Despite being the most common pelvic gynecologic malignancy in industrialized countries, no targeted therapies are available for patients with metastatic endometrial carcinoma. In order to improve treatment, underlying molecular characteristics of primary and metastatic disease must be explored.
We utilized the mass spectrometric-based mutation detection technology OncoMap to define the types and frequency of point somatic mutations in endometrial cancer. 67 primary tumors, 15 metastases corresponding to 7 of the included primary tumors and 11 endometrial cancer cell lines were screened for point mutations in 28 known oncogenes. We found that 27 (40.3%) of 67 primary tumors harbored one or more mutations with no increase in metastatic lesions. FGFR2, KRAS and PIK3CA were consistently the most frequently mutated genes in primary tumors, metastatic lesions and cell lines.
Citation: Krakstad C, Birkeland E, Seidel D, Kusonmano K, Petersen K, Mjøs S, et al. (2012) High-Throughput Mutation Profiling of Primary and Metastatic Endometrial Cancers Identifies KRAS, FGFR2 and PIK3CA to Be Frequently Mutated. PLoS ONE 7(12): e52795. https://doi.org/10.1371/journal.pone.0052795
Editor: Wayne A. Phillips, Peter MacCallum Cancer Centre, Australia
Received: September 28, 2012; Accepted: November 21, 2012; Published: December 27, 2012
Copyright: © 2012 Krakstad et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The Western Norway Regional Health Authority (grant number: 911351 and 911624), Norwegian Research Council (grant number: 193373 and 205404), and The Norwegian Cancer Society (grant number: 628837). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Despite being the most common pelvic gynecologic malignancy in industrialized countries, no targeted therapies are available for patients with metastatic endometrial carcinoma. Although 75% are treated at an early stage, 15% to 20% recur. For patients with advanced disease at diagnosis or recurrent disease, outcome is poor. In order to improve treatment, underlying molecular characteristics of primary and metastatic disease must be explored. Furthermore, improved tools for correct stratification of patients according to risk-groups and improved definitions of potential targets for novel therapeutics are of great importance and much work is undertaken to develop better criteria to select patients for individualized therapies .
To assess the risk of recurrent disease, traditionally endometrial cancer has been divided into two subgroups, type I and type II carcinomas . Type I endometrial carcinoma is associated with good prognosis, low grade, endometrioid histology and rarely metastasize to regional and distant sites . In addition, type I endometrial cancers are often hormone receptor positive with PTEN and KRAS mutations. Type II endometrial carcinomas are associated with poor prognosis, non-endometrioid histology, high grade, loss of hormone receptors and altered expression of p53 and p16. Still, the value of this classification to predict prognosis and for treatment stratification is limited as 20% of type I endometrial cancers recur and 50% of type II cancers do not .
Currently, conventional chemotherapy regimens and anti-hormonal treatment are basis for adjuvant and systemic treatment of recurrent or metastatic endometrial cancer as targeted therapies are not yet available in the clinic. However, mutational profiles are applied for selection of targeted therapeutics for several other cancers and also applied for clinical trials stratification. Our previous screening of a smaller number of endometrial cancer patients identified somatic mutations in FGFR2, KRAS, PIK3CA, PTEN, PT53 and CTNNB1 . However, this study did not rule out possible mutations in other known oncogenes that could be potentially interesting for targeted treatment of endometrial cancer. Thus, the current study was undertaken to screen for a large panel of known oncogenic mutations in a series of primary and metastatic lesions from endometrial cancer patients using the high-throughput method OncoMap , . OncoMap provides a unique opportunity to simultaneously interrogate a large number of known mutations in a large number of genes, thus providing the opportunity to characterize the molecular subgroups of endometrial cancer with a potential relevance for targeting novel therapeutics.
All parts of the study have been approved according to Norwegian legislation as well as international demands for ethical review. The study was approved by the Norwegian Data Inspectorate, Norwegian Social Sciences Data Services, and the Western Regional Committee for Medical and Health Research Ethics, REC West (NSD15501; REK 052.01). Patients were included in the study after written informed consent approved by the ethics committee (REK West).
