Issues in Personalized Medicine in Cancer: Intratumor Heterogeneity and Branched Evolution Revealed by Multiregion Sequencing

Intratumor Heterogeneity and Branched Evolution Revealed by Multiregion Sequencing[1]

Curator and Reporter: Stephen J. Williams, Ph.D.

UPDATED 3/04/2023

UPDATED 3/11/2023

Updates on detection of tumor heterogeneity and tumor evolution on specific tumor types are found at end of the introduction to this article

Genomic instability is considered a hallmark and necessary for generating the mutations which drive tumorigenesis. Multiple studies had suggested that there may be multiple driver mutations and a plethora of passenger mutations driving a single tumor.  This diversity of mutational spectrum is even noticed in cultured tumor cells (refer to earlier post Genome-Wide Detection of Single-Nucleotide and Copy-Number Variation of a Single Human Cell).  Certainly, intratumor heterogeneity has been a concern to clinicians in determining the proper personalized therapy for a given cancer patient, and has been debated if multiple biopsies of a tumor is required to acquire a more complete picture of a tumor’s mutations.  In the New England Journal of Medicine, lead author Dr. Marco Gerlinger in the laboratory of Dr. Charles Swanton of the Cancer Research UK London Research Institute, and colleagues reported the results of a study to determine if intratumoral differences exist in the mutational spectrum of primary and metastatic renal carcinomas, pre- and post-treatment with the mTOR (mammalian target of rapamycin) inhibitor, everolimus (Afinitor®)[1].

The authors compared exome sequencing of multiregion biopsies from four patients with metastatic renal-cell carcinoma who had been enrolled in the Personalized RNA Interference to Enhance the Delivery of Individualized Cytotoxic and Targeted Therapeutics clinical trial of everolimus (E-PREDICT) before and after cytoreductive surgery.

Biopsies taken:

  • Multiregion spatial biopsy of primary tumor (representing 9 regions of the tumor)
  • Chest-wall metastases
  • Perinephric metastases
  • Germline DNA as control

Multiple platforms were used to determine aberrations as follows:

  1. Illumina Genome Analyzer IIx and Hiseq: for sequencing and mutational analysis
  2. Illumina Omni 2.5: for SNP (single nucleotide polymorphism)-array-based allelic imbalance detection for chromosomal imbalance and ploidy analysis
  3. Affymetrix Gene 1.0 Array: for mRNA analysis

A phylogenetic reconstruction of all somatic mutations occurring in primary disease and associated metastases was  performed to determine the clonal evolution of the metastatic disease given the underlying heterogeneity of the tumor.  Basically the authors wanted to know if the mutational spectra of one metastasis could be found in biopsies taken from the underlying primary tumor or if the mutational landscape of metastases had drastically changed.


Multiregion exon-capture sequencing of DNA from pretreatment biopsy samples of the primary tumor, chest wall metastases, and perinephrous metastasis revealed 128 mutations classified as follows:

  • 40 ubiquitous mutations
  • 59 mutations shared by several but not all regions
  • 29 mutations unique to specific regions
  • 31 mutations shared by most primary tumor regions
  • 28 mutations shared by most metastatic regions

The authors mapped these mutations out with respect to their location, in order to determine how the metastatic lesions evolved from the primary tumor, given the massive heterogeneity in the primary tumor.  Construction of this “phylogenetic tree” (see Merlo et. al[2]) showed that the disease evolves in a branched not linear pattern, with one branch of clones evolving into a metastatic disease while another branch of clones and mutations evolve into the primary disease.

One of the major themes of the study is shown by results that an average of 70 somatic mutations were found in a single biopsy (a little more than just half of all tumor mutations) yet only 34% of the mutations in multiregion biopsies were detected in all tumor regions.

This indicated to the authors that “a single biopsy was not representative of the mutational landscape of the entire bulk tumor”. In addition, microarray studies concluded that gene-expression signatures from a single biopsy would not be able to predict outcome.

Everolimus therapy did not change the mutational landscape.  Interestingly, allelic composition and ploidy analyses revealed an extensive intratumor heterogeneity, with ploidy heterogeneity in two of four tumors and 26 of 30 tumor samples containing divergent allelic-imbalances.  This strengthens the notion that multiple clones with diverse genomic instability exist in various regions of the tumor.

 The intratumor heterogeneity reveals a convergent tumor evolution with associated heterogeneity in target function

Genes commonly mutated in clear cell carcinoma[3, 4] (and therefore considered the prevalent driver mutations for renal cancer) include:

Only VHL mutations were found in all regions of a given tumor, however there were three distinct SETD2 mutations (frameshift, splice site, missense) which were located in different regions of the tumor.

SETD2 trimethylates histones at various lysine residues, such as lysine residue 36 (H3K36).  The trimethylation of H3K36 is found on many actively transcribed genes.  Immunohistochemistry showed trimethylated H3K36 was reduced in cancer cells but positive in most stromal cells and in SETD2 wild-type clear-cell carcinomas.

Interestingly most regions of the primary tumor, except one, contained a kinase-domain activating mutation in mTOR.  Immunohistochemistry analysis of downstream target genes of mTOR revealed that mTOR activity was enhanced in regions containing this mutation.  Therefore the intratumoral heterogeneity corresponded to therapeutic activity, leading to the impression that a single biopsy may result in inappropriate targeted therapy.   Additional downstream biomarkers of activity confirmed both the intratumoral heterogeneity of mutational spectrum as well as an intratumoral heterogeneity of therapeutic-target function.

The authors conclude that “intratumor heterogeneity can lead to underestimation of the tumor genomics landscape from single tumor biopsies and may present major challenges to personalized-medicine and biomarker development”.

In an informal interview with Dr. Swanton, he had stressed the importance of performing these multi-region biopsies and the complications that intratumoral heterogeneity would present for personalized medicine, biomarker development, and chemotherapy resistance.

Q: Your data clearly demonstrates that multiple biopsies must be done to get a more complete picture of the tumor’s mutational landscape.  In your study, what percentage of the tumor would be represented by the biopsies you had performed?

Dr. Swanton: Realistically this is a very difficult question to answer, the more biopsies we sequence, the more we find, in the near term it may be very difficult to ever formally address this in large metastatic tumours

Q:  You have very nice data which suggest that genetic intratumor heterogeneity complicates the tumor biomarker field? do you feel then that quests for prognostic biomarkers may be impossible to attain?

Dr. Swanton: Not necessarily although heterogeneity is likely to complicate matters

Identifying clonally dominant lesions may provide better drug targets

Predicting resistance events may be difficult given the potential impact of tumour sampling bias and the concern that in some tumours a single biopsy may miss a relevant subclonal mutation that may result in resistance

Q:  Were you able to establish the degree of genomic instability among the various biopsies?

Dr. Swanton:  Yes, we did this by allelic imbalance analysis and found that the metastases were more genomically unstable than the primary region from which the metastasis derived

Q: I was actually amazed that there was a heterogeneity of mTOR mutations and SETD2 after everolimus therapy?   Is it possible these clones obtained a growth advantage?

Dr. Swanton: We think so yes, otherwise we wouldn’t identify recurrent mutations in these “driver genes”

Dr. Swanton will present his results at the 2013 AACR meeting in Washington D.C. (http://www.aacr.org/home/scientists/meetings–workshops/aacr-annual-meeting-2013.aspx)

The overall points of the article are as follows:

  • Multiple biopsies of primary tumor and metastases are required to determine the full mutational landscape of a patients tumor
  • The intratumor heterogeneity will have an impact on the personalized therapy strategy for the clinician


  • Metastases arising from primary tumor clones will have a greater genomic instability and mutational spectrum than the tumor from which it originates


  • Tumors and their metastases do NOT evolve in a linear path but have a branched evolution and would complicate biomarker development and the prognostic and resistance outlook for the patient

A great video of Dr. Swanton discussing his research can be viewed here


Everolimus: an inhibitor of mTOR

The following information was taken from the New Medicine Oncology Database (http://www.nmok.net)




Approved/Filed Indications

Novartis PharmaCurrent as of: August 30, 2012 Generic Name: Everolimus
Brand Name: Afinitor
Other Designation: RAD001, RAD001C
RAD001, an ester of the macrocytic immunosuppressive agent sirolimus (rapamycin), is an inhibitor of mammalian target of rapamycin (mTOR) kinase.Administration Route: intravenous (IV) • PO • solid organ transplant
• renal cell carcinoma (RCC), metastatic after failure of treatment with sunitinib, sorafenib, or sunitinib plus sorafenib
• renal cell carcinoma, advanced, refractory to treatment with vascular endothelial growth factor (VEGF)-targeted therapy
• treatment of progressive neuroendocrine tumors (NET) of pancreatic origin (PNET) in patients with inoperable, locally advanced or metastatic disease

Marker Designation
Gene Location

Marker Description


5’-AMP-activated Protein Kinase (AMPK)AMPK beta 1 (beta1 non-catalytic subunit) • HAMPKb (beta1 non-catalytic subunit) • MGC17785 (beta1 non-catalytic subunit) • AMPK2 (alpha1 catalytic subunit) • PRKAA (alpha1 catalytic subunit) • AMPK alpha 1 (alpha1 catalytic subunit) • AMPKa1 ( AMPK is a member of a metabolite-sensing protein kinase family found in all eukaryotes. It functions as a cellular fuel sensor and its activation strongly suppresses cell proliferation in non-malignant cells and cancer cells. AMPK regulates the cell cycle by upregulating the p53-p21 axis and modulating the TSC2-mTOR (mammalian target of rapamycin) pathway. The AMPK signaling network contains a number of tumor suppressor genes including LKB1, p53, TSC1 and TSC2, and modulates growth factor signaling involving proto-oncogenes including PI3K, Akt and ERK. AMPK activation is therefore therapeutic target for cancer (Motoshima H, etal, J Physiol, 1 Jul 2006; 574(Pt 1): 63–71).AMPK is a protein serine/threonine kinase consisting of a heterotrimeric complex of a catalytic alpha subunit and regulatory ß and gamma subunits. AMPK is activated by increased AMP/ATP ratio, under conditions such as glucose deprivation, hypoxia, ischemia and heat shock. It is also activated by several hormones and cytokines. AMPK inhibits ATP-consuming cellular events, protein synthesis, de novo fatty acid synthesis, and generation of mevalonate and the downstream products in the cholesterol synthesis pathway (Motoshima H, etal, J Physiol, 1 Jul 2006; 574(Pt 1): 63–71). – ovarian cancer
– brain cancer
– liver cancer
– leukemia
– colon cancer
CREB regulated transcription coactivator 2 (CRTC2)TOR complex 2 (TORC2, mTORC2) • RP11-422P24.6 • transducer of regulated cAMP response element-binding protein (CREB)2 • transducer of CREB protein 2 • TOR1Location: 1q21.3 The mammalian target of rapamycin (mTOR) exists in two complexes, TORC1 and TORC2, which are differentially sensitive to rapamycin. cAMP response element-binding protein (CREB) regulated transcription coactivator 2 (CRTC2) or TORC2 is a multimeric kinase composed of mTOR, mLST8, mSin1, and rictor. The complex is insensitive to acute rapamycin exposure and functions in controlling cell growth and actin cytoskeletal assembly.TORC2 controls gene silencing, telomere length maintenance, and survival under DNA-damaging conditions. It is primaily located in the cytoplasm but also shuttles into the nucleus (Schonbrun M, etal, Mol Cell Biol, Aug 2009;29(16):4584-94). – brain cancer
Hypoxia inducible factor 1 alpha (HIF1A)HIF1-alpha (HIF-1 alpha) • HIF-1A • PASD8 • MOP1 • bHLHe78Location: 14q21-q24 The alpha subunit of the hypoxia inducible factor 1 (HIF-1alpha) is a 826 amino acid antigen consisting of a basic helix-loop-helix (bHLH)-PAS domain at its N-terminus. HIF-1alpha is rapidly degraded by the proteasome under normal conditions, but is stabilized by hypoxia resulting in the transactivation of several proangiogenic genes. HIF-1alpha is responsible for inducing production of new blood vessels as needed when tumors outgrow existing blood supplies. HIF-1alpha serves as a transcriptional factor that regulates gene expression involved in response to hypoxia and promotes angiogenesis.HIF-1alpha is a proangiogenic transcription factor induced primarily by tumor hypoxia that is critically involved in tumor progression, metastasis and overall tumor survival. HIF-1alpha functions as a survival factor that is required for tumorigenesis in many types of malignancies, and is expressed in a majority of metastases and late-stage tumors. HIF-1alpha is overexpressed in brain, breast, colon, endometrial, head and neck, lung, ovarian, and pancreatic cancer, and is associated with increased microvessel density and/or VEGF expression – prostate cancer
– bladder cancer
– nasopharyngeal cancer
– head and neck cancer
– kidney cancer
– pancreatic cancer
– endometrial cancer
– breast cancer
Mammalian target of rapamycin (mTOR)FK506 binding protein 12-rapamycin associated protein 1 • RAFT1 • FK506 binding protein 12-rapamycin associated protein 2 • FRAP • FRAP1 • FRAP2 • RAPT1 • FKBP-rapamycin associated protein • FKBP12-rapamycin complex-associated protein 1 • rapamycin target protein • TOR • FLJ44809 • MTORC1 • MTORC2 • RPTOR • RAPTOR • KIAA1303 • mammalian target of rapamycin complex 1Location: 1p36.22 The mammalian target of rapamycin (mTOR) is a large serine/threonine protein (Mr 300,000) having heat repeats, and protein-protein interaction domains at its amino terminus, and a protein kinase domain at its carboxy terminus. mTOR is a member of the phosphoinositide 3-kinase (PI3K)-related kinase (PIKK) family and a central modulator of cell growth. It regulates cell growth, proliferation and survival by impacting on protein synthesis and transcription. mTOR is present in two multi-protein complexes, a rapamycin-sensitive complex, TOR complex 1 (TORC1), defined by the presence of Raptor and a rapamycin insensitive complex, TOR complex 2 (TORC2), with Rictor, Protor and Sin1. Rapamycin selectively inhibits mTORC1 by binding indirectly to the mTOR/Raptor complex via FKBP12, resulting in inhibition of p70S6kinase but not the mTORC2 substrate AKTSer473. Selective inhibition of p70S6K attenuates negative feedback loops to IRS1 and TORC2 resulting in an increase in pAKT which may limit the activity of rapamycin.In a hypoxic environment the increase in mass of solid tumors is dependent on the recruitment of mitogens and nutrients. As a function of nutrient levels, particularly essential amino acids, mTOR acts as a checkpoint for ribosome biogenesis and cell growth. Ribosome biogenesis has long been recognized in the clinics as a predictor of cancer progression; increase in size and number of nucleoli is known to be associated with the most aggressive tumors and a poor prognosis. In bacteria, ribosome biogenesis is independently regulated by amino acids and energy charge. The mTOR pathway is controlled by intracellular ATP levels, independent of amino acids, and mTOR itself is an ATP sensor (Kozma SC, etal, AACR02, Abs. 5628). – breast cancer
– pancreatic cancer
– multiple myeloma
– liver cancer
– brain cancer
– prostate cancer
– kidney cancer
– lymphoma
Signal transducer and activator of transcription 3 (STAT3)Stat-3 • acute-phase response factor (APRF) • FLJ20882 • HIESLocation: 17q21 Signal transducer and activator of transcription 3 (STAT3) is a member of the STAT protein family. STAT3, plays a critical role in hematopoiesis. STAT3 is located in the cytoplasm and translocated to the nucleus after tyrosine phosphorylation. In response to cytokines and growth and other activation factors, STAT family members are phosphorylated by the receptor associated kinases and then form homo- or heterodimers, which translocate to the cell nucleus where they act as transcription activators. – multiple myeloma
– hematologic malignancy
– lymphoma
Sonic hedgehog homolog (SHH)Shh • HHG1 • HHG-1 • holoprosencephaly 3 (HPE3) • HLP3 • SMMCILocation: 7q36 Sonic hedgehog, a secreted hedgehog ligand, is a human homolog of the Drosophila segment polarity gene hedgehog, cloned by investigators at Harvard University (Marigo V, etal, Genomics, 1 Jul 1995;28 (1):44-51).The mammalian sonic hedgehog (Shh) pathway controls proliferation of granule cell precursors in the cerebellum and is essential for normal embryonic development. Shh signaling is disrupted in a variety of malignancies. – pancreatic cancer
– CNS cancer


1.         Gerlinger M, Rowan AJ, Horswell S, Larkin J, Endesfelder D, Gronroos E, Martinez P, Matthews N, Stewart A, Tarpey P et al: Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. The New England journal of medicine 2012, 366(10):883-892.

