DNA methylation as a

prognostic marker in clear cell

Renal Cell Carcinoma

Emma Andersson Evelönn

Department of Medical Biosciences

Department of Surgical and Perioperative Sciences

Umeå 2020

This work is protected by the Swedish Copyright Legislation (Act 1960:729) Dissertation for PhD ISBN: 978-91-7855-166-8 ISSN: 0346-6612 New Series Number: 2066 Cover design by Elin Larsson. Electronic version available at: http://umu.diva-portal.org/ Printed by: UmU Print Service, Umeå University Umeå, Sweden 2020

Framgångens hemlighet är att aldrig släppa målet ur sikte.

Till Hugo och Hanna, ni är det bästa i mitt liv

vii

Table of Contents

Abstract ............................................................................................ iii

Populärvetenskaplig sammanfattning ............................................. iv

Abbreviations ................................................................................... vi

Introduction ....................................................................................... 1 Clear Cell Renal Cell Carcinoma ...................................................................................... 1 Genomic traits in ccRCC.................................................................................................. 4

Genomic aberrations................................................................................................. 4 Common mutations in ccRCC ................................................................................... 4

Epigenetics ........................................................................................................................ 5 Histone modifications ................................................................................................ 7 RNA epigenetic modifications .................................................................................. 8 DNA methylation ....................................................................................................... 9 DNA methylation in cancer ..................................................................................... 10 DNA methylation in ccRCC ...................................................................................... 12 Methylation biomarkers for ccRCC ....................................................................................... 12

Aims ................................................................................................. 14 Paper I ............................................................................................................................. 14 Paper II............................................................................................................................ 14 Paper III .......................................................................................................................... 14

Materials and Methods .................................................................... 15 Workflow describing patients and methods included in this thesis ............................. 15 Patients and tissue samples ........................................................................................... 15

Tissue samples included in paper I ......................................................................... 16 Tissue samples in papers II and III ......................................................................... 16 Tumor samples from the Tumor Cancer Genome Atlas (Paper II)....................... 16 Blood analysis .......................................................................................................... 16

DNA extraction ............................................................................................................... 16 Methylation analysis ....................................................................................................... 17

Bisulfite conversion (Papers I-III) .......................................................................... 17 Methylation array analysis (Papers I-III) ............................................................. 17 Data preprocessing ................................................................................................................. 18

Array data analysis ................................................................................................................ 18

The building of prognostic classifiers ............................................................................ 18 Promoter Methylation Classification (PMC) Panel (Paper II) .............................. 19 Triple Classifier (Paper III) ..................................................................................... 19

Genetic analysis ............................................................................................................. 20 Genetic aberrations (Papers I and II) .................................................................... 20 VHL status (Paper I) ............................................................................................... 20

Gene expression analysis ............................................................................................... 20 RNA preparation (Paper II) ................................................................................... 20 Gene expression arrays (Paper II) ......................................................................... 20

ii

Statistical analysis........................................................................................................... 21

Results ............................................................................................ 22 Demographic overview of the ccRCC cohorts............................................................... 22 Loss of chromosome arm 3p and VHL genetic and epigenetic status (Paper I) ......... 23

Methylation analysis of the VHL promoter (Paper I). ......................................................... 23

Genomic aberrations and ccRCC disease progression (Paper II) ................................ 24 DNA methylation classification and its correlation to clinicopathological parameters

(Papers I and II) ............................................................................................................. 25 Methylation alterations associated with tumor progression (Paper I)................ 27 A stepwise increased DNA methylation level during tumor progression (Papers I

and II). ..................................................................................................................... 28 Inter- and intra- heterogeneity in DNA methylation profiles (Paper II). ........... 28

Methylation and genetic aberration and their correlation with biological and mitotic

age (Papers I and II) ...................................................................................................... 29 New prognostic classifiers in ccRCC (Papers II-III). ................................................... 30

Promoter Methylation Classifier (PMC) panel (Paper II). ................................... 30 Genes represented in the PMC panel...................................................................................... 31

Triple Classifier (Paper III). ................................................................................... 32

Discussion ....................................................................................... 34 General conclusions and future perspectives ............................................................... 40

Acknowledgment ............................................................................ 42

References ...................................................................................... 45

vii

Abstract

Clear cell renal cell carcinoma (ccRCC) is the most common type of renal cell

carcinoma worldwide. Metastatic ccRCC is correlated to poor prognosis whereas

non-metastatic disease has a 5-year survival rate up to 90%. Due to increased

accessibility to different types of diagnostic imaging the frequency of metastatic

ccRCC at diagnosis has decreased since the beginning of the 21st century. This has

led to an earlier detection of primary tumors before patients present symptoms.

However, 20-30% of the non-metastatic patients at diagnosis will progress and

metastasize within five years of primary nephrectomy. Identifying patients at

high risk of tumor progression at an early stage after diagnosis is of importance

to improve outcome and survival. Currently, in Sweden, the Mayo scoring system

is used to divide tumors into low, intermediate or high risk for tumor progression.

DNA methylation has been associated with tumor development and progression

in different malignancies. In this thesis, Illumina Infinium HumanMeth27

BeadChip Arrays and Human Meth450K BeadChip Arrays have been used to

evaluate the relationship between methylation and clinicopathological variables

as well as ccRCC outcome in 45 and 115 patients.

Our studies identified an association between higher level of promoter-associated

DNA methylation and clinicopathological variables in ccRCC. There was a

significant stepwise increase of average methylation from tumor-free tissue, via

non-metastatic tumors to metastatic disease. Cluster analysis divided patients

into two distinct groups that differed in average methylation levels, TNM stage,

Fuhrman nuclear grade, tumor size, survival and tumor progression. We also

presented two prognostic classifiers for non-metastatic tumors; the promoter

methylation classifier (PMC) panel and the triple classifier. The PMC panel

divided tumors depending on the methylation level, PMC low or PMC high, with

significantly worse prognosis in the PMC high group. This data was verified in an

independent, publically available cohort. The triple classifier was created using a

combination of clinicopathological variables, previously identified CpGs

biomarkers and a novel cluster analysis approach (Directed Cluster Analysis).

The triple classifier had a higher specificity compared to the clinically used Mayo

scoring system and predicted tumor progression with higher accuracy at a fixed

sensitivity.

The identification of two epigenetic classifiers that predicted outcome in non-

metastatic ccRCC further establishes the role of DNA methylation as a prognostic

marker. This knowledge can contribute to identification of patients with a high

risk of tumor progression and can be of importance in the decision regarding

adjuvant treatment post-nephrectomy.

iv

Populärvetenskaplig sammanfattning

Klarcellig njurcancer är den vanligaste form av njurcancer. I Sverige

diagnostiseras ca 1 000 individer årligen med sjukdomen. Idag upptäcks ofta

njurcancer när patienter undersöks med bilddiagnostik av buken av andra

anledningar, exempelvis vid oklar buksmärta och vid trauma. Detta gör att

tumörerna upptäcks innan de hunnit ge symtom och andelen patienter med

spridd sjukdom vid diagnos är under 20%, jämfört med över 30% i början av

milleniet.

Den enda botande behandling som finns för njurcancer är total eller partiell

nefrektomi, vilket innebär att hela njuren eller delen av njuren med tumör

opereras bort. Om sjukdomen upptäcks tidigt, innan den hunnit sprida sig till

kringliggande organ, är prognosen god och 90% av patienterna lever minst fem

år efter diagnos. Är sjukdomen spridd vid diagnos finns det idag behandlingar

som förlänger överlevnadstiden, och ny immunbehandling som också kan bota

sjukdomen.

