FIELD OF THE INVENTION

The present invention relates to a combined therapy against cancer which has proven to be particularly useful in the prevention and treatment of pancreatic cancer.

BACKGROUND

Pancreatic ductal adenocarcinoma (PDAC) is the third cause of cancer deaths in the US and is projected to be second after non-small cell lung cancer (NSCLC) by 2030 (Rahib et al., 2014) ¹ . The 5-year survival remains below 7 % due to the lack of effective treatments. Gemcitabine, a nucleoside analogue approved in 1997, is still the standard of care ^{2 3} , and its combination with nab-paclitaxel or erlotinib has shown only modest improvements ^{4 5} . Other therapies such as FOLFIRINOX are very toxic and can only be administered to selected patients ⁶ . The main genetic drivers of PDAC have been identified ⁷ . Whereas mutations in K-RAS appear to be the main initiating event, additional mutations in several tumor suppressors including TP53, CDKN2A, SAMD4, BRCA2 and TGFfiR contribute to tumor progression. Unfortunately, none of these cancer drivers are currently druggable, thus making it difficult to devise effective therapies against PDAC. Only a small percentage of clinically relevant mutations may benefit from available targeted therapies. So far, therapeutic strategies in genetically engineered mouse (GEM) PDAC models have failed to achieve tumor regression. Only tumor models driven by a doxycycline-inducible K-RAS ^G12D undergo tumor regression upon silencing of K-RAS ^G12D expression ⁸⁹ . On the other hand, K-RAS oncogenic signaling is mediated by two signaling cascades made up of druggable kinases ¹¹ . Previous studies have shown that ablation of the EGF receptor (EGFR) delayed PDAC development in K-Ras/Tp53 driven GEM tumor models ^{12 13} . More recently, we have identified c-Raf as a suitable therapeutic target for K-Ras/Tp53 driven lung adenocarcinoma ¹⁴ . Here we describe that combined, systemic ablation of EGFR and c-Raf expression in advanced K-Ras/Tp53 mutant PDAC tumors results in complete regression of a significant percentage of tumors with tolerable toxicities. These observations may open the door to the development of rational targeted therapies for PDAC tumor patients.

WO2015087279 A1 describes the triple combination of a B-RAF inhibitor (dabrafenib), a MEK inhibitor (trametinib) and a EGFR inhibitor (panitumumab) and its use in the treatment of cancer. It has been described that the inhibition of B-Raf is not useful because it inhibits the complete pathway and its toxicity is quite high (Blasco et al 201 1 Cancer Cell 19, 652- 663).

Inhibiting K-Ras and MEK also implies intolerable side effects due to the toxicity of the treatment.

DESCRIPTION OF THE INVENTION

The inventors have shown that the simultaneous deletion of Egfr and c-Raf in PDAC results in a significant therapeutic effect with an extremely low toxicity.

The present invention represents a very promising therapy in the treatment of pancreatic cancer since it leads to a complete tumor regression while its side effects are much smaller than other therapies described up to date.

In a first aspect, the present invention relates to a pharmaceutical composition comprising an inhibitor of the expression, activity and/or function of c-Raf and an inhibitor of the expression, activity and/or function of EGF Receptor (EGFR). Preferably, wherein the inhibitor of the expression, activity and/or function of c-Raf does not inhibit C-RAF kinase activity.

In a preferred embodiment of the fist aspect, MAPK and Pi3K pathways are not affected by the pharmaceutical composition of the invention. The expression“are not affected” as used herein means that the MAPK and Pi3K pathways are not inhibited by the pharmaceutical composition of the invention.

In a preferred embodiment, the inhibitor of the expression, activity and/or function of c-Raf comprises a c-Raf inhibitor compound, a c-Raf inhibitor antibody or an antigen binding fragment thereof, a peptide or a nucleotide sequence. As illustrated in the examples, the inventors have shown that inhibiting the expression of c-Raf in combination with inhibiting Egfr leads to a complete tumor regression. The inhibition of c-Raf expression achieved by shRNA is very effective and specific.

In a preferred embodiment, the inhibitor of the expression, activity and/or function of c-Raf does not inhibit C-RAF kinase activity. In a preferred embodiment, the inhibitor of the expression, activity and/or function of c-Raf inhibits the expression of c-Raf or promotes C- RAF degradation. In another preferred embodiment, the inhibitor of the expression, activity and/or function of c-Raf blocks C-RAF kinase independent activities. In a preferred embodiment, the present invention relates to a pharmaceutical composition comprising an inhibitor of the expression, activity and/or function of c-Raf and an inhibitor of the expression, activity and/or function of EGF Receptor (EGFR), wherein the inhibitor of the expression, activity and/or function of c-Raf:

(i) does not inhibit C-RAF kinase activity;

(ii) inhibits the expression of c-Raf;

(iii) promotes C-RAF degradation; and/or

(iv) blocks C-RAF kinase independent activities.

In a preferred embodiment, the inhibitor of the expression, activity and/or function of c-Raf is other than sorafenib, vemurafenib, dabrafenib y LY3009120.

In a preferred embodiment, the inhibitor of the expression, activity and/or function of c-Raf does not inhibit the expression, activity and/or function of b-Raf.

In a preferred embodiment, the inhibitor of the expression, activity and/or function of c-Raf promotes C-RAF degradation. In a more preferred embodiment, said inhibitor is a proteolysis targeting chimera (PROTAC).

In a preferred embodiment, the inhibitor of the expression, activity and/or function of EGFR comprises an EGFR inhibitor compound, an EGFR inhibitor antibody or an antigen binding fragment thereof, a peptide or a nucleotide sequence.

In a preferred embodiment, the nucleotide sequence is or codifies for a guide RNA, an interfering RNA or a micro RNA.

In a preferred embodiment, the inhibitor of the expression, activity and/or function of EGFR is selected from afatinib, erlotinib, gefitinib, brigatinib, icotinib, neratinib, lapatinib, vandetanib, osimertinib, cetuximab, panitumumab, necitumumab, nimotuzumab, zalutumumab, matuzumab and combinations thereof.

In a preferred embodiment, the pharmaceutical composition comprises a therapeutically effective amount of an inhibitor of the expression, activity and/or function of c-Raf, a therapeutically effective amount of an inhibitor of the expression, activity and/or function of EGFR and optionally a pharmaceutically acceptable excipient.

In another aspect, the present invention relates to the pharmaceutical composition according to any one of the preceding claims for use in the prevention and/or treatment of cancer. Preferably, for use in the prevention and/or treatment of pancreatic cancer. More preferably, for use in the prevention and/or treatment of pancreatic intraepithelial neoplasia or of pancreatic ductal adenocarcinoma. In a preferred embodiment, the pharmaceutical composition is for use in tumor regression in a subject afflicted with cancer. Preferably, for use in tumor regression in a subject afflicted with pancreatic cancer. More preferably, for use in tumor regression in a subject afflicted with pancreatic intraepithelial neoplasia or of pancreatic ductal adenocarcinoma.

An aspect of the present invention is a method for the prevention and/or treatment of cancer comprising the administration of a therapeutically effective amount of an inhibitor of the expression, activity and/or function of c-Raf and a therapeutically effective amount of an inhibitor of the expression, activity and/or function of EGF Receptor (EGFR). Preferably, the method of the present invention is for the treatment and/or prevention of pancreatic cancer, more preferably of pancreatic intraepithelial neoplasia (PanIN) or of PDAC.

DESCRIPTION OF THE DRAWINGS

Figure 1. Effect of c -Raf and Egfr ablation in initiation of K-Ras driven pancreatic lesions a, Number of low- and high-grade PanIN lesions and PDAC tumors in one year old KeC mice carrying either wild type c-Raf (solid circles, n=12) or conditional c -Raf ⁰ * (open circles, n=8) alleles. Horizontal bars indicate the average number of lesions per mouse b, Survival of control KPeC (black, n=20), KPeC ; Cc/ 4 ^K35M/K35M (dark grey, n=14), and KPeC;Egfr ^lox/lox; Cc//i4 ^K35M/K35M (light gray, n=1 1 ) mice. All mice died of PDAC tumors at the indicated times c, Survival of control KPeC ((i), black, n=20), KPeC; Raf 7 ^|0X/|0X ((ii), n=13), KPeC ;Egf/ ^j0X/l0X ; Raf 7 ^+/lox ((iii), n=10), KPeC;Eg/H ^0X ;fiaf7 ^l0X/l0X ((iv), n=5) and

KPeC;Egf/ ^j0X/l0X ;Raf7 ^l0X/l0X ((v), n=14) mice. All mice died of PDAC tumors at the indicated times d, PCR analysis of c -Raf and Egfr alleles using DNA extracted from K-Ras ^G12V expressing, X-Gal positive acinar cells isolated by Laser Capture Microdissection. Migration of c -Raf- and Egfr- null alleles (lane 1 ) c-Raf ^{ox and} Egff ^ox conditional floxed alleles (lane 2) and c-Raf ^{+ and} Egfr wild type alleles (lane 3). DNA extracted from X-Gal positive (lane 4) and X-Gal negative (lane 5) acinar cells of KPeC;c -Raf ^oxnox Egff ^{oxnox m} ' ^ce Lane 6 is a blank control. Lane M depicts a DNA ladder. Expected size for each DNA fragment is indicated.

Figure 2. Ablation of c-Raf and Egfr expression induces acceptable toxicities. a, Western blot analysis of EGFR, c-RAF, pERK1 /2, ERK1/2, pAKT and AKT expression in tissues obtained from Egfr ^l+ Raf1 ^+l+ ,Tg. UBC-CreERT2 and Egff ^oxllox ; Raf / ^loxi|ox : Tg . L/SC-C re E RT2 mice after 3 weeks of TMX exposure. GAPDH served as a loading control b, Body weight change in grams (g) of male (upper) and female (lower) c-Raf ^+/+ Egfr ^/+ g.hUBC- CreERT2 ^+/T (solid circles, n=5) and c-Raf ^0Xll0X ,Egff ^0Xll0X ,Tg.hUBC-CreERT2 ^+IT mice (open circles n=5) exposed to TMX for the indicated length of time c, Representative H&E and Toluidine-blue staining of skin sections from c-Rafr ⁺ ;Egfr ^+/+ ;Tg.h L/EC-CreERT2 ^+/T and c- Raf ^0Xd0X ;Egfr ^0X/l0X ;Tg.h L/EC-CreERT2 ^+/T mice exposed to TMX for 15 weeks. Scale bars represent 100 pm (H&E) and 20 pm (toluidine blue) d, Representative H&E and cleaved Caspase-3 staining in sections of small intestine of c-Rafr ⁺ ;Egfr ^+/+ ;Tg.h L/EC-CreERT2 ^+/T and c-Raf ^0Xd0X ;Egfr ^0X/l0X ;Tg.h L©C-CreERT2 ^+/T mice exposed to TMX for 15 weeks. Scale bars represent 100 pm (H&E) and 20 pm (cleaved Caspase-3).

Figure 3. c-Raf and Egfr ablation induce regression of a fraction of PDAC tumors a, Total tumor volume visualized by weakly ultrasound monitorization of KPeFC; c -Raf ^+l+ ,Egfr ^+l+ control mice (n=10 mice, 15 tumors) exposed to a TMX diet for the indicated time. Each color represents a different mouse b, Total tumor volume visualized by weakly ultrasound monitorization of KPeFC; c-Raf ^0Xll0X Egf ^0Xll0X mice (n= 6 mice, 7 tumors) exposed to a TMX diet for the indicated time. Each color represents a different mouse (R1 -R6). c, Representative ultrasound images of the regression of a large tumor (23.5 mm ³ ) present in the R3 mouse after 3 and 6 weeks of TMX exposure. Visible lesions are outlined. Tumor volumes are indicated. ND: not detectable d, (left column) Representative H&E stained paraffin sections of the pancreata of control KPeFC;c -Raf ^+l+ ,Egfr ^+/+ C1 ^and KPeFC;c- Ra/’ ^ox/lox ;Eg / ^ox/lox R1 and R3 mice after six weeks of TMX exposure. The tumor present in the C1 mouse is outlined by a dotted line. Scale bar represents 1000 pm. Box insets, indicated by arrowheads, mark areas shown at higher magnification in the images shown to the right stained with H&E (H/E) and Cytokeratin19 (CK19). Scale bar represents 100 pm. e, Same as in d for tumors present in KPeFC;c -Ra† ^oxlox \Eg ^oxlox R4-R6 mice f, Higher magnification of representative H&E, Egfr, pErk and Ki67 stained paraffin sections from KPeFC;c -Raf ^+l+ ,Egfr ^+l+ < ^C1 > ^and KPeFC;c -Ra†° ^xl '° ^x ,Egff ^oxl ' ^ox (R1 , R3-R6) mice. Scale bar represents 20 pm. The R2 mouse did not had a scar lesions g, Representative IHC staining of Cleaved Caspase 3 of sections of a pancreatic tumor (“Regressor”) from a KPeFC,Egfr° ^x/ '° ^x Raf1'° ^xn ° ^x mouse that decreased 30% in volume after two weeks of TMX exposure and of tumors (“Non Responder”) from two independent KPeFC;Egfr ^0X/l0X ;/¾f7 ^l0X/l0X mice that continued growing during the same period of time. Scale bar represents 20 pm.

Figure 4. A fraction of PDAC tumors do not respond to c-Raf and Egfr ablation a, Total umor volume visualized by weakly ultrasound monitorization of KPeFC;c-Raf ^0X/l0X ;Egfr ^0X/l0X mice (n=4 mice, 4 tumors) exposed to a TMX diet for the indicated time. Each color represents a different mouse (NR1 -NR4). b, Representative ultrasound images of the progression of the tumor present in the NR2 mouse after 3 weeks of TMX exposure. Visible lesions are outlined in white. Tumor volumes are indicated c, Western blot analysis of c- Raf and Egfr expression in lysates obtained from PDAC tumors present in control KPeFC;c- Raf ^+,+ ,Egfr ^,+ mice depicted in Figure 3a (C1 -C3) and in NR1 -NR3 KPeFC;c -Raf ^ox/lm -,Egff ^ox,lox mice exposed to TMX for six weeks. Expression levels of Erk1/2 pErk1/2, Akt and pAkt are also shown. Gapdh served as a loading control d, (left column) Representative H&E stained paraffin sections of the pancreata of control C2 KPeFC;c -Raf ^+l+ ,Egfr ^+/+ mouse ⁽ depicted in Fig 3a> and of non-responder NR1 and NR2 KPeFC;c -Raf ^mnm Eg ^mnm mice after six weeks of TMX exposure. Tumors are outlined by dotted lines. Box insets, indicated by arrowheads, mark areas shown at higher magnification in the adjacent images shown to the right stained with H&E, Cytokeratin19 (CK19), pErk and Ki67. Scale bar represents 50 pm.

Figure 5. Differential RNA expression profiles of PDAC tumor cells sensitive and resistant to c-Raf and Egfr ablation a, Colony formation assay of tumor cell lines established from individual tumors of two KPeFC;c -Raf ^+l+ ,Egfr ^+l+ control mice (C1 and C2) and of six KPeFC;c-Raf ^ox/lox ;Egfr ^o,</lo,< animals These tumor cell lines are designated as“Responder cells” (RC1 -RC3) or Non responder cells (NRC1 -NRC3) based on their proliferative properties after ablation of c-Raf and Egfr expression with Ad-Cre particles. Infection with Ad-GFP particles was used as a negative control. Cells were fixed with glutaraldehyde and stained with crystal violet b, Quantification of the number of colonies expressed as percentage of the number of colonies observed in cells infected with Ad-Cre and those infected with Ad-GFP particles. Error bars indicate standard deviation c, Heat map representing color-coded expression levels of differentially expressed genes in NRC vs. RC cell lines infected with Ad-GFP particles d, GSEA pathway analysis of NRC vs. RC cell lines infected with Ad-GFP particles. The normalized enrichment score (NES) ranking was generated by the GSEA. e, Heat map comparing the transcriptional profiles of RC and NRC cells with those of PDAC tumors as reported by Bailey and cols. e2, Genes selected from the two thousand genes differentially expressed between RC and NC cells based on their involvement in signaling pathways known to participate in the development and/or progression of PDAC. Genes are ordered according to the log2 fold change f, Western blot analysis of Egfr, c-Raf, pAkt, Akt, pErk1/2, Erk1/2, pCofilin, pStat3 and Stat3 protein expression in whole cell extracts of the indicated cell lines obtained 5 days after (top) Adeno-GFP or (bottom) Adeno-Cre infection. Gapdh served as a loading control g, pStat3 staining in paraffin-embedded sections of PDAC tumors of KPeFC;c -Raf ^+l+ ,Egfr ^+l+ C1 mouse and KPeFC,c- Raf ^mllm Egff ^{mnm mice} that harbored tumors that regressed (R3), progressed (NR1 , NR3) or relapsed (RT1 , RT2) upon c-Raf and Egfr ablation. Scale bar represents 20 pm. h, Quantification of pStat3 positive tumor cells in paraffin-embedded PDAC tumor sections from KPeFC;c-/¾fr ⁺ ;Egfr ^+/+ control mice (C, solid bar, n=3), KPeFC;c- Raf ^o xii ^o x-Egfr ^o x ^/ i ^o x _mjce responded to c -Raf and Egfr ablation (R, open bar, n=3). KPeFC;c-Raf ^0X/l0,< ;Egfr ^0,</l0,< mice that did not respond to c-Raf and Egfr ablation (NR, grey bar, n=3) and KPeFC;c-Raf ^0X/l0,< ;Egfr ^0,</l0,< mice that relapsed after their initial response to c- Raf and Egfr ablation (RT, striped bar, n=2). Error bars indicate standard deviation. P-values were calculated using the unpaired Student’s t test. ^*** p < 0.001 .

Figure 6. c-Raf and Egfr expression are required for proliferation of patient-derived PDAC xenografts. Left, Western blot analysis of Egfr and c-Raf protein expression in whole cell extracts obtained from the indicated PDX cell line (PDX1 -4) expressing a scramble shRNA (-) as well as shRNAs against Egfr (E), c-Raf {R) and Egfr + c-Raf { E/R). Gapdh served as a loading control. Center, Tumor growth after subcutaneous implantation in nude mice (n=4) of the corresponding PDX cell lines expressing a scramble shRNA (black) as well as shRNAs against Egfr (blue), c-Raf (red) and Egfr + c-Raf (green). Right, Quantification of tumor growth at the end of the experiment. Error bars indicate standard deviation. P-values were calculated using the unpaired Student’s t test. ^* p < 0.05, ^** p < 0.01 , ^*** p < 0.001 . n.s: no significant.

