Cancer Therapeutics, Targets, and Chemical Biology Research
Chromosomal Instability Confers Intrinsic Multidrug Resistance
Alvin J.X. Lee1, David Endesfelder1,4, Andrew J. Rowan1, Axel Walther5,6, Nicolai J. Birkbak7, P. Andrew Futreal8, Julian Downward2, Zoltan Szallasi7,9, Ian P.M. Tomlinson5, Michael Howell3, Maik Kschischo4, and Charles Swanton1,6
Abstract Aneuploidy is associated with poor prognosis in solid tumors. Spontaneous chromosome missegregation events in aneuploid cells promote chromosomal instability (CIN) that may contribute to the acquisition of multidrug resistance in vitro and heighten risk for tumor relapse in animal models. Identification of distinct therapeutic agents that target tumor karyotypic complexity has important clinical implications. To identify distinct therapeutic approaches to specifically limit the growth of CIN tumors, we focused on a panel of þ colorectal cancer (CRC) cell lines, previously classified as either chromosomally unstable (CIN ) or diploid/near- diploid (CIN ), and treated them individually with a library of kinase inhibitors targeting components of signal þ transduction, cell cycle, and transmembrane receptor signaling pathways. CIN cell lines displayed significant intrinsic multidrug resistance compared with CIN cancer cell lines, and this seemed to be independent of somatic mutation status and proliferation rate. Confirming the association of CIN rather than ploidy status with multidrug resistance, tetraploid isogenic cells that had arisen from diploid cell lines displayed lower drug sensitivity than their diploid parental cells only with increasing chromosomal heterogeneity and isogenic cell þ line models of CIN displayed multidrug resistance relative to their CIN parental cancer cell line derivatives. In þ a meta-analysis of CRC outcome following cytotoxic treatment, CIN predicted worse progression-free or disease-free survival relative to patients with CIN disease. Our results suggest that stratifying tumor responses according to CIN status should be considered within the context of clinical trials to minimize the confounding effects of tumor CIN status on drug sensitivity. Cancer Res; 71(5); 1858–70. 2011 AACR.
Introduction instability (CIN), resulting in ongoing numerical and struc- tural chromosomal aberrations in cancer cells, leading to Colorectal cancer (CRC) is associated with at least 2 intratumoral heterogeneity (2, 3). The less common pattern distinct patterns of genomic instability (1). The more com- is microsatellite instability (MIN) caused by a deficiency in mon form of genomic instability in CRC is chromosomal the mismatch repair apparatus. Because of the accumulation of replication errors, MIN results in length variation of þ microsatellite sequences in DNA. Most MIN CRC cell lines 1 are near-diploid (4, 5) and chromosomally stable (CIN ). In Authors' Affiliations: Translational Cancer Therapeutics Laboratory, þ 2Signal Transduction Laboratory, and 3High Throughput Screening contrast, CIN CRC cell lines are aneuploid and display a Laboratory, Cancer Research UK London Research Institute, London, higher frequency of chromosomal missegregation errors United Kingdom; 4University of Applied Sciences, Mathematics and 5 during each mitosis relative to diploid cells (2). In human Techniques, Remagen, Germany; Molecular and Population Genetics, þ The Wellcome Trust Centre for Human Genetics, Oxford, United Kingdom; CRC, CIN is widely inferred through the measurement of 6 Royal Marsden Hospital, Department of Medicine, Sutton, United King- tumor DNA ploidy (6); normal diploid cells are defined with dom; 7Center for Biological Sequence Analysis, Technical University of Denmark, Lyngby, Denmark; 8Cancer Genome Project, Wellcome Trust a DNA index of 1.0 (7) and thus an increase in DNA index Sanger Institute, Cambridge, United Kingdom; and 9Children's Hospital infers polyploidy or aneuploidy. Approximately 25% of Informatics Program at the Harvard-MIT Division of Health Sciences and human CRC are both CIN and MIN (6). Technology, Harvard Medical School, Boston, Massachusetts þ Consistent molecular mechanisms responsible for the CIN Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/). phenotype and hence means to target this pattern of genome instability in colorectal and other solid tumors remain poorly A.J.X. Lee and D. Endesfelder contributed equally to the work. defined. Putative mechanisms that may contribute to CIN Corresponding Author: Charles Swanton, Translational Cancer Therapeu- ticsLaboratory, Cancer Research UKLondonResearch Institute,44Lincoln's include weakening of the spindle assembly checkpoint (SAC; Inn Fields, London WC2A 3LY, United Kingdom. Phone: 44-20-7269-3463; refs. 8, 9), defective sister chromatid cohesion (10), merotelic Fax 44-20-7269-3094; E-mail: [email protected] sister chromatid attachments (11), defective cytokinesis (12), doi: 10.1158/0008-5472.CAN-10-3604 centrosome amplification (13), and chromosome breakage– 2011 American Association for Cancer Research. fusion–bridge cycles (14).