Även om klarcellig njurcancer som är begränsad till njuren botas när tumören

avlägsnas är det ungefär en tredjedel av dessa patienter som drabbas av en ny

njurtumör eller spridning till andra organ inom fem år efter diagnos. Behovet av

att identifiera dessa patienter är stort när de är i behov av tilläggsbehandling. Idag

utgår klinikerna från hur tumören ser ut när de bestämmer hur patienterna ska

följas efter kirurgi, vid karakteristika som talar för en hög risk för spridning följs

patienterna tätare och under längre tid.

Genom att analysera hur ccRCC tumörer ser ut på cellnivå har skillnader i

genuttryck och DNA-sekvenser kunna identifieras. Att utnyttja dessa skillnader

är viktiga för att identifiera tumörer med ökad risk för spridning.

I min avhandling har vi analyserat DNA metylering i klarcellig njurcancer vid

diagnos. DNA metylering är en epigenetisk förändring, en förändring på DNA

nivå som inte påverkar DNA sekvensen men kan påverkar vilka gener som

uttrycks i cellerna. Nivåerna av DNA metylering skiljer sig mellan olika

prognostiska grupper ccRCC. De tumörer som vid diagnos redan spridit sig till

kringliggande organ har en högre grad av metylering jämfört med de tumörer

som växer endast i njuren. Vi har även kunnat visa på skillnader mellan de lokala

tumörer som senare sprids. Detta gör att DNA metylering kan användas som

prognostisk markör i ccRCC.

vii

Genom att utnyttja skillnaderna i DNA metylering har vi kunnat identifiera två

olika sätt att förutspå risk för spridning av lokal tumörsjukdom vid diagnos. Först

utnyttjade vi likheterna i metylering mellan de tumörer som är spridda vid

diagnos och de som sprids med tiden. Vi skapade med detta en ’Promoter

Methylator Classifier (PMC) panel’. Denna metod kan användas för att bedöma

risk för spridning och vi kunde verfiera resultaten i ett stort ccRCC tumörmatrial

från USA.

Vår andra diagnostiska modell byggde på en kombination av DNA metylering och

kliniska variabler (analys av blodprov och kunskap om tumörens utseende).

Genom att kombinera skillnader i metylering tillsammans med de kliniska

variabler, som är kända att öka risken för tumörspridning, kunde vi ytterligare

förbättra klassificieringen, den metoden kallades för ’the triple classifier’.

Båda modellerna kunde identifiera 85% av tumörerna som spreds inom fem år

efter diagnos, det är identiskt med den metod som används vid klinisk

klassificiering av ccRCC, Mayo system. Största skillnaden mellan våra båda

kliniska modeller och Mayo låg i antalet som bedömdes ha en ökad risk för

spridning men sedan inte drabbas. För Mayo låg denna på 50%, för PMC panelen

på 45% och slutligen för the triple classifier på 36%. Detta indikerar att vår triple

classifier var bäst på att både identifiera patienter med ökad risk för

tumörspridning utan att därför fånga in en alltför stor andel ”friska” som

bedömdes ha en ökad risk.

Sammanfattningsvis så har vi i denna avhandling visat att DNA metylering är en

stabil epigenetisk förändring som kan korreleras till risk för spridning vid

klarcellig njurcancer. Genom att utnyttja DNA metylering med eller utan kliniska

variabler har vi kunnat bygga två prognostiska modeller för att bedöma

spridningsrisken av lokaliserad njurcancer vid tidpunkten för diagnos. Ett

framtida mål, är att vidareutveckla dessa modeller för att ingå i vårdprogrammet

för njurcancer och förbättra livskvaliteten hos dessa patienter.

vi

Abbreviations

CIMP – CpG Island Methylator Phenotype

CIP – Cumulative Incidence of Progress

CpG – Cytosine-phosphate-Guanine

CSS – Cancer-Specific Survival

ccRCC – clear cell Renal Cell Carcinoma

CNV – Copy Number Variations

DCA – Directed Cluster Analysis

HRP – Hig Risk for Progress

LRP – Low Risk for Progress

M0 – Non-metastatic tumor at diagnosis

M0-PF – Non-metastatic tumor at diagnosis that will not progress, ‘true’ M0.

M0-P – Non-metastatic tumor at diagnosis that later will progress.

M1 – Metastatic tumor at diagnosis

PCA – Principal Component Analysis

PI-CpGs – CpGs previously shown to be associated with ccRCC

PMC – Promoter Methylation Classifier

PFS – Progression-Free Survival

RCC – Renal Cell Carcinoma

T-tissue – tumor tissue

TF-tissue – tumor-free tissue

vii

Original Papers

List of original papers included in this thesis:

I. DNA methylation status defines clinicopathological parameters

including survival for patients with clear cell renal cell carcinoma

(ccRCC).

Evelönn EA, Degerman S, Köhn L, Landfors M, Ljungberg B,

Roos G

Tumor Biol. (2016) 37:10219-10228

II. DNA methylation associates with survival in non-metastatic clear

cell renal cell carcinoma.

Evelönn EA, Landfors M, Haider Z, Köhn L, Ljungberg B, Roos G,

Degerman S.

BMC cancer (2019) 19:65

III. Combining epigenetic and clinicopathological variables improves

prognostic prediction in clear cell Renal Cell Carcinoma

Evelönn EA*, Vidman L*, Källberg D, Landfors M, Liu X, Ljungberg

B, Hultdin M, Degerman S#, Rydén P#

Manuscript

Reprints were made with permission from the respective publishers.

*# Contributed equally

viii

Published articles not included in the thesis:

• Wilms’ tumor 1 can suppress hTERT gene expression and telomerase

activity in clear cell renal cell carcinoma via multiple pathways.

Sitram RT, Degerman S, Ljungberg B, Andersson E, Oij Y, Sugiyama H,

Roos G, Li A.

Br J Cancer. 2010 Oct 12;103(8):1255-62

• Immortalization of T-cells is accompanied by gradual changes in CpG

methylation resulting in a profile resembling a subset of T-cell leukaemia.

Degerman S, Landfors M, Siwicki JK, Revie J, Borssén M, Evelönn EA,

Forestier E, Chrzanowska KH, Rydén P, Keith WN, Roos G.

Neoplasia. 2014 Jul;16(7):606-15.

1

Introduction

Clear Cell Renal Cell Carcinoma

Kidney cancer is the 14th most common type of cancer worldwide and 8th most

common in the western world (1). Renal cell carcinoma (RCC) accounts for

approximately 80% of kidney cancers and the most common subtype is clear cell

RCC (ccRCC). In Sweden, approximately 1 200 individuals are yearly diagnosed

with kidney cancer (2). Over 80% of patients diagnosed with ccRCC have

localized disease at diagnosis and therefore have a good prognosis. Out of these,

30% will later be burdened with tumor progression. One of the challenges today

is to identify these patients and give them the best outcome and quality of life.

The incidence of RCC differs between countries, the highest incidence is seen in

the Czech Republic and the lowest in China, Thailand, and African countries (2,19

vs. <2 per 100,000, age-standardized rate for males) (3). However, differences

between parts of countries and between ethnicities within countries are also seen

(1,3). RCC is more common in men than in women (3).

In a review by Scelo and Larose (2018) a list of risk factors for kidney cancer were

evaluated (3). Tobacco smoking, excess body weight, hypertension, and chronic

kidney disease are the most common and well-defined risk factors (3). Hereditary

traits have also been observed for kidney cancer, with 3-5% of cases occurring

within families (4). For ccRCC, there are at least six different syndromes

associated with the disease, with von Hippel Lindau (VHL) disease as the most

common (1,4,5).

The classic triad of symptoms associated with kidney cancer, including ccRCC,

are flank pain, haematuria, and a palpable abdominal mass. However, these

symptoms often occur late in tumor development. Other diffuse symptoms of

ccRCC are paraneoplastic syndromes with hypercalcemia, unexplained fever,

erythrocytosis and/or Stauffer’s syndrome with cholestasis unrelated to tumor

infiltration of the liver. Today, the diffuse symptoms associated with kidney

cancer in combination with more extensive use of diagnostic imaging in the clinic

have led to over 50% incidentally detection of kidney cancer. Patients that seek

medical attention due to abdominal symptoms often get examined with

computed tomography (CT), ultrasound or magnetic resonance imaging (MRI)

and small tumor masses on the kidneys can then be detected even though they

are not the cause of the problem.