2.         Merlo LM, Pepper JW, Reid BJ, Maley CC: Cancer as an evolutionary and ecological process. Nature reviews Cancer 2006, 6(12):924-935.

3.         Varela I, Tarpey P, Raine K, Huang D, Ong CK, Stephens P, Davies H, Jones D, Lin ML, Teague J et al: Exome sequencing identifies frequent mutation of the SWI/SNF complex gene PBRM1 in renal carcinoma. Nature 2011, 469(7331):539-542.

4.         Dalgliesh GL, Furge K, Greenman C, Chen L, Bignell G, Butler A, Davies H, Edkins S, Hardy C, Latimer C et al: Systematic sequencing of renal carcinoma reveals inactivation of histone modifying genes. Nature 2010, 463(7279):360-363.

Updated 3/04/2023 articles on specific tumor type intratumoral heterogeneity

Complex Patterns of Genomic Heterogeneity Identified in 42 Tumor Samples and ctDNA of a Pulmonary Atypical Carcinoid Patient

Source: Tamsin J. Robb, Peter Tsai, Sandra Fitzgerald, Paula Shields, Pascalene S. Houseman, Rachna Patel, Vicky Fan, Ben Curran, Rexson Tse, Jacklyn Ting, Nicole Kramer, Braden J. Woodhouse, Esther Coats, Polona Le Quesne Stabej, Jane Reeve, Kate Parker, Ben Lawrence, Cherie Blenkiron, Cristin G. Print; Complex Patterns of Genomic Heterogeneity Identified in 42 Tumor Samples and ctDNA of a Pulmonary Atypical Carcinoid PatientCancer Research Communications 3 January 2023; 3 (1): 31–42. https://doi.org/10.1158/2767-9764.CRC-22-0101

Tumor evolution underlies many challenges facing precision oncology, and improving our understanding has the potential to improve clinical care. This study represents a rare opportunity to study tumor heterogeneity and evolution in a patient with an understudied cancer type. A patient with pulmonary atypical carcinoid, a neuroendocrine tumor, metastatic to 90 sites, requested and consented to donate tissues for research. 42 tumor samples collected at rapid autopsy from 14 anatomically distinct sites were analyzed through DNA whole-exome sequencing and RNA sequencing, and five analyzed through linked-read sequencing. Targeted DNA sequencing was completed on two clinical tissue biopsies and one blood plasma sample. Chromosomal alterations and gene variants accumulated over time, and specific chromosomal alterations preceded the single predicted gene driver variant (ARID1A). At the time of autopsy, all sites shared the gain of one copy of Chr 5, loss of one copy of Chr 6 and 21, chromothripsis of one copy of Chr 11, and 39 small variants. Two tumor clones (carrying additional variants) were detected at metastatic sites, and occasionally in different regions of the same organ (e.g., within the pancreas). Circulating tumor DNA (ctDNA) sequencing detected shared tumor variants in the blood plasma and captured marked genomic heterogeneity, including all metastatic clones but few private tumor variants. This study describes genomic tumor evolution and dissemination of a pulmonary atypical carcinoid donated by a single generous patient. It highlights the critical role of chromosomal alterations in tumor initiation and explores the potential of ctDNA analysis to represent genomically heterogeneous disease. Significance: DNA sequencing data from tumor samples and blood plasma from a single patient highlighted the critical early role of chromosomal alterations in atypical carcinoid tumor development. Common tumor variants were readily detected in the blood plasma, unlike emerging tumor variants, which has implications for using ctDNA to capture cancer evolution.

In this article Robb et. al analyzes the tumor evolution of variants and mutations over time in a single patient.  Tumor samples were obtained from liquid biopsy as ctDNA, which is unlike the previous study in renal cancer, where multiple fine needle aspirates needed to be conducted to determine intratumoral heterogeneity with respect to spatial distribution within a solid tumor.  In this current study the authors were able to study the evolution of tumor heterogeneity in a single patient.

Tumor evolution underlies many of the most pressing challenges facing clinical precision oncology today, and a better understanding of this process has the potential to improve clinical care. We present a unique example of tumor evolution in an uncommon tumor type not previously featured in such studies, a pulmonary atypical carcinoid [an intermediate grade neuroendocrine tumor (NET)].

NETs arise from hormone-producing cells of the neuroendocrine system located throughout the body, and are highly heterogeneous, both genetically and pathologically. While once considered rare tumors (1) and thought to be indolent in nature (2), we now recognize that NETs have an age-adjusted incidence of 6.2 cases per 100,000 in the country where this study was undertaken, New Zealand (3), similar to the incidence of ovarian and cervical cancers (4). Pulmonary atypical carcinoids are mitotically active well-differentiated NETs, with one in five presenting with distant metastatic disease at diagnosis (5). They feature relatively few DNA variants, with recurrent variants occurring in chromatin remodeling genes such as MEN1 and ARID1A [reported in 25% and 10% of studied atypical carcinoids, respectively (6)]. In general, NETs feature few small DNA variants and often instead have large-scale chromosomal changes; for example, around 25% of pancreatic NETs lose a suite of 10 chromosomes, and a further 40% feature the loss of chromosome (Chr) 11 (7). Few genomic studies have been completed on pulmonary atypical carcinoids; however, comparative genomic hybridization studies identified recurrent deletions in Chr 11q (harboring MEN1) in around 60% of cases (8–10).

Tumor development is an evolutionary process with similarities to Darwinian natural selection, where tumor cells may be under multiple simultaneous “selective pressures” (11), including immune attack (12) and drug treatment (13). There are multiple debated models for the generation, propagation, and selection of variants throughout tumor development, including linear, branching, punctuated, and neutral evolution (11, 14), each of which may occur in different tumors or at different timepoints within the same patient. Tumor evolution has been associated with key clinical challenges, including cancer metastasis (15), immune evasion (12, 16), and drug resistance (13, 17, 18).

Sequencing the cell-free DNA (cfDNA) in a patient’s blood plasma to identify ctDNA variants derived from tumor cells has been postulated to better represent the total disease burden (including potential tumor heterogeneity) than single tumor biopsies or resections (19, 20). However, we do not yet fully understand the detectability of ctDNA shed from different anatomic sites around a patient’s body—and early results suggest that some tumor sites may be more easily detected than others (21).

In this study, we completed multimodal genomic analysis on samples collected at autopsy from 41 metastatic tumor sites from a single patient with an uncommon pulmonary atypical carcinoid, alongside a blood plasma sample and clinical biopsies, to catalog driver genomic alterations and hypothesize their order of accumulation. We compared the genomic variants identified in the tumors with those detected in the patient’s blood plasma to consider how well the ctDNA analysis represented the genomic tumor heterogeneity, and in turn, infer metastatic sites that clinical ctDNA assays may poorly detect.  A potential limitation, though, as stated by authors

“studies that analyze many tumors from a single patient, while revealing detailed and valuable insights about that patient’s cancer, cannot necessarily be generalized to other patients. Furthermore, all samples except for two clinical biopsies were collected at a single timepoint (at autopsy), limiting the extent of computational tumor evolution inferences that could be made, and the small size of these biopsies limited the scope of genomic assays that could be applied

In their conclusions they allude to the branched evolution that was seen with the renal cancer study:

Models of Tumor Evolution and Metastatic Dissemination

We can hypothesize a putative genomic evolution process (Fig. 7) that is consistent with the genomic heterogeneity observed at autopsy and in the two clinical biopsies. The early lung tumor represented by the diagnostic biopsy featured gain of Chr 5, loss of Chr 21 as well as nine small variants, none of which are known or predicted cancer drivers. Genomic features of the tumors collected at autopsy are consistent with further mutation after the lung biopsy was undertaken, followed by an evolutionary event such as a selective sweep, that fixed into the genomes of subsequent tumors a loss of function variant in ARID1A as well as 29 small non-driver variants and additional chromosomal alterations including Chr 11 chromothripsis and Chr 6 loss. There are plausible selective advantages to the tumor of each, including Chr 6 LOH of the HLA genes responsible for presenting peptides to T cells, hypothetically reducing the immune system’s ability to recognize neoantigens presented by the tumor cells (47). The data are consistent with subsequent branching events, including the accumulation of a second somatic “hit” to EPS8L2 (where the first was somatic LOH on Chr 11), before two parallel lineages developed, metastatic group 1 and 2 (Fig. 7). In metastatic group 1, some variants of unknown significance deserve further study. For instance, truncation of tumor suppressor ETV6 [a common fusion gene partner in breast and thyroid cancer (48)], following the LOH of a region of Chr 12 may be functionally significant given this gene’s role in development (49). The missense SLIT1 variant was the only additional protein-affecting variant carried by tumors in metastatic group 2 and may play a role in angiogenesis and migration (50). Interestingly, the two metastatic groups in the pancreas appear to have arisen by two independent seeding events. However, in the literature there is not overwhelming evidence for consistent genetic drivers of metastatic progression (51, 52) and we have no evidence in our data to suggest the additional genomic variants accumulated by each of the two metastatic groups (e.g., in EPS8L2ETV6, and SLIT1) were necessary to drive their metastasis. Given a general propensity of cancers to metastasize to lung (53), it is possible that the presence of multiple metastatic groups in the lung may be the result of tertiary metastatic seeding events returning tumor cells that have evolved elsewhere to the site of the lung primary tumor.


UPDATED 3/11/2023  Articles on evolution of tumor heterogeneity in ovarian cancer

Evaluating Tumor Evolution via Genomic Profiling of Individual Tumor Spheroids in a Malignant Ascites

An Author Correction to this article was published on 21 November 2018

This article has been updated


Epithelial ovarian cancer (EOC) is a silent but mostly lethal gynecologic malignancy. Most patients present with malignant ascites and peritoneal seeding at diagnosis. In the present study, we used a laser-aided isolation technique to investigate the clonal relationship between the primary tumor and tumor spheroids found in the malignant ascites of an EOC patient. Somatic alteration profiles of ovarian cancer-related genes were determined for eight spatially separated samples from primary ovarian tumor tissues and ten tumor spheroids from the malignant ascites using next-generation sequencing. We observed high levels of intra-tumor heterogeneity (ITH) in copy number alterations (CNAs) and single-nucleotide variants (SNVs) in the primary tumor and the tumor spheroids. As a result, we discovered that tumor cells in the primary tissues and the ascites were genetically different lineages. We categorized the CNAs and SNVs into clonal and subclonal alterations according to their distribution among the samples. Also, we identified focal amplifications and deletions in the analyzed samples. For SNVs, a total of 171 somatic mutations were observed, among which 66 were clonal mutations present in both the primary tumor and the ascites, and 61 and 44 of the SNVs were subclonal mutations present in only the primary tumor or the ascites, respectively. Based on the somatic alteration profiles, we constructed phylogenetic trees and inferred the evolutionary history of tumor cells in the patient. The phylogenetic trees constructed using the CNAs and SNVs showed that two branches of the tumor cells diverged early from an ancestral tumor clone during an early metastasis step in the peritoneal cavity. Our data support the monophyletic spread of tumor spheroids in malignant ascites.


Epithelial ovarian cancer (EOC) is a silent but mostly lethal gynecologic malignancy. The most common histological EOC subtype is high-grade serous carcinoma, and the current treatment strategy involves a primary debulking surgery followed by chemotherapy to reduce the tumor burden1,2. Recent advances in genomics have revealed the presence of extensive intra-tumor heterogeneity (ITH) in many cancers, including ovarian cancer3,4,5. The presence of extensive clonal diversity increases the capacity of a given tumor to survive upon an expected strike in the microenvironment and thus is thought to be responsible for a reduced response to current chemotherapy and to contribute to chemoresistance development6,7,8.

Unlike other solid tumors, the primary route of metastasis in EOC patients is the transcoelomic metastasis route, which is a passive process and involves dissemination of tumor cells from the primary tumor tissue into the peritoneal cavity9. Thus, early disseminating clones may exist in the malignant ascites tumor microenvironment (TME) and may form an independent subclonal lineage and contribute to ITH. Both protumorigenic and antitumorigenic factors are known to be enriched in the malignant ascites TME10. However, genetic differences between tumor cells in the primary tissue and tumor cells surviving in the ascites TME are not yet fully understood. Multi-region sequencing of both the primary tumor and associated metastases in ovarian cancer has provided insights into spatial heterogeneity and has shown that metastatic tumors maintain the genetic alterations found in the primary tumor and arise with little accumulation of genetic alteration5. However, the extent of the genetic heterogeneity within and between the primary tumor and tumor cells found in ascites remains underestimated.

Here, to uncover the genetic heterogeneity of tumor cells in malignant ascites, we introduced a genetic profiling method for individual tumor spheroids which are the common form of tumor cells floating in malignant ascites. Inspired by single-cell analysis, we hypothesized that genetic profiling of individual tumor spheroids might uncover the heterogeneity within and between the primary tumor and tumor cells in ascites. We isolated individual tumor spheroids through a laser-aided isolation technique. Then, we performed low-depth whole-genome sequencing (WGS) and high-depth whole-exome sequencing (WES) for ten tumor spheroids and eight primary tumor samples from a high-grade serous (HGS) EOC patient. We explored somatic copy number alterations (CNAs) and single-nucleotide variants (SNVs) to determine the tumor evolution and ITH between the primary tissues and the tumor spheroids from the malignant ascites. This study reports the feasibility of analyzing tumor cells in malignant ascites to detect early disseminating EOC clones.


Preparation and isolation of single tumor spheroids from the ascites of an ovarian cancer patient

A malignant ascites was collected during a primary debulking surgery. The tumor spheroids in the malignant ascites were purified, fixed and prepared on a discharging layer-coated glass slide (Fig. 1A). Single tumor spheroids on the slide were isolated by an infrared (IR) laser pulse as described in our previous publication11. Briefly, the discharging layer consisted of indium tin oxide (ITO), which vaporizes when irradiated by an IR laser pulse. The ITO vaporization generates pressure, by which cells in the irradiated area are discharged from the slide. From the prepared sample on the slide, we isolated ten individual tumor spheroids, which were tens of micrometers in diameter and contained hundreds of cells (Fig. 1B,C). Isolating and capturing each tumor spheroid took less than 1 second on average, which means this technique is feasible for analyzing a large number of samples and could be implemented in a routine procedure. The isolated single tumor spheroids were collected in PCR tubes for further reactions.

