimental in PDGF-CC of targeting or pharmacological Genetic state. receptor–negative mone a hor with associated was cancer of breast epithelium malignant the of with breast patients cancer, we in found expression that PDGF-CC cohorts well-characterized and large several In response. treatment of type breast cancer, which is a of determinant patient prognosis and signaling paracrine for through PDGF-CC in CAFs in the of specification the sub role molecular unappreciated previously a delineate CC CC expression was not in cells ( detected epithelial luminal myoepithelial cells and endothelial cells in capillaries, whereas PDGF- to basal limited was tissue breast in healthy of PDGF-CC Expression stromal stromal tissue and malignant epithelium ( between border the at conspicuous most was immunoreactivity CC (Supplementary Table 4 CC was correlated with tumor grade and associated with poor survival Supplementary Table 3 in the PDGF family, PDGF the in PDGFR- i.e.,

Our findings were corroborated in a second cohort of 550 patients patients 550 of cohort second a in corroborated were findings Our s e l c i t r 2 , b - asso neither was for PDGF-CC immunoreactivity ). Stromal 0 ; for patient characteristics, see see characteristics, patient for ; ) ) and correlated these findings to param clinicopathological β , signaling pathway activity is determined through the the through determined is activity pathway signaling , Supplementary Fig. 1c Supplementary Fig. 1 in vivo in e and Table ), in which high epithelial expression of PDGF- and specificity of one of these PDGF isoforms, isoforms, PDGF these of one of specificity Table 1 Supplementary Fig. 1d ). 1 ). Notably, multivariable Cox regres α α and PDGFR- / 21 β , and 2 heterodimers α 2 ; ; for patient characteristics see -positive state that conferred conferred that state -positive Fig. Fig. 1 Table Supplementary Table 1 Table Supplementary Fig. Fig. 1 d n vitro in 1 ). ). Next, we evaluated β ). ). In contrast, mod c , were, prominently ). ). Notably, PDGF- 18 ). Both receptors Supplementary Supplementary Supplementary Supplementary , 1 9 tde sug studies . Herein, we we Herein, . Fig. Fig. 1a , b α ). ). ). ). ------)

average tumor size to 96 96 to size tumor average Pdgfc that human tumors revealed breast from cancer,analyses histological ( of mice the survival prolonged as well as latency tumor longer significantly gle copy or both copies of the the of copies both or copy gle and and relevant mouse MMTV-PyMT engineered mouse model of breast cancer based on the widely studied context of we gland tumorigenesis, generated mammary a genetically the in expression PDGF-CC of aspects functional the Toinvestigate breast cancer PDGF-CC is functionally important for growth of experimental ofcells the tumor microenvironment signaling. through PDGF cells engage in functional paracrine communication with mesenchymal evaluable tumors, respectively).of (100%) These findings 480 indicateof 480 that and malignant (98.6%) 489 of 482 in CAFs by expressed PDGF-CC encoding gene the inactivates that cassette lacZ a of insertion an ing type (WT) mice with tumors that were grade-matched to those of of those the to grade-matched were that tumors with mice (WT) type wild- of cohort a as disease, of onset delayed the to due likely most 14-week-old ( in mice signaling tumor-bearing PDGF-CC of absence the in metastases ( cell lines from tumors derived from MMTV-PyMT; MMTV-PyMT; from derived mice, tumors from FVB/n lines of cell pads fat mammary the into transplantation topic ortho following exponentially grew that tumors to rise gave readily tumors in and mice WT (PeRo-Bas1 cells) MMTV-PyMT PeRo-Bas2 from derived lines cell isolated independently two whereas addition, mice. Masson’s trichrome staining of tumor sections demonstrated demonstrated sections tumor of staining Masson’strichrome mice. nontransgenic of transplanted all Whereas mice. FVB/n pads fat mammary the into orthotopically mice tumors from MMTV-PyMT; MMTV-PyMT; from tumors of fragments transplanted we defects, developmental to due not was ( lacking lacking tecture between MMTV-PyMT; Histological analyses revealed considerable differences in tumor archi response in the tumor microenvironment Pdgfc as palpable tumors ( tumors palpable as (PeRo-Lum1 establish and cells to PeRo-Lum2 failed cells) efficiently planted fragments from from fragments trans planted the of half about only tumors, growing as established 21) of Lund cohort ( mice ( mice tumors in of the growth MMTV-PyMT mammary impacted severely PDGFR- and PDGF circuitry between malignant cells and stromal fibroblasts fibroblasts stromal and cells ( malignant between circuitry PDGF paracrine a of existence the establishing thus cancers, breast human in seen pattern expression the of recapitulation consistent onstrated hampered as compared to that of the the of that to compared as hampered control control mice ( setting, growth of the transplanted of growth transplanted the setting, gave rise to tumors. In accordance with our in findings the transgenic size of 220 220 of size Supplementary Fig. 2a Fig. Supplementary Fig. 1 Fig. i. 1g Fig. To that determine the delayed tumor development in Pdgfc −/− deficiency results in a blunted fibrotic and angiogenic Fig. 1 Fig. f Pdgfc ). Genetic deficiency for for deficiency Genetic ). mice were of a lower grade than tumors from age-matched age-matched from tumors than grade lower a of were mice , j −/− , k 2 Supplementary Fig. 3a Fig. Supplementary ± 5 ). We also observed a trend toward fewer pulmonary pulmonary fewer toward trend a observed also We ). cohort displayed a similar metastatic burden as the mice mice as the burden metastatic a similar displayed cohort ( f 47 mm 47 ( Supplementary Fig. 1e β ). Control mice presented with tumors of an average average an of tumors with presented mice Control ). Pdgfc Fig. Fig. 1g Supplementary Fig. 3c Fig. Supplementary expression in tumors of MMTV-PyMT mice dem mice MMTV-PyMT of tumors in expression in vivo advance online publication online advance −/− Supplementary Fig. 3d Fig. Supplementary 3 – (mean (mean mice). Visualization of PDGF-CC, PDGFR- PDGF-CC, of Visualization mice). i Supplementary Fig. 3c Fig. Supplementary ) and had significantly larger areas ) of and necrosis had significantly – Pdgfc ± d 41 mm 41 ). Strikingly, genetic deficiency for for deficiency genetic Strikingly, ). ± Pdgfc −/− s.e.m.), whereas deficiency for a sin a for deficiency whereas s.e.m.), Pdgfc Pdgfc mice (11 of 21; 21; of (11 mice Pdgfc 23 Pdgfc , 3 +/+ b , and 87 87 and allele significantly reduced the the reduced significantly allele , f 2 +/+ ). In line with the findings in in findings the with line In ). ; PDGFR- 4 ). and MMTV-PyMT; MMTV-PyMT; and Pdgfc intercrossed with intercrossed mice carry Pdgfc and MMTV-PyMT; was also associated with a with associated also was −/− tumors was dramatically tumors was dramatically , +/+ +/+ ± e ). 71 mm 71 ). However, this was was this However, ). tumor fragments (21 fragments tumor α

tumors ( tumors and PDGFR- nature medicine nature χ 2 test, test, 3 Pdgfc Pdgfc , respectively respectively , Fig. 1 Fig. P = 0.003) 0.003) = −/− Pdgfc Pdgfc −/− β mice mice Pdgfc mice l were ). In ). −/− −/− α ------

© 2018 Nature America, Inc., part of Springer Nature. All rights reserved. Pdgfc equal variance Student’s Student’s variance equal ( cells (for scoring scheme, see see scheme, scoring (for cells according to absence (negative; (negative; absence to according 40 bars, Scale border. invasive the on PDGF-CC of expression high particularly with cells tumor identify arrows and boundary, epithelium–stroma the represents line dashed The reproducible. was tissue human independent experiments for each group). *** group). each for experiments independent ( mice WT of pad fat mammary the into transplanted mice MMTV-PyMT from derived MMTV-PyMT; of tissue sections from tumors from 14-week-old 14-week-old from tumors from sections tissue of arbitrary units. ** units. arbitrary in tumor lysates of tumors from 14-week-old MMTV-PyMT mice ( mice MMTV-PyMT 14-week-old from tumors of lysates tumor in breast tissue ( tissue breast one tumor of each genotype, but is representative of analysis of 5 mice per group. Scale bars, 500 500 bars, Scale group. per 5 mice of analysis of representative is but genotype, each of tumor one ( protein 1 ( for as markers such CAFs, prototypical encoding genes of transcripts mRNA of abundance qRT-PCRa showed of reduced idea, support this analysis group. Scale bars, 500 500 bars, Scale group. of vascular endothelial growth factor-A expression in the absence absence the in expression factor-A growth endothelial vascular of levels in reduction 65% a revealed qRT-PCR analysis accordance, In ( hypoxia of regulator transcriptional ter hypoxia, as through immunostaining evidenced for HIF-1 Pdgfc of of matrix the in collagen intratumoral of deposition reduced severely a 1 Figure nature medicine nature fibroblasts in the breast tumor microenvironment ( microenvironment tumor breast the in fibroblasts stromal of activation and/or recruitment the promotes signaling CC f Fig. 1 Fig. a ) Average volume (with s.e.m.) of tumors from 14-week-old MMTV-PyMT mice ( mice MMTV-PyMT 14-week-old from tumors of s.e.m.) (with volume ) Average

Pdgfc Healthy breast +/+ −/− o l f MMTV-PyMT , , versus

−/− ; Student’s ; Epithelial expression of PDGF-CC is associated with a poor outcome in patients with breast carcinoma. ( carcinoma. breast with patients in outcome a poor with associated is PDGF-CC of expression Epithelial χ 3 genotype 2 Tumor volume (mm ) S100a4

1,000 1,500 3 tumors, which is consistent with the notion that PDGF- that notion the with consistent is which tumors, test. ( test. Tumor volume (mm ) PDGF-CC 500 100 150 200 250 300 a 0 50 , Pdgfc b 0 02 30 20 10 0 Pdgfc α ) or breast carcinoma ( carcinoma breast ) or P h -smooth muscle actin ( actin muscle -smooth ) ) and PDGFR- = 0.007 ( = 0.007 Pdgfc –

k Pdgfc Pdgfc t +/− ) H&E staining of tumors from from tumors of staining ) H&E -test, -test, advance online publication online advance −/− µ , , +/+ m. ( m. t χ mice exhibited significantly increased levels of *** –/– +/+ -test. ( -test. 2 Time (d) Pdgfc

test. *** test.

P o Pdgfra *** ) Results from qRT-PCR analysis showing the average expression (with s.e.m.) of markers for cancer-associated fibroblasts fibroblasts cancer-associated for markers of s.e.m.) (with expression average the showing qRT-PCR from analysis ) Results < 0.01 for each marker). Tumors from from Tumors marker). each for 0.01 < Supplementary Fig. 1a Fig. Supplementary n b g +/– = 438 samples) or presence (positive, defined as a score of 1+ to 3+; 3+; to 1+ of a score as defined (positive, presence or samples) = 438 ) Grading of tumors from 14-week-old MMTV-PyMT mice ( mice MMTV-PyMT 14-week-old from tumors of ) Grading α Pdgfc ( ), ), P Pdgfra P = 3 × 10 = 0.003 ( = 0.003 c PDGF-CC –/– , d ) for PDGF-CC. Results were validated on >10 independent samples to ensure the staining pattern on pattern staining the ensure to samples independent >10 on validated were Results PDGF-CC. ) for Supplementary Fig. 4a Fig. Supplementary ), ), in Acta2 *** MMTV-PyMT P = 6 × 10 −11 g Pdgfc genotype Pdgfc Percentage of tumor S100a4 ; distribution of of ; distribution

), fibroblast-specific fibroblast-specific ), tissue (cumulative) 100 Pdgfc ) of individuals with breast carcinomas from the Zürich cohort. ** cohort. Zürich the from carcinomas breast with individuals ) of 10 20 30 40 50 60 70 80 90 0 +/+ m −/−

−6 MMTV-PyMT Pdgfc

( ), ), tumor samples Fig. 1m Fig. +/+ m ; two-sided, unpaired, equal variance Student’s Student’s variance equal unpaired, ; two-sided, P ) and ) and ( c = 0.009 ( = 0.009 Pdgfc +/+ α h Breast carcinoma , , , the mas j n ) and ) and Pdgfc = 3 mice in each group; analysis was performed independently 3 times). a.u., a.u., 3 times). independently performed was analysis group; each in = 3 mice Pdgfc +/+ µ *** Pdgfc , n m. ( m. Pdgfc ). In ). +/– – Pdgfc c Acta2 −/− PDGF-CC e **** ). ). −/− Pdgfc - ) Kaplan–Meier analysis of breast cancer–specific survival dichotomized dichotomized survival cancer–specific breast of analysis ) Kaplan–Meier n

( versus versus n −/− ) MMTV-PyMT mice. Stains are representative of of representative are Stains mice. ) MMTV-PyMT transcriptional analysis of tumors derived from MMTV-PyMT; MMTV-PyMT; from derived tumors of analysis transcriptional To elucidate the molecular signature of basal-like molecular subtype Expression of PDGF-CC in breast tumors is associated with a angiogenesis. and recruitment CAF growth, tumor breast in ing of blockade corroborate logical PDGF-CC a role for signal paracrine f rhtpc aa-ie D-B21 uos ( tumors Fig. 5a MDA-MB-231 response basal-like angiogenic orthotopic and of growth the delayed mice tumor- SCID in 6B3 bearing antibody inhibitory the of administration through ( hemorrhagic tumors were severely of area, these abundance necrotic the increased ( PDGF-CC of ); two-sided, unpaired, equal variance Student’s Student’s variance equal unpaired, two-sided, ); –/– ( n i ,

= 10 mice in each group). *** group). each in mice = 10

Necrosis Carcinoma Normal fat Hyperplasia Adenoma Late carcinoma k –/– ) MMTV-PyMT mice at 14 weeks of age. Images shown are from from are shown Images age. of weeks 14 at mice ) MMTV-PyMT Pdgfc n – = 21 mice in each group, comprising comprising group, each in mice = 21 c ). ). Taken together, and from results genetic pharmaco these +/+ d , , n χ = 5 mice in each group). * group). each in = 5 mice Supplementary Fig. 4e Fig. Supplementary 2 Supplementary Fig. 4d Fig. Supplementary o test. *** test.

Relative expression (a.u.) MMTV-PyMT h j 0.0 0.5 1.0 1.5 µ PDGF-CC m. ( m. Pdgfc Pdgfc n l P dfaS0a Acta2 S100a4 Pdgfra ) Average volume (with s.e.m.) of tumors tumors of s.e.m.) (with volume ) Average Pdgfc = 452 samples) of PDGF-CC in tumor tumor in PDGF-CC of samples) = 452 = 0.00052; distribution of of distribution = 0.00052; t -test. ( -test. ** –/– +/+ +/+

P a = 0.0001; two-sided, unpaired, unpaired, two-sided, = 0.0001; m – d , n ) Immunostaining of healthy healthy of ) Immunostaining ) Masson’s trichrome staining staining trichrome ) Masson’s P Pdgfc ** ). Neutralization of PDGF-CC of PDGF-CC Neutralization ). = 0.002; log-rank test. test. log-rank = 0.002; e k i P ). Furthermore, in line with with line in Furthermore, ).