2

Early detection of cancer is of importance since the opportunity to treat with

curative intention decreases with tumor progression. In ccRCC small tumors

located in the kidney without spread to regional lymph nodes, vena cava and/or

distant organs have the best prognosis. Localized ccRCC has a 5-year survival rate

of up to 90% whereas if metastasized the prognosis is poor with a survival rate of

10%. The high frequency of incidentally diagnosed tumors has led to a decrease

in metastatic disease, from 30% at the beginning of the 21st century to 18% around

the year 2010 in Sweden. At diagnosis, all patients are treated according to a

standardized process described in the Swedish Health Care Program for kidney

cancer (2).

The diagnosis is set for each patient, based on results from CT scans and later

from histological examination of the dissected tumor. From CT scans tumor

diameter is measured, the localization and growth pattern of the tumor is

determined and involvement of lymph nodes, signs of tumor growth into the

renal veins and signs of distant metastasis are also analyzed. This information is

then used to stage tumors according to the TNM classification; Tumor (T), Node

(N) Metastasis (M) (Figure 1) (6).

Non-metastatic disease is treated with either partial or total nephrectomy. Small

(<7 cm in diameter) tumors can be treated with partial nephrectomy (PN) to save

the non-affected kidney tissue and thereby spare kidney function. If the tumor is

locally advanced or if the tumor is disadvantageously located in the kidney,

radical nephrectomy (RN) is performed. Local lymph nodes and the adrenal

gland are vacated if there are signs of involvement from CT scans or during the

clinical evaluation in surgery. The adrenal gland is saved if there is no sign of

tumor overgrowth (2).

Figure 1. Simplified representation of Tumor, Node, Metastasis (TNM) staging in ccRCC. TNM I:

Small tumor (T), <7 cm in diameter, limited to the kidney; TNM II: Larger tumor, > 7 cm in

greatest diameter, limited to the kidney; TNM III: Tumor of any size that involve major veins

and/or adrenal glands, limited by Gerota’s fascia surrounding the kidney and might involve one

lymph node (N); and TNM IV: Tumor growing beyond Gerota’s fascia or involve more than one

lymph node or have distant metastasis (M).

3

If the patient has metastatic disease at diagnosis, the best possible treatment plan

is finally determined by a team composed of a urologist, oncologist, radiologist

and sometimes a pathologist. Some patients benefit from nephrectomy even

though it is not with a curable intention but to prolong survival and quality of life.

These patients might also benefit from systematic treatment and/or surgical

removal of metastasis (2).

After nephrectomy, tumors are analyzed by pathologists to determine the type of

kidney cancer the patient is burdened by. In the case of RCC, there is further

subclassification into clear cell (70-80%), papillary (10-15%), chromophobic

(approx. 5%) or one of the more uncommon collecting duct carcinoma or

medullary carcinoma that constitute for 0.5% of RCC cases each. If evaluated as

a ccRCC the tumor is staged according to Fuhrman grade (7), level of sarcoid

tissue, presence of necrosis and vessel infiltration. The pathologist also evaluates

the tumor size and determine if there are abnormalities in the surrounding tissue

to verify that the tumor has been radically removed.

Even if a patient has non-metastatic disease at diagnosis, approximately 30% of

patients will later develop metastasis. Tumor progression is most common within

three to five years post-nephrectomy with 80-90% tumor progression events

occurring within this time period. Therefore, these patients are scored according

to the Mayo Scoring System that includes information of the primary tumor (i.e.

T-stage), tumor size, involvement of regional lymph nodes (i.e. N-stage),

Fuhrman Grade, and the presence/absence of necrosis (8). Depending on the

Mayo score, patients are divided into three groups with different risk of tumor

progression; low, intermediate or high risk. Post-surgical follow up differs

between these risk groups, with more frequent and longer follow up in the

intermediate and high-risk groups in comparison to the low-risk group. At the

moment, only one study indicates that adjuvant treatment is beneficial in non-

metastatic ccRCC before tumor progression. It is therefore, not recommended to

enroll patients without progress in adjuvant treatment (4,9,10).

In contrast to non-metastatic ccRCC, treatment for metastatic ccRCC has shown

to be beneficial throughout the years with prolonged overall survival when

combining nephrectomy and/or systematic treatment. Radiation and

chemotherapy have low or no effect in treating ccRCC whereas the development

of systematic treatment with tyrosine kinase inhibitors (TKI), mTOR-inhibitors

and immunological treatment have been of importance to prolonged survival (11-

13).

4

Genomic traits in ccRCC

Genomic aberrations and mutations are present in ccRCC tumors in a

heterogeneous pattern and at least 90% of tumors show at least one genetic

alteration (14-16). Even though several different copy number variations and

genetic mutations are published, only a small number of them are common within

the ccRCC population. Moreover, few studies have shown a correlation between

genetic aberrations and aggressiveness or outcome in a reproducible matter (17).

Studies investigating intratumoral heterogeneity have shown that the genetic

landscape can differ widely depending on where in the tumor the sample is taken

from (17,18).

Genomic aberrations

The most common genomic aberration in ccRCC is the allelic loss of the

chromosome arm 3p, which is present in approximately 90% of ccRCC tumors

(15,19). Other genetics aberrations are presented in variable frequencies; loss of

the whole chromosome 4, 9, 19, 20 and 22, as well as losses of regions within 1p,

4p, 4q, 9p, 9q, 13q, and 14q, have been reported. Chromosomal gains are reported

in less quantity with a gain of chromosome arms 1q, 3q, 5q, 7q, 8q, and 20q, as

well as gain of whole chromosome 5, as the most commonly amplified regions

(17,20-25).

Köhn et al., (2014) analyzed a subset of ccRCC patient samples included in this

thesis for genomic aberrations and identified twelve genomic regions altered in

at least 20% of samples (22). Loss of chromosome 3p was the most common

alteration present in 88% of the ccRCC samples. The gain of a region on

chromosome arm 7q and a loss of regions on chromosome arms 9p and 9q were

all correlated to poor outcome (22).

Common mutations in ccRCC

The most common genetic mutations present in ccRCC are that of VHL, PBRM1,

SETD2, BAP1, KMD5C, TERT, PTEN, mTOR, and TP53 genes, which are present

in a range from 2% (PTEN and TP53) to 90% (VHL) of cases (14,16).

The VHL, PBRM1, SETD2, and BAP1 genes are all located on chromosome 3p.

One allele of the VHL gene is lost as an early event during tumor development,

whereas the silencing of the other allele occurs later in tumor progression via

either a frameshift mutation or epigenetic alterations (14). Correlation between

VHL gene inactivation and clinical outcome has not been reported (5).

5

The PBRM1 gene encodes for a subunit in the nucleosome remodeling complex

and its mutation rate is reported in up to 50% of ccRCC tumors (26). Mutations

in PBRM1 can result in abnormal/malfunctioning gene product resulting in

unchecked cell growth (27). SETD2 encodes for a histone H3K36 lysine

trimethyltransferase (28) and has a mutation rate of up to 30% in ccRCC (26).

Inactivation of SETD2 is correlated to poor overall survival and increased risk of

tumor progression in the TCGA-KIRC cohort (29).

BAP-1 mutations are present in up to 15% of tumors and it is known that BAP1 is

involved in cellular proliferation and regulation of DNA damage repair.

Mutations have been reported to be correlated to a worse prognosis (26).

The telomerase reverse transcriptase (TERT) gene plays an important role in

telomere lengthening and is mutated in up to 14% of ccRCC tumors (26). Relative

telomere length in tumor tissue has been positively correlated to larger tumor

diameter but not to the clinical outcome (30).