Figure 1
figure 1

An overview of individual tumor spheroid isolation from malignant ascites. (A) A malignant ascites was collected during a primary debulking surgery. Tumor spheroids in the malignant ascites were purified, fixed and prepared on a discharging layer (Indium Tin Oxide (ITO), 100 nm in thickness)-coated glass slide. (B) The laser isolation technique was used to isolate individual tumor spheroids. This technique utilizes an IR pulsed laser, which vaporizes the discharging layer on the glass slide. Using this technique, ten individual tumor spheroids were isolated from the slide. The isolated cells underwent WGA and sequencing. (C) The images before and after isolation demonstrate that the targeted tumor spheroids in the malignant ascites were specifically isolated without disturbing the neighboring cells. The scale bars represent 100 μm.

Whole-genome amplification of the isolated individual tumor spheroids

The isolated single tumor spheroids were lysed by proteinase K. Then, the samples underwent multiple displacement amplification (MDA, Fig. 2A). The amplification was monitored via real-time PCR. The results showed that all the isolated samples yielded successful amplification (10/10). Additionally, comparing the amplification plots between the tumor spheroids and controls showed that there was no or a negligible amount of carry-over contamination (Fig. 2B). Every reaction yielded over 2 µg of amplified DNA, which was enough to conduct WGS and WES.

Figure 2
figure 2

WGA of the isolated tumor spheroids and several quality metrics of the amplified products. (A) MDA was performed to amplify the DNA in each tumor spheroid. MDA amplified tumor spheroid DNA 103– to 104-fold. (B) The amplification process was monitored by observing the fluorescence signal in each reaction. A non-template control was included in the reaction to testify carry-over contamination. The results showed that there was no or a negligible amount of carry-over contamination. (C,D) The distributions of the normalized read depth and VAF reflect the quality of the WGA products. Compared with the distributions of the amplified products from single cells, the distributions of the tumor spheroids were similar to those of the primary tissues. This indicated that the amplified products from the tumor spheroids had a negligible amount of WGA artifacts.

Next, we calculated and plotted the distributions of the normalized read depth (Fig. 2C) and variant allele frequency (VAF, Fig. 2D) based on the sequencing data to evaluate the amplification uniformity of the MDA reaction. In the Fig. 2C,D, the distributions of the MDA products from single cells were used for comparison. Normalized read depth indicates the uniformity of the number of sequencing reads throughout the whole-genome. The DNA from bulk tumor samples showed normal-like distributions with small variance, but whole-genome amplified DNA from single cells presented a skewed distribution because of non-uniform amplification. In contrast, the distributions of the tumor spheroids were similar to the distributions of the tumor bulk samples, rather than the whole-genome amplified products from the single cells. This result suggests that the effect of non-uniform amplification during MDA was minimized because hundreds of cells were included in the individual tumor spheroids. Similarly, the VAF distributions of the tumor spheroids were similar to those of the bulk tumor samples but not to the distributions of the single cells. This result supports the presumption that the MDA products of the tumor spheroids present a balanced allele amplification without losing one of the two alleles.

Low-depth WGS reveals the somatic CNAs and genetic subclones

First, we assessed the somatic CNAs of the primary ovarian cancer tissues and the tumor spheroids from the ascites (Supplementary Table S1). We carried out low-depth WGS using the Illumina platform to produce 8.53 ± 0.879 (×106) sequenced reads for each sample. As a result, we generated CNA profiles based on which we performed a hierarchical clustering analysis (Fig. 3A). The clustering yielded three distinct genetic subgroups. The primary ovarian cancer tissues (RO 1–7 and LO, named “Primary clone” and colored red) were clustered together. In contrast, the tumor spheroids from the ascites were divided into two clusters, one of which showed a primary-like CNA profile (AC 1–3 and 7–8, named “Ascites clone 1” and colored yellow), but the other presented a normal-like profile (AC 4–6 and 9–10, named “Ascites clone 2”, colored green).

Figure 3
figure 3

CNA analysis based on the genetic subclones of the tumor cells identified via low-depth WGS. (A) A genome-wide CNA analysis was performed using the low-depth WGS data. Each row represents each sample, and the samples were reordered by the hierarchical clustering method. The clustering analysis generated three major clusters, which were named Primary clone (red), Ascites clone 1 (yellow), and Ascites clone 2 (green). The clear differentiation of the CNA profiles between the Primary clone and Ascites clones implied that the tumor spheroids in the Ascites clones were not derived from the tumor cells in the Primary clone but from another independent tumor lineage. (B) Representation of the CNA profiles in detail at several regions for RO1, AC1, and AC4. The three samples exhibited both shared and exclusive CNAs. For example, deletion of FAT1 (1st column) and amplification of MYC, CYC1, and PARP10 (2nd column) were shared in every sample. However, the amplification of KDM5A (3rd column) and NOTCH3 (4th column) was exclusive to the Primary clone. This might indicate that the FAT1, MYC, CYC1, or PARP10 alterations conferred a growth advantage to the common ancestor of the Primary clone and Ascites clones. In contrast, the KDM5A or NOTCH3 amplifications might cause branching from the common ancestor and proliferation of the Primary clone.

Interestingly, the CNA profiles showed that deletion of FAT1 and amplification of MYC, PARP10, and CYC1 were shared by most of the samples (Fig. 3B). These genes are reported to be recurrently deleted (FAT1) or amplified (MYC, PARP10, and CYC1) in pan-cancer data12. These facts suggest that the shared CNAs might be the driving alterations at the first stage of cancer initiation. However, the primary clone had exclusive focal amplifications of KDM5A and NOTCH3 (Fig. 3B), which are known as recurrently amplified genes in ovarian cancer12,13. These focal amplifications of KDM5A and NOTCH3 might allow the primary clone to overwhelm the other subclones and finally dominate the left and right ovaries. However, we did not find a critical focal amplification or a deep deletion exclusive to Ascites clone 1. This implied that other types of alterations might drive Ascites clone 1 to survive or propagate in the peritoneal fluid.

WES reveals somatic SNVs and genetic subclones

To identify the somatic SNVs, the samples underwent WES. For each sample, the sequencing run generated 134 ± 21.4 depth of data, covering the whole exome of the human genome. As a result, 171 somatic SNVs were identified by variant calling from all the samples (Supplementary Table S2). The results shown in Fig. 4A,B revealed that 38.6% of the SNVs were common to the primary tumor and tumor spheroids from the ascites, and 35.7% of the SNVs exclusively belonged to primary-only and 25.7% to ascites-only mutations. The exclusive mutations in the Ascites clone suggest that this clone evolved by accumulating mutations independent from the Primary clone. Interestingly, the Ascites clone had a nonsynonymous mutation in the KRAS gene (p.G12D). The single nucleotide substitution results in an activating KRAS mutation that is a well-known oncogenic mutation associated with the anchorage-independent growth of tumor cells through the acquisition of anoikis resistance in various malignancies14,15. Therefore, the mutation in KRAS in the Ascites clone might provide an additional fitness gain for anchorage-independent survival in the ascites TME. However, both the Primary and Ascites clones shared somatic SNVs in TP53 and ARID1A, which are well-known driver mutations in ovarian cancer16,17. At the initial stage of tumorigenesis, these mutated genes might be tumor-initiating SNVs in conjunction with the CNAs of FAT1, MYC, PARP10, and CYC1. In addition to these somatic variants, the patient had germline variants in BRCA1 (NM_007294.3:c.1511dupG) and TP53 (NM_001126118:c.C98G), which are well-known susceptibility genes of ovarian cancer and are likely to predispose individuals to ovarian cancer and promote carcinogenesis (Supplementary Table S3)18,19.

Figure 4
figure 4

SNV analysis based on the WES data. The WES data from the primary tissue samples and tumor spheroids were used to analyze the SNVs. The results showed that a significant portion of the SNVs was shared in the Primary clone and Ascites clone 1. At the same time, the Primary clone and Ascites clone 1 had unique mutations. This result suggests that the two clones might have branched from a common ancestor. Ascites clone 2 was excluded from the analysis because the tumor spheroids in Ascites clone 2 were presumed to contain a large number of normal cells in each tumor spheroid. The full list of variants is listed in Supplementary Table S5.

Cellular composition of the tumor spheroids

Regarding the CNAs, Ascites clone 2 had no alteration except for amplification of the 8q24 region. Concerning the SNVs, Ascites clone 2 had fewer mutations than the other clusters. Based on these facts, we examined the possibility that normal cells exist in a tumor spheroid. We assumed that the VAF distribution of Ascites clones 1 and 2 would be similar if the two subclones had a similar proportion of normal cells. However, the VAF of Ascites clone 2 would be low if a single tumor spheroid from the clone included a high proportion of normal cells. We tested this idea by plotting the VAF distribution of each sample (Fig. 5). The results showed that most of the VAF distributions from the Primary clone and Ascites clone 1 were located at a higher range than those from Ascites clone 2. Therefore, we concluded that the small number of CNAs and SNVs in Ascites clone 2 was not due to their true characteristics but because the proportion of tumor cells in the tumor spheroid was small. Consequently, we excluded Ascites clone 2 from the following phylogenetic analysis.

Figure 5
figure 5

Analysis of the allele frequency to infer the cellular composition of each sample. The VAF distribution was plotted for each sample from the (A) Primary site and (B,C) Ascites. The mutations were categorized into common, primary-only or ascites-only mutations. Common mutations were somatic SNVs, which were detected in both the Primary clone and Ascites clones, and primary-only and ascites-only mutations, which were shared somatic SNVs detected only in the Primary clone and Ascites clones, respectively. The results showed that most of the VAF distributions from Ascites clone 2 were located at a much lower range than those from the Primary clone and Ascites clone 1. This suggests that the tumor spheroids in Ascites clone 2 had a large proportion of normal cells in each tumor spheroid.

In addition to the presence of normal cells in the samples, we examined the possibility of the presence of heterogeneous tumor cells in the samples. By comparing the allele frequency distributions of the common and primary-only mutations for each sample, we found that the allele frequencies of the common mutations were higher than those of the primary-only mutations for the primary tissues (7 of 8 samples, p < 0.01). This result implies that each of the primary tissues (except RO3) had two or more subclones sharing common mutations but not subclonal mutations. In contrast, the allele frequencies of the common mutations were similar to those of the ascites-only mutations for the tumor spheroids (8 of 10 samples). This result can be interpreted to indicate that, compared with the primary tissue samples, each tumor spheroid was comprised of genetically homogeneous tumor cells. Also, this hypothesis can be supported by analyzing the variant allele frequency according to the occurrence of the variants (Supplementary Fig. S2).

Constructing phylogenetic trees based on the somatic CNAs and SNVs

The phylogenetic trees were constructed from the CNA and SNV data. We achieved a CNA-based phylogeny analysis by identifying the common chromosomal breakpoints, calculating a trinary event matrix, and constructing a maximum parsimony tree20. The phylogenetic tree showed that an ancestral cancer clone accumulated CNAs and divided into two clones, which gained additional exclusive CNAs (Fig. 6A). Notably, these two genetic clones were composed of tumor spheroids from ascites and tumor tissues. Potentially, physically separated and biologically distinct TMEs might drive cancer cells into different alteration statuses.

Figure 6
figure 6

Constructing phylogenetic trees and inferred evolutionary history of the tumor. Phylogenetic trees were constructed using both the (A) CNA profiles and (B) SNV profiles. The two trees presented similar topologies and indicated that the Primary clone and Ascites clone 1 were derived from one ancestral clone at the early stage of cancer development. In addition, the phylogenic trees indicate early monoclonal and unidirectional seeding of tumor spheroids in Supplementary Fig. S3 malignant ascites and no additional clonal seeding from the primary site in this patient. (C) Based on the sequencing data from the primary tissue samples and the tumor spheroids from the ascites, the evolutionary trajectory was inferred. The tumor was initiated at the right ovary to generate the ancestral clone. With the further accumulation of mutations, the ancestral clone evolved into two subclones, the first of which was found in the right ovary and metastasized to the left ovary. The second subclone shed into the ascites TME and became extinct or dominated by the first subclone in the right ovary. Eventually, the Ascites subclone moved to the peritoneal cavity.

Maximum parsimony tree generation using the CNA data has a couple of limitations. First, this approach needs to set thresholds to define the amplified, neutral, and deleted status. The resultant tree is significantly affected by thresholds, and there is no golden rule to set the thresholds. Second, the proportion of normal cells in a sample has a substantial impact on a tree because the CNA status might be incorrectly assigned according to the normal cell portion. For example, the VAFs of RO6 (Fig. 5) show that the sample had a large number of normal cells. In this case, the copy number value of RO6 was close to the normal value (Fig. 3A), although the overall pattern was not similar to that of the normal sample. Thus, the thresholding led RO6 to be the same as the normal sample. For this reason, we excluded RO6 when constructing the maximum parsimony tree based on the CNA data.

Next, we constructed a phylogenetic tree from the SNV data. This approach does not use manual thresholding, and a phylogenetic tree is less affected by a normal cell portion. Therefore, we expected that, compared with the CNA-based approach, this approach would provide a more accurate result. The results showed that the cancer cells accumulated mutations as a single clone and divided into two independent clones (Fig. 6B). Moreover, with the full advantage of the SNV information, the phylogenetic tree presented the sequential creation of RO3, LO, and the rest of the Primary clones. Overall, the phylogenetic tree based on the SNV data rather than the CNA data presented a more stable and biologically explainable result.

Inferring the evolutionary trajectory of the primary ovarian cancer and the single tumor spheroids in the ascites

This patient harbored a bilateral ovarian tumor at the time of the primary debulking surgery. It is important to note whether these bilateral tumors arise independently or are the result of metastasis. The clonal evolution of the tumorigenesis theory provides two mechanisms of bilateral ovarian tumor development. If bilateral ovarian tumors arise from independent ancestral clones, they would have distinct genomic profiles without sharing somatic alterations. In contrast, bilateral tumors would have an identical set of somatic variants if they resulted from metastasis21. The somatic CNAs and SNVs of the left and right primary ovarian tumor in this study displayed comparable genomic profiles, strongly indicating a monoclonal origin of the bilateral tumor in this patient. This was further confirmed by calculating the clonality index (CI) based on previous reports21,22 revealing that the bilateral ovarian tumors were clonally related (CI1 = 1.0).