We have studied a total of 69 patients for mutations in 28 known oncogenes (Table 1). 23 of the included patients had previously been screened for fewer oncogene mutations by another method . The patients were recruited from a population based patient series of 701 patients with endometrial cancer prospectively collected at Haukeland University Hospital, Norway. Age at diagnosis, FIGO stage, histological subtype and grade, treatment and follow-up was registered as previously reported . Distribution of clinico-pathologic variable for the 69 investigated cases did not differ significantly from the larger (n = 701) unselected patient cohort (Table 2). Tissue was available from 67 primary tumors and 15 metastatic lesions from 9 patients of which 7 had corresponding tissues from primary lesions available for comparison. The majority of selected lesions were verified by frozen sections to contain >80% malignant epithelial component with a minimum cut off for inclusion of 50% purity.
Endometrial cancer cell lines Hec1A, Hec1B, KLE, RL95-2, ECC1 were purchased from ATCC-LGC Standards, London, UK, MFE-280, MFE-296, MFE3-19, EFE-184, AN3-CA were from DSMZ, Germany and Ishikawa from Sigma-Aldrich, St.Louis, MO. All cells were maintained in medium as recommended by the supplier, supplemented with Penicillin/Streptomycin (Sigma-Aldrich, St.Louis, MO).
OncoMap and DNA sequencing
DNA from primary and metastatic lesions was extracted from fresh frozen biopsies. DNA was isolated by digestion over night at 65°C in lysis buffer containing proteinase K, followed by a standard ethanol precipitation. DNA from 11 endometrial cancer cell lines was extracted using Qiagen Tissue DNA kit according to manufacturers protocol. DNA quantity was measured using the Quant-iT™ Picogreen® Assay (Invitrogen) and high quality of the DNA assured on a 0.7% agarose gel before genomic DNA was amplified using the Repli-g Midi Kit (Qiagen, Germany) according to manufacturers' instructions. Amplified DNA was diluted 1∶10 in 1xTE buffer (pH 8.0) and after hydration for 24 h at room temperature further diluted to a working concentration of 5 ng/µl in water. Mutations were detected in genome-amplified DNA using a mass spectrometry-based single base extension technique (Sequenom, Inc.) as previously described . Primers for additional assays to detect mutations described in several cancer studies since 2008 , , , ,  were designed using the Sequenom Assay Design Software. Following amplification and mutation site specific probe elongation analytes were spotted on SpectroCHIPs I and masses detected using a Bruker matrix-assisted laser desorption/ionization–time of flight mass spectrometer (Sequenom). Spectra were manually reviewed using the Typer 4.0 Software (Sequenom). A list of the mutations included in OncoMap and the corresponding amino acid changes is given in Table S1.
To validate the proportion of the most frequently mutated oncogenes detected by OncoMap, genomic DNA was extracted from freshly frozen primary tumor tissue from 199 additional patients. In total 264 patients were screened for point mutations in KRAS (exon 2) and PIK3CA (exon 9 and 20) as described . Details regarding primers and conditions are available upon request. Sequencing reactions were analyzed on an ABI Prism 3100 genetic analyzer using the Sequencing Analysis software, version 3.7.
Oligonucleotide DNA microarray analyses
A microarray dataset corresponding to the 69 primary tumor samples included in the OncoMap screen was generated. RNA was extracted using the RNeasy Mini Kit (Qiagen, Hilden, Germany) and hybridized to Agilent Whole Human Genome Microarrays 44k (Cat.no. G4112F), according to the manufacturers instructions. Arrays were scanned using the Agilent Microarray Scanner Bundle and data were imported and analyzed in J-Express software (Molmine, Norway). Median spot signal was used as intensity measure. Expression data were normalized using quantile normalization. Microarraydata have been deposited in the ArrayExpress Archive database, http://www.ebi.ac.uk/arrayexpress/ (ArrayExpress accession: E-MTAB-1358).
A SAM (Significance Analysis of Microarray) analysis between grade I–II and grade III was performed to identify significantly differentially expressed genes according to histologic grade. 306 genes were significantly differentially expressed (FDR<0.01) between the two groups. Hierarchical clustering was performed on this list of genes using weighted average linkage and Pearson correlation as similarity measures. Clinico-pathological data and mutational status were mapped manually to the cluster-tree to visualize the distribution of mutation across the patient population.