1858 Cancer Res; 71(5) March 1, 2011
Downloaded from cancerres.aacrjournals.org on September 30, 2021. © 2011 American Association for Cancer Research. CIN Confers Intrinsic Multidrug Resistance
Increasing evidence suggests that CIN is associated with analysis were present within the CLP database, and a total of poor prognosis in solid tumors (6, 15, 16). It has been sug- 20 of the 61 genes resequenced in the project were found to gested that adverse outcome associated with CIN may be have somatic mutations in at least 1 of those 21 cell lines. related to increased tumor cell heterogeneity, driving the Additional information regarding the somatic mutation status ability of tumors to adapt to environmental stresses (17– of APC, CTNNB1, KRAS, MLH1, PIK3CA, and TP53 were 19). Consistent with a hypothesis whereby CIN may enhance obtained from both published (32–35) and internal laboratory þ tumor adaptation, transient initiation of CIN, following the data. Isogenic HCT116 MAD2 / cell lines (9) and HCT116 brief induction of MAD2 expression in activated KRAS- PTTG1 / cell lines (10) were donated courtesy of Drs. initiated mouse lung tumor models, is associated with a high Benezra and Vogelstein, respectively. To generate tetraploid frequency of tumor recurrence following withdrawal of the HCT116 cells, naturally occurring tetraploid cells were iso- KRAS oncogenic stimulus (20). Preclinical studies have shown lated from the parental cell line and single cell sorted using that CIN is associated with the rapid acquisition of multidrug flow cytometry. Clonal FISH was performed with centromere resistance in cell line systems (21) and intrinsic taxane resis- enumeration probes against centromeres on chromosomes 2 tance in vitro and in vivo (22). Recently, we and others have and 15. proposed the existence of a CIN survival phenotype that þ allows CIN tumor cells to tolerate the impact of excessive Calbiochem Kinase Inhibitor Library and 5-FU screen chromosome gains and losses (22–24) that may in turn impact Calbiochem Kinase Inhibitor Libraries I and II (EMD Bios- upon altered drug sensitivity. ciences) containing 160 inhibitors were used. Comprehensive Determining how CIN might impact upon prognosis and data for these inhibitors including references documenting how this pattern of genomic instability might be specifically target inhibition or downstream signaling cascade inactiva- targeted remains an important research area (23, 25). Evi- tion can be found on the manufacturer's Web site (36). dence in lower eukaryotes has shown that aneuploid Sacchar- Cells were plated into 96-well tissue culture microplates at omyces cerevisiae are dependent on increased glucose an initial seeding density of 4,000 cells per well. After 24 hours, utilization and are more sensitive to both heat shock protein cells were treated with the inhibitors at a final concentration of 90 and proteosome inhibitors (26). Polyploid S. cerevisiae are 10 mmol/L per well. This concentration was selected following dependent upon increased expression of genes involved in drug titration test analysis to give an optimal range of mean sister chromatid cohesion and mitotic spindle function (27). relative surviving cells following inhibitor treatment across all Roschke and Kirsch have shown the existence of anticancer cell lines. After 72 hours of treatment, cell viability was assayed compounds that may specifically target karyotypically com- using the Celltiter-Blue Cell Viability Assay (Promega) accord- plex cancer cells (25). These observations indicate that ing to the manufacturer's instructions. The fluorescent readout karyotypic instability may be specifically targeted in eukar- for each inhibitor treated well was normalized to vehicle yotic organisms and suggest that CIN might be an exploitable control wells. For the HCT116 MAD2 isogenic cell lines, a final and targetable phenotype in cancer. inhibitor concentration of 1 mmol/L per well was used to allow To identify distinct therapeutic approaches to limit the for a larger range of surviving cells across all inhibitors. þ growth of CIN tumors relative to diploid cells, we focused on 5-Fluorouracil (5-FU) was used at both 1 and 10 mmol/L for a panel of CRC cell lines that had previously been classified as a treatment length of 72 hours. The 27 CRC cell lines were þ CIN or CIN and used kinase inhibitor and cytotoxic libraries plated and assayed using the same conditions and methods as to identify agents that might be