Finally, the history of the ovarian cancer development and progression was established based on the genomic profiles to understand the tumor evolution and its direction in this patient. As noted earlier, ovarian cancer metastasis occurs through a passive process, which initially involves physical shedding of tumor cells from the primary tumor into the peritoneal cavity, and the accumulation of ascites facilitates distant seeding of tumor cells along the peritoneal wall. Given a fixed chance of evolution, two scenarios are possible, either a monoclonal or polyclonal seeding process. If only certain clones from the primary tumor are fit to survive in the ascites TME, distinct clones, which may have diverged early, may be selected and progress over time in the primary and ascites TMEs, showing a tendency toward independent tumor evolution driven by different TMEs. In contrast, if tumor evolution is entirely driven by clonal dominance and the physical shedding of tumor cells from the primary tumor occurs by chance, then dominant clones expand in size and others may remain unchanged or become extinct over time at the primary tumor site. As the tumor grows, multiple clones may shed from the primary tumor into ascites. The ascites TME then acts as a reservoir of clonal lineage, and tumor cells in the ascites would represent the entire mutational landscape of a given tumor. For our case, we observed significant genetic differences in the CNAs and SNVs among the primary tissue samples and tumor spheroids. The dominant clones found in the right ovary were absent in the ascites TME, and we found 44 tumor spheroid–specific somatic SNVs (Supplementary Table S2). Furthermore, the comparable allele frequencies between the common mutations and tumor spheroid–specific mutations suggest that the tumor spheroids in the ascites TME are comprised of genetically homogeneous tumor cells compared with the primary tissues. Therefore, we conclude that the tumor spheroids were from a single subclonal lineage, supporting a mono- and early-seeding origin of the tumor spheroids in this patient. Based on these perspectives, we drew a potential evolutionary trajectory of the tumor from the patient (Fig. 6C). The tumor was initiated at the right ovary to generate the ancestral clone. With further accumulation of mutations, the ancestral clone evolved into two subclones, the first of which was found in the right ovary and metastasized to the left ovary. The second subclone shed into the ascites TME and became extinct or dominated by the first subclone in the right ovary. Eventually, the Ascites subclone moved to the peritoneal cavity. To validate that the Ascites subclone did not exist in the right ovary, four additional primary tissue samples were screened. We selected eight different loci where ascites clone specific mutations were located and analyzed genotypes of the loci for the four addition primary tissue samples (Supplementary Table S4). As a result, we found that the genotyping results also support our scenario. In addition, the summary of genome-wide somatic CNAs and SNVs indicated that the tumor cells in the primary tissue and the ascites possessed exclusive alterations as well as common ones (Supplementary Fig. S3). This result shows that the tumor cells in the primary tissue and the ascites were two subclonal lineages, which branched from one ancestral lineage.


In this study, we attempted to determine the presence of genetic heterogeneity within and between a primary tumor and the associated tumor spheroids in the ascites by performing multi-region sequencing of the primary tumor and genetic profiling of the individual tumor spheroids using the laser-aided cell isolation technique. We performed both WGS and WES of the primary tumor and tumor spheroid samples. First, we discovered high ITH levels in eight primary tissues and ten tumor spheroids. We also discovered that the CNA profiles in the primary and associated tumor spheroids were separated into two distinct genetic clusters, suggesting that the TME may be operative during tumor evolution. Second, we identified somatic SNVs using WES. We discovered a total of 171 somatic SNVs from all the samples, and 66 (38.6%) of these SNVs were ubiquitous mutations that were common to the primary tumor and tumor spheroids. The rest were either primary-only (61 SNVs, 35.7%) or ascites-only (44 SNVs, 25.7%) mutations, highlighting the notion that the tumor spheroids might have diverged early and accumulated additional mutations independently from the Primary clone. Bashashati et al., reported that genetically distinct clones are found in serous ovarian cancer patients and evolved from a single ancestral lone23. Supporting this idea, both phylogenic analyses, using the CNAs and SNVs, showed that the tumor spheroids might have diverged early from an ancestral tumor clone, evolved further with distinctive genomic profiles, and formed an independent subclonal lineage, thereby contributing to the ITH.

We also assessed the normal cell contamination in both the primary tumor and tumor spheroids using the VAF distribution in each sample. Indeed, both the Primary clone and Ascites clone 1 showed higher VAF distributions than Ascites clone 2, suggesting that the normal-like CNA and SNV profiles in Ascites clone 2 were due to a high proportion of normal cells. These findings are consistent with previous data from ovarian cancer patient-derived tumor spheroids and mouse models that suggested the presence of tumor-associated macrophages in the center of tumor spheroids24.

Although we only studied a single high-grade EOC patient, our data support previous studies demonstrating early divergence of the ascites sample from the primary tumor25. Further studies are needed to compare similarities and differences between the ascites spheroids and distant metastasis samples. Our data suggest that the mutation set of ascites spheroids does not represent the entire mutational landscape of a given EOC patient. This disagrees with recent findings by Choi et al. showing that ascites tumor cells represent the entire mutational landscape of a given tumor, and no additional genetic aberrations were detected26. In contrast, our data showed the presence of genetic heterogeneity within and between the primary tumor and the associated ascites spheroids. Moreover, the primary and associated ascites spheroids diverged early in tumor development, and not all the Primary clones disseminated into the ascites TME. However, our study is limited to a single ascites TME and provided no insight into distant metastatic sites.

Additionally, our data demonstrated that, compared with the primary tissue samples, each tumor spheroid was comprised of genetically homogeneous tumor cells (Fig. 5). Recent study by McPherson et al., indicated that there are two distinct patterns of intraperitoneal seeding in ovarian cancer, monoclonal unidirectional seeding from the ovary and polyclonal spread and reseed in27. Likewise, our data support monoclonal and unidirectional seeding of tumor spheroids in malignant ascites from the primary tumor in this patient. This can be interpreted in two ways. First, the tumor cells in an ascites may have low ITH. In this case, the spheroids of the tumor cells would be genetically homogeneous. Second, the tumor cells with a similar genetic profile may form individual tumor spheroids. In this case, the tumor cells in each tumor spheroid might have the same genetic profile, but two different tumor spheroids might be genetically different. For this case, isolating and analyzing the individual tumor spheroids from ascites might be widely utilized to discover the ITH of ovarian cancer.

Our data can partly be explained by the theory of Darwinian selection. For simplicity, tumor evolution is described as a series of expansions of clones, where each expansion series is driven by additional mutation acquisition, and clone fitness is tested by Darwinian selection. This selective sweep is context-dependent, and thus, genetic variants that are beneficial at a certain point may become extinct throughout the period of tumor progression. As a consequence, these clones may be absent in a fully grown tumor28. The selective pressures are further influenced by the dynamics of the TME, thereby increasing the complexity of tumor evolution29. The presence of extensive ITH in tumor spheroids and the early divergence of these subclones from the primary tumor suggests that we are currently underestimating the tumor genomic landscape.

In addition to the importance of genetic differences between tumor cells in primary tissue and those in ascites, knowledge regarding the genetic heterogeneity within the tumor cells in ascites would be valuable. Although not thoroughly studied, the genetic diversity of tumor cells in an ascites may have a large impact on tumor relapse and metastasis, given that transcoelomic spread is the primary route of metastasis in ovarian cancer. However, there has been no attempt to discover the genetic heterogeneity of individual tumor spheroids. In this study, we evaluated 10 individual tumor spheroids, five of which contained sufficient tumor cells for the analysis. Although we observed genetic heterogeneity of the individual ascites spheroids, a follow-up study should analyze at least a few tens of individual tumor spheroids per patient to find a clear signature of the genetic heterogeneity in an ascites.


In this study, we performed genome-wide sequence analysis of the primary tumor and the associated tumor spheroids in the malignant ascites of an EOC patient. We analyzed genetic heterogeneity in the primary tumor and tumor spheroids through multi-region sequencing and the laser-aided cell isolation technique11. From the sequencing data, we discovered clonal or subclonal somatic CNAs and SNVs, based on which we constructed phylogenetic trees and inferred the evolutionary history of tumor cells in the patient. As a result, we found that the tumor cells in the malignant ascites were an independent lineage from the primary tumor. The phylogenetic analysis showed that the lineage branched before the evolution of the cancer cells at the primary tissues, which suggests that analyzing malignant ascites might be used to detect ovarian cancer or metastasis in the early stage. In summary, the genetic plasticity and similarity between a primary tumor and associated tumor spheroids are still not clear, and yet, the nature of the similarity may have profound implications for both tumor progression and therapeutic outcomes in ovarian cancer. Therefore, future prospective studies profiling the genomic information of primary ovarian tumors, distant metastatic tumors, and tumor spheroids to determine the direction of tumor evolution and metastasis of ovarian cancer are warranted.


Patient information and sample preparation

A 42 yr old female patient diagnosed with primary high-grade serous ovarian cancer (Grade 3, stage IIIC) presented with malignant ascites and peritoneal seeding. Both primary tissues and malignant ascites were collected during primary debulking surgery. Fresh primary tissues and tumor cell clusters were mounted onto ITO-coated glass slides. Six samples were taken randomly from the solid portions of right ovary and only one from left ovary. Blood was collected to serve as the normal control. Ten tumor cell clusters were collected from the malignant ascites and fixed in 10% (v/v) formaldehyde. This study was approved by the Institutional Review Board (IRB) at Seoul National University Hospital (Registration number: 1305-546-487) and performed in compliance with the Helsinki Declaration. We obtained informed consent from the patient prior to primary debulking surgery to be used in research.

Laser-aided isolation of tumor spheroids and their whole-genome amplification

Previously, we developed and published a laser-aided cell isolation technique11 and designed two different pieces of software written in Python scripts and available at Github (https://github.com/BiNEL-SNU/PHLI-seq). Isolation of tumor spheroids was performed as described in the prior publication. In brief, an infrared laser was applied to the target area, vaporizing Indium Tin Oxide (ITO) layer and discharging the targeted tumor spheroid on the region. We used glass slides with a 100-nm-thick ITO layer.

The 8-strip PCR tube caps for the retrieval of tumor spheroids were pre-exposed under O2 plasma for 2 minutes. The tumor spheroids were lysed using proteinase K (cat no. P4850-1ML, Sigma Aldrich) according to the manufacturer’s directions after the PCR tubes were centrifuged. For whole-genome amplification, we used GE’s Illustra Genomiphi V2 DNA amplification kit (cat no. 25-6600-30). We added 0.2 µl of SYBR green I (Life Technologies) into the reaction solution for real-time monitoring of the amplification (Fig. 2B). All amplified products were purified using Beckman Coulter’s Agencourt AMPure XP kit (cat no. A63880) immediately following the amplification. To prevent carry-over contamination, the pipette tip, PCR tube, and cap for the reaction were stored in a clean bench equipped with UV light and treated with O2 plasma for 2 minutes before use. Additionally, we monitored the real-time amplification of non-template controls to ensure that no contaminants were transferred.

Sequencing library preparation, whole-genome, and whole-exome sequencing

The whole-genome amplified products or genomic DNA were fragmented using an EpiSonic Multi-Functional Bioprocessor 1100 (Epigentek) to generate DNA fragments with 250-bp on average. The fragmented products underwent Illumina library preparation using Celemics NGS Library Preparation Kit (LI1096, Celemics, Seoul, Korea) for the whole-genome sequencing library preparation, and SureSelectXT (Agilent, CA, US) for whole-exome sequencing. DNA purification was performed by TOPQXSEP MagBead (XB6050, Celemics, Seoul, Korea), and DNA libraries were amplified using the KAPA Library Amplification Kit (KAPA Biosystems, KK2602). Finally, the products were quantified by TapeStation 2200 (Agilent, CA, US). We used HiSeq 2500 150 PE (Illumina) to generate 1 Gb/sample for whole-genome sequencing and 5 Gb/sample for whole-exome sequencing, respectively.

Detecting copy number alterations

We used low-depth whole-genome sequencing data and the variable-size binning method30 to estimate the CNAs of the samples. Briefly, the whole genome was divided into 15,000 variable-sized bins (median genomic length of bin = 184 kbp), in which each bin had an equal expected number of uniquely mapped reads. Then, each sequence read was assigned to each bin followed by Lowess GC normalization to obtain the read depth of each bin. The copy number was estimated by normalizing the read depth of each bin by the median read depth of the reference DNA.

Detecting Single Nucleotide Variants

GATK (v3.5-0) IndelRealigner and BaseRecalibrator were used to locally realign reads around the Indel and recalibrate the base quality score of BAM files31. Then, GATK UnifedGenotyper, Varscan, and MuTect were used and combined the results to avoid false-positive variant calls32. First, GATK UnifiedGenotyper was used with default parameters followed by GATK VariantRecalibrator to obtain filtered variants31. Data of primary tissue samples and ascites tumor spheroid samples were processed together to produce a single vcf file. dbSNP build 137, HapMap 3.3, Omni 2.5, and 1000 G phase1 were used as the training data for variant recalibration. Also, annotation data including QD, MQ, FS, ReadPosRankSum, and MQRankSum were used for the training. Variants detected in the paired blood sample of the cancer patient were removed to produce the final list of GATK called variants. Varscan233 (ver 2.3.7) and Mutect34 (ver 1.1.4) were used with default parameters to produce the lists of Varscan and MuTect called variants, respectively. Here, paired blood read data was also used to remove germline variants.

Among the variants from the three callers, variants called by at least two callers were collected to obtain intra-sample double-called sites. We could reduce false-positive variant caused by NGS errors by considering only double-called variants for subsequent analysis32. Among the intra-sample double called sites, variants found in at least two samples were collected to remove WGA (whole genome amplification) errors, and the genomic loci with the resultant variants were considered confident sites. Finally, a variant in the confident sites was considered to be true if one of the three variant callers detected the variant at the locus and the allele count of the variant was significantly larger than that of the other non-reference bases (Fisher’s exact test, p < 10−3). The overall process is visually described in Supplementary Fig. S1.

Constructing phylogenetic trees based on the somatic CNAs and SNVs

As the first step to create the phylogenetic tree based on the somatic CNAs, the common chromosomal breakpoints were identified using low depth WGS data. For this, multipcf function which is included in copynumber library in R was used (Gamma = 50). Then, a trinary event matrix was constructed. The elements of the matrix are −1, 0, or 1, which are the numeric codes for loss, neutral, or gain. The rows and columns of the event matrix represent samples and segmented chromosomes by common breakpoints. A chromosomal region of a sample was considered as loss or gain when the expected ploidy was smaller than (mean ploidy −0.7) or larger than (mean ploidy +0.7), respectively. If an expected ploidy was between (mean ploidy −0.5) and (mean ploidy +0.5), the region was considered neutral. Otherwise, we considered the element for that region as a missing value. Finally, we constructed the phylogenetic tree using maximum parsimony in R. For phylogenetic tree generation, phangorn library was used. For the phylogenetic tree based on the SNVs, all procedure is same as that of CNAs, except the way of constructing the event matrix. For SNVs, we used 1 for a mutated locus and 0 for a not mutated locus.

Ethics approval and consent to participate

This study was approved by the Institutional Review Board (IRB) at Seoul National University Hospital (Registration number: 1305-546-487) and performed in compliance with the Helsinki Declaration. We obtained informed consent from the patient prior to primary debulking surgery to be used in research.

Availability of data and materials

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Tumor evolution and intratumor heterogeneity of an epithelial ovarian cancer investigated using next-generation sequencing



The extent to which metastatic tumors further evolve by accumulating additional mutations is unclear and has yet to be addressed extensively using next-generation sequencing of high-grade serous ovarian cancer.


Eleven spatially separated tumor samples from the primary tumor and associated metastatic sites and two normal samples were obtained from a Stage IIIC ovarian cancer patient during cytoreductive surgery prior to chemotherapy. Whole exome sequencing and copy number analysis were performed. Omental exomes were sequenced with a high depth of coverage to thoroughly explore the variants in metastatic lesions. Somatic mutations were further validated by ultra-deep targeted sequencing to sort out false positives and false negatives. Based on the somatic mutations and copy number variation profiles, a phylogenetic tree was generated to explore the evolutionary relationship among tumor samples.