Figure 7. Comparative analysis of the KPeC and KPeFC tumor models a, Survival of KPeC (solid circles, n=20) and KPeFC (open circles, n=8) mice b, Representative pancreatic lesions of (top) KPeC and (bottom) KPeFC mice stained with H&E. Scale bars represent 20 pm (PanlN1 A/B, PanlN2, PanlN3) and 50 pm (PDAC).

Figure 8. PDAC tumors of KPeFC mice become resistant to c-Raf and Egfr ablation a, Total tumor volume visualized by weakly ultrasound monitorization of KPeFC;c- ^ _a ^tox/iox . g,^ ^jox/iox _mjce exposed to TMX. Each color represents a different mouse b, Western blot analysis of c-Raf and Egfr expression in lysates derived from the PDAC tumors present in KPeFC;c-Rafr ^/+ ;Egfr ^/+ control C1 mouse depicted in Figure 3a and KPeFC;c-Raf ^l0X/l0X ;Egfr ^0X/l0X RT2 mouse exposed to TMX for 1 1 weeks. Gapdh served as a loading control c, (left column) Representative H&E stained paraffin sections of the pancreata of KPeFC;c- Ra ^ox/iox j Egfr ^ox/iox RT1 and RT2 mice after 10 and 1 1 weeks of TMX exposure, respectively. Tumors are outlined by a dotted line. Scale bar represents 1000 pm. Box insets, indicated by arrowheads, mark areas shown at higher magnification in the adjacent images to the right stained with H&E, CK19, pErk and Ki67. Scale bar represents 50 pm.

Figure 9. Histological characterization of the residual scar lesions present in“Responder” mice after TMX exposure a, Low and b, high magnification of representative sections of a PDAC tumor present in control PeFC\c- RaP ^l+ ,Egfr ^l+ C1 mouse and in the scar lesions of KPeFC;c-Raf ^0X/l0X ;Egfr ^0X/l0X R3 and R4 mice stained with Masson ^' s trichrome (T. Masson), Hyaluronic Acid Binding Protein (HABP), F4/80 and CD3. Scale bars represents a. 100 pm and b, 20 pm.

Figure 10. PanIN lesions present in “Responder” mice. Representative H&E stained paraffin sections of the pancreata of KPeFC;c-/¾/ ^0X/l0X ;Egfr ^0X/l0X R2 mouse after six weeks of TMX exposure. Scale bar represents 1000 pm. Box inset, indicated by an arrowhead, marks the area shown at higher magnification in the adjacent images to the right stained with H&E, Egfr, CK19, pErk and Ki67. Scale bar represents 50 pm.

Figure 11. Differential expression of CK19 and Gata6 in PDAC tumors used to generate the“Responder” and Non Responder” cell lines. Representative H&E, CK-19, Gata6 and pStat3 stained paraffin embedded sections of the original KPeFC;c-Ra/ ^l0X/l0X ;Egfr ^0X/l0X tumors used to generate the (Top) “Responder” (RC1 -RC3) and (Bottom) “Non responder” (NRC1 -NRC3) tumor cell lines as determined by their response to c-Raf and Egfr ablation. Scale bar represents 50 pm.

Figure 12. c-Raf and Egfr are essential for proliferation of PDX cell lines in vitro a, Left, Western blot analysis of Egfr and c-Raf expression in whole cell extracts obtained from the corresponding PDX cell line (PDX1 -4) using a scramble shRNA (-), two shRNA against Egfr (E1 , E2), two shRNAs against c-Raf (R1 , R2) and the combination of shRNAs against Egfr and c-Raf (E/R). Gapdh served as loading control. Center, Cell proliferation assays of the corresponding PDX cell line expressing sceamble shRNA (black), shRNAs against Egfr, shRNAs against c-Raf and the combination of shRNAs against Egfr and c-Raf. Proliferation was determined by MTT and is expressed as arbitrary units (a.u.). Error bars indicate standard deviation. Right, Quantification of cell proliferation at the end of the experiment. Error bars indicate standard deviation. P-values were calculated using the unpaired Student’s t test. ^* p < 0.05, ^** p < 0.01 , ^*** p < 0.001 . n.s: no significant c, EGFR and c-RAF expression is essential for in vitro proliferation of PDAC cells-derived from PDX tumor models. (Left) Cell proliferation of the indicated PDX-derived cells infected with a scramble shRNA (black) or with shRNAs against EGFR (light and dark blue), RAF1 (red and pink) and EGFR plus RAF1 (green). (Right) Western blot analysis of EGFR and c-RAF expression in whole cell extracts obtained from the indicated PDX-derived cells using either a scramble shRNA (-), shRNAs against EGFR (E1 , E2), RAF1 (R1 , R2) and EGFR plus RAF1 (E1/R1 ) (right). Proliferation was determined by MTT and expressed as fold increase in the number of cells determined at each of the indicated days. Error bars indicate mean ± SD. GAPDH served as loading control.

Figure 13. Pharmacologic inhibition of EGFR in combination with c-RAF knockdown inhibits proliferation of PDAC cells derived from PDX tumor models. Cell proliferation of the indicated PDX-derived cells infected with scramble shRNA (black), infected with shRNA R1 against c-RAF (red), infected with shRNA R1 against c-RAF and exposed to Gefitinib (IC ₅ o) (green), infected with shRNA R1 against c-RAF and exposed to Erlotinib (IC ₅ o) (blue) (left) and Western blot analysis of c-RAF expression in whole cell extracts obtained from the indicated PDX-derived cells using either a scramble shRNA (-) or a shRNAs against c-RAF (R1 ) (right). Proliferation was determined by CellTiter-Glo and expressed as fold increase in the number of cells determined at each of the indicated days. Error bars indicate mean ± SD. GAPDH served as loading control in Western blot.

Figure 14. Mice not included in the trial a, Spurious expression of the FlpO recombinase. Fluorescent labeling in representative sections of PDAC, healthy pancreas, skin and intestine of KPeFC, Rosa26 ^{/CAG dJom3lo~ EGFP} mice exposed to TMX for 4 weeks. Dotted white line indicates a small EGFP ⁺ papilloma. Arrows indicate two recombinant cells in intestine. Scale bar represents 50 pm. b, Limited Cre-mediated cleavage of Egff° and Raff ^ox alleles, as determined by the expression of EGFR and c-RAF proteins, in representative KPeFC and KPeFC; Eg / ^j0X/l0X ; Raff ^ox/lox mice after TMX exposure. GAPDH served as a loading control.

Figure 15. a, Generation of a conditional c-Raf kinase dead allele b, Lack of therapeutic effect of a c-Raf ^D468A kinase dead isoform on K-Ras /TP53 mutant tumors c, In vitro c-Raf kinase assay, c-Raf active kinase domain was expressed in baculoviral systems. Negative control 1 : only MEK. Negative control 2: only c-Raf.

Figure 16. Lack of significant inhibition of PDX tumor cell growth in vitro by four independent c-Raf kinase inhibitors: a, MLN2480, b, GW5074, c, PLX8394 and d, LSN3074753. PDX- derived cell lines (PDX dd and PDX dc2) were treated with 2 pan-Raf Kinase inhibitors (MLN2480 and LSN3074753), a c-Raf Kinase inhibitor (GW5074) and a paradox breaker inhibitor (PLX8394) for 72 hours to determine their respective IC50. Only the pan-Raf inhibitor LSN3074753 has a significant effect in cell viability. Figure 17. Lack of tumor regression in a PDX tumor model in vivo by a pan-Raf kinase inhibitor. Pan-Raf (LSN3074753) treatment of K-Ras ^G12C mutant lung adenocarcinoma PDX model: Pulm 24. Immunocompromised mice were implanted subcutaneously with the lung tumor PDX1 model. Once the animals reached tumors of approximately 200 mm ³ they were randomized in 2 treatment groups. Whereas one group received vehicle (n=7 mice/8 tumors) (solid circles) the other was treated with the pan-Raf inhibitor LSN3074753 (40 mg/kg, PO, BID for 28 days) (n=13 mice/15 tumors) (white circles). Tumor growth was monitored twice a week by caliper measurements. Tumor volumes are shown as mean ± SEM.

Figure 18. Comparative effect of treating K-Ras/TP53 genetically engineered mice with a c-Raf kinase inhibitor (a) and genetically ablating c-Raf expression (b). a, tumor volume of vehicle (9 mice/17 CT+ tumors, black bars) and c-Raf inhibitor LSN3074753 inhibitor (10 mice/20 CT+ tumors, white bars) treated mice b, _fusion of figures 3b and 3c of Sanclemente et al. 2017 to show the tumor regression achieved by genetically ablating c- Raf expression (K-/¾s ^+/FSFG12V : TP53F ^F c- Raf ^{+ +} in black bars and K-Ras ^+/FSFG12V ; 7P53 ^F/F ;c- Raf ^lox/iox in grey bars).

EXAMPLES

The inventors demonstrate that combined inhibition of EGFR and c-RAF expression is a very effective therapy against PDAC, both in mutant Kras/Trp53- driven GEM tumor models as well as in human PDXs. Also, they have found the surprising fact that systemic elimination of these targets results in tolerable toxicities, primarily resulting from the lack of EGFR activity.

Human PDACs are genetically more complex than those of GEM tumor models. Yet, the inhibitory effect of EGFR and c-RAF knockdown in nine out of ten independent PDX tumor models illustrates that this therapeutic strategy is plausible to be also effective in the clinic. It is importante to highlight the surprising fact that human PDX-derived tumor cells are more sensitive than the corresponding mouse tumors in spite of their more complex mutational profile.