preferentially lethal toward described earlier. þ þ CIN cells. Both isogenic and nonisogenic CIN cell lines displayed intrinsic multidrug resistance in vitro relative to Biolog Anti-Cancer Agent Microplate screen CIN cell lines. Importantly, consistent with the proposal that Biolog Anti-Cancer Agent Microplates M11–M14 consisting þ CIN is associated with intrinsic multidrug resistance, in a of 92 anticancer agents at 4 increasing concentrations þ meta-analysis of patient outcome in CRC, CIN was asso- between 0.1 and 25 mmol/L per agent were used. Cells were ciated with significantly worse clinical outcome relative to plated at an initial seeding density of 5,000 cells per well. After diploid cancers in both early- and late-stage disease following 72 hours, the number of surviving cells per well was assayed cytotoxic therapy. using Biolog Redox Dye Mix MA according to manufacturer's instructions and normalized to the negative control wells. Materials and Methods Cell proliferation assay Cell lines and FISH analysis We measured the proliferation rate of all 9 CIN and 18 þ 27 CRC cell lines (Table 1; Supplementary Table 1), pre- CIN CRC cell lines, using the IncuCyte Long-term Cell viously characterized for numerical/structural CIN, MIN sta- Imaging System. Cell lines were plated in 96-well plates at tus (2, 28–30), and subject to Affymetrix SNP 6.0 Array analysis an initial plating density of 4,000 cells per well and phase- where available (20/27 cell lines; Wellcome Trust Sanger contrast images were obtained every 2 hours over 70 hours, Institute), were used. allowing measurements of cell monolayer confluence. Outliers We used publicly available somatic mutation data from the were removed manually, and growth curves were fitted by Sanger Institute Cancer Cell Line Project (CLP) and COSMIC splines with the R package grofit (37) with smoothing para- þ database (31). Fifteen CIN and 6 CIN cell lines used in our meter smooth.gc ¼ 0.7.
www.aacrjournals.org Cancer Res; 71(5) March 1, 2011 1859
Downloaded from cancerres.aacrjournals.org on September 30, 2021. © 2011 American Association for Cancer Research. 1860
acrRs 15 ac ,2011 1, March 71(5) Res; Cancer Table 1. Cell lines used in this study al. et Lee Downloaded from cancerres.aacrjournals.org Cancer Research. on September 30, 2021.©2011AmericanAssociation for acrResearch Cancer
NOTE: Black indicates presence of mutation; white indicates absence of mutation; and gray indicates mutation status not known. CIN Confers Intrinsic Multidrug Resistance
Meta-analysis of clinical studies Results Survival data were summarized using a log hazard ratio for þ þ comparison between CIN and CIN groups. Data from Classification of CIN cell lines and relationship with individual studies were extracted using the methods described ploidy status and structural chromosomal complexity þ by Parmar and colleagues (38) and pooled to generate the We selected 9 CIN and 18 CIN CRC cell lines (Table 1). summary statistic and CIs using a fixed-effects model with CIN status for these cell lines had been previously described in inverse variance weighting. All meta-analyses were performed terms of numerical and structural chromosomal aberrations using Stata 10.1 (Stata Corp). (2, 28–30). To confirm the utility of published approaches for defin- Statistical methods ing the CIN status of the cell lines, where SNP Array data All tests were performed as 2-sided unless otherwise men- were available, we estimated the ploidy status of the cell tioned. To remove outliers, drugs resulting in relative number lines by using weighted mean integer copy numbers derived of cells greater than 1.4 were eliminated from analysis. A from the PICNIC (Predicting Integral Copy Numbers In Kolmogorov–Smirnov test was performed to test for overall Cancer) algorithm (ref. 40; Fig. 1A). Ploidy estimates classi- þ differences in the distribution of relative cell numbers follow- fied cells as CIN if they surpassed a threshold of ploidy þ ing inhibitor treatment between CIN and CIN cell lines. greater than 2.2 (corresponding to an estimated DNA index Inhibitors that showed a fraction of surviving cells greater of 1.1). In addition, modal chromosomal number as deter- than 0.8 in more than 75% of the cell lines were excluded from mined previously using high-quality metaphase spreads (28, the analysis. For comparisons of 2 cell lines, drugs resulting in 29) correlated well with ploidy estimates derived from relative number of cells greater than 0.8 in both cell lines were weighted mean PICNIC copy number analysis using the removed from analysis. A Wilcoxon signed-rank test was used SNP Array data [Pearson's correlation coefficient (CC) ¼ þ to test for differences between CIN and CIN cells (for each 0.94, Pearson's correlation test P < 0.0001; Fig. 1B]. These concentration of drug in Biolog microplates). For the Biolog data support the utility of SNP Array data for estimating microplates, each concentration of drug was used in duplicate; cell line ploidy status and are consistent with ploidy esti- therefore, the replicates which showed the least difference in mates by traditional measures, confirming the CIN status of cell number between the 2 cell lines was used for further cell lines used in this analysis. analysis. Next, we addressed the relationship between CIN status and The maximal slope of the growth curve (m) for each cell structural chromosomal complexity, using a summary struc- line was used to test whether the difference in drug sensi- tural chromosomal complexity score (SCCS) derived from the þ tivity between CIN and CIN colorectal cancer cell lines SNP Array data sets. This SCCS was determined by summar- þ was not solely due to different proliferation rates. CIN and izing (i) the number of break points, (ii) LOH events as CIN cell lines with m < 1 were tested with a one-sided predicted using PICNIC (40), and (iii) the Genome Integrity Wilcoxon–Mann–Whitney test. Next, we corrected for the Index (GII; ref. 41) into a single value for each cell line (Fig. 1C). influence of the proliferation rates of each cell line. We There was a highly significant correlation between ploidy estimated a linear regression model for all cell lines with status and the SCCS (Pearson's CC ¼ 0.746, P ¼ 0.0002). m < 1, where m was used as an independent variable and the Taken together, these analyses confirm that cells classified as þ mean fraction of surviving cells over all inhibitors as a CIN have significantly greater ploidy and structural chromo- dependent variable. The residuals of this analysis were somal complexity than CIN cells. þ tested for significant differences between CIN and CIN þ with a one-sided Wilcoxon–Mann–Whitney test (39). An CIN status is associated with intrinsic multidrug analysis of covariance model with interactions was applied, resistance where the mean fraction of surviving cells over all inhibitors Next, we aimed to determine whether a specific kinase þ was used as a dependent variable, m as a linear, independent inhibitor could be identified to selectively target CIN cell variable, and CIN status as a factor variable. The interaction lines. We used a small molecule library (Calbiochem Kinase termwasutilizedtotestforsignificant differences in slopes Inhibitor Library I and II) that included 160 inhibitors to treat þ þ between CIN and CIN cell lines. the 18 CIN and 9 CIN cell lines. Preliminary drug titration þ To test for significant differences in sensitivity to thymi- experiments revealed that the majority of the CIN cell lines dylate synthase inhibitors comparing HCT116 diploid par- were resistant to concentrations up to 1 mmol/L (Supplemen- þ ental and PTTG1 / or MAD2 / cells, we corrected for the tary Fig. 1) and therefore 10 mmol/L was selected as the influence of different concentrations by estimating a linear optimal drug concentration for cell growth inhibition across regression model for near linear correlations between con- the majority of cell lines in order to attempt to identify drugs þ centration and sensitivity or a one-way ANOVA otherwise. In that were specifically active in CIN cells. No specific inhibitor each case, the concentration was used as the independent or inhibitor family was found to be preferentially active in þ variable and the resulting residuals were tested for differ- CIN cell lines compared with CIN cell lines. In contrast, þ þ ences between CIN and CIN cells with a Wilcoxon–Mann– CIN cancer cell lines were significantly more resistant to the Whitney test. inhibitors (Kolmogorov–Smirnov test, P < 0.0001; Fig. 2A–C). All statistical analyses were performed in R and can be Following correction for multiple testing hypotheses, 45 inhi- found in the Supplementary Sweave document. bitors were identified which showed significantly greater
www.aacrjournals.org Cancer Res; 71(5) March 1, 2011 1861
Downloaded from cancerres.aacrjournals.org on September 30, 2021. © 2011 American Association for Cancer Research. Lee et al.