The OncoMap screen for 387 oncogenic mutations in 28 commonly mutated genes in cancer (Table S1) was applied in 67 primary and 15 metastatic endometrial carcinoma lesions as well as 11 endometrial carcinoma cell lines and detected mutations in 7 of the investigated genes. We found that 27 patients (40.3%) had point mutations in one single gene, while 4 patients (6.0%) had mutations in 2 genes. Among the seven genes with detected somatic mutations in primary and metastatic lesions, KRAS (17.9%), PIK3CA (14.6%) and FGFR2 (10.4%) were the most frequently mutated, while mutations in BRAF (1.5%), EGFR (1.5%), HRAS (1.5%) and NRAS (1.5%) were rare. The frequencies of KRAS and PIK3CA mutations were validated by DNA sequencing in 264 primary tumors (Table 3). FGFR2 mutation frequency had been validated previously . The most common single mutation found by OncoMap screening was FGFR2 aaS252W (9.0%), however the most frequently mutated gene was KRAS (17.4%) (Table 3). The OncoMap screen of the 11 established endometrial cancer cell lines identified as expected KRAS G12D and PIK3CA G1049R mutations in both Hec1A and Hec1B, while FGFR2 mutation S252W was found in MFE280 and MFE319. Additionally, two PIK3CA mutations were identified in MFE280 and MFE296 (E545K and P539R, respectively). We did not find any of the cell lines to have mutations in any of the other genes included in the OncoMap panel.
To explore a possible link between type of mutations and gene expression patterns in primary tumors, a hierarchical cluster analysis of 306 genes significantly differentially expressed (SAM analysis, FDR<0.01) according to histologic grade was performed. We found that there was no significant association between specific oncogene mutations and patient clusters based on transcriptional signatures (Figure 1). This finding appears to be in line with our previous report on a smaller data set applying an earlier generation of mRNA genearrays, with no enrichment for PIK3CA mutations in the patient cluster capturing aggressive phenotype .
A hierarchical clustering of 306 significantly differentially expressed genes between grade I–II and grade III was mapped with clinico-pathological data and mutational status to visualize the distribution of mutation across the patient population. Green square color indicate good prognosis groups (Grade I–II, FIGO I–II, endometrioid type, ERα positivity) and no detected mutation in indicated gene, Red square color indicate poor prognosis groups (Grade III, FIGO III–IV, non-endometrioid types, ERα negativity) and detected mutation in indicated gene. Black square: data missing.
To further investigate if mutation pattern changed during disease progression, 15 metastatic lesions from 9 patients from which seven had primary tumors available for comparison, were analyzed for mutations. KRAS, PIK3CA and FGFR2 were found to be the most frequently mutated genes also in metastatic lesions, with no significant increase in mutation frequency (Table 4). In two cases, mutations were detected in the metastatic lesions but not in the primary lesion, while one case with mutation in the primary lesion had no detectable mutation in the metastatic lesion. The small sample set available for this analysis, tumor heterogeneity and differences in stromal contamination should call for caution in the conclusions.
Activating mutations in specific proto-oncogenes may confer oncogene-addiction. Such mutations have been identified in several genes and may drive malignant disease progression. This principal for oncogene-addiction can be exploited to develop new targeted therapies . Currently, mutational profiles are applied for selection of targeted therapeutics for e.g. BRAF inhibitors in malignant melanoma  and BRAF and EGFR targeting in lung- and colorectal cancers , . For endometrial cancer, none of the novel targeted therapeutics is available in the clinic at present. However, several ongoing clinical trials aim at exploiting targets supported by recent comprehensive molecular profiling of primary endometrial carcinoma lesion , dominated by trials targeting the phosphatidylinositol 3-kinase (PI3K)/AKT/mammalian target of rapamycin (mTOR) or FGFR2. However, to our knowledge, no previous study has reported as comprehensive mutational data for a large panel of oncogenes in endometrial cancers including metastatic lesions.
A large number of oncogene mutations has been identified to be important in cancer development. Recently, several papers have reported the usefulness of the high-throughput genotyping platform OncoMap to screen for mutations in a large panel of known cancer oncogenes , , , . The high degree of concordance between our findings using OncoMap for the investigated genes and the validated frequency in the present study as well as previously published mutation frequencies in endometrial cancer samples based on traditional sequencing, is assuring. Using OncoMap we found that 40.3% of the analyzed endometrial cancer samples harbored at least one mutation. Of the 28 oncogenes included, mutations were only found at high frequency (>10%) in KRAS, PIK3CA and FGFR2. These genes have been linked to endometrial cancer previously, both by us ,  and others , .
In the present study, the S252W mutation in FGFR2 was identified as the most frequent single mutation (9%) in endometrial cancer. The somatic FGFR2 mutations include the S252W and P253R alleles, where autosomal dominant mutations are associated with the congenital developmental disorder Apert syndrome . We, and others , have previously linked these mutations to endometrial cancer, through increased tumor cell survival and anchoring independent growth in endometrial cancer cell lines, and indicated the potential for FGFR2 inhibitors in mutated cell lines . It has also been reported that FGFR2 inhibitors induce cell death in endometrial cancer cells despite PTEN inactivating mutations . The frequency of FGFR2 mutations detected in the present OncoMap screen of 10.4% is in concordance with our previous findings from 122 endometrial cancer patients from the same region, finding FGFR2 to be mutated in 12.3% . Recently, a frequency in this range of 10.3% was also published by others .
Several of the PIK3CA mutations were detected at relatively low frequencies (<3%), however the total frequency of any detected PIK3CA mutations was 13.4%. We have validated this frequency of point mutations in PIK3CA (exon 9 and 20) in 14.6% in a cohort of 264 endometrial cancer patients. This is consistent with the reported mutational frequency of PIK3CA in endometrial carcinoma in the COSMIC database for PIK3CA mutations tested for in OncoMap . A potential relevance for targeting therapy in patients harboring PIK3CA mutations was recently supported in a study demonstrating higher response rate to PI3K/AKT/mTOR inhibitors for patients with mutated compared to wild type PIK3CA in breast and gynecologic malignancies .
KRAS mutations were found in 17.9% of the cases, with high frequency of point mutations in exon 2 (G12A, G12C, G12D and G13D), validated in 264 endometrial cancer patients (14,7%; ) and also in line with previous studies (18%; ). KRAS mutations have been associated with low grade, and endometrioid histologic subtype, although not with prognosis , . Interestingly, KRAS and FGFR2 mutations were found to be mutually exclusive, in line with a previous report . In terms of therapy KRAS mutational status has been linked to EGFR inhibitor resistance in colorectal cancer , but further studies are needed in endometrial carcinoma to explore such potential link.
In line with the present study, we previously reported a low frequency (2%) of mutations in BRAF in endometrial cancer . Interestingly, with the exception of a few mutations in NRAS, HRAS, EGFR and BRAF (1.5%), no other hot-spot mutations were identified in the remaining 21 oncogenes screened for, neither in primary tumors nor in metastatic lesions.
The present work used a version of OncoMap covering 387 mutations in a total of 28 different oncogenes. In endometrial carcinomas, the oncogenes CTNNB1 and tumor suppressor genes PTEN and P53 have also been reported to be frequently mutated ,  but were not included in the present sceen and can therefore not be accounted for. Among the included genes and mutations, we have identified and validated KRAS, PIK3CA and FGFR2 to be the most frequently mutated oncogenes in endometrial cancer. Although transcriptional signature pattern according to histologic grade did not identify any distinct subgroups linking any of the mutations to phenotype, PIK3CA, KRAS and FGFR2 mutations may still be of relevance for targeting novel therapeutics in endometrial cancer. Nevertheless, more knowledge regarding functional aspects of the different mutations and their implications for response to drugs will be important to guide further selection of patients for molecularly based clinical trials.
The authors thank Britt Edvardsen, Ingjerd Bergo, Gerd Lillian Hallseth, Bendik Nordanger, Beth Johannessen and Hua My Hoang for technical assistance. Roman Thomas is thanked for providing lab support and facilities to run the Oncomap screen.
Conceived and designed the experiments: CK EB HBS. Performed the experiments: CK EB DS SM EAH MKH AMØ. Analyzed the data: CK EB SM EAH KK EW HBS. Contributed reagents/materials/analysis tools: KP MKH KHK HMJW JT. Wrote the paper: CK EB DS HBS.
- 1. Salvesen HB, Haldorsen IS, Trovik J (2012) Markers for individualised therapy in endometrial carcinoma. Lancet Oncol 13: e353–361.
- 2. Bokhman JV (1983) Two pathogenetic types of endometrial carcinoma. Gynecol Oncol 15: 10–17.
- 3. Fujimoto T, Nanjyo H, Fukuda J, Nakamura A, Mizunuma H, et al. (2009) Endometrioid uterine cancer: histopathological risk factors of local and distant recurrence. Gynecol Oncol 112: 342–347.
- 4. Dedes KJ, Wetterskog D, Ashworth A, Kaye SB, Reis-Filho JS (2011) Emerging therapeutic targets in endometrial cancer. Nat Rev Clin Oncol 8: 261–271.
- 5. Dutt A, Salvesen HB, Chen TH, Ramos AH, Onofrio RC, et al. (2008) Drug-sensitive FGFR2 mutations in endometrial carcinoma. Proc Natl Acad Sci U S A 105: 8713–8717.
- 6. MacConaill LE, Campbell CD, Kehoe SM, Bass AJ, Hatton C, et al. (2009) Profiling critical cancer gene mutations in clinical tumor samples. PLoS One 4: e7887.
- 7. Thomas RK, Baker AC, Debiasi RM, Winckler W, Laframboise T, et al. (2007) High-throughput oncogene mutation profiling in human cancer. Nat Genet 39: 347–351.
- 8. Trovik J, Wik E, Stefansson IM, Marcickiewicz J, Tingulstad S, et al. (2011) Stathmin overexpression identifies high-risk patients and lymph node metastasis in endometrial cancer. Clin Cancer Res 17: 3368–3377.
- 9. Ding L, Getz G, Wheeler DA, Mardis ER, McLellan MD, et al. (2008) Somatic mutations affect key pathways in lung adenocarcinoma. Nature 455: 1069–1075.
- 10. Greenman C, Stephens P, Smith R, Dalgliesh GL, Hunter C, et al. (2007) Patterns of somatic mutation in human cancer genomes. Nature 446: 153–158.
- 11. Hammerman PS, Sos ML, Ramos AH, Xu C, Dutt A, et al. (2011) Mutations in the DDR2 kinase gene identify a novel therapeutic target in squamous cell lung cancer. Cancer Discov 1: 78–89.
- 12. Weir BA, Woo MS, Getz G, Perner S, Ding L, et al. (2007) Characterizing the cancer genome in lung adenocarcinoma. Nature 450: 893–898.
- 13. Wood LD, Parsons DW, Jones S, Lin J, Sjoblom T, et al. (2007) The genomic landscapes of human breast and colorectal cancers. Science 318: 1108–1113.
- 14. Birkeland E, Wik E, Mjos S, Hoivik EA, Trovik J, et al. (2012) KRAS gene amplification and overexpression but not mutation associates with aggressive and metastatic endometrial cancer. Br J Cancer
- 15. Salvesen HB, Carter SL, Mannelqvist M, Dutt A, Getz G, et al. (2009) Integrated genomic profiling of endometrial carcinoma associates aggressive tumors with indicators of PI3 kinase activation. Proc Natl Acad Sci U S A 106: 4834–4839.
- 16. Yap TA, Workman P (2012) Exploiting the cancer genome: strategies for the discovery and clinical development of targeted molecular therapeutics. Annu Rev Pharmacol Toxicol 52: 549–573.
- 17. Chapman PB, Hauschild A, Robert C, Haanen JB, Ascierto P, et al. (2011) Improved survival with vemurafenib in melanoma with BRAF V600E mutation. N Engl J Med 364: 2507–2516.
- 18. Corcoran RB, Ebi H, Turke AB, Coffee EM, Nishino M, et al. (2012) EGFR-mediated re-activation of MAPK signaling contributes to insensitivity of BRAF mutant colorectal cancers to RAF inhibition with vemurafenib. Cancer Discov 2: 227–235.
- 19. Prahallad A, Sun C, Huang S, Di Nicolantonio F, Salazar R, et al. (2012) Unresponsiveness of colon cancer to BRAF(V600E) inhibition through feedback activation of EGFR. Nature 483: 100–103.