Replacement of EGFR ablation by pharmacological inhibition of EGFR kinase activity yielded similar results. However, inhibition of c-RAF with c-RAF kinase inhibitors does not reproduce the results obtained and disclosed herein. Surprisingly, only inhibiting the expression of RAF1 or c-RAF leads to the results disclosed herein.

Importantly, the inventors have used conditional Raf1 kinase alleles in a lung“therapeutic model” that show that the therapeutic effect observed upon loss of c-RAF expression is not be mediated by its kinase activity.

Moreover, the inventors gave found that Pan-RAF kinase inhibitors have limited anti-tumor activity (they do not induce tumor stasis or tumor regression) in several assays, including: in vitro PDX cell lines, in vivo PDX tumor model, in vivo K-Ras/TP53 genetically engineered mouse models. In vivo evidence that expression of a kinase dead isoform of c-Raf (c-Raf ^D468A ) has no therapeutic activity against genetically engineered mouse tumors induced by K-Ras and TP53 mutations

Figures 3B and 3C of Sanclemente et al 2017 show genetically engineered mouse tumors induced by K-Ras and TP53 mutations and the therapeutic effect of c-RAF ablation on K- Ras/TP53 mutant tumors.

Figure 15A depicts the strategy that was used by the inventors for generating a conditional c-Raf kinase dead allele. Said c-RafD468A kinase dead isoform lacked therapeutic effect on K-Ras/TP53 mutant tumors (Fig. 15B).

Importantly, the c-Raf ^D468A isoform does not have kinase activity neither in vitro as determined by measuring its kinase activity in a baculovirus expression system or in vivo as determined by its ability to block the“paradoxical effect” caused by expression of a B- Raf kinase mutant (Fig. 15C).

Nieto et al 2017 showed that Expression of the c-Raf ^D468A kinase dead isoform reverts the paradoxical effect induced by B-RAF kinase mutants (fig 2 of Nieto et al, 2017). c-Raf kinase inhibitors do no inhibit tumor cell growth but genetically ablating c-Raf expression does

Four independent c-Raf kinase inhibitors: MLN2480, GW5074, PLX8394 and LSN3074753 were used to treat PDX-derived cell lines (PDX dd and PDX dc2). Two of them are pan- Raf Kinase inhibitors (MLN2480 and LSN3074753), one is a c-Raf Kinase inhibitor (GW5074) and the fourth one is a paradox breaker inhibitor (PLX8394). The treatment lasted 72 hours and their respective IC ₅ o was determined. Only the pan-Raf inhibitor LSN3074753 has a significant effect in cell viability (Fig. 16).

Importantly, pan-Raf (LSN3074753) treatment of K-Ras ^G12C mutant lung adenocarcinoma PDX model (Pulm 24) did not lead to tumor regression in vivo. Immunocompromised mice were implanted subcutaneously with the lung tumor PDX1 model. Once the animals reached tumors of approximately 200 mm ³ they were randomized in 2 treatment groups. Whereas one group received vehicle (n=7 mice/8 tumors), the other was treated with the pan-Raf inhibitor LSN3074753 (40 mg/kg, PO, BID for 28 days) (n=13 mice/15 tumors). Tumor growth was monitored twice a week by caliper measurements and no tumor regression could be observed (Fig. 17).

Also, in order to compare the effect of treating K-Ras/TP53 genetically engineered mice with c-Raf kinase inhibitor LSN3074753 and genetically ablating c-Raf expression, K- Ras ^+/LSLG12V ;p53 ^l0X/l0X mice were infected with 10 ⁶ pfu of Ad-Cre at 8 weeks of age and tumor development was monitored by CT measurements. Once the animals developed at least one tumor bigger than 3 mm ³ they were treated with the pan-Raf LSN3074753 inhibitor (25 mg/kg, PO, BID for one month). Figure 18 A represents the tumor volume of individual CT+ tumors treated with vehicle (black bars) and LSN3074753 (grey bars).

For comparing these results with those of the genetic ablation of c-Raf expression, figure 18 B is presented. This figure is a combination of the data presented in figures 3B and 3C of Sanclemente et al 2017, showing the tumor volume in K-Ras ^+IFSFG ™, TP53F ^IF , c-Raf ^+/+ mice (black bars) and in K-Ras ^+IFSFG ^, TP5^ ^IF , c-Raf ^lox/lox mice (grey bars).

Induction of pancreatic lesions by K-Ras ^G12V requires c-Raf expression

We have previously shown that c-Raf is essential for initiation and progression of K-Ras ^G12V driven lung adenocarcinomas ¹⁴ ’ ¹⁵ . To investigate the contribution of c-Raf to K-Ras ^G12V driven pancreatic tumors, we added conditional c-Raf floxed alleles to the K- Pas ^+/LSLG12V9eo ;E/as-tTA/7e/0-Cre strain, a GEM tumor model that induces expression of K- F?as ^G12V in acinar cells during late embryonic development. These mice, designated as KeC, developed PanINs lesions with complete penetrance as well as PDAC tumors in about 20% of the animals at one year of age ¹⁶ . As illustrated in Fig. 1 a, elimination of c-Raf alleles completely prevented PanIN formation as well as PDAC tumors. These results illustrate that c-Raf is essential for acinar cell transformation by K-Ras oncogenes.

Ablation of Tp53 overcomes the requirement for c-Raf expression in K-Ras ^G12V driven PanIN and PDAC lesions

Development of aggressive K-Ras driven PDAC tumors requires additional oncogenic insults such as deletion or inactivation of the Tp53 or p16/p19 tumor suppressors ^17-20 . To examine the effect of abrogating c-Raf signaling in the absence of Tp53, we added the floxed c-Raf alleles to the K- Ras ^+/LSLG12Vgeo : Tp53 ^{ox lox} : E/as- tT A/ TetO-Gre strain (from now on KPeC mice). As illustrated in Figure 1 b, deletion of c-Raf had no effect on the development of K-Ras/Tp53-driven PDAC tumors. All animals, regardless of whether they carried wild- type (n=20) or floxed (n=13) c-Raf alleles succumbed to pancreatic tumors around 30 weeks of age (Fig. 1 b). Efficient deletion of c-Raf alleles was verified by PCR analysis of PDAC cells isolated from tumor specimens. Therefore, ablation of Tp53 completely subverted K- Ras oncogenic signals to make them independent of c-Raf. Alternatively, the absence of Tp53 may activate additional oncogenic pathways that cooperate with K-Ras ^G12V signaling in a c-Raf independent manner. Combined Egfr and Raf1 ablation completely inhibits PDAC development in a tumor initiation model

We and others have previously reported that Egfr ablation prevented the formation of Pan IN lesions in oncogenic Kras- driven GEM PDAC models. Furthermore, the absence of EGFR delayed PDAC development in the absence of p53 (Ardito et al., 2012; Navas et al., 2012). To identify effector molecules that could cooperate with Egfr ablation in preventing mutant Kras/Trp53- driven PDAC development, we added conditional floxed alleles to the ras ^+/LSLG12V9eo ; rrp55 ^lox/lox ; E/as- tT A/ Te/O-Cre strain. This strain has been designated as KPeC to indicate that the driver mutations are selectively induced by the elastase gene promoter in the acinar cell compartment instead of in all pancreatic cell lineages as in the classical KPC model (Hingorani et al., 2005).

Among those K-RAS effectors likely to cooperate with EGFR in mediating PDAC development, we selected the CDK4 cell cycle kinase and the c-RAF kinase based on our prior observations that they are essential for the development of K-RAS ^G12V driven lung tumors (Blasco et al., 201 1 ; Puyol et al., 2010; Sanclemente et al., 2018). Moreover, ablation of CDK4 or c-RAF does not induce unacceptable toxic effects such as those observed upon ablation of the MEK1 /2 and ERK1/2 kinases (Blasco et al., 201 1 ; Puyol et al., 2010). To interrogate whether tampering with CDK4 activity could cooperate with EGFR ablation in preventing PDAC development, we mutated the endogenous Cdk4 to encode a kinase dead K35M isoform to better recapitulate pharmacological treatments.

As illustrated in Figure 1 B, control KPeC mice (n=20) succumbed to PDAC at the average of 15 weeks of age. Expression of the kinase dead CDK4 ^K35M (n=14) did not prevent PDAC development, but it increased the median survival of the tumor-bearing mice to similar to that observed in the absence of EGFR (Ardito et al., 2012; Navas et al., 2012). Combined ablation of Egfr and expression of CDK4 ^K35M did not decrease the rate of PDAC development or further extended survival (n=1 1 ) (Figure 1 B). Surprisingly, in contrast to the results obtained with lung tumors, ablation of Raf1 had no effect on PDAC development and all animals (n=13) succumbed to pancreatic tumors with a latency similar to that of KPeC mice (Figure 1 C). However, concomitant ablation of Egfr and Raf1 completely prevented PDAC development (n=14), up to two years of age (Figure 1 C). Detailed histological analysis of serial sections of their pancreata failed to identify PanIN lesions or even metaplasias. These mice retained K-RAS ^G12V expression in their acinar cell compartment as determined by the presence of b-galactosidase, a surrogate marker for K- RAS ^G12V expression (Guerra et al., 2003). Isolation of these cells by laser-capture microdissection confirmed efficient recombination of Raff ^ox and Egfr ^ox alleles (Figure 1 D). No such recombination was observed in adjacent acinar cells negative for b-galactosidase expression (Figure 1 D). Thus, EGFR and c-RAF, but not CDK4, must signal through independent pathways essential for initiation and development of pancreatic tumors. Finally, inhibition of PDAC development requires complete absence of EGFR and c-RAF expression because different combinations of floxed and wild-type Egfr and Raf1 alleles in KPeC mice delayed, but did not prevent, PDAC development (Figure 1 C).

Mutational complexity of PDAC tumors driven by K-Ras/Tp53 mutations

GEM tumor models of PDAC driven by K-F?as ^G12V and loss of Tp53 closely reproduce the histopathology of human tumors ¹⁹ ’ ²⁰ . Yet, their mutational complexity appears to be more limited ²²²³ . Indeed, exomic next generation sequencing (NGS) of KPeC tumors revealed a similar number of miscoding mutations (13.6 mutations/tumor) (Table 1 ) similar to those reported elsewhere (Chung et al., 2017), a significant mutational complexity albeit more limited than that of their human counterpart (Bailey et al., 2016; Biankin et al., 2012; Jones et al., 2008; Raphael et al., 2017; Waddell et al., 2015; Witkiewicz et al., 2015). Interestingly, none of the mutated genes identified in this genomic analysis appeared in different tumors (T able 1 ) . Bioinformatic analysis revealed that almost half of these mutated genes felt within the twelve signaling pathways identified in human PDAC tumors ²³ . Interestingly, these mouse tumors display a wide heterogeneity since none of the 146 mutated genes identified in our analysis appeared in more than one tumor (Table 1 ). Thus, limiting our therapeutic options to those targeting K-RAS signaling pathways.

Table 1. Genes mutated in PDAC tumors of KPeCmice analyzed by exomic NGS sequencing.

^* Numbers in parentheses indicate distinct mutations appearing within the corresponding gene.

Generation of a“therapeutic” PDAC model for the genetic evaluation of anti-tumor and toxic effects of therapeutic targets

Target ablation at the time of tumor initiation does not reflect therapeutic intervention in the clinic. Moreover, in most studies, targets are selectively ablated in selected tissues or in those cells that express the oncogenic insult(s) (Drosten et al., 2017; Perez-Mancera et al., 2012). These strategies fail to provide information regarding the toxic effects that might occur in the clinic when the targets are inhibited via systemic administration of the corresponding inhibitors. Therefore, we have developed a GEM strain that separates temporally and spatially tumor development from target ablation/inhibition. This strain, ras ^{+ FSFG12V} : Trp5 ^M : E/as- tT A/ 7efO- FI pO :Tg . /SC-CreERT2, designated as KPeFC, incorporates two distinct recombinases, FlpO and CreERT2. FlpO, responsible for tumor induction, is expressed by the same Tet-Off system used in the KPeC strain. Indeed, KPeFC and KPeC mice develop PanIN lesions and PDACs with complete penetrance and similar kinetics (Figure 7A, 7B). Expression of the tamoxifen (TMX)-inducible CreERT2 recombinase is driven by the promoter of the human Ubiquitin C gene ( UBC ), a locus expressed in all adult tissues (Ruzankina et al., 2007). Thus, exposure of KPeFC mice to a TMX containing diet allows the systemic recombination of any conditional floxed allele added to this strain.

Systemic ablation of EGFR and c-RAF expression in adult mice induces tolerable toxicities

Many therapies fail in the clinic due to unacceptable toxic effects (Gewirtz et al., 2010; Hwang et al., 2016). Thus, we examined whether concomitant, systemic ablation of EGFR and c-RAF expression could be well tolerated in mice. To this end, we exposed 12 week old Egf^ ^mnm RafV ^mnm ,Tg. UBC-CreERT2 mice to a TMX containing diet for 15 weeks ^{(n=5 males} _’ n=s females) This treatment resulted m efficient recombination of both floxed alleles (Figure 2A). Egfr ^+l+ ,Raf1 ^+l+ g. UBC-CreERJ2 siblings were used as controls Mice lost weight during the initial treatment, yet they recovered a few weeks later (Figure 2B). Egf^^RafV^ ^x g L/SC-CreERT2 mice developed skin alterations such as hyperplasia and disorganization of the epidermis, hyperkeratosis, folliculitis and inflammation with increased numbers of mast cells and significant hair loss (Figure 2C). Moreover, these animals occasionally developed ulcers and scabs (Figure 2C). These toxic effects were similar to those previously observed in mice lacking EGFR in keratinocytes (Franzke et al.,

2012). These skin defects are highly reminiscent of the acneiform rash and folliculitis observed in human patients treated with EGFR inhibitors (Owczarczyk-Saczonek et al.,

2013). We also observed a slight disorganization of the crypts in the small intestine with increased numbers of apoptotic cells, however the overall architecture of the tissue was not affected (Figure 2D). No significant toxicities were observed in mice upon ablation of c-RAF expression. Taken together, these observations suggest that combined inhibition of EGFR and c-RAF signaling might be well tolerated by patients.

Previous studies have shown that systemic ablation of the MEK1/2 and ERK1/2 kinases results in the rapid degeneration of the intestinal and colonic crypts leading to death within two weeks of TMX exposure (Blasco et al, 201 1 ). Similar results have been observed upon ablation of the three members of the RAF kinase family but not when the systemic targeting was limited to c-RAF (Sanclemente et al., 2018). As illustrated in Figure 2A, concomitant elimination of Egfr and Raf1 in a variety of tissues did not affect either MAPK nor PI3K signaling, two of the main pathways responsible for homeostatic RAS signaling, an observation that may explain the minimal toxic effects observed upon ablation of EGFR and c-RAF expression.

Regression of advanced PDAC tumors upon systemic ablation of EGFR and c-RAF expression

Next, we assessed the consequences of systemically ablating EGFR and c-RAF expression in mice carrying advanced Kras/Trp53 mutant PDACs. Tumor bearing KPeFC (n=14) and KPeFC] Egfi ^0Xll0X ;Raf1 ^l0Xll0X mice (n=45) carrying lesions ranging from 2 to 50 mm ³ were exposed to a TMX-containing diet. Unfortunately, 4 out of 14 KPeFC and 14 out of 45 KPeFC] Egf '^iRaf ¹⁰ ^ animals had to be eliminated due to various circumstances including the appearance of unrelated tumors, mainly sarcomas and papillomas. To determine whether these tumors were a consequence of spurious expression of the FlpO recombinase, we introduced a Rosa26 ^{CAG tdTomatorEGFP} allele in KPeFC mice. These mice displayed FlpO-mediated recombinant activity in skin as well as in other tissues as revealed by the presence of dTomato ⁺ cells (Figure 14A). Moreover, most of these dTomato ⁺ cells became green upon TMX exposure due to expression of the EGFP marker mediated by the CreERT2 recombinase (Figure 14A). In addition, a significant percentage of KPeFC ]Egfi ^oxnox ;RafV ^oxllox mice (n=19) could not be included in the study due to inefficient Cre-mediated recombination (Figure 14B). As a consequence, only 10 control KPeFC and 12 KPeFC] Egf ' ;Raf†° ^x,lox mice could be evaluated in the trial.

Control KPeFC mice (n=10) died between 2 to 8 weeks following TMX exposure (Figure 3A). To our surprise, 8 of 12 KPeFC ]Egfi ^0Xll0X ;Raf1 ^l0Xll0X mice displayed a rapid decrease in tumor volume upon TMX exposure. Six mice, designated as“Regressors” (R), became tumor-free by micro-ultrasound analysis after six weeks of TMX exposure (Figure 3B, 3C). Four of these“Regressor” animals (R1 -R4) were sacrificed after six weeks of TMX exposure whereas the remaining animals, R5 and R6, were allowed to survive for 10 additional weeks. No tumor reappearance, as evaluated by micro-ultrasound analysis, was observed during this time period (Figure 3B). Detailed histological examination of their pancreata revealed normal tissue architecture (Figure 3D). One“Regressor” mouse (R2) did not display any lesion at the location where the tumor was formerly located. Yet, the other“Regressor mice (R1 , R3 and R4) exhibited single tiny scars, presumably remnants of their original tumor (Figure 3D). These scars appeared as very small fibrotic lesions measuring between 0.05 mm ³ to 0.5 mm ³ , reflecting a reduction in tumor volume over 5,000-fold (Figure 3D). They were mostly composed of a dense network of organized collagen fibers, along with a significant content of hyaluronic acid (Figure 9A, 9B). We also observed signs of chronic inflammation characterized by the presence of macrophages and T lymphocytes at their edges (Figure 9A, 9B). Scars of R1 and R6 mice contained a small percentage of Ki67 ⁺ proliferating cells that expressed significant levels of CK19 and pERK and retained EGFR expression, suggesting that they represent residual unrecombined tumor cells (Figure 3D, 3E, 3F). Indeed, some of these cells displayed atypia and loss of cellular architecture. Scars present in the remaining“Regressor” mice (R3-R5) also contained CK19 ⁺ epithelial cells organized in ductal-like structures. However, they express low levels of pERK and no EGFR. Although a few of these cells also stained for Ki67, they did not present atypia suggesting that they may not be neoplastic cells (Figure 3F).

Tumor regression appeared to be mediated by apoptotic cell death. Immunohistochemical analysis of a tumor that regressed around 30% during the first two weeks of TMX exposure (“Regressor”) revealed a 7% of cleaved Caspase 3 expression (Figure 3G). In contrast, tumors of two independent mice that continued growing during the same period of time (“Non Responder”) only displayed a 0.5% of cleaved Caspase 3 expression (Figure 3G).

Finally, the pancreata of these“Regressor” mice contained low-grade PanINs (3 to 10 per mouse) including the R2 mouse in which the original PDAC had completely disappeared. Most of these lesions expressed EGFR (Figure 10). Whether these PanINs are derived from cells that were not able to progress or represent late events during the course of the study, remains to be determined.

PDAC tumors“Resistant” to combined ablation of Egfr and Raf1

Tumors present in two mice that initially regressed with kinetics similar to those of the “Regressor” mice, started to grow rapidly after 6 to 8 weeks of TMX exposure killing them 4 to 5 weeks later (Figure 8A). Western blot analysis of tumor tissue revealed the absence of EGFR and c-RAF, indicating that tumor progression was not due to incomplete recombination of the conditional Egfr and/or Raf1 alleles (Figure 8B). Therefore, we have designated these mice as “Resistant” (T). Whether tumor progression was due to the acquisition of new mutations or to the emergence of clones that did not require EGFR and c-RAF signaling, remains to be determined. Indeed, the tumor present in the T2“Resistant” mouse had a distinct sarcomatoid phenotype as illustrated by the lack of expression of CK19 and pERK (Figure 8C).

Moreover, four KPeFC;Eg// ^j0X/l0X ,7 ^: ?af7 ^l0X/l0X mice did not respond to the TMX diet. Tumors present in these animals, designated as“Non Responders” (N), progressed similarly to those present in control KPeFC mice (Figure 4A, 4B). Histopathological analyses did not reveal significant differences with tumors present in control animals or in KPeFC;Eg / ^j0X/l0X ; F?af/' ^ox/ ' ^ox mice not exposed to TMX diet (Figure 4D). These tumors did not express EGFR or c-RAF, yet they retained active MAPK and PI3K AKT signaling pathways (Figure 4C). Thus, we hypothesized that these tumors must have undergone additional alterations that made them independent of EGFR/c-RAF signaling. Alternatively, they may have originated from a putative distinct type of acinar cell that does not require these signaling pathways for proliferation. Transcriptional differences between cells derived from “Regressor” and “Non Responder” pancreatic tumors

To gain insights into the mechanisms responsible for the differential responses of these PDAC tumors to EGFR and c-RAF ablation, we generated tumor cell lines from 15 KPeF;Eg / ^j0X/l0X ;/ ^: ?af1 ^l0X/l0X mice that were not enrolled in the preclinical trial because they lacked the UBC- CreERT2 transgene. Elimination of EGFR and c-RAF expression upon infection with AdCre particles led to apoptotic cell dead, as determined by cell cycle analysis, in 4 of these 15 cell lines designated as“Regressors”, RC. In contrast, other cell lines (4 out of 15) were completely resistant to cell death. They were designated as“Non Responder” cells, NO. The results obtained with 3 RC and 3 NC cell lines are illustrated in Figure 5A. The remaining seven cells lines displayed a mixed phenotype with various percentages of cells undergoing cell death upon AdCre infection. Thus, suggesting the existence of intratumoral heterogeneity in these experimental tumors (McGranahan and Swanton, 2017).

Analysis of known K-RAS effectors in NC and RC cells failed to demonstrate significant differences in the phosphorylation levels of the ERK and AKT kinases as well as in pCOFILIN (Figure 5F), suggesting that the MAPK, PI3K and ROCK1 pathways might not be responsible for the proliferation of NC cells in the absence of EGFR and c-RAF expression. Interestingly, ablation of EGFR and c-RAF expression in NC cells induced increased phosphorylation of STAT3 at the canonical Tyr705 residue (Figure 5F). These observations were further substantiated by IHC analysis (Figure 5G, 5H). Tumors present in“Non Responder” mice as well as in mice that became resistant to Egfr and Raf1 ablation, constitutively expressed high levels of nuclear pSTAT3 (Figure 5G, 5H). No increase in pSTAT3 expression was observed in control KPeFC tumors or in those few ductal-like cells present in the residual scars of“Regressor” mice (Figure 5G, 5H).

Next, we determined the transcriptional profiles by RNAseq analysis of three RC cell lines and three NC cell lines. As illustrated in Figure 6A, RC and NC cells displayed distinct transcriptional profiles that included more than two thousand differentially expressed genes. Gene Set Enrichment Analysis (GSEA) (Liberzon et al., 2015) identified several pathways enriched in RC vs NC cells (Figure 5D). The most significantly enriched gene signatures in RC cells included those corresponding to“bile acid, cholesterol, xenobiotic and fatty acid metabolism”,“apoptosis” and“p53 pathway” (Figure 5B). Significantly enriched gene sets in the NC cells were those corresponding to“E2F targets”,“EMT” and“MYC targets”. Other enriched pathways included the “PI3K/AKT/mTOR” and “IL6/JAK/STAT3” signaling pathways (Figure 5D). Comparison of data obtained by RNAseq analysis with a transcriptional classification of human PDACs (Bailey et al., 2016), revealed that NC cells displayed a transcriptional profile most similar to the“squamous subtype”. In contrast, RC cells fit best with the other classifications,“Immunogenic”,“ADEX” and“Progenitor” (Bailey et al., 2016) (Figure 5E). A list of 57 genes selected from pathways known to play relevant roles in PDAC development and progression is highlighted in Figure 5E2. EGFR and c-RAF are essential for proliferation of patient-derived pancreatic tumor xenografts (PDX)

To determine whether combined inhibition of EGFR and c-RAF signaling could provide therapeutic benefit to PDAC patients, we knocked down their expression in cells derived from nine PDX tumor models harboring KRAS and TP53 mutations (Table 4). A tenth PDX tumor model (PDX-10) carrying a wild-type TP53 was also included in the study. Individual knockdown of EGFR or c-RAF expression with two independent shRNAs for each locus reduced their proliferative properties to various extents. However, combined knockdown of EGFR and c-RAF expression completely interfered with the proliferative capacity of those cell derived from nine out of the ten PDX tumor models (Table 2 and figure 12). Only those cells derived from PDX-6 were partially inhibited upon EGFR or c-RAF knockdown. Four of the PDX-derived tumor cells that fully responded to EGFR or c-RAF knockdown (PDX-1 to 4) were injected into immunocompromised mice. Again, only the combined knockdown of EGFR and c -RAF effectively inhibited growth of these human PDAC tumor cells in vivo (Figure 6). These observations suggest that combined inhibition of EGFR and c-RAF expression may have significant therapeutic activity in human PDAC tumors.

Table 2. Original designation and K-RAS and TP53 mutations of the PDX tumors used in this study.

Finally, we determined whether pharmacological inhibition of the kinase activities of EGFR and c-RAF led to similar results. Unfortunately, none of three different c-RAF kinase inhibitors (MLN2480, GW5074 and PLX8394) displayed significant inhibitory activity in various in vitro and in vivo assays (Sanclemente et al., 2018). Thus, we had to inhibit c-RAF by knocking down its expression with specific shRNAs as described above. To block EGFR activity, we used two independent inhibitors, Gefitinib and Erlotinib (Burotto et al., 2015; Lim et al., 2014). Exposure of cells derived from the ten independent PDX tumors described above to Gefitinib at IC ₅ o concentrations determined in short term cultures (Table 3) along with the two independent c-RAF shRNAS resulted in the complete inhibition of their proliferations, including those cells derived from the PDX-6 tumor model that were partially resistant to EGFR and c-RAF knockdown (Figure 13). Moreover, cells derived from 7 out of 10 PDX tumor models underwent cell death resulting in a reduced number of cells at the end of the 12-day experiment (Figure 13). Similar results were obtained with Erlotinib, except for PDX-3 and PDX-6-derived cells that only displayed partial inhibition (Figure 13). These results, taken together, suggest that pharmacological intervention may also result in significant inhibition of human PDAC tumors in the clinic. Table 3. IC50 of Gefitinib and Erlotinib for each of the PDX-derived cells.

Mice

Elas-tT N 7efO-C re /Kras ^{+/LSLG12 Vgeo} , Trp53 ^oxllox , Eg ^oxnox , Raf1 ^loxnox , ras ^+/FSFG12V , Trp53f ^Mn , Tg. L/SC-CreERT2 and Rosa26 ^{+AmF3(CAGddTomat} ° ^rEGFP)Pien/J strains have been previously described (Guerra et al., 2007; Jesenberger et al., 2001 ; Jonkers et al., 2001 ; Lee et al., 2012; Natarajan et al., 2007; Nieto et al., 2017; Plummer et al., 2015; Ruzankina et al., 2007). The Cc//c4 ^K35M/K35M strain was obtained by a Cre-dependent FLEx switch strategy that replaced expression of the wild-type CDK4 protein by a CDK4 ^K35M kinase dead isoform (Schnutgen et al., 2003) (Schnutgen et al., 2003). The transgenic Te/O-FlpO strain was generated by pronuclear injection of CMV- Te/O-FlpO DNA into B6.CBA zygotes (Pease and Saunders, 201 1 ). All mice were maintained in a mixed 129/Sv-C57BL/6 background. Immunodeficient NU -Foxnlnu mice (females, 5-weeks-old) were purchased from Harlan Laboratories. All animal experiments were approved by the Ethical Committees of the Spanish National Cancer Research Centre (CNIO) and they were performed in accordance with the guidelines stated in the International Guiding Principles for Biomedical Research Involving Animal, developed by the Council for International Organizations of Medical Sciences (CIOMS). All strains were genotyped by Transnetyx (Cordova, Tennessee, USA). Mouse treatments and tumor monitorization

Tumors were measured with a micro-ultrasound system (Vevo 770, Visualsonics) with an ultrasound transducer of 40 MHz (RMV704, Visualsonics). To this end, mice were anesthetized with a continuous flow of 1 % to 3% isoflurane in 100% oxygen at a rate of 1 .5 liter/min. Hypothermia associated with anesthesia was avoided using a bed-heater. Abdominal hair was removed by depilation cream to prepare the examination area. Tumor size was calculated as Length x Width ² /2. Recombination of the Egff° and Raff ^ox conditional alleles was mediated by activation of the inducible CreERT2 recombinase with TMX. To this end, mice were fed with a TMX-containing diet (Teklad CRD TAM400 diet, Harlan) ad libitum. Control mice carrying the corresponding wild-type alleles were also fed with the same diet.

Histopathology and immunohistochemistry.

For routine histological analysis, specimens were fixed in 10% buffered formalin (Sigma) and embedded in paraffin. For histopathological analysis, tissues were serially sectioned (3pm thick) and stained by conventional H&E every ten sections. Antibodies used for immunostaining included those raised against: CK-19 (CNIO Monoclonal Antibodies Core Unit), cleaved Caspase-3 (Cell Signaling Technology; 9661 ), CD3 (Santa Cruz Biotechnology, M-20), Egfr (Abeam, ab52894), pErk (Cell Signaling Technology; 9101 ), F4/80 (ABD Serotec, Cl: A3-1 ), Gata6 (R&D Systems, AF1700), Ki67 (Master Diagnostica, 00031 10QD), HABP (Millipore, 38591 1 ) and pStat3 (Cell Signaling Technology; 9145). Stained slides were scanned using the Mirax scanner (Zeiss). Images were analyzed by Zen2 software and photos were exported using Zen 2 software (Zeiss).

X-Gal staining, Laser Capture Microdissection and PCR analysis.

X-Gal staining, laser capture microdissection and Egfr PCR analysis have been previously described ¹² . c-Raf wM type, floxed and null alleles were identified with forward c-fiaf 1 F (SEQ ID NO: 1 : 5 ^' -CTGATTGCCCAACTGCCATAA-3 ^' ), c-Raf 3F (SEQ ID NO: 2: 5 ^' - GAGTCAGCAAATGCACTGAAATG-3 ^' ) and reverse c-Raf 1 R (SEQ ID NO: 3: 5 ^' - ACTGATCTGGAGCACAGCAAT-3 ^' ) primers at 94 °C for 1 minute, followed by 35 cycles of denaturation at 94 °C for 30 seconds, annealing at 60 °C for 30 seconds and extension at 72 °C for 30 s, and finally, followed by a long extension at 72 °C for 10 minutes. These primers yielded DNA products of 196 bp, 270 bp and 143 bp for wild-type, floxed and null c-Raf alleles respectively.

EGFP and tdTomato fluorescence imaging

Tissues were fixed overnight by immersion in 4% paraformaldehyde (PFA) in 0.01 M phosphate-buffered saline (PBS) at 4°C and rinsed in PBS before equilibration in 30% sucrose in PBS for 48 h at 4°C. Samples were thereafter included in O.C.T.™ compound (Sakura) and frozen. Cryosections of the samples were stained with Dapi for nuclei detection (ThermoFisher), mounted with Prolong Gold antifade reagent (ThermoFisher) and visualized with a TCS-SP5 laser scanning confocal microscope (Leica) equipped with AOBS and both 10X/0.4NA and 20X/0.7NA dry objectives. A z-stack was acquired and the maximum projection is shown.

Apoptosis Assay

After 96 hr of infection with AdCre particles cells were harvested by trypsinization and fixed with 70% (v/v) ethanol at 4°C overnight. Fixed cells were incubated in phosphate-buffered saline (PBS) containing 100 pg/ml RNase A for 30 minutes at 37°C, followed by staining with 0.003% of Propidium Iodide for 30 minutes on ice. Thereafter, cells were collected on a nylon mesh filter (pore size, 40 mm), and cell cycle was assayed by flow cytometry (FACSCalibur) at excitation of 488 nm and at emission of 585 nm, and analyzed using a FACSDiva Version 6.1 .2 (BD Bioscience).

Mouse PDAC cell cultures

To generate mouse PDAC explants, freshly isolated tumors were minced with sterile razor blades, digested with collagenase P (1 .5pg/ml) in Hank’s Balanced Salt Solution (HBSS) for 30 min at 37 ^Q C, and cultured in DMEM with 10% of fetal bovine serum (FBS) and 1 % Penicillin/Streptomycin. All studies were done on cells maintained in culture for less than ten passages. Their corresponding genotypes were verified by PCR analysis. PDAC cells explants were infected with Adeno-Cre particles (multiplicity of infection, 100) and seeded for colony formation assay 5 days after. Adeno-GFP particles were used as negative controls. Cells were seeded in equal cell numbers (5x 10 ³ ) and allowed to form colonies for 2 weeks. Plates were fixed with 0.1 % glutaraldehyde (Sigma) and stained with 0.5% Crystal Violet (Merck). Colonies were counted and quantified.

PDX tumor models.

PDX tumors models were used include Panc-1 , Panc-2, Panc-4, Panc-185, Panc-198, H- PDAC-H-X132, H-PDAC-M-X3 and H-PDAC-M-X7 (Table 4). Table 4. KRAS and TP53 mutations in PDX tumor models.

Panc-1 , Panc-2, and Panc-4 were obtained from patients who underwent surgical resection at the Kog University Hospital, Istanbul, Turkey with approval by the Ethical Committee (CEI 60-1057-A068). Panc-185, Panc-198, H-PDAC-H-X132 were obtained from Hospital HM Sanchinaro, Madrid, Spain, with approval by the Ethical Committee (CEIC HM Hospitales, FHM.06.10). H-PDAC-M-X3 and H-PDAC-M-X7 were obtained from the Hospital Virgen de la Arrixaca, Murcia, Spain, with approval by the Ethical Committee (CEIC HCUVA-2013/01 ). Specific informed consent for PDX model generation was obtained from all patients. PDAC003T and PDAC013T tumor models have been already described (Nicolle et al., 2017).

PDX knockdown assays

Cells derived from these PDX tumor models were infected with lentiviral supernatants expressing shRNAs against EGFR (E1 , TRCN0000121203 and E2, TRCN0000121206), c- RAF (R1 , TRCN0000001065 and R2, TRCN0000001068). The E1 , E2 and R2 shRNAs were cloned in a plasmid that carries a Puromycin resistant cassette. Instead, the R1 shRNA was cloned in a plasmid that conferred Blasticidine resistance. A scrambled shRNA control vector was used as a negative control. Infected PDX cells were seeded in 96-well plates at a density of 1 ,500 cells per well and proliferation was assessed using the MTT assay. For in vivo studies, infected cells (0.5x10 ⁶ ) derived from PDAC003T, PDAC013T, Panc-1 and Panc-4 tumor models were injected 1 : 1 in PBS:Matrigel Matrix (Corning, 354234) into dorsal flanks of immunodeficient mice. Tumor growth was measured every 3 days with a caliper and calculated as Length x Width ² /2 until humane end point.

Western blot analysis

Protein extracts (25 pg) obtained from tumor tissue or cell lines were separated on SDS/PAGE gels (Thermo Fisher Scientific), transferred to a nitrocellulose membrane and blotted with antibodies raised against Egfr (Abeam, ab52894), c-Raf (BD Biosciences, 610151 ), Erk-1 (BD Biosciences 554100), Erk-2 (BD Biosciences, 610103), pErk1/2 (Cell Signaling, 9101 ), Akt (Cell Signaling, 9272), pAkt (Cell Signaling, 4060), Stat3 (Cell Signaling, 9132), pStat3 (Cell Signaling, 9131 ), pCofilin (Santa Cruz, sc-21867-R) and Gapdh (Sigma, G8795). Primary antibodies were detected against mouse or rabbit IgGs (HRP, Dako and Alexa Fluor 680, Invitrogen) and visualized with ECL Western blot detection solution (GE Healthcare) or Odyssey infrared imaging system (LI-COR Biosciences).

Next generation sequencing

Genomic DNA obtained from 1 1 paired tumor and tail tissue was enriched in protein-coding sequences using the SureSelect Mouse All Exon kit (Agilent Technologies). The resulting target-enriched pool was amplified and subjected to paired-end sequencing (2 c 100 bp) using HiSeq2000 sequencing instruments at the Beijing Genomics Institute (BGI). Sequencing reads were mapped to the reference genome (mm9) using the Burrows- Wheeler Aligner (BWA) (Li and Durbin, 2010) alignment tool version 0.5.9. Sites that differed from the reference genome (variants) were identified and empirical priors were constructed for the distribution of variant frequencies in each sample independently. High- credibility intervals (posterior probability > 1—1 e-5) were obtained for the observed frequency of the variants using the statistical algorithm for variant identification (SAVI) algorithm (Trifonov et al., 2013). Variants were considered absent if their allele frequency was <2% and present if detected with an allele frequency above 15%, corresponding with the sensitivity threshold of direct Sanger sequencing. Variant total depth was also required to be >10x and <300x. Variants were excluded if present in mouse dbSNP database, detected in any of the normal samples, or observed in only one strand. Finally, candidate protein altering somatic variants (nonsense, missense, and small insertions and deletions) were identified when variants were absent in the normal and present in the tumor with at least 1 % change in frequency from normal with high posterior probability (> 1 - 1 e-5). RNAseq and GSEA analysis

RNA from PDAC cell explants was extracted with Qiagen RNeasy Mini Kit. 1 pg of total RNA was used for further analysis. PolyA ⁺ fraction was purified and randomly fragmented, converted to double stranded cDNA and processed through subsequent enzymatic treatments of end-repair, dA-tailing, and ligation to adapters as in lllumina's "TruSeq Stranded mRNA LT Sample Prep Kit". The adapter-ligated library was completed by PCR with lllumina PE primers. The resulting purified cDNA library was applied to an lllumina flow cell for cluster generation and sequenced on an lllumina NovaSeq 6000 instrument by following manufacturer's guidelines. 101 bp single-end reads were analyzed with the Nextpresso pipeline (Grana et al., 2017) as follows: sequencing quality was verified with FastQC vO.1 1 .0 (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Reads were aligned to the mouse genome (NCBI37/mm9) with TopHat-2.0.10 (Trapnell et al., 2012) using Bowtie 1 .0.0 (Langmead et al., 2009) and SAMtools 0.1 .19 (Li et al., 2009), allowing 2 mismatches and 20 multihits. Differential expression was tested with DESeq2 (Love et al., 2014) using the Mus musculus NCBI37/mm9 transcript annotations from https://ccb.ihu.edu/software/tophat iqenomes.shtml. GSEAPreranked (Subramanian et al., 2005) was used to perform a gene set enrichment analysis of the described gene signatures on a pre-ranked gene list, setting 1000 gene set permutations. Gene Set Variation Analysis (GSVA) (Hanzelmann et al., 2013), was used to estimate the variation of pathway activity over the samples in an unsupervised manner. Heatmaps presented in this study were built with GENE-E software package (https://software.broadinstitute.org/GENE-E/index.html). IC ₅ o Determinations

PDX cell lines were plated at 5,000 cells per well in triplicates in 96-well plates and grown for 24 hours. Cells were treated with a dilution series of Gefitinib (Cymit Quimica SL), Erlotinib (LC laboratories). Control cells were incubated with media containing DMSO. Cell viability was assessed with CellTiter Glo Luminescent Cell Viability Assay after 72 hours of treatment. Luminescence counts were read in a Victor Instrument (Perkin Elmer) with the recommended settings. To calculate the IC ₅₀ , values were plotted against the inhibitor concentrations and fit to a sigmoid dose-response curve using GraphPad Software.

Gefitinib and Erlotinib treatment of PDX-derived cells

For pharmacologic studies PDX-derived cells were seeded in 96-well plates at 1 ,500 and 3,000 cells/well in triplicates, and incubated for 24 hours in DMEM media supplemented with 10% FBS, 2 mM L-glutamine, 50 U/ml penicillin and 50 pg/ml streptomycin (GIBCO- Invitrogen) before adding the IC ₅₀ of the corresponding IC50 concentration of inhibitor in DMSO. The same concentration of DMSO was used as a control. Cells were exposed to the corresponding inhibitor for 12 days, in the presence or absence of a c-RAF shRNA (R1 ) changing medium and drug every two days. Cell viability was assessed with CellTiter Glo Luminescent Cell Viability.

Data and software availability

Next generation sequencing data have been deposited in the Gene Expression Omnibus (GEO) database under the accession number GSE1 12434 as well as in the NCBI’s Sequence Read Archive (SRA) under a BioProject with the accession number PRJNA462276.

Quantification and statistical analysis

Data are represented as mean ± SD. Significance was calculated with the unpaired Student ^' s t test using GraphPad Prism software. A p value that was less than 0.05 was considered to be statistically significant for all data sets. Significant differences between experimental groups were: ^* p< 0.05, ^** p< 0.01 or ^*** p< 0.001 .

REFERENCES

1 . Rahib, L. et al. Cancer Research 74, 2913-2921 (2014).

2. Hidalgo, M. N. Engl. J. Med. 362, 1605-1617 (2010).

3. Burris III, H. A. et al. 15, 2403-2413 (1997).

4. Moore, M. J . et al. J. Clin. Oncol. 25, 1960-1966 (2007).

5. Von Hoff, D. D. et al. N. Engl. J. Med. 369, 1691 -703 (2013).

6. Garrido-Laguna, I. & Hidalgo, M. Nat. Rev. Clin. Oncol. 12, 319-334 (2015).

7. Maitra, A. & Hruban, R. H. Annu. Rev. Pathol. 3, 157-88 (2008).

8. Ying, H. et al. Cell 149, 656-670 (2012).

9. Collins, M. A. et at. J. Clin. Invest. 122, 639-653 (2012).

10. Ostrem, J. M. et al. Nature 503, 548-551 (2013).

1 1 . Cox, A. D. et al. Nat. Rev. Drug Discov. 13, 828-851 (2014).

12. Navas, C. et al. Cancer Cell 22, 318-30 (2012).

13. Ardito, C. M. et al. Cancer Cell 22, 304-317 (2012).

14. Sanclemente, M. et al. Cancer Cell 1 -12 (2018). doi:10.1016/j.ccell.2017.12.014

15. Blasco, R. B. et al. Cancer Cell 19, 652-663 (201 1 ).

16. Guerra, C. et al. Cancer Cell 11 , 291-302 (2007).

17. Aguirre, A. J. et al. Genes Dev. 17, 31 12-3126 (2003).

18. Bardeesy, N. et al. Proc. Natl. Acad. Sci. U. S. A. 103, 5947-5952 (2006).

19. Hingorani, S. R. et al. Cancer Cell 7, 469-483 (2005).

20. Guerra, C. et al. Cancer Cell 19, 728-739 (201 1 ).

21 . Hingorani, S. R. et al. Cancer Cell A, 437-450 (2003).

22. Chung, W.-J. et al. Proc. Natl. Acad. Sci. U. S. A. 114, E10947-E10955 (2017).

23. Jones, S. et al. Science (80-. ). 321 , 1801 -1806 (2008).

24. Drosten, M. et al. Cold Spring Harb. Perspect. Med. a031542 (2017). doi:10.1 101 /cshperspect.a031542

25. Perez-Mancera, P. A. et al. Gastroenterology 142, 1079-1092 (2012).

26. Ruzankina, Y. et al. Cell Stem Cell 1 , 1 13-26 (2007).

27. Owczarczyk-Saczonek, A. et al. Postep. Dermatologii i Alergol. 30, 195-198 (2013).

28. Bailey, P. et al. Nature 531 , 47-52 (2016).

29. Kamerkar, S. et al. Nature 546, 498 (2017).

30. Todoric, J. et al. Cancer Cell 32, 824-839. e8 (2017).

31 . Chio, I. I. C. et al. Cell 166, 963-976 (2016).

32. Moffitt, R. A. et al. Nat. Genet. 47, 1 168-1 178 (2015). 33. Collisson, E. A. et al. Nat. Med. 17, 500-503 (201 1 ).

34. Mueller, S. et al. Nature 554, 62-68 (2018).

35. Raphael, B. J. et al. Cancer Cell 32, 185-203. e13 (2017).

36. Torres, C. & Grippo, P. J. Ann. Med. 1 -1 1 (2018).

37. Jonkers, J. et al. Nat. Genet. 29, 418-25 (2001 ).

38. Natarajan, A. et al. Proc. Natl. Acad. Sci. U. S. A. 104, 17081 -6 (2007).

39. Jesenberger, V. et al. J. Exp. Med. 193, 353-364 (2001 ).

40. Nieto, P. et al. Nature 548, 239-243 (2017).

41 . Lee, C.-L. et al. Dis. Model. Mech. 5, 397-402 (2012).

42. Pease, S. & Saunders, T. L. (Springer, 201 1 ).

43. Nicolle, R. et al. Cell Reports 21 , 2458-2470 (2017).

44. Graha, O. et al. Curr. Bioinform. 12, (2017).

45. Trapnell, C. et al. Nat. Protoc. 7, 562-578 (2012).

46. Langmead, B. et al. Genome Biol. 10, R25 (2009).

47. Li, H. et al. Bioinformatics 25, 2078-9 (2009).

48. Love, M. I. et al. Genome Biol. 15, (2014).

49. Subramanian, A. et al. Proc. Natl. Acad. Sci. 102, 15545-15550 (2005).

50. Hanzelmann, S. et al. BMC Bioinformatics 14, 1-20 (2013).

51 . Aguirre, A.J. et al. (2018). Cancer Discov. CD-18-0275.

52. Biankin, A. V. et al. (2012). Nature 491, 399^105.

53. Franzke, C.-W., et al. (2012). J. Exp. Med. 209, 1 105-1 1 19.

54. Gewirtz, D.A. et al. (2010). Curr Opin Investig Drugs 11, 612-614.

55. Guerra, C. et al. (2003). Cancer Cell 4, 1 1 1 -120.

56. Hwang, T.J. et al. (2016). JAMA Intern. Med. 176, 1826-1833.

57. Li, H., and Durbin, R. (2010). Bioinformatics 26, 589-595.

58. Liberzon, A. et al. (2015). Cell Syst. 1 , 417-425.

59. Plummer, N.W. et al (2015). Development. 142(24) :4385-93.

60. Puyol, M. et al. (2010). Cancer Cell 18, 63-73.

61 . Schnutgen, F. et al. (2003). Nat. Biotechnol. 21 , 562-565.

62. Tiriac, H. et al. (2018). Cancer Discov. CD-18-0349.

63. Trifonov, V. et al. (2013). BMC Syst. Biol. 7, S2.

64. Waddell, N. et al. (2015). Nature 518, 495-501 .

65. Witkiewicz, A.K. et al. (2015). Nat. Commun. 6, 6744.