Only 6% of the somatic mutations were present in every sample of a given case with TP53 as the only known mutant gene consistently present in all samples. Two non-spatial clusters of primary tumors (cluster P1 and P2), and a cluster of metastatic regions (cluster M) were identified. The patterns of mutations indicate that cluster P1 and P2 diverged in the early phase of tumorigenesis, and that metastatic cluster M originated from the common ancestral clone of cluster P1 with few somatic mutations and copy number variations.


Although a high level of intratumor heterogeneity was evident in high-grade serous ovarian cancer, our results suggest that transcoelomic metastasis arises with little accumulation of somatic mutations and copy number alterations in this patient.

Peer Review reports


Epithelial ovarian cancer is the fifth leading cause of cancer death among women in the USA [1]. The major reason for the poor prognosis is the fact that more than 75% of patients are diagnosed with advanced stage disease characterized by metastasis to the peritoneal cavity. The metastatic patterns of ovarian cancer differ from those of most other malignant epithelial disease. Transcoelomic is the most common route of metastasis in epithelial ovarian cancer and contributes to the significant morbidity and mortality associated with this cancer [2]. Given the high recurrence rate and poor long-term survival of women with advanced stage disease, there is a strong need to document the unique metastatic patterns of epithelial ovarian cancer by comparing the differences in genetic profiles between primary and metastatic lesions.

With the recent development of next-generation sequencing (NGS) technology, the Cancer Genome Atlas (TCGA) researchers have identified molecular abnormalities related to the pathophysiology, clinical outcome, and potential therapeutic targets in high-grade serous ovarian cancer (HGSC) [3]. The TCGA study provides a large-scale integrative view of the aberration in HGSC with extensive heterogeneity between individual tumors. However, it is not certain whether the genomic alterations found in single tumor biopsy samples from primary tumors are maintained in metastatic lesions. Furthermore, intratumor heterogeneity has been proposed as the main cause of treatment failure and drug resistance in ovarian cancer and other primary cancers [4]. Recently, NGS technology has led to progress in the evaluation of intratumor heterogeneity in various cancers [58]. In the field of HGSC, intratumor heterogeneity has been evaluated within primary tumors and associated metastatic sites, and the divergence of genetic variants was observed [5]. Despite evident intratumor heterogeneity within individual patients, little is known about how metastatic tumors further evolve compared to primary sites. The aim of this study was to explore the mutational profiles of primary tumors and associated metastatic lesions, and to identify the evolutionary relationship between primary and metastatic clones with NGS technology.


Patient information and sample preparation

A 71-year-old female was diagnosed with stage IIIC ovarian cancer at the time of sample collection. She had no family history for breast or ovarian cancer. She underwent BRCA1/2 germline mutation testing (Integrated BRACAnalysis®) and no mutation was found. Preoperative CA-125 level was 336 U/mL. She underwent cytoreductive surgery followed by platinum-based chemotherapy. During cytoreductive surgery, a right ovarian cystic mass of 10 × 9 × 8 cm in size was found. Multiple solid lesions were found inside the right ovarian cystic mass. A left ovarian tumor measuring 6 × 5 × 4 cm and consisting mostly of solid tissue was also found. Seven samples were taken randomly from the solid portions of both ovaries with a certain distance retained between each sample. All tissues consisted of >70% high-grade (FIGO grade 3) serous adenocarcinoma cells based on pathological review. Adjacent normal tissues from the left fimbriae and blood were also collected to serve as normal controls. Eleven tumor samples were collected from the ovaries, right fimbriae, bladder peritoneum, and omental lesions during surgery under the supervision of our pathologist (Min A Kim) (Figure 1A).

Figure 1
figure 1

Intra-tumoral mutational profiles of HGSC. (A) Sampling sites of tumor and normal control tissues. (B) Phylogenetic tree of somatic mutations. (C) Phylogenetic tree of somatic copy number variations. (D) Patterns of somatic mutations across samples. HGSC: high grade serous ovarian cancer, RO: right ovary, RF: right fimbriae, LO: left ovary, LF: left fimbriae, BP: bladder peritoneum, OM: omentum.

This patient had no evidence of recurrence at the time of publication and 12 months had passed since the completion of first-line treatment. This was a platinum-sensitive case (>6 months after first-line treatment completion). This study was approved by the Institutional Review Board (IRB) at Seoul National University Hospital (Registration number: C-1305-546-487) and performed in compliance with the Helsinki Declaration. We obtained informed consent for samples to be used in research. Written informed consent was obtained from the patient for publication of the case report including any accompanying images and disclosure of sequence data.

Library construction, exome capture, and sequencing

Genomic DNA was extracted separately from each sample (Qiagen, Valencia, CA, USA) and shotgun libraries were constructed by shearing 3 μg of genomic DNA. The SureSelect Human All Exon V4 + UTRs kit (Agilent, Santa Clara, CA, USA) was used to capture 71 Mbps of exons and UTRs, according to the manufacturer’s protocol, which were subsequently sequenced on an Illumina HiSeq2500 (Additional file 1: Tables S1 and S2). Sequencing data are accessible at Sequence Read Archive (SRA, accession number SRS823287).

Analysis of whole exome sequencing data

Short reads were aligned to the reference human genome (hg19) using Novoalign V2.07.18 with the default options. PCR and optical duplicates were removed using Picard v1.67 MarkDuplicates. Local realignment around the known indels in dbSNP135 and base quality score recalibration were performed using the Genome Analysis Toolkit (GATK) v2.6-4 [9]. Somatic mutations were identified by muTect 1.1.4 with the default options [10], and manually inspected by using Integrative Genome Viewer (IGV) [11]. The variants were annotated using the SeattleSeq Annotator, and then the variants listed in dbSNP132 and in repetitive regions were removed (repeatMasker, tandemRepeat column in SeattleSeq). Intronic, intergenic, near-gene, and synonymous mutations were also excluded. The germline mutations were identified by the GATK Unified Genotyper with the blood sample. Small indels were detected by Dindel v1.01 [12]. To avoid false positive somatic indels, only indels validated manually by IGV and confirmed by multiplex PCR were considered real variants. Candidate driver mutations and functional germline mutations were called based on the results from seven functional prediction algorithms and three conservation score algorithms using ANNOVAR [13] and dbNSFP v2.0 [14] (Additional file 2). All URLs for the analysis programs are listed in Additional file 2.

Somatic copy number alteration (SCNA) analysis

Genomic DNA (~600 ng) from each sample was processed with SNP chip analysis using the Genome-wide Human SNP Array 6.0 (Affymetrix, Santa Clara, CA, USA) according to the manufacturer’s instructions (Additional file 1: Table S1). Raw data were processed with the Affymetrix SNP6 Copy Number Inference Pipeline developed by Broad Institute using GenePattern modules [15]. Briefly, raw data was calibrated to signal intensities, called genotypes, and then the signal intensities were converted to copy number calls. After refinement of the copy numbers, somatic copy number alterations (SCNA) were called by subtracting the signals in the tumor sample from those in the normal sample. The segments of the SCNA were identified by circular binary segmentation.

For omental samples, whole exome sequencing data was used to detect SCNA. Pair-end read data was processed by the Varscan2 copynumber and copyCaller [16] with whole exome sequencing data of blood and the following non-default parameters: max-segment-size, 250; data-ratio 0.301 for OM1 and 0.306 for OM2. These raw segment data were smoothed and segmented using the ‘DNAcopy’ R package [17] with alpha = 0.01, nperm = 10,000, and trim = 0.025, then the segment values were magnified three times. All SCNA were visualized using Circos plot v0.64 [18].

Validation of somatic mutations and indels

Since quality control for false negatives is crucial for exploring intratumor heterogeneity, we selected 122 loci primers for multiplex PCR with HiSeq2500 for validation. Primer pairs were designed and synthesized based on column-based methods, pooled, and multiplex PCR was performed with 50 ng of genomic DNA from each sample (Celemics, Seoul, Korea). Subsequently, each product was indexed, mixed, and deeply sequenced on HiSeq2500. Raw data was deindexed and mapped to the reference human genome (hg19) using NovoAlign. Mutascope was used to call somatic mutations, and compared to the whole exome sequencing data [19]. Only the loci with at least 500 reads of both normal and tumor tissue and >5% allelic fraction were used for validation.

Phylogenetic tree construction and variant classification

A phylogenetic tree was generated to assess the tumor evolutionary patterns in terms of somatic mutations. The phylogenetic analysis followed the method described in a previous report [5]. All point mutations were converted to binary data (0 = no mutation, 1 = somatic mutation) for each sample, and a matrix with sample names in rows and loci in columns was generated. Next, we calculated Pearson correlation coefficients (ρxy) between samples x and y, and 1-ρxy was considered the distance between x and y. The Neighbor-Joining method [20] and the Unweighted Pair Group Method with Arithmetic Mean (UPGMA) method were applied to cluster samples and construct the phylogenetic tree. We used the ‘ape’ R package [21] for these analyses.

Samples were segregated by cluster P1, cluster P2, and cluster M for further analysis (Figure 1). If any somatic mutation was found in at least three samples in ‘cluster P1’ or at least two samples in ‘cluster P2’ and ‘cluster M’, we concluded that the mutation truly existed in that respective cluster. A mutation was classified as “Common” when it was found in all clusters, as “Shared” when found in any two clusters, as “Cluster-specific” when found in only one cluster, otherwise as “Sample-specific”.

Similar to the somatic mutation analysis, somatic copy number alterations were also converted to weights as follows; δmax [log10 L, 1], where L is the segment length, δ = 1 if the segment was amplified, −1 if deleted, or 0 otherwise. A matrix with sample names in rows and altered regions in columns were constructed. Pearson coefficients were calculated, and a phylogenetic tree was generated as described above.

The segments were classified as cluster P1, cluster P2, and cluster M as well. Initially, the cut-off values (log2 ratio) for amplified and deleted segments were set to 0.2 and −0.2, respectively. We decided that the segment was altered, either amplified or deleted, only if all samples in each cluster were amplified or all samples were deleted. If any sample in a cluster was altered differently, the segment was neglected. We classified the segments as “Common” when all three clusters had the same sample variation, “Shared” when any two clusters had the same variation, and “Specific” when variation was found in only one cluster. Coding genes (RefSeq database) within each segment were collected and functional analysis was performed using the DAVID functional annotation tool [22] and the GO_BP (Gene Ontology, Biological Process) and KEGG pathway databases.


Whole coding exons and untranslated regions (71 Mbp) in genomic DNA from seven ovarian tumor sites, three metastatic lesions, and two normal control samples (including a blood sample) were sequenced (Figure 1A). The mean coverage was 92× for tumor tissue and 65× for normal tissue. We sequenced more deeply on two omental tumor samples (211×, 199×) to thoroughly explore the variants in metastatic lesions (Additional file 1: Table S2). A total of 2,248 somatic mutations (3.2/Mb for each sample on average) were identified, and the average number of non-synonymous or splicing site mutations was 122 per sample (range: 77–167) (Additional file 1: Table S3). To avoid overestimation of intratumor heterogeneity, we randomly selected 122 somatic mutations (Additional file 1: Table S4) and performed multiplex PCR followed by ultra-deep re-sequencing (median coverage: 9,647×) for eight samples (Additional file 1: Table S1). The precision, false negative rate, and false positive rate of mutation calling in whole exome sequencing were 93%, 6%, and 1%, respectively. We found no pathologic BRCA1 and BRCA2 germline mutation in this patient. Other germline mutations are listed in Additional file 1: Table S5.

Phylogenetic trees were generated with somatic mutation data on 634 loci that were found at least once in the tumor samples. The samples from primary sites were segregated into two clusters (clusters P1 and P2), and the samples from metastatic lesions formed cluster M (Figure 1B). Based on the evolutionary tree, clusters P1 and P2 diverged earlier than cluster P1 and M. Interestingly, clusters P1 and P2 were not united according to the spatial position of sampling sites. These patterns were also observed in the phylogenetic tree based on copy number variations (Figure 1C).

Next, we classified 313 non-synonymous or splicing site mutations into four groups: Common, Shared, Cluster-specific, and Sample-specific (Figure 1D). Only 19 mutations (6%) were found in most samples, the Common group, which showed higher intratumor heterogeneity than previous studies across various cancers [58]. Ten non-synonymous mutations in genes including TP53KIF13A, and SPIC were identified (Table 1), indicating that those mutations were acquired in the early stage of tumorigenesis. Eighty-two (26%) somatic mutations were in the Shared group. All mutations in the Shared group were discovered in both cluster P1 and cluster M, supporting a common evolutionary origin. Also, 25 nonsynonymous mutations were considered as the candidate driver mutations. TP53 Y220C and SPIC E152K in the Common group are only mutations listed in COSMIC database (Table 1). We could not identify any anti-neoplastic therapeutic agents that interact with candidate driver mutations except PRKCQ C281S, which was found to interact with sophoretin [23]. However, this mutation is only detected in Cluster P2.

Table 1 Candidate driver mutations affecting characteristics of ovarian cancer

Only 11 somatic mutations were identified in the Cluster-specific group in cluster M, much fewer than those in clusters P1 and P2 (39 and 54, respectively). The mutations classified in cluster M-specific group were dominantly found in most samples of cluster M but not in other clusters. However, all 11 cluster M-specific mutations were also found in at least one sample from cluster P1. In contrast, most cluster P2-specific mutations were found only in cluster P2 (Figure 1D). False negative calling of cluster M-specific mutations was less likely, since the omental samples were deeply exome-sequenced and further validated by multiplex PCR followed by deep re-sequencing. The false negative rate of mutation calling in omental samples calculated with validation sequencing was less than 10%. Therefore, it seems that cluster M diverged from the common ancestry clone of cluster P1 with few additional somatic mutations.

To identify the branching mutation related to the origin of cluster P2, we focused on a subset of cluster P2-specific mutations found in non-cluster P2 samples (Additional file 1: Table S6). The allele frequency of each sample determined by ultra-deep re-sequencing was normalized to the mean allele frequency. The normalized allele frequencies were comparable between cluster P2 and non-cluster P2 samples, but that of ARNT2 S457* in cluster P2 was about ten-fold higher than in the right fimbriae. This finding supports the notion that the mutation was obtained upon the divergence of cluster P2.

SCNA were derived from six tumor samples and a normal sample. The analysis showed that the genomic architectures of samples in cluster M were similar to the patterns in cluster P1, but an arm-scale deletion on chromosome X was observed in cluster M (Figure 2). In contrast, the copy number patterns of cluster P2 were quite different from those of cluster P1 or M, supporting the conclusion that clones in cluster P2 diverged earlier than cluster M. Similar to the somatic mutation classification, we classified “Common segments” when the amplified/deleted segments were observed in all samples, suggesting that the segments formed in the early phase of tumorigenesis. Forty-four Common segments spanning 101 Mb were amplified, and 168 Common segments spanning 245 Mb were deleted. These segments covered 354 genes and 1,835 genes, respectively. Then, we characterized functional pathways affected by these genes. The genes related to ‘skeletal system development’ were enriched in amplified Common segments, and those related to ‘embryonic development ending in birth or egg hatching’ and ‘chemokine signaling pathway’ were enriched in deleted Common segments (Additional file 1: Table S7). The segments were considered a “Shared segment” when amplification or deletion was found in samples from any two clusters. We determined that 154 Shared segments (227 Mb) were amplified and 248 Shared segments (287 Mb) were deleted in both clusters P1 and M. Pathways previously reported to be altered in ovarian cancer, such as the JAK/STAT signaling pathway and the Cytokine-Cytokine Receptor pathway [24], were also identified in these Shared segments, but not in the Common segments. Interestingly, the genes related to blood vessel morphogenesis (31 genes, Benjamin-Hochberg (BH) score 0.094) were deleted Shared segments between cluster P1 and M, but not cluster P2. In contrast, the genes related to hemophilic cell adhesion (50 genes, BH score 9.9×10−16) were enriched in the amplified segments found only in cluster P2.