= 0.036; distribution of distribution = 0.036; Cumulative survival 0.5 0.6 0.7 0.8 0.9 1.0 n -deficiency, we performed = 7 mice from 3 from = 7 mice 0 n t -test. 1 = 10 mice from each each from mice = 10 Pdgfc 2 ** 3 s e l c i t r a –/– Time (years) 4 Pdgfc Supplementary Supplementary PDGF-CC (+) PDGF-CC (–) 5 6 +/− 7 8 **P =0.002

versus versus 9

10

11

12  - -

© 2018 Nature America, Inc., part of Springer Nature. All rights reserved. from a panel of 51 breast tumor–derived cell lines cell tumor–derived breast 51 of panel a from and of absence in the of cancer breast models in mouse to found was upregulated be ( analysis MMTV-PyMT; MMTV-PyMT; from lysates whole-tumor in higher 17.2-fold average, on was, sion Age Age at diagnosis sion of of sion carcinomas from MMTV-PyMT; MMTV-PyMT; from carcinomas corroborated at the protein level in lysates whole-tumor of mammary further was PDGF-CC of loss and expression FOXA1 between tion studies sion of n bes cne o te uia sbye tmr fo MMTV- from tumors subtype, PyMT; luminal the of cancer breast and ited robust expression of ER of expression robust ited Pdgfc HER2 G pN Parameter regulated gene was found to be forkhead box A1 ( A1 box forkhead be to found was gene regulated differentially most The progression. and development tumor breast in of panel to a analyze that targeted genes are designed instrumental expression of expression Pdgfc Pdgfc CK5/6, grade; histological Nottingham G, involved; nodes 5/6. lymph of cytokeratin number pN, stage; tumor pT, primary analyzed; not NA, interval; confidence CI, risk; relative RR, Table 1 A  Menopausal Menopausal status lines of the basal-like subtype but not in cells originating from the the from originating cells in not but subtype basal-like the of lines meaningful express of not levels did lines cell basal-like whereas subtype, of expression and 6.5-fold higher in and from higher cell lines isolated 6.5-fold CK5/6 PgR ER pT subtype ( subtype The Cancer Genome Atlas (TCGA) network (TCGA) Atlas Genome Cancer The within collected tumors breast 1,086 of profiles transcriptional the PDGF-CC PDGF-CC stroma PDGF-CC epithelium

1+ 1+ vs. 2+, 3+ 0 vs. 1–3+ negative vs. positive 0–2+ vs. 3+ negative vs. positive negative vs. positive 1–3 0–3 1–4 pre- vs. post- <60 years vs. Next, we set out to determine the functional implications of high high of implications functional the determine to out set we Next, α s e l c i t r PDGFC −/− +/+ +/+ FOXA1 1 Pdgfc

PDGFC , mice ( mice tumors ( tumors and MMTV-PyMT; MMTV-PyMT; and 27 Cox Cox regression analysis of risk factors for breast cancer death in the Zurich cohort Supplementary Fig. 6c Fig. Supplementary FOXA1 Supplementary Fig. 6b Fig. Supplementary , 2 Pdgfc 8 in human breast cancer. Transcriptional data mined mined data Transcriptional cancer. breast human in . In accordance with the association between between association the with accordance In . −/− FOXA1 ≥ Foxa1 was highly correlated with a non-basal-like molecular molecular was with correlated a highly non-basal-like 60 60 years Supplementary Fig. 6b Fig. Supplementary mice with high levels of FOXA1 protein also exhib also protein of FOXA1 levels high with mice was predominantly observed in breast tumor cell cell tumor breast in observed predominantly was Pdgfc , as expected ( expected as , , we investigated the correlation between between correlation the investigated we , Fig. 2 Fig. in tumors from breast was a specific feature was a of specific cell lines of the luminal −/− a mice than in MMTV-PyMT; MMTV-PyMT; in than mice and and α Supplementary Fig. 6a Fig. Supplementary Pdgfc , as determined through western blot blot western through determined as , Fig. 2 Fig. 1.542 1.542 (1.170–2.032) 0.931 0.931 (0.708–1.225) 1.520 (1.156–1.998) 1.496 (1.010–2.217) 1.941 (1.373–2.745) 0.511 (0.388–0.673) 0.494 (0.364–0.670) 1.669 (1.341–2.077) 1.681 (1.423–1.986) 1.611 (1.421–1.825) 1.355 (0.965–1.902) ), confirming results from previous previous from results ), confirming ). Considering the fact that that fact the Considering ). Pdgfc −/− RR (95%CI) mice using a qRT-PCR array array qRT-PCR a using mice b ). +/− ). In contrast, high expres high contrast, In ). mice and MMTV-PyMT; MMTV-PyMT; and mice Pdgfc Pdgfc 2 6 revealed that expres that revealed Univariable analysis −/− Foxa1 −/− mice. Analysis of Analysis mice. 2 tumors than in 9 ). The associa The ). revealed that that revealed Pdgfc ); its expres its ); +/+ FOXA1 FOXA1 Foxa1 mice mice <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 P value 0.002 0.931 0.003 0.045 0.079 - - - - -

of the existence of an activated ER activated an of existence the of of tumors in luminal the absence signifying ( documented for enrichment in ER resulted cells these in treatment tamoxifen that demonstrated (GSEA) analysis enrichment set gene Fig. 6d ( ( 1 receptor by estrogen encoded factors transcription luminal-defining the for sites ing (E2)-stimulated (E2)-stimulated in this panel of cell lines ( lines cell of panel this in ited a statistically significant inverse correlation with that of that with correlation inverse significant a statistically ited ( subtype luminal n oh C7 n T7 cls ( cells. T47D and MCF7 both in human to encoded orthologous proteins genes of mouse by sites binding for enriched significantly were genes underexpressed and over- the genes, 1,000 of lists permuted cells relative to the the to relative cells cells T47D and MCF7 nal of lumi analyses (ChIP–seq) sequencing and chromatin immunoprecipitation from sets data public Using tumors. signify breast program luminal ing gene regulatory genome-wide a with associated overexpressed overexpressed and the top 1,000 underexpressed genes in the of ER tutive activation consti in suggesting cells, to PeRo-Bas1 as compared cells higher PeRo-Lum1 was genes these of level expression basal the cells, Lum1 of expression of induction further demonstrate to unable were we Although cells. was exclusively observed for for observed exclusively was ( tumors breast basal-like from derived lines cell in family PDGF the GATA3 Fig. 2 Fig. upeetr Fg 7 Fig. Supplementary α -target genes or genes with documented sensitivity to tamoxifen to tamoxifen sensitivity or genes documented with genes -target e – ). We next investigated whether the absence of of absence the whether investigated We ). next ) in the genes most differentially expressed between estradiol ) in estradiol the between expressed genes most differentially g ), indicating activation of a global gene regulatory program of gene regulatory a activation global ), indicating Esr1 advance online publication online advance Pdgfc 1.577 1.577 (1.041–2.388) 0.603 (0.411–0.886) 1.211 (0.757–1.937) 1.410 (1.067–1.862) 1.415 (1.173–1.708) 1.624 (1.378–1.913) 1.437 1.437 (1.035–1.995) 1.484 1.484 (1.035–2.127) 1.457 (0.891–2.382) Fig. 2 Fig. -target genes following E2 stimulation in PeRo- in stimulation E2 following genes -target Pdgfc RR (95%CI) ESR1 α −/− signaling. We the top 1,000 signaling. determined next c ). Indeed, the expression of of expression the Indeed, ). NA NA PeRo-Lum1 cells and and cells PeRo-Lum1 +/+ ). Additionally, tamoxifen treatment treatment tamoxifen Additionally, ). 3 Fig. 2 Fig. ), ), 0 , we analyzed the enrichment of bind of enrichment the analyzed we , cells. Compared to 100,000 randomly randomly 100,000 to Compared cells. FOXA1 PDGFC Multivariable analysis d ). The enrichment for expression expression for enrichment The ). i. 2f Fig. α , and GATA binding protein 3 protein binding GATA and , and not for other members of of members other for not and pathway in PeRo-Lum1 cells, cells, PeRo-Lum1 in pathway ESR1 – Pdgfc k and and , ,

FOXA1 Pdgfc . In further support . support In further nature medicine nature <0.001 <0.001 P value 0.031 0.424 0.016 0.032 0.133 Supplementary Supplementary 0.01 0.03 NA NA PDGFC +/+ and and PDGFC PeRo-Bas1 PeRo-Bas1 Pdgfc FOXA1 GATA3

exhib was was −/− - - - - -

© 2018 Nature America, Inc., part of Springer Nature. All rights reserved. 10 the average expression and whiskers depict the range of expression with statistical outliers (>2 s.d. from the mean) indicated by circles. *** circles. by indicated mean) the from s.d. (>2 outliers statistical with expression of range the depict whiskers and expression average the FOXA1 genes in the PDGF family in in family PDGF the in genes in in ESR1 height of the box, respectively. respectively. box, the of height the 3× of outside fall that outliers extreme and whiskers the of outside fall that outliers statistical represent triangles and Circles box. the of height the 1.5× depict whiskers and expression median the represents midline the range, interquartile the represents box The cohort. Lund the in included carcinomas ( carcinomas Correlation coefficient and and coefficient Correlation performed independently 3 times; mean mean 3 times; independently performed between expression of of expression between qRT-PCR analysis of expression of the luminal subtype marker marker subtype luminal the of expression of qRT-PCR analysis Figure 2 Figure nature medicine nature node metastases (lymph node met) ( met) node (lymph metastases node or asynchronous distant relapses ( relapses distant asynchronous or P a as used then was instances resample of number total the by divided list gene query the of rank The lists. gene sampled randomly the of average the represents dot green the signature; query the to size equal of lists gene sampled randomly 100,000 of a total of out one represents dot blue each genes; value for the probability to acquire the level of enrichment by chance. ( chance. by enrichment of level the acquire to probability the for value Pdgfc −6 ; two-sided, unpaired Welch’s unequal variances variances Welch’s unequal unpaired ; two-sided, , , ( FOXA1 b

−/− Expression of PDGF-CC in breast carcinomas is associated with the hormone receptor–negative, basal-like molecular subtype. ( subtype. molecular basal-like receptor–negative, hormone the with associated is carcinomas breast in PDGF-CC of Expression ) and ) and PeRo-Lum1 cells as compared to to compared as cells PeRo-Lum1 l l e d a : : , and , and Primary tumor ERα score PDGFC n Microarray feature intensity PDGFC (%) = 470 independent samples; *** samples; independent = 470 100 100 200 300 400 500 600 (log (expression)) 20 40 60 80 2 Relative expression (a.u.) –3 –2 –1 0 0 0 1 2 3 4 10 20 30 GATA3 –3

0 eaie1 +3+ 2+ 1+ Negative PDGFA ( advance online publication online advance FOXA1 c Primary tumorPDGF-CCscore ) in a panel of 51 breast cancer cell lines cell cancer breast 51 of a panel ) in –2 FOXA1 (log , as derived from public data sets from ChIP–seq analyses in MCF7 cells, in the 1,000 most up- and downregulated genes genes downregulated and up- most 1,000 the in cells, MCF7 in analyses ChIP–seq from sets data public from derived , as Pdgfc Pdgfc P n value from Pearson correlation analysis of of analysis correlation Pearson from value PDGFB and and = 26 basal cell lines and and lines cell basal = 26 –1 Foxa1 P =0.0001 r =–0.52 –/– +/+ n 2 PDGFC (expression)) : : PDGFC *** 0 m n = 29 independent samples; ** samples; independent = 29 : : *** n ± = 132 independent samples; *** samples; independent = 132 1 in a panel of 51 breast cancer cell lines divided according to basal-like or luminal-like molecular subtype. subtype. molecular luminal-like or basal-like to according divided lines cell cancer breast 51 of a panel in s.e.m. depicted). *** depicted). s.e.m. PDGFD Basal Luminal Luminal Basal Pdgfc *** 2 P +/+ = 2.1 × 10 = 2.1 PeRo-Bas1 cells. The red dot represents the query signature of 1,000 up- or downregulated downregulated or up- 1,000 of signature query the represents dot red The cells. PeRo-Bas1 t b c -test. ( -test.

n PDGFC FOXA1 m = 25 luminal cell lines. The box represents the interquartile range, the midline represents represents midline the range, interquartile the represents box The lines. cell luminal = 25 Lymph node met ERα score (log (expression)) (log (expression)) i f 2 2 –3 –2 –1 (%) ChIP score ChIP score –3 –2 –1 100 0 1 2 3 4 2 3 0 1 500 700 500 700 f 20 40 60 80 – 0 Foxa1 Number ofgenes(down) −14 k P Number ofgenes(up) ) Enrichment for binding sites for luminal-defining transcription factors encoded by encoded factors transcription luminal-defining for sites binding for ) Enrichment 0 400 300 ESR1 300 Basal ESR1 BT-20 BT-20 < 0.001; two-sided, unpaired, equal variance Student’s Student’s variance equal unpaired, two-sided, < 0.001; Basal ; two-sided Jonckheere–Terpstra test for ordered alternatives), synchronous lymph lymph synchronous alternatives), ordered for test Jonckheere–Terpstra ; two-sided 2 eaie1 +3+ 2+ 1+ Negative BT-549 BT-549 9

Lymph nodemetPDGF-CCscore HBL100 HBL100 divided according to basal-like or luminal-like molecular subtype. ( subtype. molecular luminal-like or basal-like to according divided in tumors from 14-week-old MMTV-PyMT mice ( mice MMTV-PyMT 14-week-old from tumors in

P HCC1143 HCC1143

= 0.0015; two-sided Jonckheere–Terpstra test for ordered alternatives) alternatives) ordered for test Jonckheere–Terpstra two-sided = 0.0015; HCC1187 HCC1187 Luminal n P <1×10 P <1×10 HCC1500 HCC1500 Luminal = 26 basal cell lines and and lines cell basal = 26 l HCC1569 HCC1569 – P

n HCC1937 HCC1937 400 = 1.5 × 10 = 1.5 ) Correlation between expression of ER of expression between ) Correlation HCC1954 HCC1954 HCC2157 HCC2157 –5 –5 HCC3153 HCC3153 HCC38 HCC38 HCC70 HCC70 HS 578T HS 578T g j MCF-10A MCF-10A ChIP score ChIP score MCF-12A MCF-12A −8 300 400 300 400 MDA-MB-157 MDA-MB-157 MDA-MB-231 MDA-MB-231 *** ; two-sided Jonckheere–Terpstra test for ordered alternatives) alternatives) ordered for test Jonckheere–Terpstra ; two-sided Number ofgenes(down) MDA-MB-435 MDA-MB-435 Number ofgenes(up) FOXA1 FOXA1 2 280 220 2 280 220 MDA-MB-436 MDA-MB-436 MDA-MB-468 MDA-MB-468 SUM-1315MO2 SUM-1315MO2 SUM-149PT SUM-149PT SUM-159PT SUM-159PT n P <1×10 P <2×10 SUM-190PT SUM-190PT = 25 luminal cell lines. ( lines. cell luminal = 25 n SUM-225 SUM-225 Distant relapse ERα score AU565 AU565 (%) BT-474 BT-474 100 BT-483 BT-483 20 40 60 80 –5 –4

