Background, Materials and methods, Antibodies and reagents, Animal procedures, Reverse-phase protein Array, Real-time PCR, Human transcriptome array, PKH26 staining, Lentiviral infection, Flow cytometry, cell cycle analysis and cell sorting, Immunofluorescence, Western blotting, RNA interference, Cell viability assay, Statistical analyses, Results, Global pathway analysis shows the activation of chemoresistance-related factors in QCSCs, ZEB2, pCRAF and pASK1 are coexpressed upon chemotherapy and coregulated in CRC cells, ZEB2 expression induces tumor transition towards a slow growing chemoresistant state, ZEB2 expression correlates with worse prognosis and CMS4 in CRC patients, Discussion, Conclusions, Supplementary information

Background

The existence of cancer cells able to survive antineoplastic drugs and to regenerate  a local or distant tumor  undermines the  effectiveness  of  cancer   therapies.   Drug  resistance   is tightly  connected  to  the  presence   of  cancer   stem   cells (CSCs) responsible  for  tumor  progression,  metastatization and  recurrence [1, 2]. Therapy-resistant cells with features of stalled/delayed  cycling have been identified in solid and hematologic tumors including melanoma, glioblastoma, me- dulloblastoma,  leukemia,  lung, breast,  pancreatic  and  ovar- ian cancer [3–14], suggesting that a population of quiescent/ slow proliferating  cancer stem cells (QCSCs) may represent an  essential  tool  by which  tumors  resist  to  external  chal- lenges. Additionally, quiescence  is typical also of tumor  cells present  in the bloodstream, disseminated in the bone  mar- row  or  within  lymph  nodes  (that  altogether   account   for minimal residual disease), suggesting that quiescent cells represent a crucial therapeutic target [15]. In colorectal can- cer (CRC), QCSCs were identified as cells able to reactivate upon  serial transplantation [16, 17], to survive chemother- apy and to endure  metabolic stress [18, 19]. Recently, two distinct  populations of slow cycling cells were identified  in CRC with different strategies. A label-retaining approach identified  dormant CRC cells as a differentiated  population with enhanced clonogenic  capacity and  high levels of Wnt and Hedgehog  signaling [20]. Differently, a histone  2B-GFP (H2B-GFP) pulse-chase  approach  identified a population of slow cycling cells characterized by expression  of the TET2 dioxygenase and by enhanced chemoresistance [21]. The quiescent/drug  resistant   state   in  solid  tumors   is  tightly linked to tumor  heterogeneity and in particular  to the ability of cancer cells to undergo  epithelial-to-mesenchymal transi- tion  (EMT), an  epigenetic  programme that  crucially regu- lates the stemness, chemoresistance and invasive ability of cancer cells [22]. According  to its pleiotropic  effects on cel- lular phenotype  and function, EMT recruits a series of genes with  multiple  functions  in embryogenesis  and  carcinogen- esis such as ZEB1, ZEB2, SNAI1, SNAI2 and TWIST1  [23]. Among  these,  ZEB2 has been  shown  to regulate  epithelial cell plasticity and proliferation,  but also to balance stemness and  differentiation,  standing  as a master  regulator  of cell state transitions [24–26]. Notably, ZEB2 was also recently recognized  as a factor implicated  in drug resistance  in CRC through  FBXW7 E3-ubiquitin ligase binding [27]. Moreover, ZEB2 expression  was associated  with  poor  oncologic  out- come and distant recurrence, emerging as a new clinical bio- marker  in CRC [28]. In this study, we aimed to isolate and characterize  a population of cells with combined  features of quiescence  and  therapy  resistance  that  is  present   in  un- treated   colorectal   tumors   and  becomes  largely  prevalent upon  chemotherapy treatment.  In  line  with  our  previous studies  showing  that  PKH-retaining tumor cells were endowed  with higher tumorigenic capacity and chemother- apy resistance  [14, 16], we undertook an in-depth molecular characterization of PKH26+   cells isolated  from  CRC xeno-grafts through  gene expression analysis and reverse-phase proteomic arrays, providing for the first time a combined picture  of both transcriptional circuits and activated protein pathways. New insights on the molecular factors that or- chestrate  quiescence  programs  will likely open  new thera- peutic  avenues  to eradicate  non-proliferating cancer  cells, both in primary tumors  and at premetastatic sites.

Materials and methods

Primary colorectal cancer cells and cell lines

Colorectal cancer (CRC) specimens were obtained from patients  undergoing surgical resection  upon informed consent   and  approval  by the  Sapienza-Policlinico   Um- berto I Ethical Committee (RIF.CE: 4107 17/10/2016). Tissue  samples  were collected  by a pathologist  immedi- ately after surgery, quickly washed 2–3 times in cold phosphate buffered  saline (PBS) and  then  transferred in Dulbecco’s modified Eagle’s medium  (DMEM; Thermo Fisher  Scientific) containing  3% penicillin-streptomycin- amphotericin  B solution   (Lonza)  until  processing.   For tissue  dissociation,  CRC samples  were  first washed  3–4 times  in  PBS,  then   cut  by  forceps   and/or  scalpel  in pieces  of  approximately   0.5 mm  or  smaller.  Fragments were further  washed twice by centrifugation at 150 g for 3 min, then  incubated  in DMEM with 1.5 mg/ml  collage- nase  type  II  (Thermo  Fisher  Scientific)  and  20 mg/ml DNAse (Roche Diagnostics)  for 1 h at 37 °C under  shak- ing.  The  cell  suspension   was then filtered through a 100 μm nylon mesh  and washed by 2 further  centrifuga- tion steps in DMEM. Pellets were resuspended in Colo- rectal   Cancer   Spheroid   Cells  (CCSCs)   medium   [16] supplemented with 10 mM nicotinamide, 1 mM Y-27632 (both  from  Sigma-Aldrich),  20 ng/ml  human   EGF  and 10 ng/ml   human   basic  fibroblast   growth   factor   (both from  PeproTech).   The  resulting  suspension was  plated in ultra-low attachment tissue culture flasks (Corning Costar), and cultured  in humidified  atmosphere at 37 °C, 5% CO2.  Every 2 to  3 days, half of the  culture  medium was refreshed.  Clusters  of proliferating  cells became  evi- dent  after a variable length  of time, ranging  from  5 days to  3 weeks.  Cultures  in  which  no  proliferating   clusters were detected  after  4 weeks were discarded.  The  result- ing multicellular spheroid cultures were then passaged weekly  and  used  for  in  vitro  and  in  vivo experiments within the 12th passage. Genomic DNA was routinely extracted  from  CCSCs and patient-matched nontumoral tissues with the Dnasy Mini Kit (Qiagen) and used for mutation analysis  [29]  and  for  Short  Tandem   Repeats (STR) analysis. The latter was performed  with the AmpFlSTRIdentifiler  Plus Kit (Applied Biosystems) and used  to generate  a unique  STR profile for each primary CRC cell line, which  was used  to monitor purity  of the line over time and to confirm  its matching  with the ori- ginal patient  material.  CCSCs were then  routinely  tested for their ability to produce colon adenocarcinomas his- tologically identical  to the human  tumors  of origin when injected   into   NOD.Cg-Prkdcscid    Il2rgtm1Wjl/SzJ   (NSG) mice  (The  Jackson  Laboratory)  as  previously  described [29]. Primary CRC cells used in this study were obtained from  a 63 years  old  male  CRC patient  undergoing sur- gery for G3 TNM IIIC right colon tumor and displayed mutated APC, TP53, PI3KCA and  KRAS and  from  a 65 years old female CRC patient  undergoing surgery for G2 TNM   IIA  right  colon  tumor   with  mutated  APC and wild-type  KRAS, TP53, PI3KCA. SW480  cells were pur- chased  from the American  Type Culture  Collection (ATCC) and cultured  in DMEM supplemented with 10% heat-inactivated fetal bovine  serum,  100 U/ml  penicillin, and 10 μg/ml streptomycin (Thermo Fisher Scientific) at 37 °C in a 5% CO2  atmosphere. Cultured  cells (both  pri- mary and commercial lines) were routinely tested for mycoplasma  contamination with  the  PCR  Mycoplasma Test Kit (PanReac AppliChem).