Figure 2
figure 2

Genomic profiles of somatic copy number alterations (SCNA). (a) Common segments (green) and Shared segments (grey). (b, c, d) Cluster P1 samples, RO1, RF, and LO4. (e, f) Cluster M samples, OM1, and OM2. (g) Cluster P2 sample, LO3. Overall the pattern of SCNA in cluster M was similar to the pattern in cluster P1 except for a large deletion on chromosome X. Cluster P2 showed distinct SCNA patterns compared to other clusters. red: amplification, blue: deletion.

Phylogenetic tree analysis based on somatic mutation and copy number variation was used to study the clonal relationship between different regions of primary and metastatic tumors (Figure 3). The findings indicate that metastatic lesions derive from a common, ancestral clone within the primary tumor. In cases of bilateral ovarian tumors like the one assessed here, metastatic potential may be gained in the early stages of tumorigenesis.

Figure 3
figure 3

Evolutionary model of non-spatial clustered metastatic ovarian cancer. (CCR: cytokine-cytokine receptor pathway, SCNA: somatic copy number alteration).

Based on the SCNA data, we focused on the frequently detected focal SCNA reported in the TCGA data (Additional file 1: Table S8) [3]. Among the top 20 most frequently observed focal amplifications, 12 segments were altered in our study, and only MECOM, TERT, and MYC segments were found among Common amplified regions. Although regions containing KRAS, ID4, MYCL1, and SOX17 were observed as tightly localized amplification peaks, these peaks were observed only clusters P1 and P2 for the patient in our study. Also, among the top 20 most frequent focal deletions, we found that 15% (3 of 20) of focal deletions including RB1 and PPP2R2A were commonly observed in our patient. NF1 is one of the genes shown to be related to intratumor heterogeneity in a previous study [5]. However, NF1 deletions were observed in both clusters in our study. This finding suggests that intratumor heterogeneity might appear differently in each patient. Lastly, we annotated the copy number variation patterns of 22 drug targets listed in the TCGA project for this patient [3]. Although 15 targeted genes were altered in this patient, only 40% (6 of 15) of the targeted genes were altered in all clusters (Additional file 1: Table S9).


Using NGS technology followed by confirmative validation, we were able to identify the clonal evolution of multiple samples collected from both ovaries and metastatic lesions in a single patient. Even though only 6% of somatic mutations were present in all samples, the vast majority of somatic variants found in the metastatic samples were present in the primary tumor samples. All 11 cluster M-specific mutations were found in at least one sample in cluster P1, and no somatic mutation was further accumulated in cluster M. In addition, SCNA showed that the genomic architecture of samples in cluster M were similar to the patterns in cluster P1. These findings suggest that peritoneal seeding arises with little accumulation of somatic mutations and copy number alterations in this patient. We also observed that non-spatial clusters of the primary ovarian cancer samples (cluster P1 and P2) shared a small number of genetic variations (Common mutations and segments), which indicates that metastatic potential developed at an early stage, and tumor clones in the peritoneal fluid were already able to implant in ovarian tissues at that moment.

Our analysis demonstrated that all metastatic samples from this patient were related to cluster P1, not P2, suggesting that the metastatic ability of ancestry clones was more accelerated in cluster P1. Based on this connection, we found that different cancer-related pathways were altered in the early divergent clones (cluster P1 and M vs. cluster P2). JAK/STAT signaling pathway genes including JAK2, known to be related to tumor migration through the epithelial-mesenchymal transition (EMT) [24], were only amplified in clusters P1 and M, supporting the hypothesis that clones in these clusters might be under migration pressure. In contrast, genes involved in cell adhesion pathways were only amplified in cluster P2, indicating that the clones in cluster P2 might be under an opposite pressure to clusters P1 and M.

Whether metastasis requires mutations beyond those required to drive the primary tumor is controversial [25]. In oropharyngeal squamous cell carcinoma, phylogenetic reconstruction according to somatic point mutations showed that metastatic samples arose as a late event [26]. In pancreatic cancer, seeding metastasis may require driver mutations beyond those required for primary tumors, and phylogenetic trees across metastases show organ-specific branches [27]. On the contrary, in HGSC, peritoneal seeding may arise with little accumulation of somatic mutations and copy number alterations. We could not identify the known driver variants causing transcoelomic metastasis in our patient.

In our study based on exome sequencing, all metastatic clones (cluster M) diverged together at a late stage, and the clusters of the primary tumor were distributed in both ovaries (non-spatial clusters). Our results provide a clue that some clones in the primary tumor can have metastatic potential, and that transcoelomic metastasis might be a simple spreading process using existing metastatic ability rather than supporting the previous tumor evolution models (linear [28], parallel [29], or mixed [30]). Regarding the clinical importance of transcoelomic metastasis in HGSC, it is surprising that few additional mutations were found in peritoneal seeding samples. This finding indicates the possibility that the microenvironment, including factors such as stromal cells, might play a role in fostering peritoneal implantation and cancer cell growth by secreting inducing factors [31].

Our study may help to further our understanding of tumor progression during HGSC. The data suggest that clones in peritoneal implants may not be more resistant than primary tumors in some patients. With the increasing clinical use of bioinformatics, developing methods that utilize the large amount of data to categorize patients into prognostic and treatment groups has become increasingly important [32]. This study suggests that patterns of intratumor heterogeneity between primary and metastatic clones might be the key for identifying the most appropriate treatment strategies for patients. In cases with metastatic patterns similar to the patient in this study (e.g., transcoelomic metastasis arising with little genetic alteration accumulation compared with primary tumors), debulking surgery might be useful to achieve optimal cytoreduction through adjuvant chemotherapy. If we identify those groups where seeding metastasis may require driver mutations beyond those required for primary tumorigenesis, debulking surgery might not be useful. In these instances, we should focus instead on the targeted therapy associated with driver mutations in metastatic lesions.

This study may provide important information for those who would like to evaluate tumor evolution in a larger cohort. For future studies evaluating clonal evolution in epithelial ovarian cancer, the following should be considered. First, the presence of mutations identified concurrently in most samples should be validated in a large number of cohorts in order to identify the key regulators in early tumorigenesis. Second, the clonal relationship between various metastatic sites from peritoneal seeding should be evaluated to identify the role of the microenvironment. Further studies are required to document the differences in genomic profiles between various metastatic sites such as the omentum, diaphragm, spleen, and pelvic peritoneum. This approach may elucidate the key regulators in the distinct metastatic characteristics of epithelial ovarian cancer. Third, genomic alterations other than somatic mutations and copy number changes should be considered to identify the unveiled driver variant causing tumor progression. Recently, the microRNA expression profile of an omental metastatic tumor was found to differ from that of the primary tumor in epithelial ovarian cancer, suggesting that microRNA might play role in tumor progression in metastatic tissues [33]. Another group reported that the genomic rearrangement landscapes of metastatic lesions differ from those of primary ovarian cancer [34].


We performed whole exome sequencing and copy number analysis for multiple primary and metastatic samples within an individual patient. Our research showed that HGSC has diverse intratumor heterogeneity in terms of somatic mutation and copy number variation profiles, but transcoelomic metastasis arises with little accumulation of genetic alterations in this patient.

Distinct evolutionary trajectories of primary high-grade serous ovarian cancers revealed through spatial mutational profiling

First published: 18 June 2013
Citations: 294

No conflicts of interest were declared.


High-grade serous ovarian cancer (HGSC) is characterized by poor outcome, often attributed to the emergence of treatment-resistant subclones. We sought to measure the degree of genomic diversity within primary, untreated HGSCs to examine the natural state of tumour evolution prior to therapy. We performed exome sequencing, copy number analysis, targeted amplicon deep sequencing and gene expression profiling on 31 spatially and temporally separated HGSC tumour specimens (six patients), including ovarian masses, distant metastases and fallopian tube lesions. We found widespread intratumoural variation in mutation, copy number and gene expression profiles, with key driver alterations in genes present in only a subset of samples (eg PIK3CACTNNB1NF1). On average, only 51.5% of mutations were present in every sample of a given case (range 10.2–91.4%), with TP53 as the only somatic mutation consistently present in all samples. Complex segmental aneuploidies, such as whole-genome doubling, were present in a subset of samples from the same individual, with divergent copy number changes segregating independently of point mutation acquisition. Reconstruction of evolutionary histories showed one patient with mixed HGSC and endometrioid histology, with common aetiologic origin in the fallopian tube and subsequent selection of different driver mutations in the histologically distinct samples. In this patient, we observed mixed cell populations in the early fallopian tube lesion, indicating that diversity arises at early stages of tumourigenesis. Our results revealed that HGSCs exhibit highly individual evolutionary trajectories and diverse genomic tapestries prior to therapy, exposing an essential biological characteristic to inform future design of personalized therapeutic solutions and investigation of drug-resistance mechanisms. © 2013 The Authors. Journal of Pathology published by John Wiley & Sons Ltd on behalf of Pathological Society of Great Britain and Ireland.


Ovarian cancer is the leading cause of death due to gynaecological malignancies in the developed world. High-grade serous ovarian cancer (HGSC) represents the most common histology (70%) and is responsible for the majority of advanced-stage cases. Although initially highly responsive to platinum compounds 1, the majority of HGSCs recur and affected women ultimately succumb to their disease.

TP53 mutation is believed to be the earliest tumourigenic driver event, with presence in nearly all cases 23, including pre-invasive serous tubal intraepithelial carcinomas (STICs), suggesting an HGSC aetiology in the fallopian tube 1249. Compromised homologous recombination due to BRCA dysfunction, through inherited germline polymorphism, somatic mutation or epigenetic silencing 31011, occurs in 50% of HGSC cases. The prevalence of TP53 mutations and BRCA deficiency leads to incompetent DNA repair and likely contributes to chromosomal instability, resulting in severely aberrant karyotypes. Additional recurrent but low-frequency somatic mutations have been reported in NF1CDK12 and RB1; however, in general, intertumoural heterogeneity of mutation profiles is widespread 3.

The concept of intratumoural heterogeneity has long been blamed for treatment failure in ovarian carcinoma and other primary cancers. Rooted in the clonal evolution model of tumour progression, cancers are assumed to originate from a monoclonal composition (the ancestral clone), with subsequent evolution leading to selective expansions of genomically distinct subclones 12 with distinct phenotypic properties 1216. As such, the divergence of clonal cell populations confers a heterogeneous genomic tapestry with implications for clinical endpoints, including the acquisition of metastatic potential 1721 and chemotherapeutic resistance 2223. Models of clonal evolution have been suggested for HGSC through low-resolution genomic profiling 2426 and serial mutation profiling of primary and recurrence-paired samples 27. However, little is known about the intrinsic diversity of mutational landscapes in primary tumours prior to therapeutic intervention.

Next-generation sequencing technology has enabled the study and quantification of clonal evolution and intratumoural diversity in cancer 20232830, including inference from distributions of digital allelic representation of mutations in single samples 2830, along with serial 182023 and regional comparisons of mutation profiles 31. In addition, sequencing of cell-free circulating tumour DNA (ctDNA) extracted from plasma has been demonstrated to be an effective non-invasive tool for monitoring tumour burden 3233.

Given these advances, we set out to measure precisely the clonal diversity of HGSCs present at diagnosis. Our results show how HGSCs each exhibit unique evolutionary trajectories with widespread regional diversity in mutational, copy number and gene expression landscapes prior to treatment intervention.

Materials and methods

Six women (Table 1) with histologically diagnosed high-grade serous cancer were included in this study (see supplementary material, Appendix). Tissue was obtained from a total of 31 tumour sites (range 3–10 samples/patient, including two fallopian tube lesions) with matched normal DNA for each patient (Table 1). In addition, for a subset of cases (four patients), plasma (pre-operative and pre-anaesthetic) was used to identify the presence of clonally dominant and subclonal mutations through deep sequencing of ctDNA. Thirty-three samples (27 tumour, six normal) were selected for Affymetrix SNP 6.0 genotyping arrays to determine genomic architecture and copy number alterations (CNAs), and the exomes of 25 samples (19 tumour, six normal; Table 1) were sequenced to a median exonic coverage of 70.7-fold across different samples (see supplementary material, Table S1 and Figure S1). The 19 tumour samples were used as index discovery samples to identify sets of mutations for interrogation by deep amplicon sequencing 28 (median > ×5000 coverage) to: (a) confirm predicted mutations; (b) determine the presence/absence of mutations in specific samples; and (c) infer clonal diversity estimates both within and between samples from the same patient. For 29 tumour samples, RNA was extracted for gene expression profiling by Affymetrix U133 2.0 arrays.

Table 1. Case descriptions for the six HGSC patients profiled in this study
Case no. Age Stage Treatment BRCA status Progression-free survival (months) Overall survival (months) Patient status Number of samples Index samples (exome) Index samples (Affy SNP 6.0) Index samples (Affy U133 expression)
Spatial 1 56 IV IV CP×6 Unknown (declined) 24 29 Alive, recurrent disease, 3rd line chemo 4 (right ovary, quadrants 1a–d) 4 (All and normal) 4 (All and normal) 4 (right ovary, quadrants 1a–d)
2 59 IIIA IV/IP CP×3, IV×3 Screen negative 27 27 Alive, no evidence of disease 4 (right ovary, quadrants 2a–d) 4 (All and normal) 4 (All and normal) 4 (right ovary, quadrants 2a–d)
3 64 IIIC IV/IP CP×6 Unknown (moved) 9 9 Unknown (moved) 3 (3a, right ovary; 3b, left ovary; 3c, posterior cul de sac) 3 (All and normal) 3 (All and normal) 3 (3a, right ovary; 3b, left ovary; 3c, posterior cul de sac)
4* 79 IIIC IV CP×6 Screen negative 15 17 Alive, recurrent disease, 2nd line chemo 10 (4a–e, right ovary; 4f–i, left ovary; 4j, left fallopian tube) 1 (4a and normal) 9 (4a–i, and normal) 9 (4a–e, right ovary; 4f–i, left ovary)
5* 73 IIIC IV CP×6 Unknown (declined) 17 17 Alive, no evidence of disease 8 (5a, right ovary; 5b–e, left ovary; 5f, left iliac node; 5g, left para-aortic node; 5h, left fallopian tube) 5 (5a–c, f, g and normal) 5 (5a–c, f, g and normal) 7 (5a, right ovary; 5b–e, left ovary; 5f, left iliac node; 5g, left para-aortic node)
Temporal 6 65 IIIC Primary Rx: IV CP×6 Recurrence Rx; 21 cycles (multiple agents) Screen negative 13 74 Dead of disease 2 (6a, omentum (primary surgery); 6b, left ovary (recurrence)) 2 (both and normal) 2 (both and normal) 2 (6a, omentum (primary surgery); 6b, left ovary (recurrence))

Additional details of all methods are described in the Appendix (see supplementary material).