0 CAMA-1 CAMA-1 HCC1007 HCC1007 HCC1428 HCC1428

eaie1 +3+ 2+ 1+ Negative HCC202 HCC202 HCC2185

Distant relapsePDGF-CCscore HCC2185

k h LY2 LY2

α ChIP score ChIP score MCF-7 MCF-7 1,000 1,200 1,000 1,200 and PDGF-CC in primary breast breast primary in PDGF-CC and MDA-MB-134VI MDA-MB-134VI n MDA-MB-175VII MDA-MB-175VII = 5 mice in each group; analysis analysis group; each in = 5 mice MDA-MB-361 MDA-MB-361 Number ofgenes(down)

Number ofgenes(up) MDA-MB-415 MDA-MB-415 0 600 500 0 600 500 GATA3 GATA3 t MDA-MB-453 MDA-MB-453 -test. ( -test. SK-BR-3 SK-BR-3 SUM-185PE SUM-185PE e

) Expression of all all of ) Expression SUM-44PE SUM-44PE SUM-52PE SUM-52PE b P <1×10 P <1×10 T-47D T-47D

, UACC-812 UACC-812 c s e l c i t r a

) Expression of ) Expression X600MPE X600MPE ZR-75-1 ZR-75-1 ** ZR-75-30 ZR-75-30 d –5 –5 a ZR-75-B ZR-75-B ) Correlation ) Correlation ) Results from from ) Results P = 1 ×

 © 2018 Nature America, Inc., part of Springer Nature. All rights reserved. expressing FOXA1, ER FOXA1, expressing tumors receptor-positive hormone with associated was of PDGF-CC expression low whereas (EGFR), receptor factor growth epi dermal and 5/6 (CK) cytokeratin markers basal-like the with relation Pdgfc from a tumor isolated in features a MMTV-PyMT; luminal with cells carcinoma mammary PeRo-Lum1 of sensitivity the reduced cantly (ref. CAF2 line cell CAF breast immortalized the with conditioned medium in Culture tumors. of breast features basal-like with was associated PDGF-CC by signaling epithelium-derived crine We next investigated the mechanisms molecular through which - para cells by PDGF-CC determines the molecular subtype of breast tumor A paracrine signaling circuit in stromal fibroblasts established patients. human of in cohorts tumors breast receptor–negative and hormone PDGF-CC of expression between association close a revealed have we together, of triple-negative tumors were 71.9% positive ( whereas epithelium, tumor primary the in positive PDGF-CC were for cancers breast of subgroup A luminal the of tumors ( distant metastases asynchronous and metastases node lymph synchronous established in also was relationship the which in cohort, Lund the in and PDGF-CC between hormone receptors was corroborated further 9 Fig. Supplementary Zurich the ( in immunostaining using tumors markers breast human of cohort luminal and analy basal-like through of cancer expression of breast sis of subtype molecular the with sion ER functional by acterized of of expression absence the between association an to point strongly approaches, of set diverse ( naling expression were part of data sets related to inhibition of ER of inhibition to related sets data of part were expression to correlated positively were that genes whereas genes, get related inversely were to that genes accordance, In GSEA. to TCGA of expression the to relation 7 Fig. PDGF-CC PDGF-CC epithelium tion tion of functional ER manifesta demonstrating progression, cell related cycle to decreased FOXA1 of of Correlation coefficientsinboldarestatisticallysignificant. Zurich Cohort Table 2 A  statistically significantly upregulated genes (fold change > 1.4) was was 1.4) > change (fold genes upregulated significantly of list statistically A analysis. transcriptional global performed and PDGF-CC therapy.with cells CAF2 to Next, we endocrine stimulated resistance confer fibroblasts from stromal factors that secreted suggests finding

Next, we further investigated the association of PDGF-CC expres of PDGF-CC association the investigated we Next, further Pdgfc s e l c i t r PDGFC −/− ). Finally, we also submitted a list of genes ranked by their cor by their ranked of genes a list submitted we also Finally, ).

Supplementary Fig. 8 Fig. Supplementary −/− mouse to tamoxifen-induced growth arrest ( arrest growth tamoxifen-induced to mouse Spearman’s rank correlations of PDGF-CC expression and FOXA1 expression with biomarkers for molecular subtypes in the expression were part of data sets identifying ER identifying sets data of part were expression breast cancer cells induced expression of a gene profile profile gene a of expression induced cells cancer breast Fig. 2l Fig. – n Correlation Correlation coefficient P n P Correlation Correlation coefficient α n value value signaling in PeRo-Lum1 cells ( and and ). Strikingly, PDGF-CC showed a strong cor strong a showed PDGF-CC Strikingly, ). α and PgR ( PgR and α Supplementary Table 5 Table Supplementary signaling. PDGFC PDGFC ). Jointly, our analyses, which used a used which analyses, our Jointly, ). Table and a luminal phenotype char phenotype a luminal and Supplementary TableSupplementary 6 in breast tumors included in in included tumors breast in 2 ). The inverse correlation correlation inverse The ). − FOXA1 0.002 1.000 0.180 315 288 NA ). Only 7.8% of of 7.8% Only ). Supplementary Supplementary Fig. 3 Fig. Table 2 3 1 ) signifi ) ). ). Taken PDGFC a ). This This ). α α -tar sig and and 0.0001 0.0001 − 0.454 ------0.485 ERα 285 856 of of features luminal-like the suppressed and STC1 substantially IGFBP3 to that of of that to and other each to correlated modestly, albeit significantly, all were ciated with the basal-like phenotype basal-like the with ciated tor alone affected the expression of luminal markers encoded by by encoded markers luminal Esr1 of expression the affected alone tor and and of expression the observations, our with line in and contrast, By tumors. human in regulation their in commonality also and and ( stroma tumor the in STC1 and IGFBP3 HGF, of expression the between correlation close a displayed mens speci cancer breast triple-negative 43 of cohort a result, this of port genes marker–encoding TCGA cohort, the levels of expression of of expression of levels the cohort, TCGA ( immunostaining phenotype ( phenotype ER basal-like, a with and another one with markers these of each of expression the of correlation close a demonstrated STC1 stromal and HGF, IGFBP3 stromal stromal PDGF-CC, epithelial for and a reduced sensitivity to tamoxifen. to sensitivity reduced a and ER of lack with denoted phenotype cell malignant a to leading tumors, luminal from CAFs breast in STC1 ing through PDGF-CC stimulates the expression of HGF, IGFBP3 and type into an ER an into type with HGF, IGFBP3 and STC1 converted their ER their converted STC1 and HGF, IGFBP3 with ( the sensitivity of PeRo-Lum1 cells to tamoxifen-induced growth arrest HGF,factors of the CAF-derived and cocktail IGFBP3 STC1 reduced of h 48 after stimulation ( cells CAF2 in PDGF-CC by induction robust ensure ( 3 protein binding factor selection of 58 human in breast the selection tumors Tam2Yincluded cohort representative a of scoring and immunostaining Similarly, shown). expression indicates distinct mechanisms of gene regulation (data not in correlation close of lack a but epithelium, tumor the by expressed from MMTV-PyMT; MMTV-PyMT; from tumors of stroma the populating CAFs by STC1 and HGF,IGFBP3 niocalcin 1 ( niocalcin genes, namely upregulated of stan The tamoxifen. most significantly action the to cells cancer breast luminal of sensitivity the regulate to ability their for tested were PDGF-CC, by expres induced which highly was for sion proteins, secreted candidate and PDGF-CC, effect of the mediated factors to which used concerning was hypotheses genes generate of list resulting The region’. ‘Extracellular and space’ ‘Extracellular terms ontology gene for filtered and generated Fig. 3 Fig. Pdgfc IGFBP3 , , Supplementary Table 7 Table Supplementary Foxa1 b −/− ). We also assessed whether stimulation of PeRo-Lum1 cells cells PeRo-Lum1 of stimulation whether We assessed ). also FOXC1 mammary carcinoma cells ( cells carcinoma mammary and and Supplementary Table 8 Supplementary were inversely correlated with expression of luminal the expression with correlated were inversely Stc1 Supplementary Fig. 10 Supplementary 0.0001 0.0001 − advance online publication online advance 0.299 0.301 ), hepatocyte growth ), factor ( hepatocyte Gata3 α 309 854 PgR , which encodes a transcriptional regulator asso regulator transcriptional a encodes which , Fig. Fig. 3f -negative phenotype. Indeed, although each fac each although Indeed, phenotype. -negative Pdgfc to some degree, the concerted action of HGF, action concerted the degree, some to – FOXA1 Igfbp3 h ). ). Reassuringly, in breast tumors from the +/+ ). All three factors were also variably variably also were factors three All ). mice was further confirmed through through confirmed further was mice ), were validated using qRT-PCR to to qRT-PCR using validated were ), , , ESR1 0.0001 0.0001 − CK5/6 0.331 ). In summary, paracrine signal ). In paracrine summary, ). ). Strikingly, pretreatment with a 0.246 310 860 3 2 and and ( α Fig. 3c Fig. Fig. 3 Fig. and other luminal markers markers luminal other and Supplementary Fig. 11a Fig. Supplementary PDGFC GATA3 Hgf – i ), suggesting there is is there suggesting ),

e ) ) and growth insulin ). The expression of expression The ). nature medicine nature , , α HGF ( -positive pheno -positive Fig. 3 Fig. PDGFC 0.0002 and and 0.303 EGFR 854 NA NA NA α i -negative -negative ). In sup In ). IGFBP3 , , HGF – 3 c 3 ------

© 2018 Nature America, Inc., part of Springer Nature. All rights reserved. correction. correction. P PeRo-Lum1 cells in the presence of combinations of STC1, HGF and IGFBP3. IGFBP3. and HGF STC1, of combinations of presence the in cells PeRo-Lum1 gene gene correction. ( correction. P independent experiments is shown). ** shown). is experiments independent of s.e.m.) (with average (the IGFBP3 and HGF STC1, or (control) nothing by conditioned medium in 4-hydroxytamoxifen of concentrations correlation coefficient and and coefficient correlation of (out cohort TCGA the in included tumor) one represents dot red (each carcinomas breast subtype luminal in genes two the of expression the depicts factors denoting luminal-like ( luminal-like denoting factors 5 5 independent times, and representative images from from images representative and times, independent TC medium versus CAF medium at each tamoxifen concentration (1 (1 concentration tamoxifen each at medium CAF versus medium TC of tissue sections from tumors from 14-week-old MMTV-PyMT mice for STC1 ( STC1 for mice MMTV-PyMT 14-week-old from tumors from sections tissue of of PeRo-Lum1 luminal breast cancer cells derived from MMTV-PyMT; from derived cells cancer breast luminal PeRo-Lum1 of in medium conditioned by tumor cells (TC) or CAFs (the average (with s.e.m.) of of s.e.m.) (with average (the CAFs or (TC) cells tumor by conditioned medium in Figure 3 Figure nature medicine nature values from Bonferroni Bonferroni from values = 0.003; 2 = 0.003; µ M, M, L19 P n = 0.005); two-way ANOVA with with ANOVA two-way = 0.005);

= 1,086 independent tumors); the black line represents the best fit through the data. The cross-section of two genes to the right states the the states right the to genes two of cross-section The data. the through fit best the represents line black the tumors); independent = 1,086 ) of ) of CAF-derived factors whose expression is induced by PDGF-CC reduce the sensitivity of breast tumor cells to endocrine therapy. ( therapy. endocrine to cells tumor breast of sensitivity the reduce PDGF-CC by induced is expression whose factors CAF-derived h g c a Gata3 f c – Foxa1

IGFBP3 Expression (% of L19) µ e

0.005 0.010 0.015 0.020 0.025 Cell viability (a.u.) M, M,

IGFBP3 HGF STC1 STC1 ) Results from qRT-PCR analyses showing the average expression (with s.e.m., depicted as the percentage of expression of the reference reference the of expression of percentage the as depicted s.e.m., (with expression average the showing qRT-PCR from analyses ) Results HGF 0.0 0.0 0.5 1.0 : : * P ( = 0.0003; 3 = 0.0003; P

c = 0.02, ** = 0.02, , , advance online publication online advance 1 0 n = 3 independent experiments), experiments), = 3 independent post hoc post Tamoxifen concentration ( Tamoxifen concentration + P * *** *** for the two genes. genes. two the for MMTV-PyMT FOXA1 + P µ correction. correction. = 0.007, *** = 0.007, M, M, + + + + , , P ESR1 P * + + + + + = 0.00008; 4 = 0.00008; *** < 0.01, *** < 0.01, P values from Bonferroni Bonferroni from values , , *** GATA3 Esr1 Foxa1 ** µM) P = 0.001; one-way ANOVA with with ANOVA one-way = 0.001; 100 µm 100 µm : : * *** + ** n ) or basal-like ( basal-like ) or 5432 P P = 5 tumors are shown. Scale bars, 100 100 bars, Scale shown. are = 5 tumors < 0.001; control versus STC1 + HGF + IGFBP3 at each tamoxifen concentration (1 (1 concentration tamoxifen each at + IGFBP3 + HGF STC1 versus control < 0.001; = 0.01, ** = 0.01, CAF medium TC medium µ d M, M, Esr1 IGFBP3 Expression (% of L19) STC1 0.02 0.04 0.06 0.08 P i HGF 0.0 = 0.0002; 5 = 0.0002; ( d , , FOXA1 n P post hoc post = 3 independent experiments) or or experiments) = 3 independent = 0.005, *** = 0.005, FOXC1, PDGFC FOXC1, µ M, M, Pdgfc + P b 7 ×10 + = 0.0001; 2 = 0.0001; correction. ( correction. r =0.54 µ