The PI3K-mTOR pathway plays an important role in proliferation, apoptosis, and

migration in ccRCC (26). The PTEN and mTOR genes are key regulators of this

pathway and are mutated in 2-12% of ccRCC tumors. Mutations in the TP53 gene

occurs in 2-9% of ccRCC tumors, and mutations within this gene are correlated

to metastatic disease and decreased cancer-specific survival (26).

Epigenetics

Epigenetics was introduced as a biological term in the early 1940s and defined by

Waddington as “the branch of biology which studies the causal interactions

between genes and their products which bring the phenotype into being”.

Epigenetic research has evolved since then and is now known to regulate gene

expression without altering the DNA sequence (31). In the human body, there are

over two hundred different cell types and all cells contain, more or less, the same

genomic sequence. Gene expression depends on epigenetic regulation amongst

other regulatory events. Epigenetics also play an important role in the

stabilization and packing of DNA, silencing of centromeres, telomeres, and

transposable elements (32).

6

The key epigenetic regulations within a cell are histone modifications, DNA

methylation and RNA modifications (Figure 2). The interplay between epigenetic

modifications is tightly regulated and is a complex machinery with writers,

readers, and erasers. Epigenetic writers are effector proteins that modify DNA

and/or histones. Readers recognize these modifications leading to transcriptional

activation or repression of said region. Since the epigenetic modifications are

dynamic reversible processes, epigenetic erasers exists that can remove the

modifications set by writers and thereby altering transcription (32,33).

Figure 2. A schematic and simplified presentation of the three most common epigenetic

modifications, i.e. DNA methylation, histone modifications, and RNA modifications.

7

Histone modifications

The DNA strand is wrapped around an octamer of histones i.e. two units of H2A,

H2B, H3, and H4, respectively giving rise to a nucleosome with histone H1 as a

link. Nucleosomes then coil into chromatin fibers that are further condensed and

forms the chromosome (Figure 2). Each histone constitutes of a core and a

histone tail which is accessible for modifications such as acetylation, methylation,

phosphorylation, ubiquitination, and sumoylation (34).

Acetylation, methylation, and phosphorylation are the most common histone

modifications. The combination of modifications affects how tightly the DNA is

wrapped around the histones making genes more or less accessible for

transcription (34). Acetylation of the tails of histone H3 and H4 are done by

histone acetyltransferases (HATs). Acetylation of H3K9/14/18/23, as well as

H4K5/8/12/16 are all known to activate transcription by inducing euchromatin

formation and making DNA accessible to transcription factors. Acetylation is a

reversible modification and is reversed by histone deacetylases (HDACs). HDACs

can also deacetylate other proteins and in that way affect cellular functions

(34,35).

Histone tail methylation can both activate and repress transcription. Methylation

of H3K4/36/79 are all related to active transcription whereas, H3K9/20/27

causes gene silencing. Histone methylation is mediated by histone

methyltransferases (HMTs) that bind to specific DNA sequences and interacts

with Trithorax group (Trx) proteins, the polycomb group (PcG) proteins, and/or

RNA-interference (RNAi) to mediate histone methylation. One example of this is

the EZH2 protein that binds to methylated DNA and attracts polycomb target

(PcG) proteins by binding SUZ12 and EED. EZH2, SUZ12, and EED form the

polycomb repressive complex 2 (PRC2) along with associated proteins, and cause

methylation of H3K27 inducing a repressive chromatin state (34).

Phosphorylation of histones plays a crucial role during mitosis and meiosis and

also interacts with other histone modifications affecting gene transcription. All

histones can be affected by phosphorylation at different sites at their tails.

Phosphorylation of H3S10 in combination with H3S28 are important in both

mitosis and meiosis when they play a role in chromosome condensation (36).

8

RNA epigenetic modifications

Both coding and non-coding RNAs (ncRNAs) play an important role in epigenetic

regulation of gene expression and can also be subjected to epigenetic alterations

themselves. The most common and well-characterized RNA epigenetic

modification is methylation on adenosines (m6A). The writers that catalyze m6A

methylation methyltransferase-like 3 and 14 (METTL3/14) with the help of s-

adenosylmethionine (SAM). m6A methylation can alter mRNA splicing, export,

decay, stabilization, and translation (37). Epigenetic alterations of DNA

accessibility might also affect transcription of ncRNA.

Non-coding RNA (ncRNA) constitutes 80% of transcripts from the DNA and do

not form functional proteins. There are different kinds of ncRNA with different

functions, but they can all play a role in gene function and epigenetics.

The role of Xist, a long ncRNA (lncRNA), is important in X chromosome

inactivation in females. Xist binds to the X chromosome and recruits PRC2 to

induce inactivation (38).

Micro RNAs (miRNA) are single-stranded RNAs that can interact with RNA

before translation and induce its degradation. miRNA can also change the

chromatin state by altering remodeling enzyme activity (39) and by modifying

histone proteins (40). The activity of EZH2 and the formation of heterochromatin

can be regulated by miRNA and is correlated to tumor progression (41).

Piwi-interacting RNA (piRNA) binds to Piwi, a protein important in the

recruitment and binding of PcG proteins, and plays a role in euchromatin

formation (39).

RNA-induced silencing (RISC) complex constitutes small RNA and argonaute

protein and can inhibit the translation of RNA via degradation and/or non-

degradative mechanisms. The RISC complex is also involved in

heterochromatinization of pericentromeric regions when it recruits important

components, i.e. SUV39H1 and SUV39H2, to these regions (42).

9

DNA methylation

The most well studied epigenetic alteration is DNA methylation. It was first

described by Avery et al., in 1944 (43). Methylation occurs at the 5th carbon

position of cytosines (C), most often on cytosines located next to a guanine (G)

which forms a so-called CpG site (44). In embryonic stem cells, 25% of

methylated cytosines are located in the non-CpG context and seem to be

important for maintaining cellular pluripotency. They are demethylated during

cell differentiation (45). Distribution of CpGs throughout the genome varies and

depending on the amount of CG-content areas are denoted CpG Islands (CGIs),

CpG Shores, CpG Shelves, or open sea. CGIs have a higher level of CpGs then

expected by random and about 60-70% of genes have a CGI in their promoter

(46). Non-coding regions of the DNA harbors around 80% of the CpGs and are

located in intergenic regions, satellite DNA, and transposable repetitive elements.

DNA methylation in these regions acts to stabilize DNA by protecting against

transposon activity (47).

DNA methylation pattern can either be inherited or can occur de novo, and the

pattern is established by DNA methyltransferases (DNMTs). DNMT1 ensures the

maintenance of DNA methylation through cellular division and DNMT3A and

DNMT3B implement de novo methylation after the recruitment of DNMT3L

(48,49). As other epigenetic alterations, DNA methylation is a dynamic process

where demethylation is mediated by ten-eleven translocation protein (TET1-3)

via hydroxy-methylation steps (50).

The epigenetic alterations regarding both histone modifications and de novo

methylation are tightly related to each other. As reviewed by Ceder and Bergman

(2009) the interplay between histone modifiers and DNA methyltransferases is

tightly regulated (51). During embryogenesis RNA polymerase II binds to

promoter regions and initiates H3K4 methylation, which in turn inhibits

DNMT3L binding, causing CpGs to remain unmethylated (52). In

pericentromeric satellite repeats, the DNA forms heterochromatin due to H3K9

methylation that also attracts DNMT3B to further stabilize the heterochromatin

(53). Methylated DNA attracts methylation binding protein (MBP) that can

repress transcription by interfering with the binding of transcription factors as

well as induce H3K9 methylation to maintain a repressive chromatin state (54).

Histone acetylation is also correlated to DNA methylation. Methylated DNA is

wrapped around non-acetylated H3 and H4 while unmethylated DNA is

associated with euchromatin and acetylated histones (42).