Intratumoural mutational diversity in HGSCs

The somatic mutational profiles of synchronous spatially separated samples (removed during primary surgical staging) were compared in cases 1–5 (see supplementary material, Figures S2–S6). In case 6, we compared a primary sample to a recurrence obtained 42 months later, following multiple chemotherapeutic regimens (total 21 cycles) (see supplementary material, Figure S7). A total of 1349 somatic mutations in all samples were confirmed with deep amplicon resequencing (median 5079-fold coverage; see supplementary material, Table S2), of which 404 were unique mutations (334 non-synonymous, 46 synonymous and 24 small indels), ranging from 31 to 137 unique mutations per case. A panel of shared, identical mutations, indicative of a single ancestral clonal origin, was evident, with 55, 30, 16, 15, 19 and 32 mutations in cases 1–6, respectively. All cases except case 1 harboured a mutation in TP53, with the identical mutation present in all samples from that individual (see supplementary material, Table S2), supporting a monoclonal origin with a TP53 mutation in the ancestral clone.

Patterns of mutation conservation between samples from the same patient varied widely across the six cases (51.5 ± 30.7% mutations present within all samples of a case (Figure 1A–C; see also supplementary material, Table S3)). Case 5 had the lowest conservation of mutations (10.2%; Figure 1C), whereas case 6 had the highest (91.4%). Examination of mutations shared in samples of a single tumour mass, ie case 1a–d (right ovary), case 2a–d (right ovary), case 4a–e (right ovary), case 4f–i (left ovary) and case 5b–e (left ovary), had an average of 63.0 ± 29% mutations present in all samples. Case 4f–i had the highest conservation of mutations (94.4%), followed by case 4a–e (81.2%), case 1a–d (67.1%), case 2a–d (55.6%) and case 5b–e (17.6%) (see supplementary material, Table S3). Non-synonymous mutations in known cancer genes not present in all samples included PDGFRB and SH3GL1 in case 2 (see supplementary material, Figure S3) and PIK3CACTNNB1 and RBM15 in case 4 (Figure 1B). Case 5c harboured 54 (of 102 total) mutations that were not found in the other left ovary samples, which contained 66, 33 and 45 total mutations, respectively. Phylogenetic analysis (Figure 1D) revealed a common clonal ancestry of all samples in case 5, but with an ‘early’ divergence of 5c.

Details are in the caption following the image
Intratumoural mutational profiles of HGS ovarian cancer. Anatomical sites and intratumoural mutational profile for cases 3 (A), 4 (B) and 5 (C); point mutations are shown in blue, indels in green; grey indicates a predicted mutation where validation by deep sequencing was inconclusive; light blue indicates allelic frequencies (counts of non-reference allele/total depth of coverage) in the fallopian tube lesion. (D) Phylogenetic tree of mutational profiles of cases 1–6, depicting evolutionary branching patterns reflective of clonal relationships between samples. The tree was computed using distance matrices based on Pearson correlation coefficients, followed by neighbour-joining cluster analysis. The control sample represents the ‘root’ whereby data were generated with no aberrations. Neighbour-joining distances are shown along the branches of the tree, which reflect genetic distances between branching points; longer branches indicate more genomic differences.

Due to highly disparate mutational profiles in left and right samples for case 4, we correlated the findings with histopathological review. The right ovary from case 4 (diagnosed and treated as HGSC) exhibited mixed histology of predominately high-grade endometrioid pattern, with gland formation and a minor HGSC component, while the left ovary, left fallopian tube and all metastatic sites consisted exclusively of HGSC (see supplementary material, Figure S8). Although the right ovarian tumour was p53 IHC-positive regardless of subtype, the expression of WT1 was negative in the endometrioid regions and positive in HGSC areas, as expected 34. All ten tumour samples (including left fallopian tube lesion case 4j) shared the identical mutation in TP53 (chr17, 7518290, p.239N > S), along with 16 other mutations (Figure 1B), indicating a common cell of origin despite completely distinct histological profiles (see supplementary material, Figure S8). Forty-three mutations present in the right ovary (mixed histology) were not present in the samples from the left (all HGSCs; Figure 1B). As the index exome for this case was restricted to the right set of samples, we subjected all samples to the PGM Ion AmpliSeq Cancer Panel. We detected a PIK3CA p.546Q > R hotspot mutation in case 4f–i (left ovary) that was absent from case 4a–e. Consistent with endometrioid histology, we also detected a CTNNB1 mutation in case 4a–e that was absent from case 4f–i 35.

Collectively, these results reflect extensively divergent mutational profiles in spatially separated samples, with significant evolutionary selection due to regional environments and highly unique evolutionary trajectories.

Mutational events in the ancestral clone are detectable in cell-free circulating tumour DNA from plasma

Plasma samples collected pre-operatively and pre-anaesthesia were available for four cases from which ctDNA was extracted and subjected to targeted deep-sequencing of validated somatic mutations (see supplementary material, Appendix). We assayed 72 mutations in case 1, 49 in case 2, 63 in case 4 and 131 in case 5. Across the four cases, 215/312 (69%) mutations contained sequencing coverage and 38/215 (18%) were detectable with mutation allele frequency above statistical background noise (binomial exact test, false discovery rate < 0.05) (see supplementary material, Table S2). Cases 1 and 5 were enriched for the presence of mutations in the plasma that were also present in all intrapatient tumour samples, and thus part of the ancestral clone (Fisher exact test, p < 0.05; Figure 2). Among the mutations conserved in all primary tumour samples and detected in the plasma were the TP53 mutation (chr17, 7518290 T > C) in case 4, 12 mutations in case 1, two mutations in case 2 and seven mutations in case 5. Thus, in each patient at least one mutation representative of the ancestral clone was detected. Fourteen mutations (two in case 1, two in case 2, four in case 4 and six in case 5) were also detected in ctDNA, although they were only present in a subset of the primary tumour samples of a given patient, indicating that subclonal mutations might also be amenable to monitoring in ctDNA.

Details are in the caption following the image
Deep sequencing results of cell-free circulating tumour DNA from the plasma in cases 1, 2, 4 and 5. The distribution of plasma variant allelic ratio (log10 scale) is separated by the number of tumour samples in which the mutation was originally discovered. Cases 4 and 5 both also include the fallopian tube when counting the number of tumour samples. The number of mutations based on positions with coverage (c), no coverage (nc) and those with coverage that were significantly (s) detected, based on the binomial exact test (adjusted p < 0.05), are shown. The minimum variant allelic ratio detectable using the exact test for each patient is denoted by the vertical dashed line.

We noted that the proportion of mutations detected in ctDNA was non-uniformly distributed across the four cases. In particular, cases 1 and 5 exhibited higher sensitivity (16/53 and 13/82 mutations, respectively) than cases 2 and 4, suggestive of heterogeneous rates of shedding tumour DNA into the circulation.

The intratumoural variation in genomic architectures of HGSCs

Copy number analysis from high-density genotyping arrays revealed highly disrupted karyotypes in all samples, and heterogeneous variation of genomic architectures in samples from different patients (intertumoural heterogeneity) (see supplementary material, Table S4). While the overall copy number landscapes between samples from cases 1, 2 and 6 were highly similar (see supplementary material, Figure S9), we observed substantial intratumoural heterogeneity between samples from cases 3–5 (Figure 3A–C). We first focused on extreme CNAs by examining genes affected by high-level amplifications and homozygous deletions between samples of the same patient (see supplementary material, Table S5). The most extreme example of copy number diversity was found between samples from the left and right ovaries of case 4, consistent with the mutation profiles. Across the samples in case 4, 631 genes were altered by homozygous deletions and 852 genes were altered by high-level amplifications. None of these alterations was observed in all nine tumour samples. The samples from cases 4a–e and 4f–j form distinct branches of the phylogeny and are at least as divergent from each other as from other patient samples. We validated three segmental amplifications in case 4 in one representative sample from the left and the right ovary, using fluorescence in situ hybridization (FISH): a 150 kb region, chr20, 43,503,512–43,655,407, predicted to have a high-level amplification in the left ovary of case 4 with only low-level amplification in the right ovary (Figure 4); an amplified region, chr6, 47,952,581–48,122,073 (see supplementary material, Figure S10A) and a highly amplified region, chr12, 25,863,200–26,026,351 region (see supplementary material, Figure S10B).

Details are in the caption following the image
Intratumoural genomic architecture profiles of HGS ovarian cancer. Genomic copy number architecture of intrapatient samples using Circos; samples are arrayed in concentric circles as whole-genome profiles for cases 3 (A), 4 (B) and 5 (C). Colours represent the various copy number states: dark blue, segmental homozygous deletions; blue, hemizygous deletions; red, segmental gains; dark red, amplifications. Amplitude of each segment on the track represents the logR value of the segmental copy number change.
Details are in the caption following the image
(A) Copy number alteration (CNA) and (B) fluorescence in situ hybridization (FISH) comparisons between right (a–e) and left (f–i) ovaries of case 4 at chromosome 20. The 20p control probe is labelled in spectrum green (Vysis, cat. no. 30–2520200), and Region 2 using BAC RP11-241P6 is labelled with spectrum orange (Vysis, Nick Translation Kit, cat. no. 32–801300). Right ovary (a–e) shows aneuploidic gain; left ovary (f–i) shows amplification of Region 2. (C) Both populations of cells carrying the chr20 CNA are found in the molecularly fixed, paraffin-embedded, early tubal high-grade serous carcinoma of the left fallopian tube (FT). FT FISH image corresponds to box inset in the H&E serial section shown at ×20 magnification. p53 immunopositivity highlights FT lesion in serial section, consistent with the presence of the same TP53 missense mutation (g.chr17, 7577565 T > C; c.716A > G; p.N239S) in all case 4 samples. Magnifications = (IHC image) ×20; (FISH images) ×63.

Several genes in the Cancer Gene Census 36 were altered by extreme CNA events in sample subsets (see supplementary material, Table S5). Case 1a–c (but not case 1d) exhibited homozygous deletion of tumour suppressor NF1 in an approximately 190 kb region on chr17, 26,496,299–26,686,045, harbouring NF1OMGEVI2B and EVI2A (see supplementary material, Figure S11A). FISH assays probing this event revealed subpopulations of cells containing homozygous deletion of NF1 and all cells containing monosomy of chr17, confirming that homozygous deletion of NF1 was not in the ancestral clone. The same scenario was also observed in case 3, which harboured NF1 homozygous deletions in only two of the three samples (see supplementary material, Figure S11B). Observations of subclonal NF1 deletions reinforced evidence from the mutation data that important driver mutations may be acquired late in the evolutionary history of the tumour.

In addition to high-level amplifications and homozygous deletions, we compared the overall genome architecture based on loss of heterozygosity (LOH) profiles inferred from allelic intensity ratios (see supplementary material, Appendix). All intrapatient and intratumour samples in the six cases harboured LOH in chromosome 17 (see supplementary material, Table S4), supporting evidence that this is among the earliest aberrations in the tumourigenic process. To investigate events that may further indicate continual evolution of the genomic architecture, we identified chromosomal aberrations, such as copy-neutral LOH (NLOH) and amplified LOH (ALOH), that arose from at least two sequential (compound) genomic modifications. Despite the similarities of overall proportion of the genome altered by copy number events between intrapatient samples (Figure 3A–C; see also supplementary material, Figure S9), the relative proportion of the genome altered by compound copy number events varied between samples of cases 3 and 5 (Figure 5A). For cases 3b and 3c (but not 3a) we found evidence of whole-genome duplication events (Figure 5B). Segmental deletions in case 3a which were observed as NLOH events (doubling of the remaining allele) in cases 3b and 3c and diploid heterozygous regions in case 3a were observed to be doubled to four balanced copies in cases 3b and 3c (Figure 5C; see also supplementary material, Figure S12A). Overall, 17 chromosomes in cases 3b and 3c showed evidence of doubling. The remaining chromosomes (4, 8, 11, 13 and 19) appeared to have undergone concurrent segmental aneuploid events to both case 3a and 3b preceding or following genome doubling (see supplementary material, Figure S12B, C), suggesting continual accrual of genomic aberrations after clonal divergence of case 3a from the ancestral clone (Figure 5D). In case 5, samples 5b and 5f exhibited a higher proportion of the genome harbouring compound aberrations (mean 0.43) than the other samples—5a, 5c and 5g (mean 0.23) (Figure 5A). Case 5c harboured the highest number of mutations, with 54 of 102 mutations unique to that sample. As such, the evolution of genome architecture for case 5 appears to be independent of evolution at nucleotide scales. In case 3, substantial intratumoural variation in segmental copy number alteration profiles accrued with relatively conserved single nucleotide patterns. These results suggest that evolutionary trajectories of copy number and mutational profiles can vary independently, and must both be considered to form a complete picture of clonal divergence.

Details are in the caption following the image
Evolutionary sequential compound copy number analysis. (A) Analysis of proportion of the genome that was altered by sequential compound events. Compound events include copy neutral LOH (NLOH) and amplified LOH (ALOH) regions, which indicates the occurrence of more than one copy number event in sequence (eg deletion followed by amplification of remaining allele results in ALOH). (B) Pairwise comparison of copy number samples within case 3. The number of genes with a specific predicted discrete copy number (CN) is represented by the size of the dot. Genes that also have the same zygosity (LOH or heterozygous) status between the two samples are coloured red; otherwise they are grey. (C) Doubling of chromosome 18 in case 3b relative to case 3a. Deletion (green) in 18q in 3a is observed as NLOH (blue) in 3b; amplification of 18p in 3b is balanced, indicating doubling of both diploid alleles in 3a. ‘A’ and ‘B’ genotypes are used to denote the two alleles. (D) Phylogenetic tree of discrete compound events. Genes were assigned an integer value representing the weight of observing compound events: 2, ALOH; 2, NLOH; 2, homozygous deletion; 1, hemizygous deletion; 0, diploid heterozygous; 0, allele-specific amplification. Euclidean distance was computed between pairs of tumour samples and a control (which consists of zeros for all genes) and neighbour-joining cluster analysis was used to generate the tree.

The origins of fallopian tube lesions and reconstruction of HGSC evolutionary histories

For two cases (4j and 5h) we profiled fallopian tube lesions obtained at primary surgical staging. The lesion in case 4j was a small focus of invasive carcinoma with involvement of the fallopian tube mucosa, while 5h showed invasive carcinoma within the fallopian tube wall, without mucosal involvement. It is believed that HGSC may originate from the fimbriated end of the fallopian tubes and shed cells into the peritoneal cavity, spreading transcoelomically. We deep-sequenced the set of somatic mutations over all primary tumour samples for cases 4 and 5 in the FT lesions to examine their evolutionary histories. In case 4j we identified 17 mutations (26.2% of all mutations in case 4; Figure 1B), and in case 5h we found 44 mutations (32.1% of all mutations in case 5; Figure 1C). In both cases the somatic mutations in TP53 were among the set of mutations detected in the FT.