Antibodies and reagents

Monoclonal  antibodies  against PROMININ-1  (AC133 epi- tope  both  pure  #130–090-423   used  for  immunofluores- cence  and  biotinylated  #130–090-664  used  for  flow cytometry, 1:10) were obtained  from Miltenyi Biotec. Monoclonal     anti-Ki67    (Dako,    Agilent    Technologies, #M7240, 1:200) and polyclonal anti-Ki67 (Santa Cruz Bio- technology, #sc-15,402, 1:200) were used for immunofluor- escence.  EpCAM-APC  used  for flow cytometry  (#347200, 1:40) was from  Becton  Dickinson.  Monoclonal  anti-ZEB2 (#sc-271,984,   1:200)  used   for  immunofluorescence  was from Santa Cruz Biotechnology. Anti-CRAF pS338 (MA1–90087,  1:100)  used   for  immunofluorescence  was  from Thermo Fisher and anti-CRAF pS338 (#56A6, 1:1000) used for western blot was from Cell Signaling Technology.  Anti- ASK1 pS83 (#3761, 1:1000), VIMENTIN  (#5741), CADHERIN-2 (#13116), SNAI1 (#3879), SNAI2 (#9585), TCF8/ZEB1 (#3396) used for western blot were from Cell Signaling Technology,  while anti-CADHERIN-1 (#610181) was from Becton Dickinson. Monoclonal  anti-β-ACTIN (#A5316, 1:10000) used for western blot was from Sigma- Aldrich.  Secondary  mouse  IgG,  HRP-linked  (#NA931,  1:4000) and rabbit IgG, HRP-linked  (#NA934V, 1:4000) were from  GE Healthcare   Life Sciences.  Secondary  antibodies, goat   anti-mouse  IgG   Alexa   Fluor®647-conjugated (#A21235, 1:1000), goat anti-rabbit IgG Alexa Fluor®555- conjugated  (#A21428, 1:1000), streptavidin  647 (S32357, 1:250), and 4′,6-diamidino-2-fenilindole (DAPI, #D1306, 100nM) were obtained  from Thermo Fisher Scientific. PKH26 (PKH26GL, Sigma-Aldrich)  for cell membrane labeling was used diluted 1:1000 and cells were stained following manu- facturer’s   instructions.  ProLong   Gold  Antifade  (#P7481) was from Thermo Fisher Scientific. Mayer’s haematoxylin (#MHS32)   and   Eosin  (#HT110232)   were  from   Sigma- Aldrich and used according  to the manufacturer’s protocol. Etoposide   (#E1383)  and   irinotecan  (#I1406)  were  from Sigma-Aldrich, oxaliplatin and 5-fluorouracil were from Peviva.  Agarose  (SeaPlaque  GTG  agarose,  #50111)  was from   Lonza.  Crystal   violet  (#C3886)  was  from   Sigma- Aldrich  and used 0.1% in 10% MetOH.  Triton  X-100 (#1610407)  was  from  Bio-Rad  Laboratories   and  used  at 0.1%. Stripping  buffer  was from  Thermo Fisher  Scientific (#21059) and used according  to the manufacturer’s proto- col. Matrigel  (Corning® Matrigel®  Growth  Factor  Reduced (GFR) Basement  Membrane Matrix)  was purcheased from Corning (#354230).

Animal procedures

All animal  procedures were performed according  to the Italian National animal experimentation guidelines (D.L.116/92) upon  approval of the experimental protocol by the Italian  Ministry  of Health’s  Animal  Experimenta- tion   Committee  (DM  n.  292/2015   PR  23/4/2015).   6- week-old female NOD-SCID mice from Charles River Laboratories  were  used  for  PKH26  experiments and  6-week-old female NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice (The Jackson Laboratory) were used for ex- ogenous  ZEB2 expression  experiments. For PKH26 ex- periments, 5 × 105  CCSCs were  injected  subcutaneously in  the  flank  of  NOD/SCID   mice,  in  100 μl  1:1  PBS/Matrigel  (BD Biosciences). Tumors were measured  twice weekly by an external digital caliper, and volumes were calculated  using  the  following  formula:  π/6  x  d2  x  D, where d and D represent shorter  and longer tumor measurements, respectively. Mice were grouped and sacrificed  at different  time  point  (1, 3, 6 weeks after  in- jection)  for subsequent studies.  For exogenous  ZEB2 ex- pression  experiments, 104   CCSCs  or  SW480  cells transduced with pLenti-GFP  and pLenti GFP-ZEB2 were injected  subcutaneously in the flank of NSG mice as de- scribed  above. Drug treatments started  when tumor  vol- ume  reached   50–100 mm3.   Mice  were  randomized  in control  and  treatment group  and  treated  with  12,5 mg/kg 5-fluorouracil and  5 mg/kg  oxaliplatin  intraperitone- ally weekly. Control   animals  were  treated   with  vehicle only. Tumor  growth  was measured  at the indicated  time points.  Animals  were  euthanized  according   to  the  na- tional Animal Welfare Guidelines.

Reverse-phase protein Array

Following FACS separation,  CCSCs were promptly  lysed in  10 μl  extraction   buffer  [50%  2X  Tris-Glycine   SDS Sample  Buffer  (Life  Technologies),   47.5% 1X  with  T- PER reagent  (Thermo Fisher Scientific and 2.5% Tris (2- carboxyethyl) phosphine hydrochloride (TCEP) reagent (Thermo  Fisher  Scientific)].  Lysates  were  boiled  for  3 min  and stored  at − 80 °C until  further  processing.  Prior to  printing   on  nitrocellulose  slides  (GRACE  Bio-Labs Inc.) via a robotic  arrayer  (Aushon  Biosystems), samples were thawed  and  boiled  3 min.  In order  to  increase  the amount  of protein  deposited  on each slide, printing  was performed by using  5 depositions  per  spot  and  samples were printed in technical triplicates. Reference standard lysates, i.e. HeLa + Pervanadate  (Becton, Dickinson  and Company),  A431 + EGF (Becton, Dickinson  and Company), Jurkat + Etoposide (Cell Signaling Technology) and Jurkat + Calyculin  A (Cell Signaling  Technology),  were  printed  in 10-point  decreasing  mixtures  of treated  to untreated sam- ples as procedural  controls  and as positive controls  for anti- body staining. Each reference  standard curve was printed  in technical  triplicate  at a final concentration of 0.5 mg/ml.  A selected subset of the printed  microarray  slides were stained with Sypro Ruby Protein  Blot Stain (Thermo Fisher Scien- tific) to estimate  sample total protein  concentration and the remaining  slides were stored  under  desiccated  conditions at − 20 °C. Immediately  before antibody staining, printed  slides were treated  with 1X Reblot Mild Solution  (Chemicon)  for 15 min, washed 2×5 min with PBS and incubated  for 2h in blocking  solution  containing   2% I-Block  (Applied  Biosys- tems) and 0.1% Tween-20  in PBS. Immunostaining was car- ried out using a tyramide-biotin signal amplification  kit (DAKO). Primary antibody binding was detected  using a bi- otinylated goat anti-rabbit IgG H + L (diluted at 1:7500; Vec- tor Laboratories)  or rabbit anti-mouse Ig (diluted at 1:10, DAKO) followed by streptavidin-conjugated IRDye®-680LT fluorophore (LI-COR Biosciences). Primary antibodies underwent pre- and post-RPPA validation for single band specificity by western blot using complex cellular lysates. Negative   control   slides,  incubated   only  with   secondary antibody  were  included  in  each  staining  run.  All Sypro Ruby  and  immunostained  slides  were  scanned   using  a Tecan  Power Scanner™ (Tecan  Group  Ltd) at 5 μm reso- lution. Acquired images were analyzed with MicroVigene v5.2 (VigeneTech) for spot detection,  local and negative control  background subtraction,  replicate  averaging  and total protein  normalization. The “R” software packages ‘re- shape2’,  ‘ggplot2’, ‘coin’, ‘gplots’ and  ‘shiny’ were used  to carry  out  slide  quality  control,  internal  standardization, two-way hierarchical clustering (Euclidean distance and Ward.D2   method),   Kruskal-Wallis   and  Wilcoxon   Rank Sum non-parametric statistical  tests  (Benjamini  & Hoch- berg criterion  was used for multiple  comparisons adjust- ment  with  an  accepted  false discovery  rate  of  0.05). A detailed list of antibodies  used for RPPA is available in Additional file 1: Table S1.

Real-time PCR

Total RNA was extracted with TRIzol (Thermo Fisher Scientific) following manufacturer’s instructions. 1 μg of RNA was reverse transcribed with M-MLV reverse tran- scriptase  (Thermo Fisher Scientific) and  50 ng of cDNA were  used  as  template   in  the  PCR  reactions.   Specific probes  used  for  ZEB2, MKI67, BMI1, β-ACTIN  and NANOG were  all  from  Thermo  Fisher  Scientific (Additional   file  2:  Table  S2)  and  specific  primers   for ZEB1, CDH1, VIMENTIN, SNAI1, SNAI2, CDKN1B, (Additional  file 3: Table S3) were from Sigma-Aldrich. Normalization was performed using β-ACTIN as refer- ence.  RNA  from  xenografts   derived  from  pLenti-GFP and   pLenti   GFP-ZEB2-transduced  cells  was  extracted and  reverse  transcribed as described  above. To  analyze the expression of cell cycle-associated genes TaqMan® Array, Human  Cyclins & Cell Cycle Regulation,  Fast 96- well (Thermo Fisher Scientific) was used following the manufacturer’s  instructions.  Values  were  expressed   in terms  of 2-ΔΔCt where ΔΔCT = ΔCTsample−ΔCTcalibra-tor  or ΔCt. ΔCt is the difference  in threshold cycles be- tween  the  specific  RNA  and  reference  gene  amplicons given by StepOne  Plus Real-Time PCR software by nega- tive correlation with an internal reference  dye (ROX).

Human transcriptome array

PKH26+  and PKH26−  xenograft-derived CCSCs were FACS-sorted  as described  in the  flow cytometry  section and  processed  with  the  HTA  2.0 Affymetrix  array  following the manufacturer’s instructions. The data matrix having  as rows  (statistical  units)  and  as columns  (vari- ables) of the 10 samples  (5 PKH26+  and 5 PKH26−) was analysed  by  means   of  Principal   Component   Analysis (PCA) to single out an independent component allowing for the complete  partition  of PKH26+  and PKH26−  sam- ples in the loading space [30]. The transcripts having the highest  absolute  score  in  the  discriminant  component were   identified.   The   replicated   entries   of  genes   (for PKH26−    samples:   Homo sapiens piRNA  piR-43,853 complete  sequence,  transfer  RNA Gly (anticodon TCC), transfer RNA Ile (anticodon AAT), transfer RNA Leu (anticodon AAG), transfer RNA Leu (anticodon TAG), transfer  RNA Pro  (anticodon AGG),  transfer  RNA Pro (anticodon  CGG);  for  PKH26+   samples:  Homo sapiens piRNA piR-31,233 complete  sequence,  Homo sapiens piRNA piR-35,626 complete  sequence,  Homo sapiens piRNA piR-37,799 complete  sequence,  Homo sapiens piRNA piR-38,408 complete  sequence,  Homo sapiens piRNA piR-53,527 complete  sequence,  Homo sapiens piRNA piR-55,000 complete  sequence,  Homo sapiens piRNA  piR-57,434  complete   sequence)   were  both   se- lected  as  conditions-related  genes  allow  for  a  quality proof of the results.

PKH26 staining

SW480 or CCSCs (the latter previously dissociated with TrypLE  Express,  Thermo Fisher  Scientific)  were  stained for 2 min at 37 °C with PKH26 (Sigma), then  washed  ex- tensively with PBS. PKH26 staining  was assessed by flow cytometry and cells were used for subsequent experiments only when  positivity  was ≥98%. For in vivo experiments 5× 105  PKH26-stained cells were injected  subcutaneously in NSG mice, which were sacrificed at different  times for the detection  of PKH26+  cells or at 3 weeks post-injection for all the other experiments.

Lentiviral infection

CCSCs  or  SW480   cells  were  stably  transduced  with pLenti-GFP (lentiviral vector with C-terminal GFP tag, catalogue number  PS100065) or pLenti GFP-ZEB2 (cata- logue number RC215227L2) purchased  from Origene (Rockville, MD, USA).

Flow cytometry, cell cycle analysis and cell sorting

For flow cytometry  experiments,  xenografts  derived from PKH26-stained cells were cut  into  small  pieces, washed with  ice-cold  PBS, and  subsequently  digested  with  Try- pLE express  for 15 min  at 37 °C with vigorous  pipetting every 5 min. Freshly isolated cells were stained  with bio- tinylated  anti-PROMININ-1 and  anti-EpCAM  and  spe- cific secondary  antibodies.  10 μg/ml 7-aminoactinomycin D was used for dead cell exclusion.  The  cell cycle status of  CCSCs  and  SW480  xenograft  cells  transduced  with the pLenti-GFP vector or with pLenti-GFP-ZEB2 was assessed by staining  dissociated  cells with 50 μg/ml  pro- pidium  iodide dissolved in buffer 0.1% trisodium citrate, 9.65 mM NaCl, 0.1% Nonidet  P40, and 200 μg /ml RNase for  1 h  at  room   temperature.  Samples  were  analyzed with a FACSCanto flow cytometer  (Becton Dickinson) equipped   with  a  DIVA  software.  To  obtain  EpCAM+/PKH26+     and   EpCAM+/PKH26−    or   pLenti-GFP    and pLenti   GFP-ZEB2  fractions,   cells  were  sorted   with  a FACSAria (Becton Dickinson).

Immunofluorescence

CCSCs  were  centrifuged   at  low  speed  on  polylysine- coated glass slides, whereas SW480 cells were grown dir- ectly on glass slides. Cells were then  fixed in 2% parafor- maldehyde  (PFA) for  15 min  at  room  temperature and permeabilized   in  0.1% Triton   X-100  for  5 min  at  RT then,  after two washes in PBS, they were incubated  with glycine  1 M  (Sigma-Aldrich)   1 h  at  room  temperature. Glycine was removed  without  washing and,  after block- ing in 3% BSA (Sigma-Aldrich)/3%  FBS (Gibco)/PBS (Sigma-Aldrich),  cells were  incubated   overnight  at  4 °C with  primary   antibodies   anti-Ki67,   CRAF  pS338  and anti-ZEB2.  After  two  washes  in  PBS, cells  were  incu- bated with appropriate secondary antibodies in a buffer containing    DAPI,   3%  BSA,  5 μg/ml   RNAse   (Roche) diluited  in PBS for 1 h at room  temperature in the dark. Subsequently,  glasses were mounted with ProLong  Gold Antifade. Immunofluorescence staining of xenograft- derived sections  was performed  as follows: sections  were fixed   in   2%  PFA  for   15 min   at   room   temperature, washed two times in PBS and permeabilized  in 0.1% Tri- ton  X-100  for  5 min  at  room  temperature  then  incu- bated   overnight   at  4 °C  with  primary   antibodies   anti CRAF  pS338,  anti  PROMININ-1,   anti  Ki67  and   anti ZEB2. After  washing  in  PBS, sections   were  incubated with a mixture  of appropriate secondary antibodies  and DAPI as described  above. SW480 cells were seeded  5 ×104  cells/ml  and treated  after 24 h with etoposide  10 μM or  irinotecan 10 μM  for  48 h.  Cells were  processed  for immunofluorescence as described  above and stained  with anti-pCRAF,   the   appropriate  secondary   antibody   and DAPI  for  nuclear  identification.  Slides were  analyzed  at room temperature on a FV-1000 confocal microscope (Olympus) equipped with Ultraplan Apochromatic 60X N.A.1.42 and 40X N.A. 1.30 oil immersion objectives and acquired  with the Olympus Fluoview software. The result- ing images were not subjected to further  processing.

Western blotting

Cultured   cells  or  ~ 50 mg  pieces  of  frozen  xenografts were  lysed in  the  appropriate volume  of the  respective lysis buffer: for cultured  cells we used 1% NP40 lysis buf- fer (20 mM  Tris  HCl pH 7.2, 200 mM  NaCl, 1% NP40), while  for  xenograft  tissues  we  used  10 mM  Tris  pH8, 150 mM  NaCl,  60 mM  Octyl-β-Glucoside.   Both  buffers were supplemented with protease inhibitor cocktail and phosphatase inhibitor  cocktails  I and II (all from Sigma- Aldrich). Tissues were homogenized with Pro 200 Kema Keur  (Pro  Scientific Inc. Oxford)  at maximum speed  at 4 °C for 30 s. Lysate concentration was determined with the   Bradford   assay  (Bio-Rad  Laboratories)   and   equal amounts of proteins  were loaded on a 4–12% precast  gel (Thermo Fisher Scientific) and  transferred to nitrocellu- lose  membranes  (GE  Healthcare   Life  sciences).  Blots were  blocked  with  TBST  5% nonfat  dry  milk  (Bio-Rad Laboratories)  and  incubated  overnight  at  4 °C with  pri- mary antibodies  diluted  in TBST/BSA 5%, after 4 washes in TBST then  incubated  for 45 min with specific second- ary HRP-conjugated antibodies  diluted in TBST 5% nonfat dry milk. Images were taken and analyzed with Bio-Rad ChemiDoc  Imagers (Bio-Rad Laboratories).  For densitom- etry quantification immunoblot signals were acquired with ChemiDocMP (BioRad) and the relative intensity was quantified with Image Lab software. Normalization was performed using β-ACTIN as reference.

RNA interference

1.25 × 105   CCSCs  or  SW480  cells  were  seeded  on  six- well plates  in  antibiotic-free culture  medium  and  incu- bated  for  4 h  at  37 °C in  the  presence  of 320 nM  ON-TARGETplus    SMARTpool   non-targeting   siRNA   (D-001810-01-055),    human    ZEB2   siRNA   (L-006914-02-0005) (Dharmacon/Thermo Scientific) and  5 μl Lipofec- tamine   2000  (thermo   fisher  scientific).  After  4 h  the transfection mixture  was substituted with  the  appropri- ate culture medium and cells were analyzed for cell via- bility, gene and protein  expression  at the indicated  times

Cell viability assay

The  viability of CCSCs or SW480  cells was determined by CellTiter-Glo luminescent cell viability assay (Pro- mega)    according    to    the    manufacturer’s   directions. CCSCs and SW480 cells (2.5 × 103  per well) were seeded in   96-well   plates   (three   replicates   per   experimental point) in the appropriate medium  and incubated  in a hu- midified  atmosphere  at  37 °C,  5%  CO2.   For  in  v itro chemotherapy treatment, cells were treated  for 48 h with 10 μM  5-fluorouracil  or  10  μM  oxaliplatin.   Lumines- cence  was detected  with  a DTX880  multimode micro- plate reader  (Beckman Coulter).

Statistical analyses

Statistical   analyses   were   performed  using   GraphPad Prism   version   4.0  for  Windows   (GraphPad   Software) with non-paired double-tailed t test  (after verifying nor- mal  distribution  of  the  population  with  Shapiro–Wilk test)  or  with  one-way  ANOVA  where  appropriate.  Re- sults  are  presented  as  the  mean ± SD  or  mean ± SEM where appropriate. Statistical significance is expressed  as *, P < 0.05, **, P < 0.01 and  ***, P < 0.001. Statistical  ana- lysis of Affymetrix  results  and  of Reverse  Phase  Prote- omic  Arrays  (RPPA) are  described  in the  specific Supporting  Methods   sections  and/or in  the  respective figure legends. Principal Component Analysis was per- formed  by SAS version 8.1.

Results

Isolation and characterization of QCSCs responsible for chemoresistance in colorectal tumors Seminal  studies  on  the  dynamics  of  chemotherapy re-sponse  pointed  to a rare cell population that  remains  la- tent    throughout   the   life   of   untreated   tumors    and emerges  only  upon   chemotherapy treatment  [18].  We ought  to  analyze such  “pre-existing  persisters”  by using the  proliferation-sensitive  dye  PKH26,  which  incorpo- rates into lipid membranes and is progressively lost dur- ing subsequent cell divisions. In our previous studies, we previously  demonstrated that  PKH+/high  CRC cells pos-sessed a higher tumorigenic potential as compared with PKH−/low cells, indicating  that  the quiescent/slow cycling fraction  is enriched  in cells with stemness  features  [16]. The  PKH26 experimental system  gave us the  possibility to identify cells that are quiescent/slow cycling from the initial  stages  of  tumor   development  and  not  just  in  a given moment (as occurs  instead  with the H2B-GFP sys-tem).  First,  we  aimed  to  determine  whether   PKH26+ cells  survived   chemotherapy  treatment.  SW480   CRC cells  were  stained  with  PKH26,  then  allowed  to  divide for  11 days,  after  which  chemotherapy  treatment   wased.  We  monitored the  percentage of PKH26+   cells for two additional weeks, during which the positive population decreased  to 0.5% in the  original  culture  but increased  to more  than  90% in chemotherapy-treated samples  (Fig. 1a). Drug-treated samples  consisted  of ~60% live cells after 2 weeks, as shown by 7-AAD staining (Additional  file 4: Figure S1a). This observation  indicates that   cells  selectively  surviving   chemotherapy  are   the same cells that  were quiescent/slow cycling in untreated tumors  and not cells that entered quiescence upon drug treatment. Therefore,  we focused our attention on cells present  in untreated tumors  that  are destined  to survive chemotherapy  treatment  and  we  undertook  their isolation and characterization. To do this, we used mo- lecularly annotated 3D cultures of primary CRC cells (thereafter  called CCSCs, Colon  Cancer  Spheroid  Cells) that were previously shown by our group and others to faithfully reproduce original patient  tumors  when inocu- lated in immunecompromised mice [17, 29, 31]. PKH26- stained   and   sorted   CCSCs  were   inoculated   into   the flanks of NSG mice and the percentage of PKH26+   cells was monitored over time by flow cytometry  (Fig. 1b and c). At 3 weeks post-injection we isolated from tumor  xe- nografts EpCAM+/PKH26+ and EpCAM+/PKH26− cells (Additional  file 4: Figure S1b) that  were used for further characterizations. Flow cytometry analysis showed that PKH26+  xenograft  cells were negative for Ki67 and expressed  very high  levels of PROMININ1,  indicating  a stem  cell  phenotype   (Additional  file 4:  Figure  S1c and d), in line with our  previous  observations  [16]. In order to  investigate   whether   long-term  quiescent   cells  were characterized  by  a  specific  pattern   of  gene  expression, we analyzed  PKH26+   and  PKH26−   cells freshly isolated from CRC tumor  xenografts  with the Affymetrix 2.0 hu-man  transcriptome array. The existence  of a gene signa- ture  able  to  discriminate between  the  two  populations was  investigated   through  a  purely   unsupervised  data driven approach  suitable to identify small sets of bio- logically  relevant  genes  in  an  otherwise   similar  back- ground  [30]. Principal component analysis (PCA) of the results  showed  a  sharp  distinction   between  profiles  of fast proliferating  and quiescent/slow proliferating  cells emerging  from the fourth  PCA component (PC4) which, although  accounting for only  0.15% of gene  expression variability, nevertheless  allowed for a perfect  partition of the loading component space into  PKH26+  and PKH26−areas  (Fig. 1d). Setting  two  thresholds   respectively  at  6 and 10 standard deviation  units  from the mean  (Fig. 1e), we  identified   transcripts  mostly  affected  by  PC4  and consequently more  involved into  PKH26+/  PKH26−  discrimination (detailed  in Additional  file 5: Table S4). The great majority of transcripts differentially modulated in PKH26+   and  PKH26−  cells did not  correspond to struc- tural  genes  but  rather  to  post-transcriptional regulators (microRNAs,  small  nucleolar  RNAs, piwi-interacting RNAs, long  non-coding RNAs and  tRNAs)  (Fig. 1f and Additional  file 6: Table S5), indicating  that the balance between  quiescence  and  proliferation  relies  on  the  fine tuning    of   a   basically   similar   transcription   pattern. Among  the  transcripts  more   expressed   in  QCSCs  we found the long non-coding RNA relative to the tran- scription  factor ZEB2 (zinc finger E-box binding  homeo- box  2),  previously  known  for  its  involvement   in  EMT and TGF-β-regulated processes [32–34]. Moreover, the ZEB2 mRNA had a statistically  significant  score on PC4 (− 2,34, *P < 0.01). Therefore,  we decided  to explore  the expression   and   function   of  ZEB2  in  CRC  cells.  We confirmed  the  enrichment of ZEB2 mRNA  in  PKH26+ cells isolated ex vivo from tumor  xenografts and in chemotherapy-treated  cells  (Fig.  1g  and   h),  while  in xenograft   sections   ZEB2-expressing   areas   overlapped with PKH26+  areas (Fig. 1i). ZEB2 expression  in PKH26+ tumor  cells was accompanied by an increased  expression of CRC self-renewal  factors BMI1 and NANOG [35, 36], of EMT-related genes  ZEB1, VIMENTIN,  SNAI1 and SNAI2,  of  cyclin-dependent  kinase  inhibitor   1B (CDKN1B, encoding  for p27Kip1)  and  by lower  levels of MKI67  and  CADHERIN-1 (Fig. 1l), indicating  that  the QCSCs  population in colorectal  tumors  is characterized by stemness  and EMT features. In line with these obser- vations,  we  analyzed  CCSCs  expressing  the  TOP-GFP vector as a functional  marker  of β-catenin  activity and a surrogate  marker  of CRC stem  cells [37]. Sorted  CCSCs with higher levels of TOP-GFP  (and consequently of β- CATENIN-dependent transcription) expressed higher levels of ZEB2 (Fig. 1m),  further  supporting the  stem- ness of quiescent/slow cycling CRC cells.

Fig. 1 Untreated tumors contain chemotherapy-resistant quiescent cells with an EMT/stemness phenotype and increased ZEB2 levels. a SW480 cells were stained with PKH26, treated with 2,5 μM oxaliplatin (OXA) at day 11 and monitored by flow cytometry. FACS plots are shown in Additional file 4. b FACS plots showing PKH26 positivity in CCSCs-derived tumors the day of injection (Day 0) and at 1/3/6 weeks. c Percentage of PKH26+ cells (black line) in relation to tumor size (red line). Mean ± SD or SEM (tumors), n = 6 tumors/group. d Spatial representation of principal component (PC) analysis with genes as rows (statistical units) and samples as columns (variables). n = 5 samples of 2 pooled tumors each. e Numerical PC plot identifying genes with the highest absolute score in the discriminant component. A full list of transcripts modulated in PKH26+ versus PKH26− cells is reported in Additional file 5. f Categories of transcripts enriched in PKH26+versus PKH26− cells. Replicated entries are
reported in Materials and Methods. Transcripts categories are detailed in Additional file 6. g qRT-PCR analysis of ZEB2 in EpCAM+/PKH26+ versus EpCAM+/PKH26− cells from CCSCs-derived xenografts. *P < 0.05 (two-tailed t test), mean ± SD, n = 3 pools of 6 tumors each. h qRT-PCR of ZEB2 expression in SW480 cells untreated (day 0) or treated with 2,5 μM oxaliplatin (OXA). Mean ± SD of 3 experiments. i Representative confocal image of CCSCs-derived xenograft sections showing overlapping areas of ZEB2 (red) and PKH26+ (yellow) positivity. Scale bar 80 μm. l qRT-PCR of xenograft-derived EpCAM+/PKH26+ versus EpCAM+/PKH26− cells. *P < 0.05, **P < 0.01 and ***P < 0.001 (two-tailed t test). Mean ± SD, n = 3 pools of 6 CCSCs-derived tumors each. m qRT-PCR analysis of ZEB2 expression in TOP-GFP.mcherry negative (grey) and positive (purple) CCSCs sorted from in vitro culture. Mean ± SD of 3 experiments

Global pathway analysis shows the activation of chemoresistance-related factors in QCSCs

Reverse-Phase  Protein  Array  (RPPA) allows  the  simultaneous evaluation of phosphorylated, cleaved, or un- modified proteins generating comprehensive profiles of pathway activation  in different cell or tissue samples [38,39]. RPPA was used  to  compare  PKH26+   and  PKH26− cells isolated  ex vivo from  CCSCs-derived  tumor  xeno- grafts  in  order  to  obtain  a  broad  picture   of  signaling pathways   modulated  in   these   two   populations.  Three matched  pools of ex vivo PKH26+/PKH26− cells were analyzed  with  the  antibodies   reported  in  Additional   file  1: Table S1. Hierarchical  clustering  showed that  two samples of QCSCs had a massive down modulation of most path- ways, particularly  those involved in proliferation and bio- synthesis  (Fig.  2a).  The  third   sample  of  quiescent   cells showed a down regulation  of most pathways but a simul- taneous  upregulation of a small set of phosphoproteins (c- Met,  VEGFR2, c-Abl, SGK) (Fig. 2a), indicating  the  exist- ence of multiple layers of quiescence-associated signals. Nevertheless,  principal component analysis (PCA) of RPPA results highlighted a molecular  signature  common  to quies- cent/slow  proliferating  CRC cells (Fig. 2b). Statistically sig- nificant endpoints modulated in PKH26+  and PKH26−  cells (shown in detail in Additional  file 7: Table S6) are summa- rized in Fig. 2c, where QCSCs are sharply identified by in- creased  levels of CRAF S338 phosphorylation and  ASK1 S83 phosphorylation. Importantly, pS338 CRAF and  pS83 ASK1 have been individually implicated  in protecting cells from  genotoxic  insults  [40–42],  but  they  have  also  been shown to act in concert  by forming a chemoresistance- promoting complex  at mitochondria [43]. Due to the spe- cific role of pCRAF in driving  therapy  resistance  [41], we assessed its expression  in tumor  xenografts,  where it over-lapped with PKH26+  and partially with PROMININ1+ areas (Additional  file 8: Figure S2a), and we confirmed  that  it is actually upregulated in drug-treated CRC cells (Additional file 8: Figure S2b). Fast proliferating  PKH26−  cells showed, among  others,  an  increased  expression  of phosphorylated Akt, MEK1/2,  mTOR  (and  downstream effectors  p70S6K and 4EBP1), GSK3, histone  H3 and NDRG1 (Fig. 2c). The latter is particularly interesting  as it has been reported to in- hibit EMT, stemness  and metastasis  and is related  to a fa- vorable prognosis  in CRC patients  [44]. In order to rule out the possibility of chance  correlations in the statistical  ana- lysis of  RPPA  results,  we  complemented  data  shown  in Fig. 2c with  a further  analysis having  samples  as variants and  protein  endpoints as units.  In fact, since the  samples differ only for a transient functional  state (proliferative sta- tus), they have a largely overlapping RPPA profile that translates  into a major principal  component explaining the major part (80%) of the among  samples variance [45]. This implies that discrimination of the two populations can only start from the second component on ward, getting rid only of a minor  proportion of variance.  That  said, the  loading space allowed for a posteriori  perfect discrimination among PKH26+   and  PKH26−  samples  as for Factor  2 (Additional file 8: Figure S2c), which explains  only 8,5% of total  vari- ance and represents a common regulatory  pathway within the same cell population.  We observed a remarkable  super- position  between  the two analyses, as the large majority of endpoints  are   identified   as  discriminants  in   both   ap- proaches  (Additional  file 8: Figure S2d). However, E- Cadherin  emerges from the second approach  as one of the endpoints most relevant for group discrimination, adding further significance to the hypothesis that QCSCs tend to- wards   a  mesenchymal  state.   In   summary,   our   results showed  that  the molecular  diversity among  fast proliferat- ing and quiescent/slow proliferating  CRC cells concentrates around   distinctive  pathway  profiles.  Rapidly  proliferating cells possess high levels of proteins  involved in biosyn- thetic/metabolic pathways and are shifted towards an epithelial-like  and chemosensitive status,  while QCSCs de- press  pathways  related  to  cell cycle/biosynthesis/metabol- ism  and   selectively  upregulate   factors   involved  in  self renewal,    chemoresistance   and    EMT/metastatic   ability (Fig. 2d).

Fig. 2 Reverse-phase proteomic analysis of quiescent/slow cycling xenograft cells. a Hierarchical clustering of RPPA results obtained on 3 EpCAM+/PKH26+  and EpCAM+/PKH26−  cell samples isolated from CCSCs-derived tumor xenografts. Clusters, identified for either antibodies or samples and based on optimal cut of dendrograms, are indicated by coloured bars adjacent to dendrograms. The values represented by the heatmap correspond to normalized intensities of antibodies, standardized over the sample set analyzed (z score). n = 3 pools of 12 tumors each. A list of RPPA antibodies and modulated endpoints are reported respectively in Additional file 1: Table S1 and Additional  file 7: Table  S6. b Principal component  analysis (PCA) of RPPA results showing that PKH26+  samples have a common  molecular signature. c Volcano plot showing the antilogarithm (base = 10) of the adjusted  P value versus base 2 logarithm of the ratio between PKH26+  and PKH26−  samples. Kruskal Wallis test was performed  for each  RPPA analyte  on the 3 samples stratified by PKH26 positivity.  RPPA analytes  where  Kruskal Wallis test resulted  in a statistically significant (*P < 0.05) change between  PKH26-stratified samples, underwent a further  analysis by means of Wilcoxon signed-rank test. All the resulting p values were adjusted for multiple comparisons using the Benjamini-Hochberg correction. d Schematic representation of pathways that emerged from experiments described in Figs.1 and 2 as present  in PKH26-negative fast proliferating  cells (left, red area) or in slow proliferating/quiescent  cells (right, blue area). Phosphorylation  sites are outlined  in green when they result in protein activation, in red when they inhibit protein function. Activated pathways are highlighted  in colors while inhibited pathways are depicted in light grey

ZEB2, pCRAF and pASK1 are coexpressed upon chemotherapy and coregulated in CRC cells

Having  identified  ZEB2, pCRAF  and  pASK1 as  factors upregulated in  QCSCs,  we asked  whether  their  expres- sion  was increased  upon  chemotherapy and  modulated in   a  coordinated  manner.   First,   we  analyzed   ZEB2, pCRAF and pASK1 expression in chemotherapy-treated cells and we observed a parallel increase of the three fac- tors  following  5-fluorouracil  and  oxaliplatin  treatment (Fig. 3a). Then,  we investigated  whether  the  expression of pCRAF and pASK1 was mechanistically  regulated by ZEB2 by modulating ZEB2 levels with  siRNA-mediated silencing or lentiviral overexpression in either  CCSCs or SW480  cells  and  analyzing  variations   in  pCRAF  and pASK1.   Transient   ZEB2   siRNA-mediated    silencing (Fig. 3b) induced  a decrease  in the levels of S338- phosphorylated  CRAF  and   S83-phosphorylated  ASK1 (Fig. 3c),  indicating   that  pCRAF  and  pASK1  lie  dow- stream  of ZEB2 in the  quiescence/chemoresistance pro- gram. Exogenous expression of a lentiviral GFP-ZEB2 construct  (Fig.  3d)  increased   the  expression   levels  of pCRAF and pASK1 (Fig. 3e) and resulted  in enhanced chemoresistance of both CCSCs and SW480 cells (Fig. 3f) shortly  after  cell  transduction  and  sorting.   At  longer times of culture,  however, both  CCSCs and SW480 cells transduced with ZEB2 downregulated protein  levels until they  reached  those  found  in  untreated  cultures   (Add- itional file 9: Figure S3a), where ZEB2 expression  is lim- ited to rare  Ki67-negative  cells (Additional  file 9: Figure S3b). These results indicate that ZEB2 controls  the levels of pCRAF and pASK1 and that  its levels are strictly reg- ulated in CRC cells.

Fig. 3 Coordinated expression and modulation  of ZEB2, pCRAF and  pASK1. a Left: immunoblot  analysis of ZEB2, CRAF pS338, and  ASK1 pS83 on whole lysates of SW480 cells treated  for 4 days with 5 μM 5-fluorouracil  (5-FU) or 2,5 μM oxaliplatin  (OXA). Glyceraldehyde  3-phosphate dehydrogenase (GAPDH) was used as a loading control. Right: quantification of immunoblot shown on the left. b qRT-PCR analysis of ZEB2 levels in CCSCs (left panel) and SW480 (right  panel) 24 h after siRNA-mediated silencing of ZEB2. ***P < 0.001 from two-tailed t test. Data of qRT-PCR are the mean ± SD, n = 3. c Immunoblot analysis of ZEB2, CRAF pS338, and  ASK1 pS83 on whole cell lysates 24 h upon siRNA-mediated silencing of ZEB2 in CCSCs (left  panel)  and  SW480 (right panel). The respective quantifications are shown on the right. d qRT-PCR analysis  of ZEB2 levels  in CCSCs (left panel) and SW480 (right  panel) transduced  with empty pLenti-GFP (Vector) or with pLenti-GFP-ZEB2 (ZEB2) and  sorted  on the basis of GFP expression.  ***P < 0.001 from two-tailed t test. Data are the mean ± SD, n = 3. e Immunoblot analysis of ZEB2, CRAF pS338, and  ASK1 pS83 on whole lysates of CCSCs (left panels) and SW480 cells (right  panels) transduced  with pLenti-GFP (Vector) or with pLenti-GFP-ZEB2  (ZEB2) and sorted as above. The respective quantifications  are shown on the right. f Viability of CCSCs (left) and SW480 (right) transduced  with pLenti-GFP or pLenti-GFP-ZEB2, sorted  on the basis of GFP expression and immediately treated for 48 h with 10 μM oxliplatin (OXA) and 10 μM 5-fluorouracil (5-FU). *P < 0.05 from two-tailed t test, n = 3. Data are the mean ± SD of three independent experiments

ZEB2 expression induces tumor transition towards a slow growing chemoresistant state

To investigate the effects of ZEB2 overexpression in vivo we  inoculated   freshly  sorted   ZEB2-transduced  SW480 cells in the flanks of immunecompromised mice and an- alyzed xenograft  growth, cell cycle status  and expression of cell cycle-, EMT- and stemness-related genes. ZEB2- overexpressing  tumors  grew significantly slower than vector-transduced tumors  (Fig. 4a, left panel) and dis- played  higher  ZEB2 and  lower  Ki67 levels as compared to  vector-transduced  controls  (Fig. 4a, right  panel).  Ex vivo cell cycle analysis showed that  ZEB2-overexpressing tumors  contained an increased  proportion of cells in the G0/G1  phase of the cell cycle and a lower proportion of cells in G2/M  (Fig. 4b). Assessment  of EMT and self- renewal  factors  as determined by qRT-PCR showed  that ZEB2-overexpressing    tumors    had   increased   levels  of ZEB2 itself (but  not  ZEB1), VIMENTIN,  SNAI1 and SNAI2, decreased  levels of CADHERIN1 and  increased expression  of BMI1 and  NANOG (Fig. 4c). Similar results were obtained  with CCSCs, with the difference that ZEB2- transduced-tumors had a delayed appearance as compared to  vector-transduced  tumors   (Additional   file  10:  Figure S4a-d). ZEB2-overexpressing  tumors  showed a modulation of several cell cycle-related factors including CYCLINA1, CYCLIND1, CDC2, CDC25A, HDAC9 and  HDAC5 and, importantly, a strong  upregulation of TGFB2 (Fig. 4d), in line with previous studies showing a specific role of TGFβ2 in dictating  the dormancy  of disseminated tumor  cells [46]. Then,   we  investigated   the  expression   of  ZEB2/pCRAF/ pASK1 in vivo upon  chemotherapy treatment. Vector-  and ZEB2-transduced SW480  cells were  inoculated  into  NSG mice and the resulting tumors  were treated  with oxaliplatin plus  5-fluorouracil   for  3 weeks. In  vector-transduced tu- mors,  chemotherapy induced  a growth  slowdown  associ- ated to a strong  increase of ZEB2, pCRAF and pASK1. Chemotherapy-treated control  tumors  showed  also a tran- sition towards a hybrid epithelial-mesenchymal state, as showed by the increased expression of SNAI1–2, ZEB1, VIMENTIN  and N-CADHERIN  but concomitant high ex- pression of E-Cadherin  (Fig. 4e-g). ZEB2-overexpressing  tu- mors  grew more  slowly than  controls  and  had  a baseline higher expression  of pCRAF, pASK1 and EMT-related fac- tors with decrease of E-CADHERIN, indicating  a complete EMT (Fig. 4e-g). In line with these observations,  ZEB2- overexpressing  tumors  were unaffected by chemotherapy treatment and did not change either their slow growing rate or  the  expression  of EMT/chemoresistance factors  upon drug exposure  (Fig. 4e-g). Altogether,  these results  identify ZEB2/pCRAF/pASK1  as factors  present  in rare  quiescent cells in untreated tumors  that are largely expressed upon chemotherapy treatment, thus inducing tumor  transition towards an EMT/chemotherapy unresponsive state.

Fig. 4 ZEB2 induces  a transition towards a quiescent/slow cycling and mesechymal-like state in CRC xenografts. a Left: Xenograft volume of SW480 cells transduced with pLenti-GFP (Vector, black line/triangles)  or with pLenti-GFP-ZEB2  (ZEB2, red  line/squares). Mean ± SEM, 6 tumors/ group. **P < 0.01 (two-tailed  t test). Middle: representative confocal images of Vector- and GFP-ZEB2-transduced SW480 xenografts  stained with anti-ZEB2 (red) and anti-Ki67 (white) antibodies. Scale bar 60 μm. Right: quantification  of Ki67-, ZEB2- and GFP-positive cells performed  on 3 sets composed of 5 fields/group. *P < 0.05 and **P < 0.01. Mean ± SD (two-tailed  t test, n.s. = not significant). AU, arbitrary units. b Cell cycle analysis of GFP+  cells from Vector- and ZEB2-transduced tumors. c qRT-PCR analysis  of GFP+  cells from Vector- and ZEB2-transduced tumors, n = 3 pools of 2 tumor each. *P < 0.05 and **P < 0.01 (two-tailed  t test). Mean ± SD. d mRNA levels of cell cycle genes in GFP+  cells from Vector- and GFP-ZEB2- transduced tumors. Mean ± SD, n = 3. *P < 0.05 (two-tailed  t test). e Volume of xenografts expressing pLenti-GFP (Vector, black line) or GFP-ZEB2 (ZEB2, red line), untreated or treated (Vector, gray line/triangles  and ZEB2 yellow line/diamonds) with 5-fluorouracil plus oxaliplatin (5FU + OXA). Mean ± SEM, 6 tumors  per group.  *P < 0.05 and **P < 0.01 from one-way ANOVA and Bonferroni post-tests. f Upper panels: immunoblot  analysis of ZEB2, CRAF pS338 and  ASK1 pS83 on whole tumor lysates derived from SW480 xenografts. Lower panels: densitometry  analysis of western blots, n = 3, *P < 0.05, **P < 0.01 and ***P < 0.001 (two-tailed  t test). g Left: immunoblot  analysis of EMT-related proteins on whole xenograft lysates. Every sample  is a pool of 2 tumors. Right: densitometry analysis, n = 3, *P < 0.05, **P < 0.01 and ***P < 0.001 (two-tailed  t test)

ZEB2 expression correlates with worse prognosis and CMS4 in CRC patients

Finally, we  explored  the  potential   clinical  relevance  of our  findings  by  analyzing  ZEB2 expression   in  a  CRC dataset composed  of all fresh frozen tumor  samples compiled  by the  consensus  molecular  classification  con- sortium  [47]. This set, for which consensus  molecular subtype (CMS) classification and in most cases stage is available, was separated  by TNM  stage and  analyzed for ZEB2 expression,  revealing  a  slight  but  non-significant increase  with  progressing  stage  (Additional  file 11:  Fig- ure  S5). However,  segregation  of the  patients  into  low and high ZEB2 expression  revealed a very significant in- crease in recurrence rate  in patients  with high ZEB2 ex- pression   across   all   TNM   stages   (p < 0.001,   n = 802) (Fig. 5a). Importantly, the majority  of the patients  in our dataset  could  be  faithfully  assigned  to  one  of the  four CMSs,  which  have  distinguishing   molecular,  biological and clinical features [47]. Among these, CMS4 is typified by  high  expression   of  mesenchymal  genes,  prominent TGF-β activation,  stromal  infiltration  and worse relapse- free survival [47]. In agreement with our hypothesis  that ZEB2 drives an EMT-related and  therapy-resistant  CRC phenotype  we found  a significantly increased  expression of ZEB2 in CMS4 (***P < 0.001, n = 2822) (Fig. 5b). Like- wise, consistent  with the association  of ZEB2 with a qui- escent/slowly proliferating state, MKI67 expression was reduced   in  CMS4   as  compared  to   the   other   CMSs (***P < 0.001, n = 2822) (Fig. 5c).

Fig. 5 Higher  ZEB2 expression  is linked  to CMS4 and poor prognosis in colorectal tumors. a Kaplan Meier curve showing the relapse-free survival of 802 CRC patients separated on the basis of ZEB2 expression  (red, low expression and blue, high expression). ***P < 0.001 based on log-rank test. b ZEB2 levels  in CMS4 tumors  as compared  with CMSs 1–3. ***P < 0.001 based on one-way  ANOVA, n = 2822. Outliers are depicted  as crosses. c MKI67 levels in CMS4 as compared  with CMSs 1–3. ***P < 0.001 based on one-way  ANOVA, n = 2822. Outliers are depicted  as crosses. Both  the analysis of variance and the single post-hoc pairwise comparison in b and c are highly significant

Discussion

Increasing   evidence  indicates  that  a  quiescent   state  is tightly  linked  to  drug  resistance  in  cancer  cells. How- ever,  due   to   their   rareness   and   plasticity,   quiescent cancer cells remain mostly elusive and represent a chal- lenging therapeutic target [15]. We previously demon- strated  that  stem  cells in CRC can be found  both  in the fast proliferating  (PKH26−/low) and in the quiescent/slow proliferating  (PKH26+/high) fraction,  but  PKH26high   cells are  endowed  with  a  higher  tumorigenic  potential  [16]. Now, we demonstrate that quiescent/slow cycling cells present  in untreated CRC xenografts  are  the  same  cells that resist chemotherapy treatment. Quiescent/slow cyc- ling cells isolated  from  untreated CRC xenografts  were characterized by combined  features of stemness, che- moresistance and EMT, indicating  that  quiescence  arises as a whole set of molecular  traits  covering  multiple  cel- lular processes. In fact, the connection between stemness and  EMT  was known  since early studies  by Mani  et al., who demonstrated that normal  and neoplastic  mammary cells that  underwent EMT exhibit stem cell markers  and functional  characteristics [48]. However,  further  studies also highlighted  the  implication  of quiescence  as a fea- ture  of CSCs undergoing EMT. To  cite a few, in breast cancer  Lawson and coworkers  identified a metastatic  cell population characterized by the expression  of stem  cell-, EMT-, pro-survival-, and dormancy-associated genes [9], while in acute myeloid leukemia  Ebinger et al. isolated a subset  of dormant stem  cells with  reversible  properties of  quiescence   and  therapy  resistance   [5].  In  CRC,  we found  that  PKH26+/ZEB2+  cells  were  characterized  by high  levels of PROMININ1,  by an  increased  expression of   self-renewal   factors   BMI1   and   NANOG    and   by elevated   nuclear   β-CATENIN   (as   detected   with   the TOP-GFP  assay), indicative  of enhanced stem  cell prop- erties. Notably, ZEB2 overexpression in vivo was able to recreate a QCSCs population with features of chemore- sistance  and EMT. Such phenotype  was almost  identical to  that  developed  by  chemotherapy-treated  xenografts,with the difference that  ZEB2+  tumors  appeared  to have a more complete EMT (N-CADHERINhigh/E-CADHER- INlow) as compared to chemotherapy-treated tumors  (N-CADHERINhigh/E-CADHERINhigh).   However,   different EMT  states   are  not  surprising   as  they  are  typical  of CSCs populations with enhanced plasticity [15, 49]. Altogether,  these  observations  suggest the existence  of a slow cycling/mesenchymal/stem population across dif- ferent  tumors  which may share,  at least in part,  a com- mon  molecular  signature.  Therefore,  exploring  the molecular  features  of the  dormant/stem  population will be particularly relevant for the identification  of pharma- cological strategies  aimed  at  eradicating  chemoresistant cells  or  alternatively   at  preventing   their   reactivation. Drugs  potentially   able  to  target  dormant  tumor   cells may  be  directed   against   factors   that   play  a  role  in both   EMT  and  quiescence,   such  as  those  implicated in TGFβ signaling [50]. Among  these, TGFβ2 was identified as crucial for the induction of dormancy  in disseminated tumor cells [46] and emerged as highly upregulated in ZEB2-overexpressing tumors. These findings   are   also  in   line   with   the   observation   that ZEB2 is highly  expressed  in  CMS4  tumors,  which  are also  characterized  by  prevalent   TGFβ  activation  [47]. A comprehensive characterization of pathways modu- lated   in   quiescent    CRC   cells   performed  by   RPPA showed a downregulation of main proliferative/biosyn- thetic/metabolic  pathways  together   with  an  upregula- tion   of   chemoresistance   factors   CRAF   pS338   and ASK1  pS83.  Accordingly,   CRAF  phosphorylation  in S338 was recently demonstrated to trigger a kinase- independent  mechanism  of  DNA   repair   and   thera- peutic resistance [41]. Further  underlining the tight connection  between   dormancy   and   chemoresistance, both ZEB2-overexpressing and chemotherapy-treated tumor    xenografts    acquired    increased    pCRAF   and pASK expression,  suggesting that this may represent a common stem in the transition through  an EMT/che- moresistant state.  The  finding  that  ZEB2 is  increased in  CMS4  is in  line  with  a recent  study  reporting   that this    subtype    is    characterized   by    methylation     of miR200 promoter regions and consequent increased expression of EMT-related genes [51]. Indeed, the in- creased  ZEB2 (and  decreased   MKI67)  expression   de- tected   in  CMS4  tumors   could  be  influenced   by  the abundant  stromal   infiltrate  characteristic  of  this  sub-type,  as  stromal  fibroblasts  can  also  display  a  ZEB1+/ZEB2+/miR200−/Ki67− profile [52]. In fact, the inter- actions   between   tumor   cells  and   stromal   fibroblasts have  been  shown  to  play a key role  in  defining  poor-prognosis  CRC  by exploiting  TGFβ signaling  to  drive an   aggressive   CSC-enriched   phenotype    [53].   It   is likely that  both  the  stromal  and  the  epithelial  part  of CMS4 tumors  contribute to the establishment of an aggressive phenotype  through an interplay of signals orchestrated by TGFβ, resulting in refractoriness to conventional and targeted therapies [53, 54]. This hy- pothesis  is corroborated by recent  observations  show- ing  that   budding   areas  of  the   tumor,   which  are  in close contact with the surrounding stroma, are char- acterized   by  down   regulation   of  proliferation  genes, EMT  and  switching  to  CMS4 [55].

Conclusions

Altogether, our results point to a ZEB2/pCRAF/pASK molecular signature  involved in the determination of a quiescent/slow proliferative  state  that  identifies  a subset of  cells  present   in  baseline  conditions   and  expanded both  upon  drug  treatment and  in  aggressive  CRC sub- types.  The  identification   and  characterization  of quies- cent drug-resistant CSCs may pave the way for future therapeutic strategies aimed at neutralizing  this specific population in CRC.

Supplementary  information

Supplementary information accompanies this paper at https://doi.org/10.1186/s13046-019-1505-4