- 20. Kiessling MK, Oberholzer PA, Mondal C, Karpova MB, Zipser MC, et al. (2011) High-throughput mutation profiling of CTCL samples reveals KRAS and NRAS mutations sensitizing tumors toward inhibition of the RAS/RAF/MEK signaling cascade. Blood 117: 2433–2440.
- 21. Lee J, van Hummelen P, Go C, Palescandolo E, Jang J, et al. (2012) High-throughput mutation profiling identifies frequent somatic mutations in advanced gastric adenocarcinoma. PLoS One 7: e38892.
- 22. Maeng CH, Lee J, van Hummelen P, Park SH, Palescandolo E, et al. (2012) High-Throughput Genotyping in Metastatic Esophageal Squamous Cell Carcinoma Identifies Phosphoinositide-3-Kinase and BRAF Mutations. PLoS One 7: e41655.
- 23. Byron SA, Gartside M, Powell MA, Wellens CL, Gao F, et al. (2012) FGFR2 point mutations in 466 endometrioid endometrial tumors: relationship with MSI, KRAS, PIK3CA, CTNNB1 mutations and clinicopathological features. PLoS One 7: e30801.
- 24. Peterson LM, Kipp BR, Halling KC, Kerr SE, Smith DI, et al. (2012) Molecular characterization of endometrial cancer: a correlative study assessing microsatellite instability, MLH1 hypermethylation, DNA mismatch repair protein expression, and PTEN, PIK3CA, KRAS, and BRAF mutation analysis. Int J Gynecol Pathol 31: 195–205.
- 25. Wilkie AO, Slaney SF, Oldridge M, Poole MD, Ashworth GJ, et al. (1995) Apert syndrome results from localized mutations of FGFR2 and is allelic with Crouzon syndrome. Nat Genet 9: 165–172.
- 26. Pollock PM, Gartside MG, Dejeza LC, Powell MA, Mallon MA, et al. (2007) Frequent activating FGFR2 mutations in endometrial carcinomas parallel germline mutations associated with craniosynostosis and skeletal dysplasia syndromes. Oncogene 26: 7158–7162.
- 27. Byron SA, Gartside MG, Wellens CL, Mallon MA, Keenan JB, et al. (2008) Inhibition of activated fibroblast growth factor receptor 2 in endometrial cancer cells induces cell death despite PTEN abrogation. Cancer Res 68: 6902–6907.
- 28. Bamford S, Dawson E, Forbes S, Clements J, Pettett R, et al. (2004) The COSMIC (Catalogue of Somatic Mutations in Cancer) database and website. Br J Cancer 91: 355–358.
- 29. Janku F, Wheler JJ, Westin SN, Moulder SL, Naing A, et al. (2012) PI3K/AKT/mTOR inhibitors in patients with breast and gynecologic malignancies harboring PIK3CA mutations. J Clin Oncol 30: 777–782.
- 30. Salvesen HB, Kumar R, Stefansson I, Angelini S, MacDonald N, et al. (2005) Low frequency of BRAF and CDKN2A mutations in endometrial cancer. Int J Cancer 115: 930–934.
- 31. Semczuk A, Berbec H, Kostuch M, Cybulski M, Wojcierowski J, et al. (1998) K-ras gene point mutations in human endometrial carcinomas: correlation with clinicopathological features and patients' outcome. J Cancer Res Clin Oncol 124: 695–700.
- 32. Mutter GL, Wada H, Faquin WC, Enomoto T (1999) K-ras mutations appear in the premalignant phase of both microsatellite stable and unstable endometrial carcinogenesis. Mol Pathol 52: 257–262.
- 33. Cepero V, Sierra JR, Corso S, Ghiso E, Casorzo L, et al. (2010) MET and KRAS gene amplification mediates acquired resistance to MET tyrosine kinase inhibitors. Cancer Res 70: 7580–7590.
- 34. Djordjevic B, Hennessy BT, Li J, Barkoh BA, Luthra R, et al. (2012) Clinical assessment of PTEN loss in endometrial carcinoma: immunohistochemistry outperforms gene sequencing. Mod Pathol 25: 699–708.
- 35. Wild PJ, Ikenberg K, Fuchs TJ, Rechsteiner M, Georgiev S, et al. (2012) p53 suppresses type II endometrial carcinomas in mice and governs endometrial tumour aggressiveness in humans. EMBO Mol Med 4: 808–824.