ESR1 Cell viability (a.u.) −/− P = M, M, 0.0 0.5 1.0 mice in the presence of increasing concentrations of 4-hydroxytamoxifen 4-hydroxytamoxifen of concentrations increasing of presence the in mice + + + + f –42 Foxa1 ), HGF ( HGF ), P n = 0.005); two-way ANOVA with with ANOVA two-way = 0.005); P P = 6 independent experiments is shown). ** shown). is experiments = 6 independent + + + + + * = 0.001; one-way ANOVA with with ANOVA one-way = 0.001; values from Bonferroni Bonferroni from values ) molecular subtype. The cross-section of two genes to the left left the to genes two of cross-section The subtype. ) molecular 1 0 3×10 3 ×10 GATA3 r =0.28 r =0.44 : : * Tamoxifen concentration(µM) P = P = g b P µ ) or IGFBP3 ( IGFBP3 ) or –27 ) Viability of PeRo-Lum1 cells in the presence of increasing increasing of presence the in cells PeRo-Lum1 of ) Viability –11 = 0.03, ** = 0.03, ** M, M, Esr1 ** P 1×10 µ *** 9×10 r =–0.18 r =–0.27 r =–0.21 2 ×10 = 0.00001; 3 = 0.00001; * *** *** FOXC1 + m. ( m. P = P = P = –10 –5 –7 i ) Pearson’s correlation analysis of genes encoding encoding genes of analysis correlation ) Pearson’s P e IGFBP3 Gata3 = 0.002, *** = 0.002, h Expression STC1 ). Immunostaining was performed performed was Immunostaining ). HGF P =0.023 P =0.008 r =–0.21 r =–0.11 r =–0.10 6 ×10 5 ×10 PDGFC

r =0.25 (% of L19) *** P = 0 2 4 6 P = post hoc post ( e –7 –9 µ , , 5432 M, M, ** n = 2 independent experiments) in experiments) = 2 independent P 5×10 4×10 2×10 4 ×10 r =–0.26 r =–0.31 r =–0.26 IGFBP3 + 1 ×10 P r =0.42 r =0.37 P values from Bonferroni Bonferroni from values correction. ( correction. P STC1 +HGFIGFBP3 Control = 0.00001; 4 = 0.00001; P = P = values from Bonferroni Bonferroni from values P = P = P = = 0.0003; one-way ANOVA with with ANOVA one-way = 0.0003; + –24 –10 –13 –8 –9 + + + + 7×10 8 ×10 3 ×10 1 ×10 2×10 2×10 r =–0.35 r =–0.18 r =–0.24 r =0.33 r =0.45 r =0.42 HGF P = P = P = P = P = P = P + + + + + * < 0.01, *** < 0.01, f –17 –28 –15 –24 – –8 –5 h µ ) Immunostaining ) Immunostaining M, M, s e l c i t r a P =0.037 P =0.457 P =0.517 P =0.276 P =0.933 r =–0.03 r =–0.03 9 ×10 1 ×10 r =0.05 r =0.00 r =0.09 r =0.21 r =0.19 ** Gata3 STC1 P P = P = = 0.00001; = 0.00001; n a *** –6 –6 = 3 post hoc post ) Viability ) Viability n + µ P post hoc post = 6 M, M, < 0.001; < 0.001;

 © 2018 Nature America, Inc., part of Springer Nature. All rights reserved. used clinically to mandate treatment with endocrine therapy endocrine with treatment mandate to clinically used α ER showing sample a in nuclei tumor of 1% as defined push 3 of 8 of (37.5%) PDXthe basal-like tumors above the threshold, ( cells 12a Fig. malignant of nests differentiated in focally occurred rather but uniform not was blockade PDGF-CC following ER of upregulation MMTV-PyMT; from tumors transplanted orthotopically bearing mice FVB/N WT basal-like subtype ( subtype basal-like within within by ER conferred sensitization functional revealing dently - evi initiation, treatment since by 624% grown had that tumors with reduced by an average of 17%, whereas control-treated mice presented a had was that volume signaling PDGF-CC of paracrine devoid mice ER of expression their with agreement in respectively), 30%, than partial more or complete a exhibited mice of or tumor volume by shrinkage regression as full (defined response treated the of 7) of (5 71% and tamoxifen, with treatment upon retarded severely was tumors to the progressive growth of all all of growth progressive the to ( mice control-treated of from that tumors Fig. 12e Fig. ing treatment with the aromatase inhibitor letrozole ( follow brought about a in of delay growth MDA-MB-231 xenografts ( tamoxifen of action the to sensitivity invoked indeed PDGF-CC of ER of induction the that ing tamoxifen therapy in combination with the 6B3 antibody, demonstrat following tumors MDA-MB-231 in diminished cells cancer positive ER ER Genetic or pharmacological targeting of PDGF-CC sensitizes A  ER resistant previously to therapy endocrine to sensitivity convey would PDGF-CC targeting whether investigated Wetherefore subtype. luminal the of tumors breast feature of clinical distinguishing important most the is tamoxifen, as CC in establishing a lack of ER of lack a establishing in CC PDGF- through signaling paracrine for role the this corroborated result immunostaining; through determined as 6B3, with treatment of ER levels meaningful express not did originally which tumors, MDA-MB-231 NOD cer cell line MDA-MB-231 into orthotopically immunocompromised (PDX) and 12.58 breast as 14.32 can xenografts well as the basal-like action of tamoxifen, we implanted the patient-derived triple-negative the to subtype basal-like of the tumors of breast sensitization for CC body was unable to influence the growth of fully established 14.32 14.32 ( tumors or MDA-MB-231 established fully of growth the anti influence to control unable a was body with together tamoxifen with treatment expected, MMTV-PyMT; MMTV-PyMT; late stage of malignant development, the tumors of tamoxifen-treated were tumors readily palpable. As which expected, due to at the lack of time ER the from starting tamoxifen with daily like like tumors for sensitization to endocrine therapies, such as tamoxifen in good agreement with conversion to the ER the to conversion with agreement good in 14.32 and MDA-MB-231 tumors ( tumors MDA-MB-231 and 14.32 both of retardation growth significant to led 6B3 and administration tamoxifen of combined PDGF-CC, of neutralization upon seen i. 4 Fig.

To conclusively demonstrate the utility of agents targeting PDGF- targeting agents of utility the To demonstrate conclusively s e l c i t r α α a ( expression, which confers sensitivity to endocrine therapy such therapy to endocrine sensitivity confers which expression, -negative breast tumors to hormone therapy Fig. Fig. 4 scid Pdgfc f ). Similarly, pharmacological blockade of PDGF-CC also also PDGF-CC of blockade pharmacological Similarly, ). , – f ), suggesting that the strategy is widely applicable in basal- in applicable widely is strategy the that suggesting ), d gamma mice. Indeed, analysis revealed that in 12.58 and and 12.58 in that revealed analysis Indeed, mice. gamma ). Notably, treatment with 6B3 was sufficient in itself to to itself in sufficient was 6B3 with treatment Notably, ). α b −/− ). ). At the end of tumors from the tamoxifen-treated trial, , expression of ER , expression breast tumors ( tumors breast Pdgfc Pdgfc α Fig. 4c Fig. expression in 12.58 and MDA-MB-231 tumors tumors MDA-MB-231 and 12.58 in expression +/+ +/+ or MMTV-PyMT; mice continued to grow at a rate similar to to similar rate a at grow to continued mice Fig. 4g Fig. – f and and α expression caused by neutralization neutralization by caused expression α Fig. 4 Fig. α Pdgfc was substantially upregulated upon upregulated substantially was expression in breast tumors of the the of tumors breast in expression , Supplementary Fig. 12a Fig. Supplementary h α Fig. 4i Fig. ). In contrast with this result and result this with ). In contrast -negative breast tumors tumors breast -negative b +/+ ). tumors, growth of of growth tumors, , j Fig. 4 Fig. Pdgfc ). The frequency of ER of frequency The ). α −/− -positive phenotype phenotype -positive a ). In sharp contrast contrast ). In sharp α mice were treated expression at this Supplementary Supplementary Supplementary Supplementary α positivity, positivity, signaling signaling – Pdgfc d in vivo in ). The The ). 3 4 . As . −/− α - - - - - .

normal mammary gland mammary normal the in respectively, cells, stem epithelial breast and cells progenitor cells have been thought to different cells signify of i.e., origin, luminal options. treatment improved of need in group patient a for opportunities therapeutic new offer and cancer of breast subtype regulators of CAFs as the identify functional molecular unexpectedly ER of expression the inducing through apy ther of endocrine action to the tumors breast resistant of previously sensitization prompted means, pharmacological or genetic through ( STC1 and secrete IGFBP3 to HGF, induced are that CAFs of activation through phenotype mice exhibited constitutive ER constitutive exhibited mice through regulation of FOXA1, ER of FOXA1, regulation through microenvironment tumor the of control under is subtype basal-like or luminal the for specification that demonstrates which study, our options for a large patient group in need of treatment improved therapy.effective and As such, established of use enable would subtype nal to a cancer lumi breast of receptor–negative a from hormone switch breast basal-like tumors, strategies for pharmacological the induction in alterations or epigenetic genetic targeting for drugs new searching tunities for tunities with through interference targeting paracrine therapeutic Interestingly, cell cultures that were isolated from tumors in from tumors that were isolated cultures Interestingly, cell breast carcinomas breast basal-like and luminal of source common a as proposed been have cells progenitor luminal studies, functional and analysis expression signaling between malignant cells and CAFs. and cells malignant between signaling cating cating a common progenitor cell origin efficiently to occur demonstrated been has cells tumor basal-like and luminal between Interconversion tumors. basal-like or luminal of definition the in heterogeneity and plasticity of degree overt cancer cells was only found PDGFR- in of 20% expression of as cases cohort, Lund the in carcinomas mary PDGFR-α in malignant cells is not a universal feature of human mam into a luminal state can result from epigenetic reprogramming epigenetic from result can state luminal a into ESR1 of expression through achieved be can cancer breast luminal to like lium-derived PDGF-CC orchestrates specification of an ER epithe tumor which in microenvironment, tumor breast basal-like We have revealed a paracrine signaling network that manifested in the DISCUSSION by PDGF-CC. signaling paracrine of blockade through state therapy endocrine as a consequence of conversion to an ER of ER sensitization functional establish thus ies suggesting profound microenvironmental regulation microenvironmental profound suggesting ducts, milk the or pad fat mammary the i.e., sites, anatomical ferent dif at but PDX luminal same the from grafted lesions in ER expression differential with tumors of development demonstrated study recent a the work, our of with line In programming cancer. breast of epigenetic subtype the molecular of nature the to as clues tant impor may hold tools Such to tamoxifen. sensitivity retaining while cancer breast in transition epithelial-to-mesenchymal the undergone have feature of malignant cells that have properties of stem cells and/or that to to mice implanted with clone the parental ( of expression ER with xenografted mice Conversely, action. of mechanisms diverse with letrozole, and The distinguishing features of luminal and basal-like breast cancer breast cancer features of and basal-like luminal The distinguishing Recent studies have suggested that PDGFR- that have suggested studies Recent , , FOXA1 45 , 4 6 . However, we have previously shown that expression of of expression that shown previously have we However, . Pdgfc or advance online publication online advance GATA3 37– did not respond to tamoxifen therapy in contrast in contrast therapy to not tamoxifen did respond α 4 0 -expressing luminal MCF7 cells with ectopic ectopic with cells MCF7 luminal -expressing . Moreover, a phenotypic switch from basal- from switch phenotypic a Moreover, . , whereas a transition from a basal-like state a from basal-like a transition , whereas 3 5 . However, recent studies describe a high high a describe studies recent However, . i. 4 Fig. α signaling and a luminal phenotype phenotype luminal a and signaling l α ). Blockade of PDGF-CC, either either PDGF-CC, of Blockade ). and GATA3, oppor new reveals 3 6 . . Indeed, on the basis of gene Fig. Fig. 4 α ( 4 α Fig. 4 Fig. 7 α

expression may be a may be expression . In contrast, virtually k -negative tumors to tumors -negative nature medicine nature ). ). Jointly, our stud l ). Our findings findings Our ). 4 4 in in vitro . Rather than than Rather . α α -negative -positive -positive Pdgfc , , indi 41– α −/− in in 4 α 3 ------.

© 2018 Nature America, Inc., part of Springer Nature. All rights reserved. n or a vector driving driving a vector or vector empty an carrying lentivirus with infected were that tumors MCF7 transplanted orthotopically carrying mice in s.e.m.) (with volume tumor average 6B3 in combination with oil vehicle ( vehicle oil with combination in 6B3 antibody PDGF-CC neutralizing the with treated were that tumors MDA-MB-231 transplanted orthotopically carrying mice in s.e.m.) (with volume tumor combination with vehicle ( vehicle with combination in 6B3 antibody PDGF-CC neutralizing the with treated were that 14.32 PDX triple-negative transplanted orthotopically carrying mice in s.e.m.) (with tumors that were treated with control IgG in combination with oil vehicle ( vehicle oil with combination in IgG control with treated were that tumors vehicle ( vehicle with combination in IgG2a control with treated were that 14.32 PDX triple-negative transplanted orthotopically carrying mice in s.e.m.) (with volume ( 40 bars, Scale shown. are of FVB/n mice transplanted orthotopically with tumors derived from 14-week-old MMTV-PyMT; 14-week-old from derived tumors with orthotopically transplanted mice FVB/n of oil vehicle ( vehicle oil and CAFs that is mediated through PDGF-CC results in specification of the molecular subtype and regulation of sensitivity to endocrine therapy. endocrine to sensitivity of regulation and subtype molecular the of specification in results PDGF-CC through mediated is that CAFs and negative PDX 12.58 from mice treated with control IgG2a or 6B3. Immunostaining was performed performed was Immunostaining 6B3. or IgG2a control with treated mice from 12.58 PDX negative or tamoxifen ( tamoxifen or Figure 4 Figure Student’s Student’s depicted. ** depicted. in MDA-MB-231 tumors from mice treated with control IgG2a or 6B3 in combination with oil vehicle ( vehicle oil with combination in 6B3 or IgG2a control with treated mice from tumors MDA-MB-231 in nature medicine nature n = 10 mice; mice; = 10 = 7 IgG2a-treated and 8 6B3-treated mice). *** mice). 8 6B3-treated and = 7 IgG2a-treated g e a j Tumor volume (mm3)

Tumor volume (fold) Treatment 1,000

Tumor volume (fold) 10 15 0 5

200 400 600 800 n ER -positive cell fraction (%) 20 40 60 α Genetic or pharmacological targeting of PDGF-CC induces expression of ER of expression induces PDGF-CC of targeting pharmacological or Genetic 0 = 10 mice) or tamoxifen ( tamoxifen or mice) = 10 0.5 1.5 t 0 -test. ( -test. n 2 0 1 01 20 15 10 5 0 P = 7 Pdgfc = 0.002; two-sided, unpaired, equal variance Student’s Student’s variance equal unpaired, two-sided, = 0.002; n 02 30 20 10 0 IgG2a +Tamoxifen IgG2a +Oil Time aftertherapyinitiation(d) = 12 IgG2a-treated and 12 6B3-treated mice). ** mice). 6B3-treated 12 and IgG2a-treated = 12 02 30 20 10 0 6B3 +Tamoxifen 6B3 +Oil Time aftertherapyinitiation(d) Tamoxifen Oil Time aftertherapyinitiation(d) g2 6B3 IgG2a l Pdgfc ) Schematic of the paracrine action of PDGF-CC in the breast tumor microenvironment. An active crosstalk between malignant cells cells malignant between crosstalk active An microenvironment. tumor breast the in PDGF-CC of action paracrine the of ) Schematic Pdgfc vector, vector, MMTV-PyMT; Pdgfc

MDA-MB-231 advance online publication online advance PDX 14.32 +/+ expression and were treated with oil vehicle (empty vector, vector, (empty vehicle oil with treated were and expression mice and 6 and mice n µ n = 10 mice) or tamoxifen ( tamoxifen or mice) = 10 m. ( m. *** = 8 mice). * = 8 mice). e ) Quantification of the average ER average the of ) Quantification +/+ n n = 10 mice). ( mice). = 10 = 12 mice) or tamoxifen ( tamoxifen or mice) = 12 Pdgfc P * = 0.029; empty vector + tamoxifen versus Pdgfc versus + tamoxifen vector empty = 0.029; −/− mice) or tamoxifen ( tamoxifen or mice)

3 k

Tumor volume (mm ) Treatment f P h h 100 200 300 400

= 7 × 10 ERα-positive cell count ) The average tumor volume (with s.e.m.) in mice carrying orthotopically transplanted MDA-MB-231 MDA-MB-231 transplanted orthotopically carrying mice in s.e.m.) (with volume tumor average ) The b

100 150 3

0 Tumor volume (fold) n

50 Tumor volume (mm ) = 10 mice). * mice). = 10 0 02 04 50 40 30 20 10 0 20 40 60 200 400 600 0

0 015 10 5 0 Pdgfc vector+Tamoxifen Pdgfc vector+Vehicle Empty vector+Tamoxifen Empty vector+Vehicle IgG2a Time aftertherapyinitiation(d) 02 30 20 10 0 −7 i a i Tam Oil Tam Oil Time aftertherapyinitiation(d) IgG2a +Tamoxifen IgG2a +Oil ; ; Tamoxifen Oil Time aftertherapyinitiation(d) n χ P = 13 mice). * mice). = 13 2 α = 0.003; two-sided, unpaired, equal variance Student’s Student’s variance equal unpaired, two-sided, = 0.003; n MMTV-PyMT; Pdgfc test. ( test. expression (with s.e.m.) in PDX 12.58 from mice treated with control IgG2a or 6B3 6B3 or IgG2a control with treated mice from 12.58 PDX in s.e.m.) (with expression = 6 IgG2a P MDA-MB-231 t ** MCF7 = 0.03; two-sided, unpaired, equal variance Student’s Student’s variance equal unpaired, two-sided, = 0.03; -test. ( -test. Pdgfc n f ) Quantification of the average number of ER of number average the of ) Quantification = 10 mice) or tamoxifen ( tamoxifen or mice) = 10 c 6B3 , +/+ d P ) Immunostaining for ER for ) Immunostaining = 0.025; two-sided, unpaired, equal variance Student’s Student’s variance equal unpaired, two-sided, = 0.025; mice and 7 and mice n –/– α = 9 mice; = 9 mice; and sensitizes tumors to endocrine therapy. ( therapy. endocrine to tumors sensitizes and 6B3 * 20 ** vector + tamoxifen; two-sided, unpaired, equal variance, variance, equal unpaired, two-sided, + tamoxifen; vector Endocrine treatmentresponsive Pdgfc Pdgfc Pdgfc i

n Tumor volume (fold) l d c Tumor cells = 3 independent times, and representative images images representative and times, = 3 independent 10 15 −/− +/+ 0 5 n 6B3 IgG2a vector, vector, n = 12 mice). ( mice). = 12 mice). Average tumor volume (with s.e.m.) is s.e.m.) (with volume tumor Average mice). ( = 10 IgG2a-treated and 12 6B3-treated mice) mice) 6B3-treated 12 and IgG2a-treated = 10 a PDGF-CC Luminal subtype ) or MMTV-PyMT;) or α 6B3 +Tamoxifen 6B3 +Oil 02 30 20 10 0 (brown) of tissue sections from the triple- the from sections tissue of (brown) ERα positive Time aftertherapyinitiation(d) n = 10 mice) or tamoxifen (empty vector, vector, (empty tamoxifen or mice) = 10 PDX 14.32 i ) The average tumor volume volume tumor average ) The CAFs PDX 12.58 α ERα -expressing cells (with s.e.m.) s.e.m.) (with cells -expressing PDGF-CC Endocrine treatmentunresponsive ERα negative Basal subtype Pdgfc t STC1 IGFBP3 HGF -test. ( -test. t -test. ( -test. −/− Tumor cells g a * ( ) The average tumor tumor average ) The s e l c i t r a , b b j ) mice with with ) mice ) The average average ) The ) ) Treatment t -test. ( -test. k ) The ) The 

© 2018 Nature America, Inc., part of Springer Nature. All rights reserved. iogenic therapy and to immune-checkpoint inhibitors immune-checkpoint to and therapy iogenic blockade of CAF recruitment sensitized pancreatic tumors to anti-ang tumorigenesis, aggravating despite sometimes Nevertheless, . growth tumor inhibit or promote either that subsets harboring population heterogeneous a up make CAFs that likely thus is it described, tions were able to thrive in an manner.unrestricted Given the diverse func expressing CAFs, genetically engineered pancreatic adenocarcinomas the pape the type supporting most, if not all, hallmarks of cancer progression ofcancer hallmarks notall, if most, supporting type and stromaofits intounderlying instigates activation CAFs a pheno initiation cancer repressing through tumorigenesis counteracting as described endocrine endocrine therapy in luminal breast cancer to resistance confer that CAFs CD146-negative of subset a identified study recent a Indeed, abilities. tumor-promoting their of virtue by targets drug as CAFs of utility the supports thus Abundantevidence basal-like breast cancer. breast basal-like with of for patients treatment the evaluation clinical justifies therapy tion of as PDGF-CC an combination to efficacious partner endocrine of ER CAFs as a target promising to pharmacological achieve manifestation lar subtype of breast cancer and distinguishes PDGFR- distinguishes and cancer breast of subtype lar molecu the of specification the in microenvironment tumor the for andpathways signaling sustain that malignant growth. collectively in joint treatment regimens with drugs impinging on multiple cell types carcinomas. Therapeutic targeting of CAFs may thus best be exploited combination partner to endocrine therapy in experimental mammary we found that blockade of signaling through PDGF-CC was useful as a signaling signaling or through eradication of ment and/or activation and/or ment desmoplasia, caution against blockade indiscriminate of CAF recruit severe with associated type cancer a adenocarcinoma, pancreatic of cells to a basal-like cell fate cell basal-like a to cells activation of MET signaling drives commitment of luminal progenitor reported previously reported been not has relationship causal a although established, well already is survival poor and cancers breast receptor–negative hormone with lending support to our findings our to support lending fibroblasts, normal genetically to compared fibroblasts by collagen geting of HGF, IGFBP3 and STC1. Previous studies using gene conditional tar Stimulation of breast CAFs with PDGF-CC induced the expression of the is most prominent signaling route of signal transduction autocrine in human breastnot cancers. and PDGF-CC through signaling of late D. Grabau and provision of the Tam2Y of the Further, we Fernö. by M. D. provision late and cohort Grabau the from assessments pathology with help expert Weacknowledge gratefully online ve Note: Any Supplementary Information and Source Data files are available in the the in available are references, and codes accession statements of including Methods, and data availability any associated M expressing CAFs harbored cohort PDGFR- same the in tumors breast all A 1 for HGF, has also been associated with basal-like breast cancer, in in cancer, with from results agreement our study breast basal-like with associated been also has HGF, for A 0 ck ethods Mesenchymal cells are highly plastic entities. Normal fibroblasts are The work presented here identifies a hitherto unappreciated role role unappreciated hitherto a identifies here presented work The

s e l c i t r n α o in mammary carcinomas. As such, our discovery of neutraliza rsion of the pape r w r α Pdgfra 14 . le and PDGFR- , 55 d , gme 5 6 . However, coevolution of the malignant epithelium epithelium malignant the of coevolution However, . have demonstrated a reduced expression of HGF and nt 5 4 s . . 59 β , , strongly indicating that the paracrine mode , 6 0 5 . Through use of inhibitors of hedgehog hedgehog of inhibitors of use Through . 3 . In the case of IGFBP3, the association association the IGFBP3, of case the In . 4 8 . Expression of MET, the receptor receptor MET,the of Expression . α -smooth -smooth muscle actin (α 49– 5 2 5 . . Intriguingly, constitutive 8 . . However, investigations online version of version online 59 α , 6 -expressing -expressing 0 . Similarly, . -SMA)- 11 , 5 7 ------.

develops develops inhibitory agents to PDGF-CC. study. K.P., U.E. and A.M.S. are shareholders of Paracrine Therapeutics, which application no. which is PCT/EP2016/077295, related to the findings of the current K.P., U.E. and P.R. are named inventors on Patent Cooperation Treaty (PCT) study, managed the study and wrote the manuscript. reagents and conceived the study. K.P. generated and analyzed data, conceived the data and provided exclusive reagents. U.E. analyzed data, provided exclusive provided exclusive reagents. M.H. analyzed data. H.M. L.R., and A.M.S. analyzed J.H. generated and analyzed data. E. Cordero, I.J.G.B. and E.L. generated data. A.O. H.L., G.K., S.J., S.L., J.S., C.L., P.E., C.A., S.R., E. L.H.S., C.O.-P.,Cortez, B.K.H. and P.R. generated and analyzed data and conceived the study. M. M. Bocci, Bartoschek, Australia. Government, Victorian by the provided Program Support Infrastructure Operational the and 10927888 Grant and 1084178 Fellowship (NHMRC) Council Research Medical and Health National from support funding acknowledges A.M.S. Research. for Cancer Institute Ludwig and Institutet Karolinska Society, Cancer Swedish the Council, Research Swedish the from support University.funding Lund U.E. acknowledges and BioCARE network), Linnaeus Council Research Swedish (a consortium STARGET the Society, Cancer Swedish the Council, Research Swedish the 309322), grant project, TUMORGAN (the Council Research European the from K.P.:to Grant agencies a Consolidator following the from by grants supported was herein presented at University.Lund Professor research Grosskopf The K.P. & Birgitta assistance. Göran the P.-O.is and support for statistical Bendahl technical Murone C. for and their D. O’Brien M. Cao, thank to like would 19. 18. 17. 16. 15. 14. 13. 12. 11. 10. 9. 8. 7. 6. 5. 4. 3. 2. 1. claims jurisdictional in published maps and affiliations. institutional reprints/index.htm at online available is information permissions and Reprints COMPE AU T (2001). cancer. Therapy of Early Breast Cancer 2011. Cancer Breast Early of Therapy Primary the on Consensus Expert International Gallen St. the of highlights cancer: a, R. Cao, X. Li, PDGF paracrine of Functions D. Hanahan, & G. Bergers, J., Pahler, K., Pietras, C. Anderberg, Kim, H.M., Jung, W.H. & Koo, J.S. Expression of cancer-associated fibroblast related Augsten, M. Cancer-associated fibroblasts as another polarized cell type of the tumor types cell mesenchymal of subsets Functional K. P.Pietras, Roswall, & E., Cortez, cancer. in Fibroblasts M. Zeisberg, & R. Kalluri, stroma. tumor the with interactions cancer: of Hallmarks A. Ostman, & K. Pietras, Hanahan, D. & Coussens, L.M. Accessories to the crime: functions of cells recruited Prat, A. & Perou, C.M. Deconstructing the molecular portraits of breast cancer. Voduc,K.D. treatment. to biology from cancer: breast Luminal C. Sotiriou, & M. Ignatiadis, A. Prat, A. Goldhirsch, A. Goldhirsch, T. Sørlie, Goldhirsch, A. C.M. Perou, , 369–378 (2009). 369–378 69, promotes tumorgrowthbyrecruitmentofcancer-associated fibroblasts. DF aiy ivle atvto o PDGFR- of activation involves family, PDGF 747–752 (2000). 747–752 PLoS Med. PLoS targeting. pharmacological by revealed stroma tumor proangiogenic the in signaling Med. analysis. immunohistochemical an cancer: breast metastatic in proteins microenvironment. microenvironment. tumor the in Res. Cell Exp. microenvironment. tumor the to subclasses with clinical implications. (2006). of Early Breast Cancer 2013. Cancer Breast Early of Therapy Primary the on Consensus Expert International Gallen St the of highlights (2009). 1319–1329 20, 2009. cancer breast early of therapy primary the on Consensus Expert Oncol. Rev. Clin. Oncol. Clin. Rev. J. Clin. Oncol. Clin. J. 1575–1583 (2002). 1575–1583 Nat. Cell Biol. Cell Nat. NT HOR CO T , 222 (2015). 222 13, DFC s nw rtaeatvtd iad o te PDGF the for ligand protease-activated new a is PDGF-C al. et I , 5–23 (2011). 5–23 5, N Breast lncl mlctos f h itisc oeua sbye o breast of subtypes molecular intrinsic the of implications Clinical al. et nignss tmltd y DFC, nvl ebr n the in member novel a PDGF-CC, by stimulated Angiogenesis al. et G G I ee xrsin atrs f rat acnms itnus tumor distinguish carcinomas breast of patterns expression Gene al. et , e19 (2008). e19 5, Breast cancer subtypes and the risk of local and regional relapse. regional and local of risk the and subtypes cancer Breast al. et NT oeua prris f ua bes tumours. breast human of portraits Molecular al. et

et al. Thresholds for therapies: highlights of the St Gallen International

Personalizing the treatment of women with early breast cancer: breast early with women of treatment the Personalizing al. et l RIBU , 1324–1331 (2010). 1324–1331 316, (Suppl. 2), S26–S35 (2015). S26–S35 2), (Suppl. 24 , 302–309 (2000). 302–309 2, , 1684–1691 (2010). 1684–1691 28, taeis o sbye—eln wt te iest o breast of diversity the with subtypes—dealing for Strategies al. et . . Publisher’s with neutral Nature remains regard to note: Springer advance online publication online advance aarn sgaig y ltltdrvd rwh factor-CC growth platelet-derived by signaling Paracrine al. et

ERES , 494–506 (2013). 494–506 10, Front. Oncol. Front. T T IO S N S Ann. Oncol. Ann. Cancer Cell Cancer Semin. Cancer Biol. Cancer Semin. , 62 (2014). 62 4, Proc. Natl. Acad. Sci. USA Ann. Oncol. Ann. , 2206–2223 (2013). 2206–2223 24, n - and α , 309–322 (2012). 309–322 21, a. e. Cancer Rev. Nat. , 1736–1747 (2011). 1736–1747 22,

, 3–9 (2014). 3–9 25, receptors. αβ

http://www.nature.c nature medicine nature

98, 10869–10874 AE J. FASEB Nature 392–401 6, Cancer Res. Ann. Oncol. Ann. α-receptor. J. Transl. J. 406, Mol. Nat. om/ 16,

© 2018 Nature America, Inc., part of Springer Nature. All rights reserved. 39. 38. 37. 36. 35. 34. 33. 32. 31. 30. 29. 28. 27. 26. 25. 24. 23. 22. 21. 20. nature medicine nature

luminal subtype and good prognosis. good and subtype luminal i, S. Liu, mutation-associated BRCA1 of origin of cell The M.J. Smalley, & G. Molyneux, E. Lim, P.B. Gupta, evolution. and origins cancer: Breast K. Polyak, M.E.Hammond, Chebil, G., Bendahl, P.O., Idvall, I. & Fernö, M. Comparison of immunohistochemical Ray, P.S. Kojima, Y. by reprogramming Cellular E.T. Liu, & W.K. Sung, S.L., Loh, G., Li, S.L., Kong, Neve, R.M. M.A. Thorat, S. Badve, Cancer Genome Atlas, N.; Cancer Genome Atlas Network. Comprehensive molecular H. Ding, E.Y. Lin, expression by tumors mammary Muller,W.J.Guy,of & C.T.,Induction R.D. Cardiff, A.K. Falck, (2013). 93–104 140, in primary breast cancer patients with 10 years’ follow-up. in primary tumours, synchronous lymph node metastases and asynchronous relapses Falck, A.K. J.P.Theurillat, rvs h eouin f uo-rmtn mmay toa myofibroblasts. USA Sci. Acad. stromal Natl. mammary tumor-promoting of evolution the drives alpha signaling. alpha (2007). subtypes. cancer distinct and progesteroneand receptors breastincancer. Pathologistsguidelinerecommendations immunohistochemicalfor estrogentestingof Int. J. Cancer J. Int. Biol. Neoplasia profiling. Biol. expression gene of tale cautionary a cancer: breast cancer—clinical breast primary in receptors discrepancies. for reasons and steroid associations of assay biochemical and Biol. Syst. Mol. ER of action conjoint the tumours. breast human of portraits metastatic for model mouse transgenic a disease. oncogene: T middle polyomavirus of metastases: correlation with CD8 Natl. Acad. Sci. USA Sci. Acad. Natl. in basal-like breast cancer. breast basal-like in Pathol. diseases. human for model reliable a provides model cancer breast mouse subtype A and survival. and A subtype cells in bone marrow in primary breast cancer: a prospective observational study.Cancer observational BMC prospective a cancer: breast primary in marrow bone in cells (2009). carriers. mutation BRCA1 in development tumor basal in populations of cancer cells. cancer of populations in , 2113–2126 (2003). 2113–2126 163, RA rglts ua mmay tmpoeio cl fate. cell stem/progenitor mammary human regulates BRCA1 al. et Aberrant luminal progenitors as the candidate target population for population target candidate the as progenitors luminal Aberrant al. et Mol. Cell. Biol. Cell. Mol. et al. FOXC1 is a potential prognostic biomarker with functional significance A specific requirement for PDGF-C in palate formation and PDGFR- and formation palate in PDGF-C for requirement specific A al. et rgeso t mlgac i te oym mdl T oncoprotein T middle polyoma the in malignancy to Progression al. et et al. Autocrine TGF- et al. Biomarker expression and St Gallen molecular subtype classification OA epeso i bes cne—orlto wt luminal with cancer—correlation breast in expression FOXA1 al. et et al. A collection of breast cancer cell lines for the study of functionally Analysis of and prognostic information from disseminated tumour disseminated from information prognostic and of Analysis al. et tcatc tt tastos ie ie o hntpc equilibrium phenotypic to rise give transitions state Stochastic al. et Forkhead box A1 expression in breast cancer is associated with associated is cancer breast in expression A1 box Forkhead al. et , 403 (2012). 403 12,

, 2411–2417 (2007). 2411–2417 120, NY-ESO-1 protein expression in primary breast carcinoma and carcinoma breast primary in NY-ESO-1expression al. protein et

, 51–55 (2011). 51–55 16, , 526 (2011). 526 7, Nat. Genet. Nat. t al. et

, 20009–20014 (2010). 20009–20014 107, , 1680–1685 (2008). 1680–1685 105, advance online publication online advance mrcn oit o Ciia Oclg/olg O American Oncology/College Clinical Of of Society American

Clin. Cancer Res. Cancer Clin. , 954–961 (1992). 954–961 12, , FOXA1, and GATA3 to a ligand-inducible growth state. growth ligand-inducible GATA3a and to FOXA1, α, Cancer Cell Cancer Cancer Res. Cancer , 1111–1116 (2004). 1111–1116 36, β and stromal cell-derived factor-1 (SDF-1) signaling Cell + T-cell and CD79a Nature , 633–644 (2011). 633–644 146, J. Clin. Pathol. Clin. J. , 515–527 (2006). 515–527 10, , 3870–3876 (2010). 3870–3876 70,

, 4415–4421 (2007). 4415–4421 13, J. Clin.Oncol.J. , 61–70 (2012). 61–70 490, Acta Oncol. Acta J. Clin. Invest. Clin. J. + plasmacytic/B-cell infiltration.

, 327–332 (2008). 327–332 61, , 719–725 (2003). 719–725 42, a. Med. Nat.

Breast Cancer Res. Treat. 28 , 2784–2795, (2010).

. amr Gland Mammary J.

, 3155–3163 117, 907–913 15, Am. J. Am. Proc. Proc.

48. 47. 46. 45. 44. 43. 42. 41. 40. 52. 51. 50. 49. 60. 59. 58. 57. 56. 55. 54. 53.

(2010). Cancer Cell Cancer Sci. USA Sci. cancer. breast basal human and outcome poor with associated cancer stemness and chemoresistance. and stemness cancer Horikawa, S. Horikawa, S. Jansson, F.Meng, W.L.Tam, G. Sflomos, G.M. Bernardo, Yamamoto, S. Y.Su, G. Molyneux, rve, C.R. Graveel, M.G. Ponzo, C.M. Ho-Yen, Y.J.Kim, Rhim, A.D. Rhim, B.C. Özdemir, the orchestrating Brechbuhl, H.M. fibroblasts: Carcinoma-associated T.D. cancer Tlsty, of & architects P. as Gascard, Fibroblasts S.S. McAllister, & K. Pietras, T., Marsh, the of role proposed A cancer? more get we don’t W.C.Why Hines, & M.J. Bissell, Marzec, K.A., Baxter, R.C. & Martin, J.L. Targeting insulin-like growth factor binding S. Gastaldi, ductal adenocarcinoma. ductal target for breast cancer stem cells. stem cancer breast for target cancer. soitd ih ua bsl rat cancer. (2009). 12909–12914 breast basal human with associated (2011). composition of malignancy. of composition pathogenesis. progression. cancer restraining in microenvironment (2015). protein-3 signaling in triple-negative breast cancer. cancer cells. cancer epithelial progenitors and not from basal stem cells. cells. stem basal from not and progenitors epithelial breast cancer. breast basal-like for implications progenitors: luminal mammary of commitment cell-fate Sci. Rep. Sci. (2014). inhibition in triple-negative/basal-like breast cancer. Cancer Cell Cancer induces immunosuppression and accelerates pancreas cancer with reduced survival. pteil irevrnet s eemnn o lmnl hntp ad hormone and phenotype luminal response. of determinant as microenvironment epithelial cancer. breast triple-negative in expression protein increased show 4q12, at spaced closely cancer to estrogen. to cancer the basal-like subtype? basal-like the Somatic cell fusions reveal extensive heterogeneity in basal-like breast basal-like in heterogeneity extensive reveal fusions cell Somatic al. et Cell Rep. Cell PLoS One PLoS and α PDGFR al. et MET is a potential target for use in combination therapy with EGFR with therapy combination in use for target potential a is MET al. et , 12903–12908 (2009). 12903–12908 106, , 17948 (2015). 17948 5, Cancer Cell Cancer rti kns C kinase Protein al. et Stromal elements act to restrain, rather than support, pancreatic support, than rather restrain, to act elements Stromal al. et The three receptor tyrosine kinases c-KIT,PDGFR kinases and tyrosine VEGFR2 receptor three The al. et , 762–777 (2014). 762–777 25, , 719–734 (2014). 719–734 25, A preclinical model for ER for model preclinical A al. et Met signaling regulates growth, repopulating potential and basal and potential repopulating growth, regulates signaling Met al. et e idcs amr tmr wt dvre itlge ad is and histologies diverse with tumors mammary induces Met al. et Oncogene plays a crucial role in connective tissue remodeling. tissue connective in role crucial a plays PDGFRα al. et et al. JARID1B is a luminal lineage-driving oncogene in breast cancer. Biochim. Biophys. Acta Biophys. Biochim. -e i ivsv bes cne: s hr a eainhp with relationship a there is cancer: breast invasive in C-Met al. et e idcs ies mmay acnms n ie n is and mice in carcinomas mammary diverse induces Met al. et Oncogene t al. et Depletion of carcinoma-associated fibroblasts and fibrosis and fibroblasts carcinoma-associated of Depletion al. et OA rpess h mlclr hntp o bsl breast basal of phenotype molecular the represses FOXA1 al. et et al. Fibroblast subtypes regulate responsiveness of luminal breast , 1549–1563 (2015). 1549–1563 11, , e102176 (2014). e102176 9, Clin. Cancer Res. Cancer Clin. BRCA1 basal-like breast cancers originate from luminal luminal from originate cancers breast basal-like BRCA1 , 407–422 (2016). 407–422 29, Cancer , 554–563 (2013). 554–563 32, Cancer Cell Cancer , 1428–1440 (2013). 1428–1440 32, Genes Dev. Genes play critical roles in mediating Foxq1-driven breast Foxq1-driven mediating in roles critical play β , 163–171 (2014). 163–171 120, s cnrl inln nd ad therapeutic and node signaling central a is α Cancer Cell Cancer , 735–747 (2014). 735–747 25,

, 1710–1721 (2017). 1710–1721 23, , 1070–1078 (2013). 1070–1078 1832, Cancer Res. Cancer , 1002–1019 (2016). 1002–1019 30, -positive breast cancer points to the to points cancer breast α-positive rc Nt. cd Si USA Sci. Acad. Natl. Proc. , 347–364 (2013). 347–364 24, BioMed Res. Int. Int. J. Cancer , 584–593 (2015). 584–593 75, Cell Stem Cell Stem Cell a. Med. Nat. s e l c i t r a 134, 2424–2436 rc Nt. Acad. Natl. Proc. 2015, 638526 320–329 17,

7 , 403–417 403–417 ,

106, 11 α, © 2018 Nature America, Inc., part of Springer Nature. All rights reserved. 5 yT64u/ tasei mc hv be dsrbd previously described been have FVB/N-Tg(MMTV- mice purposes. scientific transgenic for animals PyVT)634Mul/J of use for and Code care Australian the the with compliance in conducted were Experiments Lund, application M142/13 or by the Austin Health Animal Ethics Committee. in and applicationN96/11, North, Stockholm in Experimentation Animal for Mice. Cruz Santa (1:100, STC1 sc-30183). and Biotechnology sc-9028) Biotechnology Cruz Santa (1:100, IGFBP3 ab83760), Additionally, Abcam (1:100, HGF for parallel. performed in was immunostaining processed were Dako) Mib-1, clone and (1:20, Ventana) 3C6, Ki-67 clone (prediluted, EGFR UK), Novocastra, 10A7 clone HER2 (1:50, Dako), D5/16B4, clone cocktail (1:25, CK5/6 marker cytokeratin cell basal The PDGF-CC. active of excess an by blocked fully was reactivity immuno Specific system). detection HRP UView pretreatment, for CC1m (2 anti 6B3 body PDGF-CC monoclonal the samples, clinical in expression PDGF-CC arrays. tissue tumor human of analysis Immunohistochemistry specimens. evaluable 43 of consisted and ab214756) (Abcam, commercially obtained was cancers breast triple-negative human of cohort study, of The subset 58 a tumors from was representative analyzed. the cohort and (ref. 1985 1994 between Care Region Health Swedish South the from carcinoma breast 2 stage with patients postmenopausal from obtained Tam2Y the mainly were from cohort specimens The recorded. were 2010 December until events All Office). Statistics (Central Death of Causes of from Register the Swedish was retrieved and the endpoint, data on survival as used was mortality cancer–specific Breast patients. all from obtained was able; patients without follow-up data were not included not were data follow-up without patients able; avail were registry cancer cantonal the from data of follow-up all For patients, these 12-2005. StV of number Hospital reference with University study this Zürichat approved Zurich Canton the of committee ethics medical The consent. informed written gave donors sample Board– and protocols, approved Review Institutional under voluntarily enrolled patients All 1999). July (median, 2004 and 1965 between Zurich Hospital University Pathology, invasive breast at primary cancer who were the diagnosed Institute of Surgical cohorts. Patient ONLINE METHODS nature medicine nature PyMT transgene and the heterozygous or homozygous knockout of the the of knockout homozygous or Pdgfc heterozygous the and transgene PyMT MMTV- the of presence The Laboratory. Jackson The from purchased were follow-up data have previously been reported been have previously data follow-up and data pathological clinical LU247-01); and LU692-99 numbers, reference at board Lund University review ethical regional (the markers ous prognostic vari evaluating trial observational from a cancer prospective, breast primary with patients The Lund includes were samples cohort analyzed. tissue healthy A 4-mm incision under the nipple of the right abdominal mammary gland gland mammary abdominal right the of nipple the under incision 4-mm A procedure. surgical the during isoflurane under maintained and anesthetized the guidelines in ethical surpassed our permit. for survival analysis was taken as the time at which the combined tumor burden os MT-yT F 5 F, MMTV-PyMT: lows: fol as were used pairs primer The purification. and extraction acid nucleic lysis, tissue for protocol common a following biopsies tail or ear either from TTGATGAGAGAT-3 3 caliper.a Tumor using age volumeof was weeks calculated as 12 length at × width measured was mice MMTV-PyMT transgenic in Mouse experiments. mice were used as controls for the engineered genetically mice. littermatecommenced. Female WT were mice studies. in the In cases, used all backcrossed to backgroundFVB/N the for ten generations before analysis was TAGCTAGTCGATACCGTCGA-3 ′ ′ -GGAAAGTCACTAGGAGCAGGG-3 ; mutated mutated ; For transplantation experiments, FVB/N female mice aged 3 weeks were weeks 3 aged mice female FVB/N experiments, transplantation For Mice deficient for deficient Mice allele All mouse experiments were approved by the local Ethical Committees Committees Ethical local approvedwere the by experiments mouse All 2 5 in mice were verified through genotyping. DNA was prepared was DNA genotyping. through verified were mice in µ Pdgfc g/ml) was used on an automated Ventana platform (protocol (protocol platform Ventana automated an on used was g/ml) The Zurich cohort includes tissues from 890 patients with with patients 890 from tissues includes cohort Zurich The F 5 F, : The size of the tumor in each of the ten mammary glands Pdgfc ′ n R 5 R, and ′ -CTCATGTTCTCGTGACTCTGA-3 ′ were originally on the C57Bl/6 background and background C57Bl/6 the on originally were -GGAAGCAAGTACTTCACAAGGG-3 ′ ′ -AGTAGGTGAAATAAGAGGTGAACA- . ′ ; WT WT ; 21 , 2 3 Pdgfc . Written informed consent consent . Written informed 2 F 5 F, : × ( 2 3 π 0 3 / 6). The endpoint . Additionally, 69 69 Additionally, . ). ). For the current ′ -AGCTGACAT ′ n R 5 R, and To assess To assess ′ and R, R, and 2 3 and and ′ ------were implanted with tumor pieces that were 1–2 mm Patient-derived xenografts. Patient-derived ured in sedated mice twice per week using a caliper. with vehicle alone. All therapeutic administrations were open label. vehicle of ethanol and corn oil (Sigma) through heating the mixture to 55 °C or a gavagein daily, oral dissolved was via that Sigma) dose per mg (1 tamoxifen alternately assigned into the treatment groups, in which mice were treated with crine therapy, when a tumor endo was involving palpable (longest trials diameter >3 therapeutic mm), For mice were establishment. tumor of day the from delivered via intraperitoneal injection twice per week (300 clone 6B3) antibody or IgG2a isotype control antibody (Bio X Cell), which was assigned to receive treatment with anti-PDGF-C (mouse monoclonal antibody in orthotopically twice twice per week with a caliper. mice sedated in measured and monitored was Tumorgrowth d. 2 following the forand procedure surgical the of end the at intraperitoneally injected was drug, anti-inflammatory an and pain-killer a Health), Animal Pharma Orion weight; body kg per mg 5 (Rimadyl; Carpofen (Ethicon). filament polyamide with with 0.1% BSA in PBS. with 3% H quenched was activity peroxidase Endogenous HGF). and IGFBP3 (forSTC1, 6; DAKO) in a pressure cooker (for ER (pH buffer citrate in retrieval antigen by followed was rehydrated,which and formaldehyde for 12 h at 4 °C. Paraffin-embedded sections were deparaffinized paraformaldehyde. For paraffin-embedding, organs were postfixed in 4% para 4% by followed PBS with heart-perfused were mice analysis, downstream for Mouse tissue preparation, histology and immunostaining. ment (Ventana Medical Systems Inc.). the Ventana BenchMark Ultra automated immunohistochemical staining counterstainedinstruMayer’s with using haematoxylin.performed was staining The and Inc.) Systems(Ventana Kit Medical Detection DABUniversal UltraView the with incubated were sections immunostaining, the of visualization allow (Ventana Medical Systems Inc.) was applied to forsections 40 min at 36 °C. To Inc., Tuscon, AZ, USA). CONFIRM Ready-to-use anti-ER antibody, clone SP1 for 36 min at 95 °C in cell conditioning 1 (CC1) buffer (VentanaBraunschweig, Medical SystemsGermany), deparaffinized and rehydrated before antigen retrievaltumors weremounted onto (Menzel-Glaser, SuperFrost slides Plusembedded Tumors were harvested and fixed in 10% formalin. 4- (50 mg ch6B3 per kg body weight, three times per week) or vehicle control mouse.the Ten days after implantation,(PBS). receivedmice anti-PDGF-CC therapy pieces were implanted per mouse in two mammary fat pads, one on each side of ment resistant; 14.32: derived from primary tumor, treatment naive). Two tumor sion of PDGF-CC (12.58: from patients with triple-negative breast cancers with documented high expres and measured in sedated mice twice per week with a caliper. Tumor mice. monitored was FVB/N growth WT of gland mammary inguinal mice established in our laboratory were orthotopically intoinjected the fourth that was 2 × 2 mm was inserted. Suturing was performed with 6-0 Ethilon Ethilon 6-0 with performed was Suturing inserted. was mm 2 × 2 was that PyMT; created a pocket where a tumor piece (pieces were derived from either MMTV- (Vector Laboratories) to bound visualize antibodies. primary were used (ABC Elite standard kit, Vector Laboratories) with DAB as substrate appropriate biotinylatedsecondary antibodies and the ABC peroxidase system chamber.the washing, humidified After a in °C at 4 overnight performed was incubation Primary Abcam). ab83760, (1:200, HGF and Biotechnology) Cruz Santa sc-9028, (1:200, IGFBP3 Biotechnology), Cruz Santa sc-30183, (1:200, STC1against antibodies primary of incubation and blocking forthe used was tor ER (Life Technologies) CAS-block and Vectorlabs), MOM kit, diluent. The basic primary antibody Mouse against on estrogen recep (Mouse reagent blocking For all transplantation experiments, tumor growth was monitored and meas Mouse mammary cell lines derived from WT or WT from derived lines cell mammary Mouse ER For therapeutic studies, 2 × 10 α α (1:200, clone 1D5; DAKO) was incubated in MOM diluent. CAS-block staining required subsequent steps using Mouse on Mouse (MOM) Mouse on Mouse using steps subsequent required staining Pdgfc 2 O +/+ 2 in methanol for 10 min at room temperature, followed by washes mice or MMTV-PyMT; or mice scid mice. Prior to tumor inoculation, mice were randomly randomly were mice inoculation, tumor to Prior mice. BRCA2 A total of eight mice (NOD (NOD mice eight of total A mutation, derived from liver metastasis, treat 6 human MDA-MB-231 cells were inoculated α ) ) or in a water bath at 95 °C for 20 min Pdgfc −/− mice and were kept on ice) ice) on kept were and mice µ m sections of the paraffin- 3 in size and were derived Pdgfc µ doi:10.1038/nm.4494 g per week) starting scid −/− To preserve tissues MMTV-PyMT MMTV-PyMT gamma, NSG) NSG) gamma, ------© 2018 Nature America, Inc., part of Springer Nature. All rights reserved. mice mice and five tumors derived from MMTV-PyMT; FAST Mastermix and diluted with RNase-free H PCR PCR array, Qiagen). In brief, cDNA was mixed with 2× RT room temperature. Primary antibodies directed roomagainst temperature. directed PDGFR- antibodies Primary (DAKO)at FreeProtein Block min for>90 Serum using byblocking followed acetone, ice-cold in fixed were sections Frozen (HistoLab). medium perature 30% sucrose at 4 °C overnight, followed by embedding in optimal cutting tem 30 30 s, and ° 72 (3) 95 (1) of: cycles 40 for run Array. was PCR PCR The doi:10.1038/nm.4494 Quantification ofQuantification metastases. scored and samples. the in pathologist blindly a by determined was Necrosis carcinoma. late carcinoma and early adenoma, tissue, hyperplastic tissue, fat mammary of tion propor the for carcinoma’. evaluated Tumors were ‘late-stage mitotic-index ‘early high- dense, to very an invasive, and finally invasion carcinoma’ stromal with cell–dense epithelial more a and to ductal morphology) gland normal some mammary of acinar retention the (with premalignant adenoma by and characterized hyperplasia stage’ ‘precancerous a to tissue fat normal glands transformed from moves of progression stages: area following the of each at the tumors by occupied of quantification through progression of degrees different into classified was mouse) per sections tissue 10 genotype, MMTV-PyMT; MMTV-PyMT; grade. tumor of Assessment (Sigma), dehydrated and mounted in Pertex (Histolab). permeabilization solution in 3 × washing 30 min,After sectionschamber. werehumidified counterstained a in with °C Nuclear 37 Fastat Red night solution 5-bromo-4-chloro-3-indole-6- mg/ml 1 to subjected solutionwereX-Gal further (5mM K 0.04% NP-40, 0.02% deoxycholic acid sodium salt in PBS) 2 × 30 min. Sections into FVB/n mice were run in the qRT-PCR array. gated, 1:200, 12-1401 eBioscience) and PDGFR- ized to that of the housekeeping gene housekeeping the of that to ized normal was expression mRNA qPCR. for used was (KAPA Biosystems) Mix Master using Kit qPCR FAST prepared SYBR KAPA Rad). was (Bio Kit cDNA Synthesis cDNA iScript (Qiagen). Kit Mini RNeasy using isolated was 180 at min 2 for centrifuged and policeman rubber a with PBS in scraped were cultures cell °C; −80 at stored and frozen qRT-PCR. at least 15 tissue per sections lung. 25th every of staining section, the number H&E of metastatic foci (>8 cells Following in diameter) was determined in lung. entire the of sectioning serial ded in paraffin upon tissue fixation. The metastatic burden was assessed through perature for 15 min before washing in permeabilization solution (4mM MgCl b Inc.). Instruments, acquired were Images (Diagnostic software air. advanced SPOT the in using camera RTKE a SPOT using temperature room at Nikon) NA; NA; 0.50 0.75 20×, NA; 40×, 0.30 (10×, objectives Fluor Plan with equipped Nikon) containing mounting media (Vector Laboratories). to bound antibody,visualize primary and weresections mounted using DAPI- used was Technologies) (Life antibody secondary AlexaFluor488-conjugated appropriate The chamber. humidified a in °C 4 at overnight incubated were pairs were used: used: were pairs Acta2 according to the manufacturer’s instructions run (Mouse was Breast genes cancer RT cancer–specific mammary 84 analyzing array qPCR A above. qRT-PCR array. 3 TTCAGCTTGTGGATGTGCTC; ′ -galactosidase staining. -galactosidase and R, 5 For cryopreservation, the primary tumors and lungs of mice were kept in in kept were mice of lungs and tumors primary the cryopreservation, For Imaging was performed using an upright microscope (Eclipse E800; E800; (Eclipse microscope upright an using performed was Imaging , QuantiTect Primer assays (Qiagen) were used. The following primer primer following The used. were (Qiagen) assays Primer QuantiTect , ′ -GGTGGTGGTCTCGACAGTTCG-3 For RNA isolation and preparation, primary tumors were snap- were tumors primary preparation, and isolation RNA For RNA isolation and cDNA synthesis was performed as described Pdgfc C C for 30 s. Five tumors from derived MMTV-PyMT; L19 F 5 F, : +/− and MMTV-PyMT; MMTV-PyMT; and Fresh frozen sections were left to dry at room tem atroom dry to left were sections Freshfrozen ′ GGTGACCTGGATGAGAAGGA-3 - The lungs of MMTV-PyMT mice were embed Tumor tissue from MMTV-PyMT; MMTV-PyMT; from tissue Tumor Gata3 L19 : F, 5 d -galactosidase) and incubated over incubated and -galactosidase) . For . ′ - CAATGCCTGCGGACTCTACC- g 2 so they formed a pellet. RNA pellet. a formed they so Pdgfc α O and added to the RT Foxa1 (1:200, 3169S, Cell Signaling) 3 ′ . Fe(CN) Pdgfc −/− ° , , C for 15 s, (2) 60 °C for °C 60 (2) s, 15 for C Esr1 mice ( mice −/− 6 2 , , 5mM K , , SYBR Green ROX mice transplanted Pdgfra n = 5 mice per per mice 5 = α ′ (PE-conju n R 5 R, and , , 4 Fsp1 Pdgfc Fe(CN) 2 2 Pdgfc Profiler Profiler , and , +/+ +/+ ′ 6 2 ------, , ,

in in CAF2 cells following stimulation with PDGF-CC. ‘Extracellular space’terms: Ontology and ‘Extracellular(Gene proteins region’) secreted encode and to were predicted found genes to of be upregulated list a from selected were cells cancer breast of subtype molecular the on PDGF-CC centered, log In vitro In RNA-seq. and preparation cDNA library to subjected were 9 > number RNA integrity with samples and above, described as extracted was RNA Next, PDGF-CC. of ng/ml 100 with stimulation of PDGF-CC in CAFs, CAF2 starved cells were stimulated for 48 h For Belgium). analysis of transcriptional genes with expression induced by the human fibroblasts conditioned by tumor cells cells tumorof by conditioned fibroblasts human line cell immortalized an is CAF2 Collection. Type Tissue American the mouse, respectively. Human MDA-MB-231 and MCF7 cells were obtained from downloaded January 30, 2015). Using R (v3.1.1), the data were log were data the (v3.1.1), R Using 2015). 30, January ( downloaded subtype molecular to correlation and sion data from 1,086 breast tumors included in TCGA were lines. analyzed cell and tumors for breast human of gene analyses expres expression Gene with 1% penicillin–streptomycin and 10% FBS. maintained in culture in DMEM Glutamax (Invitrogen) that was supplemented obtained from ref. ref. from obtained was lines cell cancer breast 51 for data lines, cell cancer breast in expression as described before. The cells were treated with increasing concentration of concentration increasing with treated were cells The before. described as were stimulated either with CAF-conditioned medium or recombinant factors, For 4-OH-tamoxifen treatment, 15,000 cells were seeded in 96-well plates. plates. 96-well h 24 incubation After growthin medium followed cells bythe h 24 starvation, in seeded were cells 15,000 treatment, 4-OH-tamoxifen For CAF-conditioned medium through incubation for 48 h in medium. starvation produce to used was CAF2 line cell The h. 48 for medium starvation in tors fac combinationsthese orof R&D) ng/ml; (250 IGFBP3 recombinant mouse R&D), ng/ml; (30 HGF mouse recombinant BioVendor), ng/ml; (400 STC1 BSA (Sigma Aldrich). The cells were next stimulated with recombinant human were forstarved 24 h in DMEM Glutamax (Invitrogen) supplemented with 1% formed, and breast cancer subtypes were classified using the PAM50 cen PAM50 the using troids classified were subtypes cancer breast and formed, tion using Vibra-cell (Sonics & Materials, Inc.). The lysate was centrifuged 16,000 at centrifuged was lysate The Inc.). Materials, & (Sonics Vibra-cell using tion min 10 for centrifugation by 16,000 at followed °C 4 at incubation of min 30 After Technologies). (Bertin instrument 24 Precellys a using PhosSTOP) and tor inhibi protease Complete of Roche addition the with SDS 1% 100, Triton-X tion in (50 lysis buffer mM Tris-HCl, pH 7.4, 150 mM NaCl, 5 mM EDTA, 1% Western blot. Cell culture. using measured Innotech). (Alpha was system imaging Q signal FluorChem Luminescence Sciences). Life Healthcare (GE and washed Reagent was Detection Blotting Western membrane Prime ECL Amersham The with developed temperature. room at h 1 for incubated and TBST in BSA 5% in applied were A27036)) goat (Invitrogen, IgG A27014; anti-rabbit (Invitrogen, IgG anti–goat (rabbit antibodies secondary ing, Inc.; anti- Inc.; Santa anti-ER Inc.; sc-6553, Biotechnology Cruz 1:200, (anti-FOXA1, antibodies incu primary and with TBS) °C 4 in at Tween-20 bated (0.1% TBST in BSA 5% with h 1 for blocked was membrane The Technologies). Life 2, (iBlot system blotting dry a filter using nitrocellulose a to for transferred were °C proteins The 96 Technologies). (Life at denaturated Technologies), 10 min by and separated SDS-PAGE (Life on a NuPAGE gel polyacrylamide 4–12% Agent Reducing Sample NuPAGE and Buffer Sample LDS NuPAGE 5× with mixed was suspension protein The the of spectrophotometer. a on nm 562 measurement at and performed was Kit, Assay absorption Protein BCA Pierce the using mined in 12-week-old MMTV-PyMT; 12-week-old in tion routinely. PeRo-Bas1 cells and PeRo-Lum1 cells were derived from tumors 3 × 10 × 3 6 1 after centering the data around the median. For analysis of gene gene of analysis For median. the around data the centering after cell culture assays. culture cell 6 g g PR-Lum1 cells were seeded in culture medium. After 24 h, the cells cells the h, 24 After medium. culture in PR-Lum1 wereseeded cells β , and the pellet was discarded. Protein concentration was deter was concentration Protein discarded. was pellet the and , , the supernatant was collected and further subjected to sonica to subjected further and collected was supernatant the , -actin 1:5,000, ab-8227 Abcam) in 5% BSA in TBST. After wash TBST. After in BSA 5% in Abcam) ab-8227 1:5,000, -actin All cells were tested and found to be negative for mycoplasma infec 2 transformed and plotted using MedCalc v11 (MedCalc Software, A piece of tumor was minced in liquid N in liquid of A minced was tumor piece 2 9 . Probe-level data were merged on Entrez Gene IDs, IDs, Gene Entrez on merged were data Probe-level . Candidate proteins for mediating the effect of effect the mediating for proteins Candidate Pdgfc α , 1:200, sc-542 Santa Cruz Biotechnology Biotechnology Cruz Santa sc-542 , 1:200, +/+ mouse and MMTV-PyMT; and mouse http://cancergenome.nih.gov in vivo in in mice in 2 nature medicine nature before homogeniza before 3 1 . All cells were cells All . RNA-seq Pdgfc 2 trans −/− / ------,

© 2018 Nature America, Inc., part of Springer Nature. All rights reserved. Fastq ( lluminaBasecallsTo and ExtractIlluminaBarcodes each tools Picard for using 2) sample read and 1 (read files fastq two into merged and demultiplexed datawas sequencing Raw (Illumina). instrument NextSeq a usinganalysis seq RNA- to subjected and Illumina), (TruSeq, libraries prepare to templates as quality using a 2100 Bioanalyzer instrument (Agilent; RIN range, for 9.6–10), used checked were repeats individual quintuplicate in prepared Samples h. 24 EtOH was used as vehicle control, and treatment medium was replenished every (Sigma-Aldrich) for 48 h before RNA isolation (RNeasy Mini Kit, 74106, Qiagen). regulated genes with binding sites for the products of method ClosestGene the using scored were and miRNAs) and coding protein = type bio TRUE, = (gencode_basic transcripts gene Ensembl GRCh38 to assigned and the 13,000 genes with expressiondetectable in the experiment ( homology mouse and human with genes of intersection the to confined then was test enrichment ChIP–seq The 1). = homolog_orthology_confidence and ortholog_one2one, on (filtering BiomaRt using genomes mouse and human between homologous were that genes 15,140 extracted we set, data ChIP–seq top up- and downregulated genes observed in the experiment. From the human summed on the gene level and used to examine the enrichment of binding in the distribution of aroundpeaks the transcription start site. Transcript scores were scription start site and scores the onpeak the basis of the empirical cumulative Sleuth (v0.29.0) Sleuth abundance estimation, and differential expression analysis was performed with T47D cells 5 were and medium treatedusing 1×PEST). Cells starvation in (Sigma-Aldrich) for starved subsequently and 24 1×PEST) h in serum-free DMEM and (10-014-CVR, Corning; supplemented FBS with 1% BSA 10% with supplemented were seeded in normal DMEM growth medium for 8 h (10-013-CVR, Corning; Genomic analysis of breast cancer cell lines. assay at day 7. seeding. The cell proliferation reagent WST-1 (Roche) was used for the viability post 6 and 4 day at medium stimulation in Aldrich) (Sigma 4-OH-tamoxifen nature medicine nature cells tially tially upregulated and 1,000 downregulated genes in protein-coding transcript sequences. The resulting lists of the top 1,000 differen ‘–bias’ and ‘–rf-stranded’, and the transcript target was the Gencode release M15 compared to were obtained through the CistromeDB portal sets data ChIP–seq analyses. ChIP–seq from sets data available publicly using lists. The rank of the query gene list divided by the total number of resample of number total the by divided list gene query the of rank The lists. gene list with the highest score = rank 1) together with the results for the query and sum of scores in each resampled gene list was recorded and then ranked (theof 11,300 genes to provide a relevant background. The number of bound genes of randomly sampled genes. The sampled gene lists were drawn from pool the lists 100,000 to comparedwere genes expressed differentially 1,000 top the of genes). The summed score and the number of genes assigned to ChIP–seq peaks µ M estradiol (E2758, Sigma-Aldrich) and/or 5 5 and/or Sigma-Aldrich) (E2758, estradiol M 3 0 (GSM589237, GSM659787 and GSM720422, respectively) and E2-treated https://broadinstit 6 5 . The ClosestGene method assigns a ChIP–seq peak to the closest tran 3 0 (GSM589239, and GSM659795, respectively). ChIP–seq peaks were Pdgfc 6 3 . Kallisto was run with parameters ‘–bootstrap-samples 100’, ‘–bootstrap-samples parameters with run was Kallisto . +/+ PeRo-Bas1 cells were used to investigate the enrichment of ute.github.io/picard / 6 ). Kallisto (v0.43.0) PeRo-Bas1 and PeRo-Lum1 cells 4 and included E2-treated MCF-7 ESR1 Pdgfc µ M 4-hydroxytamoxifen 4-hydroxytamoxifen M , −/− FOXA1 PeRo-Lum1 cells 6 2 was used for and n = 11,300 GATA3 - - - - ,

number lated with PDGF-CC is available through the NCBI GEO database with availability. Data accession is available in the Life Sciences Reporting Summary. stated in each figure legend. six times to ensure reproducible conclusions; the exact number of repetitions is tion phase were excluded from analysis. experience of the authors. No mice entering the experimental and/or interven difference in the primary endpoint 20% of a tumor detect to growth needed rate mice of on number the the basisof estimate of an the using prior chosen was group.size Sample per mice 6 least atbut legend figure each in indicated size (refs. set (BRCA). GSEA was performed using the GseaPreranked tool in GSEA v3.0 67. experiment unless otherwise stated. ered significant. The variance was similar between experimental groups in each 66. 65. 64. 63. 62. 61. Statistical Statistical analyses. P nominal (NES), score enrichment (normalized statistics GSEA The sets. gene (c2.all.v6.1) curated and (h.all.v6.1) hallmark (MSigDB) Database Signatures that correlated with differential expression. The Pearson’s tamoxifen treatment, we ranked all genes according to the Wald test statistic for enrichment by chance. in the figure legends in a two-sided, unpaired fashion, with with fashion, unpaired two-sided, a in legends figure the in orUSA) GraphPad Prism 7 software. Statistical tests were applied as indicated culations were performed using IBM SPSS Statistics (v22.0, IBM, Armonk, NY, instances was then used as a a as used then was instances value and FDR For the analysis of gene expression in cell cultures following estradiol or estradiol following cultures cell in expression gene of analysis the For uncertainty. r codntl dwrgltd n ua diabetes. (2003). human in downregulated coordinately are Mootha, V.K.Mootha, Subramanian, A. Subramanian, Beyer,& K. Singaravelu, E.G., Christodoulou, M., W.,Sikora-Wohlfeld,Ackermann, S. Mei, H. Pimentel, P.Melsted, H., Bray,RNA- Pachter,Pimentel, Near-optimalprobabilistic & N.L., L. J.S. Parker, subtypes. (2017). mouse. and human in data accessibility based on ChIP–seq data. ChIP–seq on based A. Assessing computational methods for transcription factor target gene identification quantification. seq for interpreting genome-wide expression profiles. expression genome-wide interpreting for 15545–15550 (2005). 15545–15550 66 , 5 GSE10883 6 7 ) ) with 1,000 random sample permutations and using the Molecular itoe aa rwe: dt pra fr hPsq n chromatin and ChIP–seq for portal data a Browser: Data Cistrome al. et J. Clin. Oncol. Clin. J. Nat. Methods Nat. uevsd ik rdco o bes cne bsd n intrinsic on based cancer breast of predictor risk Supervised al. et -responsive genes involved in oxidative phosphorylation oxidative in involved genes PGC-1 α-responsive al. et ifrnil nlss f N-e icroaig quantification incorporating RNA-seq of analysis Differential al. et q Life Sciences Reporting Summary Life Sciences Data from the transcriptional analysis of CAF2 cells stimu cells CAF2 of analysis transcriptional the from Data value) are presented in the figures. Gene set enrichment analysis: a knowledge-based approach knowledge-based a analysis: enrichment set Gene al. et PDGFC . All values depicted represent mean Nat. Biotechnol. Nat.

, 1160–1167 (2009). 1160–1167 27, PLOS Comput. Biol. Comput. PLOS expression within the TCGA breast carcinoma data , 687–690 (2017). 687–690 14, P value for the probability to acquire the level of level the acquire to probability the for value Further information on experimental design In vivo r , 525–527 (2016). 525–527 34, correlation test was used to rank the genes In vitro uli Ais Res. Acids Nucleic experiments included cohorts of the

, e1003342 (2013). e1003342 9, analyses were repeated three to Proc. Natl. Acad. Sci. USA Sci. Acad. Natl. Proc. . a. Genet. Nat. ± doi:10.1038/nm.4494 s.e.m. Statistical cal 1 D658–D662 D1, 45 P < 0.05 consid 0.05 < 267–273 34,

102, - - - -

nature research | life sciences reporting summary

Corresponding author(s): Kristian Pietras

Life Sciences Reporting Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form is intended for publication with all accepted life science papers and provides structure for consistency and transparency in reporting. Every life science submission will use this form; some list items might not apply to an individual manuscript, but all fields must be completed for clarity. For further information on the points included in this form, see Reporting Life Sciences Research. For further information on Nature Research policies, including our data availability policy, see Authors & Referees and the Editorial Policy Checklist.

Please do not complete any field with "not applicable" or n/a. Refer to the help text for what text to use if an item is not relevant to your study. For final submission: please carefully check your responses for accuracy; you will not be able to make changes later.

` Experimental design 1. Sample size Describe how sample size was determined. See paragraph on Statistical analyses in section Online Methods 2. Data exclusions Describe any data exclusions. No animals were excluded from the analysis. See paragraph on Statistical analyses in section Online Methods 3. Replication Describe the measures taken to verify the reproducibility Experiments not involving animals were repeated at least three times to ensure of the experimental findings. reproducibility. For ethical reasons, only key experiments involving animals were replicated. All attempts at replication were successful. In the few cases were replication was not attempted, robust statistical analyses were applied to ensure that the results were scientifically valid 4. Randomization Describe how samples/organisms/participants were See paragraph on Animal experiments in section Online Methods allocated into experimental groups. 5. Blinding Describe whether the investigators were blinded to Tumor size measurements in therapeutic trials were performed in a blinded fashion. Also, group allocation during data collection and/or analysis. quantification of tumor parameters by histological analyses was performed in a blinded fashion. Note: all in vivo studies must report how sample size was determined and whether blinding and randomization were used.

6. Statistical parameters For all figures and tables that use statistical methods, confirm that the following items are present in relevant figure legends (or in the Methods section if additional space is needed).

n/a Confirmed

The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement (animals, litters, cultures, etc.) A description of how samples were collected, noting whether measurements were taken from distinct samples or whether the same sample was measured repeatedly A statement indicating how many times each experiment was replicated The statistical test(s) used and whether they are one- or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section. November 2017 A description of any assumptions or corrections, such as an adjustment for multiple comparisons Test values indicating whether an effect is present Provide confidence intervals or give results of significance tests (e.g. P values) as exact values whenever appropriate and with effect sizes noted. A clear description of statistics including central tendency (e.g. median, mean) and variation (e.g. standard deviation, interquartile range) Clearly defined error bars in all relevant figure captions (with explicit mention of central tendency and variation)

See the web collection on statistics for biologists for further resources and guidance.

1 Nature Medicine: doi:10.1038/nm.4494 ` Software nature research | life sciences reporting summary Policy information about availability of computer code 7. Software Describe the software used to analyze the data in this Statistical analyses were performed using GraphPad Prism 7 or SPSS version 22.0. study.

For manuscripts utilizing custom algorithms or software that are central to the paper but not yet described in the published literature, software must be made available to editors and reviewers upon request. We strongly encourage code deposition in a community repository (e.g. GitHub). Nature Methods guidance for providing algorithms and software for publication provides further information on this topic.

` Materials and reagents Policy information about availability of materials 8. Materials availability Indicate whether there are restrictions on availability of All unique materials are readily available from the authors. unique materials or if these materials are only available for distribution by a third party. 9. Antibodies Describe the antibodies used and how they were validated STC1, Santa Cruz Biotechnologies, sc-30183, http://datasheets.scbt.com/sc-30183.pdf for use in the system under study (i.e. assay and species). HGF, Abcam, ab83760, http://www.abcam.com/hgf-antibody-ab83760.html IGFBP3, Santa Cruz Biotechnologies, sc-9028, http://datasheets.scbt.com/sc-9028.pdf PDGFRA, eBioscience/ThermoFisher, 12-1401, https://www.thermofisher.com/order/ genome-database/generatePdf?productName=CD140a% 20(PDGFRA)&assayType=PRANT&detailed=true&productId=12-1401-81 PDGFRB, Cell Signaling, 3169S, https://media.cellsignal.com/pdf/3169.pdf ERa, DAKO, clone 1D5, https://www.thermofisher.com/order/genome-database/ generatePdf?productName=Estrogen%20Receptor% 20alpha&assayType=PRANT&detailed=true&productId=MA5-13191 ERa, Santa Cruz Biotechnology, sc-542, https://datasheets.scbt.com/sc-542.pdf FoxA1, Santa Cruz Biotechnology, sc-6553, http://datasheets.scbt.com/sc-6553.pdf B-actin, Abcam, ab-8227, http://www.abcam.com/beta-Actin-antibody-ab8227.pdf

10. Eukaryotic cell lines a. State the source of each eukaryotic cell line used. Cell lines were acquired from ATCC or established as primary mass cultures in house.

b. Describe the method of cell line authentication used. Authentication was performed using fingerprinting on a regular basis.

c. Report whether the cell lines were tested for All cell lines were confirmed negative for mycoplasma on a regular basis. mycoplasma contamination. d. If any of the cell lines used are listed in the database NA of commonly misidentified cell lines maintained by ICLAC, provide a scientific rationale for their use.

` Animals and human research participants Policy information about studies involving animals; when reporting animal research, follow the ARRIVE guidelines 11. Description of research animals Provide all relevant details on animals and/or See paragraph on Animals and Animal experiments in section Online Methods. animal-derived materials used in the study.

Policy information about studies involving human research participants 12. Description of human research participants Describe the covariate-relevant population All patient cohort characteristics are supplied in Supplemental Table 1 and 3 of the characteristics of the human research participants. manuscript. November 2017

2 Nature Medicine: doi:10.1038/nm.4494