The FT lesion in case 5 harboured seven mutations that were not present in all samples, indicating that the clones comprising case 5h were unlikely to have been ancestral to the tumours. Rather, phylogenetic analysis suggested that case 5h had closer clonal relationships with case 5b and 5f (left ovary and para-aortic lymph node, respectively) compared to cases 5a, 5d, 5e and 5g (Figure 1D). We suggest that case 5h is therefore more likely to be a metastatic implant in the FT rather than a precursor or early lesion. By contrast, case 4j shared all mutations present in samples taken from the left ovary with the exception of PIK3CA p.546Q > R, found at a low frequency (3% of the total 4671 reads) and only present in the region of the lesion closest to the ovary. Conversely, the fallopian tube lesion did not contain the mutation in CTNNB1 present in the right ovary samples, suggesting that it was acquired after the 17 mutations and was restricted to the endometrioid component of the right tumour.

Results from PIK3CA mutation sequencing suggested the FT lesion in case 4 was composed of heterogeneous subclones. FISH on three amplified regions of chromosomes 6, 12 and 20 (predicted in the copy number data; see above) which were specific to the right (chr12) or left (chr6, chr20) samples, revealed two coexisting populations of nuclei in the FT lesion—those that had high-level amplification on chr20 (BAC RP11-241P6) similar to the left ovary, and aneuploidic gain of the region as seen in the right ovary (Figure 4). The region profiled on chromosome 6 in the FT showed copy number patterns similar to the right ovary (see supplementary material, Figure S10A), while the chr12 region in the FT resembled the left ovary (see supplementary material, Figure S10B). Taking mutational and copy number data together, case 4 follows a monoclonal hierarchy in which two dominant clones emerged in the FT and gave rise to the divergent, histologically distinct populations in the right and left ovaries. To our knowledge, this is the first example of two distinct but related precursor clonal populations in the fallopian tube in situ lesion, with molecular data indicating that divergent cell populations in tubal mucosal lesions can evolve before extratubal spread of disease.

Intra-sample clonal diversity revealed by cellular frequency analysis of mutational profiles

We integrated genomic aberrations at all scales, including mutations, copy number and LOH, to infer the degree of clonal diversity existing within each sample. Using a Dirichlet process statistical model, PyClone 28 (see supplementary material, Appendix and Figure S13), we inferred the clonal population structure of each sample. The output of the PyClone model is an estimate of the proportion of cells in the sample harbouring each mutation in the input (cellular frequency). Each sample from cases 1–6 exhibited variance in the distribution of cellular frequencies (Figure 6), suggesting presence of intra-sample clonal diversity. Case 5 showed the highest degree of intra-sample variation with the minimum interquartile range (IQR) of 0.276 (case 5g) and the maximum of 0.510 (case 5b). Case 1 showed the lowest degree of intra-sample variation, with 0.085, 0.170, 0.197 and 0.207 IQR in samples 1c, 1b, 1a and 1d, respectively. Intriguingly, two of the case 4 samples (4h and 4i) showed comparatively tight distributions, with IQRs of 0 and 0.002, respectively. Finally, case 6b (postchemotherapy sample) showed a higher degree of variation in clonal frequency than the primary sample, suggesting that clones present in equal proportions in case 6a emerged in different proportions after chemotherapy in case 6b.

Details are in the caption following the image
Intra-sample clonal diversity spectrum of HGSCs. (A) Distribution of cellular frequency estimates over mutations in each sample (estimated using PyClone), indicating statistically significant variation both within and between samples of the same case. (B–G) Profile of cellular frequencies for all cases, where darker shades of red indicate increasing cellular frequency estimates.

Diverse genomic landscapes associate with heterogeneous transcriptional profiles

We next analysed mRNA gene expression profiles to examine the extent of intratumoural transcriptional diversity. We clustered our 29 samples with the TCGA cohort of 594 HGSC samples downloaded from the TCGA portal (see supplementary material, Appendix) to determine how our samples were distributed in the TCGA population (Figure 7A). Consistent with previous reports 337, there were four clusters of patients representing HGSC subgroups, which we mapped to the TCGA groupings: differentiated, proliferative, immunoreactive and mesenchymal (Figure 7A). Samples from cases 1, 2, 5 and 6 clustered together such that expression profiles from within one patient were found to be adjacent in the dendrogram. By contrast, cases 3 and 4 exhibited widely discrepant expression profiles between samples from the same patient. The expression profile for case 3a clustered with the proliferative group, whereas cases 3b and 3c clustered with the immunoreactive group. Thus, 3a, 3b and 3c were placed in different major branches of the dendrogram (Figure 7A), consistent with divergent copy number profiles (Figure 5). In case 4, the left ovary samples (4f–i) clustered together within the mesenchymal group, while the right ovary samples (4a–e) were clustered together with the differentiated group. These results suggest that genomic diversity observed in these two patients may elicit changes in the transcriptional programme, reflective of phenotypic change.

Details are in the caption following the image
(A) Hierarchical clustering of the expression data in a cohort consisting of 594 TCGA HGSC samples and 29 HGSC samples, representing six cases analysed in this paper. The bars show four patient groups according to the hierarchical clustering of the samples (top bar), patient labels according to Tothill et al 37 classification (middle bar), and distribution of our samples across all 623 samples. Clusters C1, 2, 4, 5 from Tothill et al correspond to differentiated, mesenchymal, immunoreactive and proliferative labels shown in the figure. (B) Simultaneous analysis of mutations and expression profiles by DriverNet, nominating the mutations that had significant impact on expression networks.

Genomic aberrations impacting transcriptional networks

Finally, we integrated gene expression and genomic mutation data to identify specific mutations impacting mRNA expression networks, using the DriverNet 38 algorithm (see supplementary material, Appendix). In total, DriverNet predicted 33 genes to be significantly associated with disrupted transcriptional networks (Figure 7B; see also supplementary material, Table S6), with TP53 as the top-ranked gene.

In case 1, mutations in SI, POLR1C, JAK2, FGF9, HIST1H2BI and PCDH12 were predicted to affect transcriptional networks (DriverNet; p < 0.05). SI and POLR1C mutations associated with disrupted expression of metabolic, TNFα and NF-κB pathways, while JAK2 associated with disrupted expression in adipocytokine, chemokine and Jak–STAT signalling, IFNγ, Notch and Sphingosine 1-phosphate (S1P) pathway genes. FGF9 and PCDH12 mutations were associated with altered expression in FGFMAPK and Wnt signalling pathway genes (see supplementary material, Table S6). These mutations are strong candidates for tumourigenic drivers in case 1, as all six were present in all samples in addition to impacting expression. Other than TP53, genes with mutations in all samples predicted to impact expression were XDHF2 and ITGB3 (case 2); STAT3C8B and COBRA1 (case 3); PDE7ACLASP1DDX23 and SI (case 4); ANGPT1 and SLC4A4 (case 5), and DIAPH1 and GRM3 (case 6). DriverNet predictions consisting of mutations present only in a subset of samples of a patient included MGAM (case 3), PIK3CACTNNB1CSF2RBRYR3DNAJ1PAK7 and PREX1 (case 4), MCCC1NUP50 and OGDHL (case 5) and PRPS1 (case 6).

These results illuminate a small subset of mutations potentially acting with TP53 in the ancestral clone (present in all samples) that alter the transcriptional programme. By contrast, mutations associated with alteration in expression of transcriptional networks present only in a subset of samples represent candidate mutations for non-tumourigenic driver mutations acquired after divergence from the ancestral clone. As such, these represent mutations potentially altering the phenotypes of only a subset of primary tumour cells.


We have shown the range of genomic diversity at nucleotide, copy number and gene expression scales present in six patients with HGSC. These samples were resected prior to treatment, revealing intratumoural diversity in the tumours’ natural evolutionary state. Our results were consistent with recent reports 39 that considerable genomic diversity is measurable in HGSCs at the level of copy number. However, in some cases (cases 3 and 5) we observed large-scale copy number differences in spatially separated samples that were not evident at the level of mutational profiling, indicative of different mutational mechanisms operating independently in different parts of a tumour. Together, copy number and mutational profiling identified well-characterized mutations in genes such as PIK3CACTNNB1 and PDGFRB, as well as homozygous deletions in NF1, that were not present in all samples of the same case. With the exception of TP53, our data suggest that well-known, actionable driver mutations may only be partially represented if only a single sample is considered per patient. Our results establish that embedded within the context of extreme intra- and intertumour heterogeneity, TP53 mutation remains the most stable genomic feature in HGSC. However, as efforts towards precision medicine informed by mutational landscapes of tumours reach maturity, it will be imprudent to ignore the degree to which regional genomic variation is present in primary samples prior to any treatment-related selective pressures. Moreover, studies of evolution in cancer through analysis of serial biopsies (eg through pre- and post-treatment samples, or comparisons of primary and metastatic samples), will need to account for evidence of divergent profiles simply due to regional sampling bias.

Mutational profiling revealed for the first time that mutations beyond TP53 are present in FT lesions. In case 4, we used the full spectrum of mutations to establish that evolutionary trajectories of two histologically and karyotypically distinct tumours within a single patient arose from a common aetiology, with 26.2% of all mutations (including TP53) conserved in the tubal lesion. Our results indicate how histologically distinct cell populations in the same patient can be linked by common aetiology in the FT. In contrast to HGSC, endometrioid tumours are thought to develop from atypical ovarian endometriosis, with the FT merely acting as a conduit for endometrial epithelium to spread to the ovary (reviewed in 40). The possibility of the FT playing a causative role in rare cases of endometrioid carcinoma, or mixed endometrioid–serous carcinoma, has not yet been explored. Mutations in case 5 revealed the FT lesion to be a metastatic implant more closely related to a lymph node metastasis than to the ovarian samples. Thus, future investigations into aetiological underpinnings of HGSC will likely benefit from mutational comparison to extra-tubal sites to definitively establish the evolutionary origin of FT lesions.

We noted an extreme case of unexpected genomic conservation. In case 6, samples were obtained at primary surgery and after 42 months, with 21 cycles of multi-agent chemotherapy in between. The two samples exhibited near-identical genomic landscapes. Coupled with the long survivorship of this patient, this observation represents an intriguing anecdote. Larger studies will be needed to establish if stable clonal population structure is a feature of long survivorship.

Finally, as proof of concept, examination of plasma ctDNA suggested that representative mutations in the ancestral clone and mutations present in subclones could be detected in ctDNA, although sensitivity to detect mutations varied from patient to patient. This implies that clinical sensitivity analyses will need to be undertaken to establish the generalizability of ctDNA analysis to clonally diverse tumours.

In summary, our study unveils the extensive genomic diversity in primary, untreated, high-grade serous cancers of the ovary prior to treatment-related selection pressures, and illuminates highly individualized evolutionary trajectories that will require detailed consideration if curative therapeutic strategies are to be achieved.


This work was supported by the Canadian Institutes for Health Research (Operating Grant No. 245779, to SPS), the Michael Smith Foundation for Health Research Career Investigator Programme (to SPS); AB holds an Eli Lilly Research Fellow award, and GH holds a National Sciences and Engineering Research Council graduate scholarship. We thank Dr Sarah Mullaly for editing the manuscript. The results published here are in whole or part based upon data generated by the Cancer Genome Atlas pilot project established by the NCI and NHGRI (Accession No. phs000178.v7p6). Information about TCGA and the investigators and institutions who constitute the TCGA research network can be found at: http://cancergenome.nih.gov/

Author contributions

Project conception and oversight, SPS; manuscript preparation, AB, GH, AT, LP, JNM, BG, DGH and SPS; data analysis and bioinformatics, AB, GH, JD, AR, JR, KS and SPS; sequencing, copy number/gene expression arrays, nucleic acid extraction and validation, WY, MM, NM, KT, TZ, RM, YZ, MAM, SY, LMP, MA and JS; tissue banking, JNM, SK and MTYL; and pathology review, BG and DGH.

Data and materials availability

All sequencing data will be deposited at the European Genome-Phenome Archive (EGA, http://www.ebi.ac.uk/ega/), which is hosted by the EBI, under accession number EGAS00001000547.

Supporting Information

Supplementary material on the internet

The following supplementary material may be found in the online version of this article:

Appendix: additional details of all methods and patient clinical histories

Figure S1. Distribution of sequence coverage of targeted exons for the 25 (19 tumour and six normal samples) exome capture libraries.

Figure S2. Intratumoural mutational profile for case 1.

Figure S3. Intratumoural mutational profile for case 2.

Figure S4. Intratumoural mutational profile for case 3.

Figure S5. Intratumural mutational profile for case 4.

Figure S6. Intratumoural mutational profile for case 5.

Figure S7. Intratumoural mutational profile for case 6.

Figure S8. Immunohistochemistry (IHC) profiles for case 4 indicate mixed histology.

Figure S9. Genomic copy number architecture of intra-patient samples using Circos.

Figure S10. Copy number alteration and fluorescence in situ hybridization comparisons between right and left ovaries and fallopian tube of case 4 at chromosomes 6 and 12.

Figure S11. Heterogeneous NF1 homozygous deletions in cases 1 and 3.

Figure S12. Comparison of cases 3a and 3b reveals genome doubling.

Figure S13. The pyClone model shown as a probabilistic graphical model.

Table S1. Median and standard deviation of coverage for the exons for the 25 (19 tumour, six normal samples) exome capture libraries.

Table S2. Omnibus table of validated somatic SNVs and indels.

Table S3. Mutation summary table.

Table S4. Copy number segments for intratumour samples of six cases.

Table S5. Extreme copy number segments for intratumour samples of six cases; homozygous deletions and high-level amplifications.

Table S6. Ranked list of candidate driver genes.

Other Articles related to this topic appeared on this Open Access Online Scientific Journal, including the following:

AMPK Is a Negative Regulator of the Warburg Effect and Suppresses Tumor Growth In Vivo

Genomics of bronchial epithelial dysplasia

Genomics in Medicine- Tomorrow’s Promise

Prostate Cancer: Androgen-driven “Pathomechanism” in Early-onset Forms of the Disease

CRACKING THE CODE OF HUMAN LIFE: Recent Advances in Genomic Analysis and Disease – Part IIC

CRACKING THE CODE OF HUMAN LIFE: The Birth of BioInformatics and Computational Genomics – Part IIB

Genome-Wide Detection of Single-Nucleotide and Copy-Number Variation of a Single Human Cell

Directions for Genomics in Personalized Medicine

LEADERS in Genome Sequencing of Genetic Mutations for Therapeutic Drug Selection in Cancer Personalized Treatment: Part 2

Paradigm Shift in Human Genomics – Predictive Biomarkers and Personalized Medicine – Part 1

Harnessing Personalized Medicine for Cancer Management, Prospects of Prevention and Cure: Opinions of Cancer Scientific Leaders @ http://pharmaceuticalintelligence.com

In Focus: Targeting of Cancer Stem Cells

Modulating Stem Cells with Unread Genome: microRNAs

What can we expect of tumor therapeutic response?


14 Responses

  1. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  2. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  3. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  4. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  5. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  6. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  7. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  8. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  9. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  10. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  11. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  12. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  13. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

  14. I actually consider this amazing blog , “SAME SCIENTIFIC IMPACT: Scientific Publishing –
    Open Journals vs. Subscription-based « Pharmaceutical Intelligence”, very compelling plus the blog post ended up being a good read.
    Many thanks,Annette

Comments RSS

Leave a Reply

%d bloggers like this: