cancers Article Tumor Expression of Cyclin-Dependent Kinase 5 (Cdk5) Is a Prognostic Biomarker and Predicts Outcome of Oxaliplatin-Treated Metastatic Colorectal Cancer Patients Vicenç Ruiz de Porras 1,2,† , Sara Bystrup 1,2,† , Sara Cabrero-de las Heras 1,2 , Eva Musulén 3,4 , Luis Palomero 5,6 , Maria Henar Alonso 6,7,8,9 , Rocio Nieto 10 , Diego Arango 10 , Víctor Moreno 6,7,8,9 , Cristina Queralt 1,2 , José Luis Manzano 1,11,12 , Laura Layos 1,11,12 , Cristina Bugés 1,11,12 and Eva Martinez-Balibrea 1,2, * 1 Program of predictive and personalized cancer medicine (PMPPC) Germans Trias i Pujol Research Institute (IGTP), Ctra. Can Ruti—Camí de les escoles s/n, 08916 Badalona, Spain; vruiz@igtp.cat (V.R.d.P.); sbystrup@igtp.cat (S.B.); scabrero@igtp.cat (S.C.-d.l.H.); cqueralt@iconcologia.net (C.Q.); jmanzano@iconcologia.net (J.L.M.); llayos@iconcologia.net (L.L.); cbuges@iconcologia.net (C.B.) 2 Program Against Cancer Therapeutic Resistance (ProCURE), Catalan Institute of Oncology, Ctra. Can Ruti- Camí de les escoles s/n, 08916 Badalona, Spain 3 Department of Pathology, Hospital Universitari Germans Trias i Pujol, Ctra. Can Ruti—Camí de les escoles s/n, 08916 Badalona, Spain; emusulen@hotmail.com 4 Department of Pathology, Hospital Universitari General de Catalunya, Grupo Quirónsalud, Pedro i Pons 1, 08195 Sant Cugat del Valles, Spain 5 Program Against Cancer Therapeutic Resistance (ProCURE), Catalan Institute of Oncology, 08908 L’Hospitalet del Llobregat, Barcelona, Spain; lpalomero@iconcologia.net 6 ONCOBELL Program, Bellvitge Institute for Biomedical Research, 08908 L’Hospitalet del Llobregat, Barcelona, Spain; mhalonso@iconcologia.net (M.H.A.); v.moreno@iconcologia.net (V.M.) 7 Oncology Data Analytics Program, Institut Català d’Oncologia (ICO), 08908 Barcelona, Spain 8 Consortium for Biomedical Research in Epidemiology and Public Health (CIBERESP), 28029 Madrid, Spain 9 Department of Clinical Sciences, Faculty of Medicine and Health Sciences, University of Barcelona, 08907 Barcelona, Spain 10 Group of Biomedical Research in Digestive Tract Tumors, CIBBIM-Nanomedicine, Vall d’Hebron University Hospital, Research Institute (VHIR), Universitat Autònoma de Barcelona, Passeig Vall d’Hebron, 119-129, 08035 Barcelona, Spain; rocio.nieto@vhir.org (R.N.); diego.arango@vhir.org (D.A.) 11 Medical Oncology Service, Catalan Institute of Oncology (ICO), 08908 Barcelona, Spain 12 B-ARGO group, Germans Trias I Pujol Research Institute (IGTP), Ctra. Can Ruti- Camí de les escoles s/n, 08916 Badalona, Spain * Correspondence: embalibrea@iconcologia.net; Tel.: +34-93-497-8684; Fax: +34-93-497-8654 † Vicenç Ruiz de Porras and Sara Bystrup contributed equally.  Received: 5 September 2019; Accepted: 9 October 2019; Published: 11 October 2019  Abstract: In recent years, an increasing number of studies have shown that elevated expression of cyclin dependent kinase (Cdk5) contributes to the oncogenic initiation and progression of many types of cancers. In this study, we investigated the expression pattern of Cdk5 in colorectal cancer (CRC) cell lines and in a large number of tumor samples in order to evaluate its relevance in this pathogenesis and possible use as a prognostic marker. We found that Cdk5 is highly expressed and activated in CRC cell lines and that silencing of the kinase decreases their migration ability. In tumor tissues, Cdk5 is overexpressed compared to normal tissues due to a copy number gain. In patients with localized disease, we found that high Cdk5 levels correlate with poor prognosis, while in the metastatic setting, this was only the case for patients receiving an oxaliplatin-based treatment. When exploring the Cdk5 levels in the consensus molecular subtypes (CMS), we found the lowest levels in subtype 1, where high Cdk5 again was associated with a poorer prognosis. In conclusion, we confirm that Cdk5 Cancers 2019, 11, 1540; doi:10.3390/cancers11101540 www.mdpi.com/journal/cancers Cancers 2019, 11, 1540 2 of 17 is involved in CRC and disease progression and that it could serve as a prognostic and predictive biomarker in this disease. Keywords: colorectal cancer; cyclin-dependent kinase 5 (Cdk5); prognostic and predictive biomarker; oxaliplatin 1. Introduction Cyclin dependent kinase (Cdk5) belongs to the cyclin-dependent kinase family of serine/threonine kinases whose activity depends on a regulatory subunit named cyclin. However, Cdk5 is considered an atypical Cdk because it binds to non-cyclin proteins p35 (CDK5R1) and p39 (CDK5R2), phosphorylation in the T-loop is not required for its activation and most of its substrates are not involved in cell cycle control. The function of this kinase was thought to be restricted to the central nervous system, where it is indispensable for normal brain development during embryogenesis and controls neuronal migration or axonal guidance, among others. Deregulation of Cdk5 is widely described in neurodegenerative diseases such as Alzheimer’s, Parkinson’s, and Huntington’s [1]. In recent years, several studies have demonstrated that Cdk5 also plays relevant physiological and pathological roles outside of the nervous system, including transcriptional regulation (through regulation of transcription factors such as p53 or signal transducer and activator of transcription 3 (STAT3)), cell migration, adhesion, angiogenesis, and apoptosis [2]. Given that some of these functions are deregulated in tumors and that a great similarity exists between the cellular and molecular mechanisms orchestrating neuronal migration during the development of the nervous system and cancer cell migration during metastasis, it is not surprising that Cdk5 plays a role in cancer. Cdk5 phosphorylates the STAT3 transcription factor on its Ser-727 residue, leading to increased proliferation and promotion of tumor formation in medullar thyroid carcinoma and in prostate cancer cells through the activation of androgen receptor [3,4] or resistance to DNA-damaging agents through the increased transcription of the DNA-repair gene essential meiotic structure-specific endonuclease 1 (EME1) in colorectal cancer (CRC) cell lines [5]. Cdk5 is involved in the regulation of DNA-damage repair through the phosphorylation of ataxia telangiectasia mutated (ATM) [6]. Elevated expression has been described in human pancreatic [7], lung [8], prostate [9], and breast [10] tumors and has been associated with a worse prognosis. The role of Cdk5 in the proliferation of tumor cells is not clear, but many studies have demonstrated its implication in cancer migration, invasion, and anchorage-independent growth. For instance, in pancreatic cancer, two studies showed that Cdk5 inhibition affected tumor malignant progression in vitro and in vivo. The rat sarcoma (RAS) pathway was reported to have an important effect on the Cdk5-mediated tumor progression on the one hand because mutant Kirsten rat sarcoma (KRAS) increased its kinase activity [7] and on the other hand, by participating in its downstream signaling through phosphorylation of Ras-related (Ral) A and B proteins [11]. In prostate and melanoma cancer models, the inhibition of Cdk5 resulted in decreased cell motility and metastatic potential [9,12] and in breast cancer, Cdk5 controlled cancer cell migration and tumor formation by regulating the phosphorylation of focal adhesion kinase (FAK) at Ser-732, as a downstream step of transforming growth factor Beta (TGF-B1) signaling, and was shown to be essential for epithelial to mesenchymal transition (EMT) and cell motility [10]. In CRC, Cdk5 functions as a tumor promoter via modulating the extracellular receptor kinase 5 (ERK5)—activator protein 1 (AP-1) axis [13] and it has been associated with oxaliplatin resistance acquisition [14]. To date, little is known about the possible role of Cdk5 as a cancer biomarker. In the present study, we studied the role of Cdk5 in several CRC cell lines, reproducing results from Zhuang et al. [13] and most importantly, we investigated the expression patterns of Cdk5 at different molecular levels in a significant number of human tumor samples from patients at different disease stages with the aim of evaluating its role as a prognostic and predictive marker in this malignancy. Cancers 2019, 11, 1540 3 of 17 Cancers 2019, 11, 1540 3 of 17 2. Results 2. Results 2.1. Cdk5 and p35 Are Broadly Expressed in CRC Cell Lines 2.1.studied We Cdk5 andprotein p35 Arelevels BroadlyofExpressed Cdk5 andinp35 CRCbyCell Lines blotting in a panel of 10 human CRC cell lines western and detectedWe both at high studied proteinlevels in of levels allCdk5 cell lines tested and p35 (Figure blotting by western 1A). In contrast, in a panel we did of 10 not detect human the p35 CRC cell calpain-cleaved lines and detected both at high levels in all cell lines tested (Figure 1A). In contrast, we did not detectCdk5 product p25, which is recognized by the same antibody. It has been reported that theisp35 activity calpain-cleaved regulated throughproduct p25, which isofrecognized phosphorylation by the Y15 [1], and thissame antibody. It hasisbeen phosphorylation reportedused commonly that Cdk5 as a measure of activity activity is regulated [5,15]. through However, basedphosphorylation of Y15 this on our experience, [1], is and notthis phosphorylation a good is approach because commonly used as a measure of activity [5,15]. However, based on our experience, Y15 is a highly conserved residue in the Cdk family [16] and the epitope against which these antibodies this is not a good approach because Y15 is a highly conserved residue in the Cdk family [16] and the epitope against are designed is highly homologous for Cdk1, Cdk2, and Cdk5, which could lead to misleading results. which these antibodies are designed is highly homologous for Cdk1, Cdk2, and Cdk5, which could As Cdk5 kinase activity is strictly dependent on the binding with p35/p25, this interaction also serves lead to misleading results. As Cdk5 kinase activity is strictly dependent on the binding with p35/p25, as a measure of activity. this interaction alsoToserves study this as abinding, measureCdk5 was immunoprecipitated of activity. To study this binding, and p25Cdk5was found was to co-immunoprecipitate immunoprecipitated and p25 was found to co-immunoprecipitate with Cdk5 in HT29, LoVo, also with Cdk5 in HT29, LoVo, HCT116, and DiFi cell lines (Figure 1B). It was foundHCT116, that p25andis more DiFi stable than cell lines p35 and (Figure 1B). leads It was to prolonged also found thatkinase p25 isactivity more stableof Cdk5 than [17], suggesting p35 and leads a Cdk5 tohyperactivation prolonged kinase inactivity these cell lines.[17], suggesting a Cdk5 hyperactivation in these cell lines. of Cdk5 Figure 1. Cyclin dependent kinase 5 (Cdk5) and p35 (CDK5R1) expression in colorectal cancer (CRC) Figure 1. Cyclin dependent kinase 5 (Cdk5) and p35 (CDK5R1) expression in colorectal cancer (CRC) cell lines. (A) Western blot analysis of Cdk5 and p35 basal expression in a panel of 10 CRC cell lines. cell lines. (A) Western blot analysis of Cdk5 and p35 basal expression in a panel of 10 CRC cell lines. Alpha-tubulin was used as endogenous control. (B) Representative western blot images showing the Alpha-tubulin was used as endogenous control. (B) Representative western blot images showing the co-immunoprecipitation co-immunoprecipitation of Cdk5 and and of Cdk5 p35/p25 in the p35/p25 indicated in the cellcell indicated lines. Co-immunoprecipitation lines. Co-immunoprecipitationwith an immunoglobin G (IgG) antibody was used as a negative control. The results with an immunoglobin G (IgG) antibody was used as a negative control. The results werewere obtained from at obtained least three from atindependent experiments. least three independent experiments. 2.2. Cdk5 siRNA-Mediated 2.2. Cdk5 Gene siRNA-Mediated Silencing Gene SilencingImpaired ImpairedCRC CRC Cells Migrationand Cells Migration andInvasion Invasion We investigated howhow We investigated thethe transient silencing transient silencingof of Cdk5 affectedthree Cdk5 affected threeimportant important characteristics characteristics of of tumortumor cells:cells: proliferation, proliferation,migration migrationand andinvasion. Forty-eighthours invasion. Forty-eight hours after after siRNA siRNA gene gene silencing, silencing, Cdk5 Cdk5 knockdown knockdown(siCdk5) (siCdk5)efficiency efficiencywas wasbetween between 90% and95% 90% and 95%atatthe the protein protein level level in HT29, in HT29, LoVo, LoVo, DiFi, and HCT116 cell lines (Figure S1A). Under these conditions, we did not observe DiFi, and HCT116 cell lines (Figure S1A). Under these conditions, we did not observe any effect in cellany effect in cell proliferation proliferation (Figure 2A(Figures 2A and S2A). and Figure S2A). However, we found significantlyreduced However, we found significantly reducedmigration and invasion migration in HT29 and and invasion LoVo siCdk5- in HT29 and LoVo transfected cell lines as compared to siRNA non-target control (siNTC) cells. In contrast, silencing of siCdk5-transfected cell lines as compared to siRNA non-target control (siNTC) cells. In contrast, Cdk5 did not decrease the migration and invasion ability of DiFi cells (Figure 2B,C). As HT29 and silencing of Cdk5 did not decrease the migration and invasion ability of DiFi cells (Figure 2B,C). As HT29 and LoVo cell lines harbor mutations in genes from the mitogen activated protein kinase (MAPK) signaling pathway [18] and the DiFi cell line is considered to be wild type [19], we wondered Cancers 2019, 11, 1540 4 of 17 Cancers 2019, 11, 1540 4 of 17 LoVo cell lines harbor mutations in genes from the mitogen activated protein kinase (MAPK) signaling pathway [18] and the DiFi cell line is considered to be wild type [19], we wondered whether whether an alteredan altered MAPK MAPK downstream downstream signaling signaling couldcould be behind be behind thethe observeddifferences. observed differences.We We took advantage of another KRAS wildtype (WT) cell line that was engineered to express a mutant form of KRAS (G12D). Cdk5 silencing caused a modest but statistically significant decrease in migration only in the SW48 KRAS G12D cell line and not in the KRAS WT. No differences in in invasion invasion ability after Cdk5 silencing were observed in either cell lines (Figure observed in either cell lines (Figure 2D,E). 2D,E). Another important important characteristic characteristicof ofcancer cancercells cellsisistheir theirability abilitytoto undergo undergo anan unlimited unlimited number number of of divisions. To study the implications of Cdk5 in their long-term clonogenic capability divisions. To study the implications of Cdk5 in their long-term clonogenic capability of CRC cells, we of CRC cells, we performed performed colony assays colony formation formation in theassays HT29 in andthe HT29 LoVo celland linesLoVo cellS2A,B). (Figure lines (Figure We observedS2A,B).a We observed decrease a decrease in the number inof the number colonies for of colonies for siCdk5-transfected siCdk5-transfected cells comparedcells compared to siNTC cellstoinsiNTC cells both lines, in both lines, however, however, these these statistically were only were only statistically significant significant for the LoVo for the LoVo cells cells (p-values (p-values 0.198 and 0.1980.046, and 0.046, respectively). respectively). Figure Figure 2. Effect of 2. Effect of Cyclin Cyclin dependent dependent kinase kinase 55 (Cdk5) (Cdk5) siRNA-mediated siRNA-mediated gene gene silencing silencing on on colorectal colorectal cancer cancer (CRC) (CRC)cells cellsproliferation, proliferation,migration migrationand and invasion. (A)(A) invasion. Graphic representation Graphic of HT29, representation LoVo,LoVo, of HT29, DiFi, DiFi, and HCT116 time-dependent cell proliferation after Cdk5 gene silencing measured by propidium Cancers 2019, 11, 1540 5 of 17 and HCT116 time-dependent cell proliferation after Cdk5 gene silencing measured by propidium iodide (PI). (B,D) Representative Boyden chamber migration and invasion assays images (4× magnification) and (C,E) bar graphs showing (mean ± SD) relative cell migration and invasion after Cdk5 silencing in the indicated cell lines. * p-value < 0.05 relative to control cells siRNA non-target control (siNTC). The results were obtained from at least three independent experiments. 2.3. Cdk5 and p35 Are Overexpressed in CRC Tumor Samples We next evaluated Cdk5 and p35 expression in human samples. We included four different cohorts: A, B, D, and F (Table 1). In cohort A, Cdk5 and p35 proteins were clearly overexpressed as detected by western blot in tumor tissues as compared to the corresponding normal adjacent tissues (Figure 3A–C). Cdk5 was detected by immunohistochemistry (IHC) in primary colorectal tumors from cohort D and scored as either negative or positive (weak and strong staining) (Figure 3D). A total of 75% of the tumors were positive for Cdk5 staining. We also observed significantly higher Cdk5 mRNA levels in tumor versus normal adjacent colon samples in two different publicly available data sets (Colonomics, cohort B, Figure 3E The cancer genome atlas (TCGA), cohort F, Figure 3F). Using TCGA data (cohort F), we analyzed the Cdk5 gene copy number variation (CNV) (region chr7:151054008–151057848). We observed a statistically significant increase (p = 7.58 × 10−7 ) in the Cdk5 copy number in primary tumors as compared to the corresponding adjacent normal colon tissues (Figure 3G). The CNV of the Cdk5 gene was strongly correlated (r = 0.55, p = 7.16 × 10−7 ) to Cdk5 expression levels in tumor tissue (Figure 3H) and also in a panel of 63 CRC cell lines (r = 0.54, p = 4.6 × 10−6 ) (Broad Institute Cancer Cell Line Encyclopedia) (Figure 3I). Cancers 2019, 11, 1540 6 of 17 Table 1. Overview of cohort used in the study. Cdk5 Origin of Molecular Clinical Data Cohort Type of Samples N Sex Age Metastasis Stage Treatment Available at Measurement Samples Data Available Available Frozen tissue Male 7 (78%) Liver 8 (89%) A WB 12 72 IV N/A Tumor Biobank N/A - - Tumor/adjacent Female 2 (22%) Lung 1 (11%) Radical Male 72 (57%) Colonomics B In silico data Micro array 98 72 (43-87) - II, MSS surgery 96 Colonomics.org RAS, CMS, etc DFS, OS Female 28 (43%) project (98%) DFS (mean Radical Male 21 (57%) Duran and 36.3 months) C Frozen tissue qPCR 37 78 (37-91) - III surgery 36 N/A - Female 16 (43%) Reynals Hospital OS (mean (97%) 42 months) 5-FU/OXA Liver 37 (71%) Male 29 (56%) (77%) DFS (mean D FFPE – TMA IHC 52 62 (37-76) Lung 5 (10%) IV Private collection ISCIII - Female 23 (44%) CAPE/OXA 9.6 months) Others 10 (19%) (23%) 5-FU/LV/IRI DFS (mean Liver 97 (70%) Male 78 (69%) (47%) 9.4 months) E FFPE – TMA IHC 139 62 (29-75) Lung 46 (33%) IV Private collection ISCIII - Female 35 (31%) 5-FU/IRI OS (mean Others 20 (14%) (53%) 20.3 months) Male 259 (54%) https: CNV, RAS, F In silico data RNA seq 473 66 (31-90) - I-IV Various TCGA project PFI, OS Female 225 (46%) //cancergenome.nih.gov CMS, etc Abbreviations: 5-FU/LV/IRI: 5-Fluorouracil/leucovorin/irinotecan; 5-FU/IRI: 5-Fluorouracil/irinotecan; 5-FU/OXA: 5-Fluorouracil/oxaliplatin; CAPE/OXA: Capecitabine/oxaliplatin; CMS: Consensus molecular subtype; CNV: Copy number variation; DFS: Disease-free survival; FFPE-TMA: Formalin-fixed paraffin-embedded tissue microarray; IHC: Immunohistochemistry; ISCIII: Instituto de Salud Carlos III; MSS: Microsatellite stable; N/A: Not available; OS: Overall survival; PFI: Progression-free interval; qPCR: Quantitative polymerase chain reaction; TCGA: The cancer genome atlas; RAS: Rat sarcoma oncogene; WB: Western blot. Cancers Cancers 2019, 154011, 1540 11, 2019, 7 of 17 9 of 17 Figure Figure 3. 3. Cyclin Cyclin dependent dependent kinase (Cdk5) kinase (Cdk5) andand p35 expression in colorectal p35 expression in cancer (CRC) cancer colorectal tumor samples. (CRC) tumor Western blot (A) and graphic representation (B,C) of Cdk5 and p35 protein expression, respectively, samples. Western blot (A) and graphic representation (B,C) of Cdk5 and p35 protein expression, in tumoral (T) and normal adjacent (N) tissues of 12 stage IV CRC patients (cohort A). Alpha-tubulin respectively, in tumoral was used (T) and as endogenous normal control. Theadjacent (N)according p-value was tissues ofto 12 stage paired IV CRC t-test. patients (cohort A). (D) Representative Alpha-tubulin was used as endogenous control. The p-value was according immunohistochemistry images of Cdk5 staining in CRC tumor tissues. The upper panel shows to paired t-test. (D) negative staining and Representative lower panel positive staining. immunohistochemistry Scaleof images bar: 100 µm. Cdk5 (E) Graphic staining representation in CRC of Cdk5 and tumor tissues. The upper panel shows negative staining and lower panel positive staining. Scale bar: 100 μm. (E) Graphic representation of Cdk5 and p35 mRNA expression in 98-paired adjacent normal and tumoral tissues from stage II microsatellite stable (MSS) CRC patients (cohort B). (F) Graphic representation of Cdk5 mRNA expression in normal and tumoral tissues of 38 I-IV CRC patients. Data were obtained from The Cancer Genome Atlas (TCGA) database (cohort F). (G) Graph representing the Cdk5 copy number Cancers 2019, 11, 1540 8 of 17 p35 mRNA expression in 98-paired adjacent normal and tumoral tissues from stage II microsatellite stable (MSS) CRC patients (cohort B). (F) Graphic representation of Cdk5 mRNA expression in normal and tumoral tissues of 38 I-IV CRC patients. Data were obtained from The Cancer Genome Atlas (TCGA) database (cohort F). (G) Graph representing the Cdk5 copy number (CNV) change between normal and tumoral tissues in 67 stage I–IV CRC patients (cohort F). (H) Correlation between Cdk5 CNV and Cdk5 gene expression in 429 stage I–IV CRC patients. p-value according to Pearson correlation test (cohort F). (I) Correlation between Cdk5 CNV and Cdk5 gene expression in 63 CRC cell lines. The p-value was according to Pearson correlation test. Data from the Broad Institute Cancer Cell Line Encyclopedia. 2.4. High Cdk5 Levels Correlate with Poorer Prognosis in CRC Patients Taking into account this data, we wondered whether Cdk5 expression could be a good prognostic and/or predictive biomarker in patients. To study this possibility, we used several cohorts in which the expression of Cdk5 was determined through different techniques (Table 1). We categorized data into high and low Cdk5 expression groups as follows: In cohorts using continuous data (cohorts B, C and F), patients were assigned to low or high groups according to Cdk5 values below or above the median, respectively; in those using categorical data (cohorts D and E), patients were grouped as low when IHC staining was negative or weak and as high when it was strong (Figure 3D). The prognostic value of Cdk5 expression was studied in two highly homogenous cohorts of localized microsatellite stable (MSS) CRC cancer patients (cohorts B and C, Table 1). Patients from cohorts B and C were diagnosed with stage II or stage III cancer, respectively, underwent surgery and did not receive any adjuvant treatment. In both the cohorts, high Cdk5 levels were associated with shorter disease free survival (DFS) (p = 0.049 95%, confidence interval (CI) 0.98–5.88 for cohort B and p = 0.048, 95% CI 0.97–7.43 for cohort C) (Figure 4A,B), and also with overall survival (OS) in cohort B (p = 0.022, 95% CI 0.96–6.48), while no differences could be observed in OS in cohort C (Figure S3A,B). To study the predictive value of Cdk5 expression in the metastatic setting, we used two different, but again, quite homogenous cohorts of patients. Cohort D consisted of stage IV CRC patients treated with oxaliplatin and 5-fluorouracil as the first-line treatment, while patients from cohort E were treated with different schedules combining irinotecan and 5-fluorouracil (Table 1). As shown in Figure 4C,D, high Cdk5 levels were associated with shorter time to progression (TTP) only in the patients that received an oxaliplatin-based treatment (median 8.095 vs. 18.23 months, p = 0.043, 95% CI 1.01–4.71) (cohort D, Figure 4C), as no differences could be observed in the patients treated with irinotecan (median 8.95 vs. 8.42 months, p = 0.45, 95% CI 0.59–1.27) (cohort E, Figure 4D). These analyses were adjusted by age and sex; however, we performed an additional analysis including metastatic location as another co-variable, which resulted in an increased significance (p = 0.018, hazard ratio (HR) = 2.72, 95% CI 1.19–6.23). Similarly, when analyzing the response rates (Chi-square), we only found differences according to Cdk5 levels in the oxaliplatin-treated patients (cohort D) as 87% (13 out of 15 patients) of the low Cdk5 group had complete or partial response, while only 53% (18 out of 34) of the high Cdk5 group responded (p = 0.029). In view of these interesting results in the metastatic setting, we wanted to analyze whether Cdk5 was also predictive for adjuvant oxaliplatin-based therapy response. As we do not have our own cohort with this patient group, we used the TCGA data (cohort F). When splitting this subgroup (n = 49) based on high or low Cdk5, we could not observe any difference in the progression free interval (PFI) (p = 0.64, HR CI) nor OS (p = 0.11, HR CI). It is worth mentioning that when analyzing the entire TCGA cohort (n = 473, stages I to IV, treated and non-treated), no differences could be observed in PFI and OS (Figure S3C,D), highlighting the importance of choosing appropriate and well-defined cohorts of patients to conduct these studies. Cancers 2019, 11, 1540 9 of 17 Cancers 2019, 11, 1540 11 of 17 Figure 4. Kaplan–Meyer analysis of disease free survival (DFS) and time to progression (TTP) depending Figure on4. Cyclin Kaplan–Meyer analysis dependent kinase (Cdk5)oflevels. disease (A) DFSfreein survival (DFS) and 98 stage II colorectal cancertime topatients (CRC) progression split (TTP) by the median of Cdk5 expression (cohort B). (B) DFS in 37 non-treated stage III depending on Cyclin dependent kinase (Cdk5) levels. (A) DFS in 98 stage II colorectal cancer (CRC) CRC patients split by the median of Cdk5 expression (cohort C). (C) TTP in 52 stage IV oxaliplatin-treated patients, grouped patients split by the median of Cdk5 expression (cohort B). (B) DFS in 37 non-treated stage III CRC depending on negative (n = 18) or positive Cdk5 (n = 34) immunohistochemistry (IHC) staining (cohort patientsD). split (D) by TTPthe median in 139 of irinotecan-treated stage IV Cdk5 expression (cohort patients C). (C) grouped TTP in 52 depending on stage IV(noxaliplatin-treated negative = 73) and patients,positive grouped(n = depending 66) Cdk5 IHCon negative staining (nE). (cohort = 18) or positive Cdk5 (n = 34) immunohistochemistry (IHC) staining (cohort D). (D) TTP in 139 stage IV irinotecan-treated patients grouped depending As we observed a difference in migration and invasion when we silenced Cdk5 in our in vitro on negative (n = 73) and positive (n = 66) Cdk5 IHC staining (cohort E). studies, we analyzed the link between Cdk5 expression and several gene sets related to EMT in the TCGA cohort. EMT is a well described process whereby cells undergo multiple biochemical changes Asthat we enable observed them a to difference in migration assume a mesenchymal and invasion phenotype, when characterized we silenced by enhanced Cdk5capacity, migratory in our in vitro studies,invasiveness, we analyzed andthe link resistance elevated betweentoCdk5 expression apoptosis. We foundand thatseveral gene only for the setssignaling TGF-β related within to EMT in the EMT didEMT TCGA cohort. high Cdk5 levelsdescribed is a well correlate with higherwhereby process pathway activation (p = 0.026, cells undergo Figure S4). multiple biochemical changes We observed a difference in the effect of Cdk5 silencing in cell lines depending on their KRAS that enable them to assume a mesenchymal phenotype, characterized by enhanced migratory mutational status when studying the migration and invasion of the cells. Furthermore, it has been capacity, invasiveness, and elevated resistance to apoptosis. We found that only for the TGF-β reported that Cdk5 is involved in the signaling cascade downstream of the epithelial growth factor signaling within receptor EMTreceptor (EGFR) did high [7]. Cdk5 levels Thus, we correlate wanted to knowwith higher whether there pathway activation was a difference (p = 0.026, Figure in the prognostic S4). We observed a difference in the effect of Cdk5 silencing in cell lines depending on their KRAS mutational status when studying the migration and invasion of the cells. Furthermore, it has been reported that Cdk5 is involved in the signaling cascade downstream of the epithelial growth factor receptor (EGFR) receptor [7]. Thus, we wanted to know whether there was a difference in the Cancers 2019, 11, 1540 10 of 17 value of Cdk5 according to KRAS mutations. Patients from cohort B were split according to KRAS mutational status (WT or mutated) and Cdk5 expression was again categorized as high or low. In the KRAS CancersWT 2019,group, 11, 1540patients with high or low Cdk5 levels had similar DFS (Figure 5A); however, in the 12 of 17 group with KRAS, mutated high Cdk5 levels predicted a statistically very significant poorer DFS see any differences compared to the Cdk5 in OS lowingroup (p = 0.004, this cohort when95% grouping according CI 1.56–36.46) to the5B). (Figure KRASWemutational did not seestatus any (Figure S3E,F). differences in OS in this cohort when grouping according to the KRAS mutational status (Figure S3E,F). Figure5.5.Kaplan–Meyer Figure Kaplan–Meyeranalysis of disease analysis free free of disease survival (DFS)(DFS) survival depending on Cyclin depending dependent on Cyclin kinase dependent (Cdk5) kinase expression and Kirsten (Cdk5) expression and rat sarcoma Kirsten oncogene rat sarcoma (KRAS) mutational oncogene status. (A) (KRAS) mutational DFS (A) status. for 72 DFSstage II for 72 colorectal cancer (CRC) stage II colorectal cancer patients (CRC)with wildtype patients with (WT) KRAS wildtype and KRAS (WT) split byand thesplit median of Cdk5 by the expression. median of Cdk5 (B) DFS for 26 expression. (B)stage DFS IIfor CRC patients 26 stage withpatients II CRC mutated KRAS with and split mutated KRASby and the median split by of theCdk5 expression median of Cdk5 (cohort B). expression (cohort B). 2.5. Cdk5 Is Associated with Consensus Molecular Subtype 1 2.5. Cdk5 Is Associated with Consensus Molecular Subtype 1 CRC has recently been classified into four consensus molecular subtypes (CMS) [20], namely, CMS1 CRC has recently been classified into four consensus molecular subtypes (CMS) [20], namely, (microsatellite instability immune), CMS2 (canonical), CMS3 (metabolic), and CMS4 (mesenchymal) CMS1 (microsatellite instability immune), CMS2 (canonical), CMS3 (metabolic), and CMS4 subtypes. The subtype with the worst survival probability is the CMS4 [20]. Taking advantage of (mesenchymal) subtypes. The subtype with the worst survival probability is the CMS4 [20]. Taking the fact that this information was available in cohort F, we wanted to explore whether there was an advantage of the fact that this information was available in cohort F, we wanted to explore whether association between Cdk5 levels and a specific CMS. First, we checked if CMS-associated OS followed there was an association between Cdk5 levels and a specific CMS. First, we checked if CMS-associated the same trend as reported previously and indeed, this was the case (Figure S5C). CMS1 displayed OS followed the same trend as reported previously and indeed, this was the case (Figure S5C). CMS1 the lowest levels of Cdk5 (Figure 6A,B). For each molecular subtype, we then compared the OS displayed the lowest levels of Cdk5 (Figure 6A,B). For each molecular subtype, we then compared between patients with high or low Cdk5 levels and did not observe any differences (Figure S5A). the OS between patients with high or low Cdk5 levels and did not observe any differences (Figure S5A). When we analyzed the progression-free interval (PFI), an endpoint similar to DFS and TTP used in our When we analyzed the progression-free interval (PFI), an endpoint similar to DFS and TTP used in previous analysis which has been shown to be the most reliable endpoint when analyzing data from our previous analysis which has been shown to be the most reliable endpoint when analyzing data the TCGA-colorectal adenocarcinoma (COAD) database [21], we found that high Cdk5 in the CMS1 from the TCGA-colorectal adenocarcinoma (COAD) database [21], we found that high Cdk5 in the correlated with poorer outcomes (p = 0.036, 95% CI 1.07–8.10) (Figure 6C), while no differences were CMS1 correlated with poorer outcomes (p = 0.036, 95% CI 1.07–8.10) (Figure 6C), while no differences found in the three remaining molecular subtypes (Figure S5B). It is worth mentioning that the majority were found in the three remaining molecular subtypes (Figure S5B). It is worth mentioning that the of the CMS1 group were non-metastatic patients (55 M0 and 4 M1). majority of the CMS1 group were non-metastatic patients (55 M0 and 4 M1). Cancers 2019, 11, 1540 11 of 17 Cancers 2019, 11, 1540 13 of 17 Figure 6. Cyclin dependent kinase 5 (Cdk5) expression in the four consensus molecular subtypes (CMS). (A) Dunn Figure 6. Cyclin and Kruskal–Wallis dependent multiple kinase 5 (Cdk5) comparison expression inofthe Cdk5 expression four consensusin CMS in 98 stage molecular II subtypes tumors. Note that only six cases were classified as CMS1 as this cohort was restricted to microsatellite (CMS). (A) Dunn and Kruskal–Wallis multiple comparison of Cdk5 expression in CMS in 98 stage II stabile (MSS) cases (B) ANOVA analysis of Cdk5 expression levels in the CMS in the cancer genome tumors. Note that only six cases were classified as CMS1 as this cohort was restricted to microsatellite atlas - colorectal adenocarsinoma - rectal adenocarcinoma dataset (TCGA-COAD-READ) including 410 stabile (MSS) stage I–IVcases (B) *ANOVA patients. p-value