10

The role of DNA methylation in gene transcription is yet to fully be established. A

high proportion of genes have CGI in their promoter, which is most often

unmethylated making the gene accessible for transcription. Methylation of a CGI

located in a gene promoter regulates gene silencing by inhibition of transcription

factor binding, initiation of an inactive heterochromatin state, and/or by

attracting the Kaiso-like family (KLF) proteins involved in gene silencing (47).

De novo methylation is affected by environmental, stochastic and/or genetic

events and might also be involved in complex diseases (55-58). DNA methylation

patterns change over time and are associated with the mitotic age of cells as well

as the chronological age of an individual. Different bioinformatics tools have been

developed to calculate the biological and mitotic age of tissues (59-61).

Some DNA methylation positions in the genome seem to be associated with

ethnicity and gender. These positions are called methylation quantitative trait

loci (mQTL) and have been identified by analyzing DNA methylation in blood

samples from several time points in large population-based cohorts (62). mQTLs

show a consistent pattern with fewer mQTLs at birth compared to childhood but

then the number decreases with increased age (62). The regulatory effect of

mQTLs is yet to be investigated but there is evidence of contributions of mQTLs

in complex diseases i.e. Crohn’s disease, hypertension, rheumatoid arthritis, and

infectious diseases (62,63).

DNA methylation in cancer

DNA methylation and its correlation to cancer was first described in 1983 when

Feinberg and Vogelstein identified a reduction of methylation of specific genes in

colon cancer (64). Since then the research area concerning methylation and

cancer progression has increased drastically. A number of methods to analyze

DNA methylation alterations have been developed, e.g. DNA methylation arrays,

bisulfite sequencing, and methylation-specific PCR. Tumor cells have a different

DNA methylation landscape compared to normal cells and the alterations

correlated to tumorigenesis include global hypomethylation and CGI

hypermethylation (65,66).

De novo hypomethylation that occurs in methylated and inactivated regions leads

to a higher risk of mutagenesis by induced genomic instability. It has been shown

that DNA hypomethylation can activate oncogenes and drive tumor progression

(67). Hypomethylation has also been reported to affect ribosomal DNA which

alters the expression of different subunits of the ribosomes and thereby nucleolar

function within cells (68).

11

Methylation alterations are known to be an early event in tumorigenesis. Arai et

al., (2012) identified differences in methylation patterns between histological

normal tissue from a tumor-bearing kidney compared to the kidney cortex from

a healthy kidney (69). Altered methylation has also been identified in histological

normal colon polyps adjacent to colorectal cancer (70) and Sato et al., (2013)

described DNA methylation patterns that differed between histological normal

lung tissue from a tumor-free lung compared to tissue from a tumor-bearing lung

(71).

Hypermethylation of unmethylated CGI located in gene promoters might affect

gene expression. Even though the genes affected by methylation alterations differ

between tumors, the gene ontologies for effected genes often are involved in cell

cycle control, apoptosis, DNA damage, immune recognition, and cell self-renewal

(33,66,72). The polycomb gene EZH2 binds to a wide set of target promoters and

these target genes are often affected by aberrant DNA methylation in different

types of tumors. Overexpression of EZH2 has been reported in prostate, bladder,

gastric, lung, hepatocellular cancer, and breast cancer and are correlated to poor

outcome (41).

The correlation between DNA methylation patterns and clinical outcome has

been studied in several types of cancer and has been evaluated in relation to

tumor progression, clinical outcome, and response to treatment. Locke et al.,

(2019) have summarized commercially available screening methods of “liquid

biopsies” that test for methylation of circulating tumor DNA (ctDNA) for several

tumor types (73). Methods have been described for bladder cancer, breast cancer,

cervical cancer, colorectal cancer, glioblastomas, liver cancer, lung cancer, and

prostate cancer with the goal to detect primary tumors, tumor progression after

treatment, and response to therapy (73). DNA methylation-based method can

also be used for the classification of Central Nervous System (CNS) tumors (74).

Capper et al., (2018) have presented a method that classifies unusual, non-

specific or non-representative histological CNS tumors (75).

The CpG Methylator Phenotype (CIMP) was first described for colorectal cancer

(CRC) in 1999 by Toyota et al., (76). CRC tumors were divided into CIMP low and

CIMP high depending on their CGI methylation levels (76). Since then, the CIMP

classification method has been used in several different tumor types, including

gastric, lung, liver, ovarian, kidney, and leukemias (77,78). CIMP panels differ

between tumor types, with different genes and/or CpGs included in the different

panels, and with a wide range in the number of included CpGs (77,78).

12

DNA methylation in ccRCC

DNA methylation plays an important role in tumorigenesis in ccRCC. More than

seventy genes, summarized by Ricketts et al., (2014), are thought to be affected

by DNA methylation (79). These epigenetically affected genes are involved in

pathways associated with RCC carcinogenesis, i.e cell cycle regulation, apoptosis,

and cell metabolism amongst others (80).

The VHL-HIF (hypoxia-inducible factor) pathway plays an important role in

renal carcinogenesis since it affects angiogenesis, proliferation, apoptosis,

survival, metabolism, and pH regulation. Important mediators in this pathway i.e

VHL, TIMP3, and GREM1 amongst others, are affected by DNA methylation

leading to its dysregulation (80).

The WNT-β-catenin signaling pathway regulates DNA-binding transcription

factors and is tightly regulated. In RCC, WNT antagonists are commonly

inactivated by DNA methylation, leading to activation of the WNT-β-catenin

pathway. These events are associated with increased levels of ongogenic target

genes and tumorigenesis (14,80).

Epithelial-mesenchymal transition (EMT) plays an important role in

embryogenesis and also in pathological processes like wound healing, tissue

fibrosis and cancer progression (81,82). In carcinomas, EMT is thought to be

necessary for the invasion-metastasis cascade. Histone modifications, DNA

promoter alterations, and post-translational activity are important for EMT by

changing the levels of transcription factors (i.e. SNAIL, SLUG, and Twist), cell

surface proteins, cytoskeleton structure, and miRNA expression (81,82). The

best-studied gene important for EMT is CDH1. The expression of CDH1 can be

inactivated by promoter methylation, inducing EMT as a result. Correlation of

hypermethylation of CDH1 and metastatic disease has been shown in breast,

bladder, lung, liver, gastric, and prostate cancer (82).

Methylation biomarkers for ccRCC

As established above, DNA methylation plays an important role in ccRCC tumor

progression. Several differently methylated genes and/or combinations of these

genes have been shown to be involved in ccRCC tumor progression and related to

clinical outcome (69,79,83-86).

Arai et al., (2012) published a CIMP panel that predicted outcome in ccRCC.

Tumor-free and tumor tissue were analyzed using the 27K arrays, differently

methylated CpGs were identified and cluster analysis was used to separate tumors

into two distinct groups. Sixteen CpGs with the highest difference in methylation

patterns between the two clusters were included in the CIMP panel (69). Later,

Tian et al., (2014) evolved this panel by using the MassARRAY system and

13

identified 23 units (more than one unit per gene) including 32 CpGs representing

seven of the primary identified CIMP genes (83).

Wei et al., (2015) created a five CpG methylation-based assay to predict overall

survival in ccRCC (84). They used two different data sets to build their classifier

and could later verify the data in three independent cohorts. This classifier was

also investigated for intratumor heterogeneity (ITH) with accurate classification

of all pieces from the same tumor in 21 of 23 patients (84).

A four-gene methylation profile was described by van Vladrop et al., (2016) that

was associated with ccRCC disease and/or poor outcome. The panel; constituted

of the GREM1, LAD1, NEFH, and NEURL genes was created using subsequent

nested multiplex methylation-specific PCR in two different cohorts (n=150 and

185 respectively). The panel was also verified in the TCGA-KIRC data set using

one single CpG site for each of the four included genes, indicating its robustness

as a prognostic marker (86).

Ricketts et al., (2014) summarized and analyzed DNA methylation of 77 genes

known to be differently methylated in kidney cancer (79). They used both 27K

and 450K arrays to analyze CpGs represented by these genes. By both methods,

methylation of SFRP1 and BNC1 were shown to be associated with poor survival

in ccRCC (79).

Joosten et al., (2017) performed a systematic review of genes methylated in RCC

(85). By this approach, nine genes were identified that correlated to progression-

free survival, disease-specific survival and/or overall survival. The conclusion of

this review highlighted the importance of using a combination of several genes to

predict outcome in ccRCC (85).

Today, the clinical use of DNA methylation as a prognostic marker for ccRCC is

not established. The importance of epigenetic alterations, including DNA

methylation, is of big interest in ccRCC where there is strong evidence of its

importance in tumorigenesis.

In this study, we identified two different methylation-based classifiers that

predict tumor progression in non-metastatic ccRCC.

14

Aims

The frequency of metastatic ccRCC at diagnosis has decreased since the beginning

of the 21st century but still, 20-30% of non-metastatic tumors progress into

metastatic disease. There is a need for better risk stratification tools to be able to

identify the non-metastatic tumors with high risk for tumor progression and treat

them accordingly to increase the quality of life for these patients.

The aim of this thesis was to evaluate the DNA methylation alterations at

diagnosis and to correlate these epigenetic modifications with clinicopathological

parameters and clinical outcome in ccRCC.

Paper I

The aim of paper I was to determine if DNA methylation can be used as a potential

prognostic marker in ccRCC.

Paper II

The aim of paper II was to evaluate the prognostic relevance of DNA methylation

in relation to clinical characteristics in ccRCC, with a special focus on non-

metastatic patients at diagnosis.

Paper III

The aim of paper III was to define a diagnostic model based on a combination of

DNA methylation profiling and clinicopathological variables for predicting tumor

progression after nephrectomy in non-metastatic ccRCC.

15

Materials and Methods

Workflow describing patients and methods included in this

thesis

Figure 3. Workflow for analysis

Patients and tissue samples

All tissue samples from clear cell Renal Cell Carcinoma (ccRCC) patients were

obtained after written, informed consent. Ethical approval for the studies was

given by the regional ethical review board in Umeå (Dnr 2011-156-31M,

20110523).

The ccRCC patients included in this thesis (summarized in Table 1) were

diagnosed between 2001 and 2009. Patients were treated with partial or total

nephrectomy without neoadjuvant treatment and followed according to the

Swedish health care program (2). Tissue samples were collected after surgery and

snap-frozen in fluid nitrogen and stored at -80°C. Both tumor (T) tissue and

histological normal tumor-free (TF) tissue from the tumor-bearing kidney were

collected. Tumors were staged according to the 2002 TNM classification system

(6) and the nuclear grade was determined according to Fuhrman et al., (1982) (7).

Survival data from all included patients were extracted in March 2014 (paper I)

and in August 2017 (paper II and III)

16

Tissue samples included in paper I

Paper I included 45 patients; 19 non-metastatic (M0) and 26 metastatic (M1)

tumors from which T samples were collected. From six M0 and six M1 patients,

TF tissue was also collected.

Blood samples from 24 M0 patients were collected and stored as buffy coats

at -80°C. Thirteen ccRCC patients were analyzed for methylation in both T tissue

and blood samples.

Tissue samples in papers II and III

Papers II and III included 115 patients; 87 non-metastatic (M0) and 28 metastatic

(M1); and 12 TF tissue samples from eight M0 and four M1 patients. Out of the

87 non-metastatic patients, 23 developed metastasis after nephrectomy (M0-P)

whereas 59 remained ‘true’ non-metastatic (M0-PF).

Thirty-nine patients included in paper I were also included in papers II and III.

For analyzing intratumor heterogeneity (paper II), multiple samples (two or

three) from the same individual were analyzed with 450K arrays (paper II).

Tumor samples from the Tumor Cancer Genome Atlas (Paper

II)

In Paper II, the publically available TCGA-KIRC dataset was used as a validation

cohort. Information from 230 non-metastatic ccRCC patients including

methylation data from 450K arrays and clinical information were downloaded

from the Broad Institute’s Genome Data Analysis Center Firehose

(http://gdac.broadinstitute.org/). All included patients were treated with partial

or total nephrectomy without neoadjuvant or adjuvant treatment.

Blood analysis

Paper III included blood samples taken within a week prior to nephrectomy. The

blood samples were analyzed for hemoglobin levels (Hb), thrombocyte particle

count (TPC), calcium, albumin, alkaline phosphatase (ALP), gamma-glutamic-

transferase (GGT), and creatinine concentrations.

DNA extraction

DNA was extracted from T, TF tissue and blood samples as described by Svenson

et al., (2009) (30) using the MagAttract technology (Qiagen).

17

Methylation analysis

Bisulfite conversion (Papers I-III)

Purified DNA from different tissues were bisulfite treated prior to DNA

methylation analysis using the EZ DNA Methylation Kit (Zymo Research, Irvine,

USA). The efficiency of the bisulfite conversion was verified by MethyLight

analysis of the ALU gene (87).

Methylation array analysis (Papers I-III)

Two hundred nanogram of bisulfite converted DNA was analyzed on either the

Illumina Infinium HumanMeth27K Bead Arrays (27K arrays, Paper I) or

HumanMeth450K BeadChip Arrays (450K arrays, papers II and III) according to

manufacturer’s protocol (Illumina In., San Diego, USA).

The 27K arrays were scanned and the fluorescence was measured using a Bead

Array Reader (Illumina) and methylation levels were analyzed using the Genome

Studio Software (version 2011.1 and the methylation analysis module (version

1.9.0). The 450K arrays were scanned on the HiScan array reader (Illumina) and

fluorescence intensities were extracted using the Genome Studio software

(V2011.1) and the Methylation module (1.9.0).

The fluorescence intensities from methylated (M) vs. unmethylated (U) probes

were used to calculate a methylation level (β-value) (β= 𝑀𝑎𝑥(𝑀, 0)/(𝑀𝑎𝑥(𝑀, 0) +

𝑀𝑎𝑥(𝑈, 0) + 100) for each CpG site ranging from 0 (completely unmethylated) to

1 (completely methylated). β-values from the 450K arrays were normalized using

the BMIQ method (88).

Technical reproducibility was analyzed in the 450K arrays by including replicate

samples, two or three samples from nine individuals. Duplicate and triplicate

samples showed a high correlation with R2 values that ranged from 0.995 to

0.997.

18

Data preprocessing

Data filtration was performed for both array types. CpGs lacking observations in

any sample and CpGs with a detection p-value greater than 0.05 were omitted.

To avoid gender bias CpGs located on chromosomes X and Y were omitted. For

the 450K arrays, CpGs aligned to multiple loci or located within 3 bp from a

known single nucleotide polymorphism (SNP) were also omitted (89) as well as

CpGs located outside the promoter region. In paper III CpGs known to be located

in a methylation quantitative trait loci (mQTLs) were excluded from the analysis

(62).

In paper II, mQTLs were excluded from the Promoter Methylation Classification

(PMC) panel.

Array data analysis

Methylation array data analysis was done using R (v2.15.0 paper I, v3.4.1 papers

II-III and v3.4.3 paper III).

Cluster analysis was performed using Ward’s method with Euclidean distance

metric (90), including all promoter-associated CpGs; 21,930 (paper I) and

155,931 (paper II).

Biological epigenetic age was calculated for T and TF tissue samples as well as

blood samples by the method presented by Horvath (2013) for patients in paper

I (60). The EpiTOC model was used in paper II to determine the mitotic age of T

and TF tissue samples (61).

Differently methylated CpGs (DM-CpGs) were identified by comparing T samples

to TF samples (∆𝛽 = 𝛽𝑇 − �̅�𝑇𝐹). Average methylation (�̅�𝑇𝐹) for included TF

samples was calculated for each study, i.e. papers I and II, and compared to each

individual T sample. A CpG site was established as DM if it had an absolute

Δβ≥0.35 (Paper I) or an absolute Δβ≥0.2 (paper II). DM-CpGs with a positive Δβ

was defined as hypermethylated and to DM-CpGs with a negative Δβ was defined

as hypomethylated.

The building of prognostic classifiers

Two different classifiers that can be used as tools for the prediction of tumor

progression in non-metastatic patients are presented in this thesis.

19

Promoter Methylation Classification (PMC) Panel (Paper II)

DM-CpGs present in at least 70% of samples within each group of samples (i.e.

M0-PF, M0-P, and M1) were selected for further analysis. DM-CpGs common for

M0-P and M1 (n=172) but not DM in M0-PF were defined as a Promoter

Methylation Classification (PMC) panel (Figure 10).

The average methylation for all 172 CpGs, with hypomethylated CpGs mirrored

(1 – average beta), was calculated for all M0 tumors. A ROC curve was created

using the average methylation as a test variable and tumor progression as a state

variable. From this curve, a Youden index was calculated to determine a cut-off

(average beta=0.688) for PMC high and low classification. The TCGA-KIRC data

set was used to evaluate the PMC panel, missing values after preprocessing were

imputed using the k-nearest neighbors’ method.

Triple Classifier (Paper III)

Classifiers for tumor progression were built by using a combination of

clinicopathological variables (clinical), CpGs previously shown to be associated

with ccRCC prognosis (PI-CpGs) and by directed cluster analysis (DCA).

The CpGs previously associated with ccRCC progression were chosen from five

original publications and one review (69,79,84-86,91). In total 64 CpGs were

chosen to be included and denoted as PI-CpGs.

The DCA method identified cluster of CpGs with potential relevance for

prognosis. Each individual CpG site was separated into two groups using 2-

means. CpGs with a skew deviation (less than 10% of samples in the smaller

group) and with an absolute mean difference (Δβ) between groups less than 0.2

were omitted from the analysis. The group labels of the remaining CpGs were

adjusted such that the groups with the highest mean methylation were all labeled

as 1 and the other group as 0. The group belongings (0, 1) were then clustered

using 40-means, giving rise to 40 DCA clusters. For each cluster, a consensus

variable was calculated as the average methylation of all included CpGs.

In total, six different classifiers were built for M0 patients; clinical, PI-CpGs,

DCA, clinical+PI-CpGs, clinical+DCA, and clinical+PI-CpGs+DCA; by using

logistic regression on the five first Principal Components. Inclusion criteria for

patients were at least five year follow up time, n=78; 58 M0-PF and 20 M0-P. To

compare the classifiers the sensitivity was set to a fixed value of 85%.

20

Genetic analysis

Genetic aberrations (Papers I and II)

To evaluate overall genetic aberrations within ccRCC both CytoSNP-12 arrays

(CytoSNP-12, Illumina) and the 450K arrays were used.

In paper I information from the CytoSNP-12, as described by Köhn et al., 2014

(22), was used to evaluate the genetic aberration of the VHL gene located on

chromosome 3p and DNA methylation in the VHL gene promoter. In paper II,

the twelve most common genetic aberrations described by Köhn et al., 2014 were

analyzed using raw data signals from the 450K arrays and the Conumee package

(v1.9.0) in R (v3.4.1). The limitations in the detection of gains and deletions

within the set regions were done through visual inspection and was set for each

individual sample.

VHL status (Paper I)

Information concerning the mutation of the VHL gene was derived from Köhn et

al., 2014 were Sanger sequencing of the VHL gene was performed (22).

Gene expression analysis

RNA preparation (Paper II)

RNA was extracted from T tissue as described in detail in paper II using the

MagAttract RNA Universal Tissue M48 Kit (Qiagen, Hilden, Germany). The

quality of RNA was analyzed using the 2100 Bioanalyzer (Agilent Technologies,

Santa Clara, CA, USA). cRNA was produced using the Illumina TotalPrep RNA

amplification kit (Ambion Inc., St. Austin, TX, USA) and the quality was

evaluated using the RNA 6000 pico kit (Agilent Technology).

Gene expression arrays (Paper II)

Genome-wide gene expression analysis was performed using Human HT12

Illumina Beadchip gene expression array (Illumina) according to the

manufacturer’s protocol and the arrays were scanned with the Illumina Bead

Array Reader (Illumina). The Illumina BeadStudio V2011.1 software was used to

normalize the data by the cubic spline algorithm, and expression levels for three

selected genes i.e. MX2 (ILMN_2231928), SMAD6 (ILMN_1767068) and SOCS3

(ILMN_1781001) were extracted.

21

Statistical analysis

Statistical analyses were performed using the Statistical Package for the Social

Sciences (SPSS Inc., Chicago, IL, USA) software version 22 (Paper I) and version

24 (papers II-III). In paper I R (v2.12.0) was used, in paper II R (v3.4.1) and in

paper III both R (v3.4.1) and R (v3.4.3) were used.

To compare differences between patient groups Chi-square test was used for

categorical variables. Mann-Whitney U test was used to compare continuous

variables between two groups and the Kruskal Wallis test when comparing three

groups. Correlations between continuous variables were analyzed using the

Pearson correlation coefficient.

Survival analysis was performed using Kaplan-Meier and the Log rank test. Five

year cancer-specific survival (CSS5yr) and cumulative incidence of progress

(CIP5yr) were obtained from Kaplan-Meier survival tables. Multivariate Cox

Regression analysis was used to analyze the combination of several predictor

variables.

Genes associated with DM-CpGs (Paper I) and CpGs included in the PMC panel

(paper II) were analyzed for protein function using the Metacore software

(GeneGO, Inc.) in Paper I and WebGestalt (92) in paper II.

Similarities within and between tumors were determined using Principal

Component Analysis (PCA) using the average beta values for 156K CpGs from the

450K arrays. PCA was performed using SIMCA version 14 (Umetrics, Umeå,

Sweden).

Twelve common regions identified as genetic aberrations by the CytoSNP-12

arrays were also analyzed using the 450K methylation arrays. Results from the

two methods were compared using Cohen’s kappa test.

Distribution of DM-CpGs was analyzed against frequent gain/loss regions in

ccRCC defined by Köhn et al., (2014) (22) to evaluate if any overrepresentation of

DM-CpGs in affected regions. The correlation was analyzed using the Chi-square

test.

Identification confirmation of multiple samples from the same individual was

verified by the 65 built-in SNP probes in the 450K arrays.

22

Results

Demographic overview of the ccRCC cohorts.

Forty-five ccRCC patients (paper I) and 115 ccRCC patients (papers II-III) were

included in these studies. Clinicopathologic variables were collected for all

included patients and are briefly summarized in Table 1 (for details see each

individual paper). Differences in tumor progression frequencies of ccRCC

patients between papers II and III are due to different criteria for follow up. In

paper III, patients with follow up time less than five years or patients that died

within five years after diagnosis, were censored. (Table 1).

Table 1. ccRCC patient overview for paper I-III.

Patients Paper I (n=45)

Paper II (n=115)

Paper III1, 2 (n=115)

Age (mean ± SD) 64.4 ± 10.1 65.0 ± 11.5 Gender

Male Female

30 15

66 49

Tumor diameter (mm, mean ± SD)

95.4 ± 40.1 70.0 ± 41.6

Morphological Grade G1 G2 G3 G4

4 13 15 13

16 46 34 18

T-stage T1 T2 T3 T4

6 14 24 1

50 25 39 1

M-stage M0 M1

19 26

87 28

TNM Stage I

II III IV

5 8 6

26

49 14 24 28

Follow Up status Alive

Alive, disease Dead, ccRCC

Dead

9 1

31 4

39 5

45 26

Tumor Progression No (M0-PF) Yes (M0-P)

M0-censored

64 23

58 20 9

1 – 5-year follow up status 2 – 5-year tumor progression status

23

Loss of chromosome arm 3p and VHL genetic and epigenetic

status (Paper I)

Genomic aberrations are commonly observed in ccRCC tumor tissue and show

considerable heterogeneity. The tumors included in this thesis were screened for

the most common genomic aberrations presented by Köhn et al., (2014) (22).

Loss of one chromosome arm, 3p, is present in approximately 90% of ccRCC

tumors and inactivation of the VHL gene due to a frame-shift mutation, located

on 3p, is a common finding in ccRCC (19).

A subset of tumor (T) samples (n=38 out of 45) within the ccRCC cohort in paper

I, was previously screened for VHL mutation in the study by Köhn et al., (2014)

(22). Genomic aberration in the VHL gene was present in 85% of analyzed T

samples, 28% had an allelic loss of one VHL copy and 56% had both one allele

loss and a frame-shift mutation in the other copy of the VHL gene (Figure 4).

Methylation analysis of the VHL promoter (Paper I).

Methylation alterations in the VHL gene have been reported at different

frequencies (14-16). Our ccRCC cohort was characterized based on VHL promoter

methylation status. Five CpGs in the VHL gene promoter were analyzed using the

Illumina Infinium HumanMeth27K BeadChip Arrays (27K arrays). The VHL

promoter methylation pattern was homogenous with two CpGs (cg27226214 and

cg16869108) hypermethylated in more than 97% of tumor samples and two CpGs

(cg03509024 and cg22782492) hypomethylated in more than 88% of samples

compared to tumor-free (TF) tissue samples. One CpG site (cg24092914) was

hypomethylated in 20% of samples and was significantly correlated to metastatic

disease (p=0.004). There was no association between the presence of genetic and

epigenetic alterations in the VHL gene (Figure 4).

Figure 4. Percentage of DM-CpGs in the VHL promoter depending on VHL status, wild type (wt),

VHL loss or VHL loss and frame-shift mutation (VHL loss+mut).

24

Genomic aberrations and ccRCC disease progression (Paper

II)

Twelve frequent genomic aberrations in ccRCC were previously described by

Köhn et al., (2014) in a subset of this cohort (Figure 5) (22). These genomic

regions were analyzed in 115 ccRCC patients by using the raw data signal from the

HumanMethy450K BeadChip Arrays (450K arrays) to identify genomic gains or

losses. The most common aberration (84% of samples) was a loss of 3p,

encompassing the VHL gene. The other 11 analyzed aberrations present in 15 to

53% of samples (Figure 5). The distribution of genetic aberrations was compared

between the groups (M0-PF, M0-P and M1) and five regions showed a significant

difference in distribution between the groups. A gain of 7p and 7q were more

common in M1 samples compared to M0-P (p=0.013 and 0.024 respectively) and

loss of 9p, 9q, and 14q were more common in M1 compared to M0-PF (p=0.001,

p<0.001, and 0.042 respectively) (Figure 5). The total number of genetic

aberrations was significantly higher in M1 samples compared to both M0-PF and

M0-P samples (p=0.024 and 0.050 respectively). The number of genetic

aberrations showed heterogeneity within all prognostic groups. All groups

included tumors with both high and low numbers of genetic aberrations.

Figure 5. Allocation of common genomic aberrations across different prognostic groups. White

color indicates no aberrations and black color indicates gain or loss of specific region.

25

DNA methylation classification and its correlation to

clinicopathological parameters (Papers I and II)

DNA methylation arrays were used to evaluate methylation alterations for the

prognostic relevance of ccRCC progression. In paper I DNA methylation was

analyzed using the 27K arrays and in paper II the 450K arrays were used.

Hierarchical clustering analysis of genome-wide promoter-associated CpG sites

was used to classify ccRCC and tumor-free (TF) tissue samples into groups

depending on similarities in methylation patterns/profiles. In papers I and II

cluster analysis was performed using 21,930 (paper I) and 155,931 (paper II)

promoter-associated CpGs. In both studies, samples were divided into two

clusters, cluster A and B, and the TF tissue samples clustered together or in close

proximity of cluster A. There was a significant difference in overall methylation

between the two clusters with higher average methylation in cluster B. There was

also a significantly higher number of hypermethylated CpGs in cluster B but no

difference in the number of hypomethylated CpGs (Figure 6A-B).

Even though the number of patients (n=45 in paper I and n=115 in paper II) and

distribution of samples in cluster A/B (58% M1 in paper I and 24% M1 in paper

II) differed between the studies the patterns of sample allocation between

clinicopathological variables were similar. Cluster B was significantly associated

with larger tumors and metastatic disease (Figure 6A-B) in both studies.

However, neither study showed a correlation of cluster status with age or gender.

Survival analysis comparing cancer-specific survival (CSS) between cluster A and

B classified tumors showed a significant difference in survival with poorer

outcome for patients belonging to cluster B in both studies. In paper II tumors

could be stratified depending on metastasis status (M0 and M1). The difference

in CSS between clusters A and B was significant in M0 tumors but not in M1 were

both groups of patients had a poor outcome (p<0.001 and p=0.840) (Figure 7A).

Investigation of the cumulative incidence of progress (CIP) in non-metastatic

tumors showed that tumors belonging to cluster B were more prone to progress

(Figure 7B).

26

Figure 6. Cluster A/B grouping based on hierarchical clustering of promoter-associated DNA

methylation in (A) 21,930 CpGs and (B) 155,931 CpGs. A gradient of average methylation (β) and

tumor diameter (mm) are shown as well as the number of hyper- and hypomethylated CpGs. Non-

metastatic (M0) and metastatic (M1) tumors are shown in white (M0) and black (M1). In (B) tumor

progression is shown with ‘true’ non-metastatic (M0-PF) in white, non-metastatic that later will

progress (M0-P) in light grey and metastatic disease (M1) in black.

27

Figure 7. Survival analysis based on a combination of Mo/M1-status and cluster belonging at

diagnosis. (A) Kaplan Meier cancer-specific survival (pCSS5yr) in 115 ccRCC patients depending on

metastatic status and cluster belonging. (B) cumulative incidence of progress (pCIP5yr) analysis of

87 non-metastatic ccRCC tumors in relation to cluster status. Log rank p-values are presented.

Methylation alterations associated with tumor progression

(Paper I).

To identify CpGs associated with ccRCC tumor progression, Δβ values for each

CpG were calculated by comparing each T sample to pooled TF tissue samples, as

described in the material and method section. CpGs with an absolute Δβ≥0.35

were defined as differently methylated CpGs (DM-CpGs). DM-CpGs present in at

least 20% of T tissue samples were chosen for further analysis to determine

protein function and its relation to polycomb target genes.

In total, 959 CpGs were defined as DM in at least 20% of the T samples

representing 840 coding genes. There was an accumulation of genes involved in

developmental processes, morphogenesis, differentiation, and cell-cell signaling.

Twenty-five percentage of genes represented by DM-CpGs were significantly

enriched by polycomb target genes, as defined by Lee et al., (2006) which were

significantly higher than expected (p<0.001) (93).

28

A stepwise increased DNA methylation level during tumor

progression (Papers I and II).

After establishing the correlation of differential methylation landscape with

clinicopathological variables, the correlation between methylation levels and

tumor progression was investigated. A significant stepwise increase in average

promoter methylation; from TF-tissue, via non-metastatic tumors (M0) to

metastatic tumors (M1) was shown in both papers I and II (Figure 8). Overall

promoter methylation levels were higher in samples analyzed by the 450K arrays

compared to the 27K arrays.

Figure 8. Average

methylation of all included

CpGs from 27K and 450 K

arrays in relation to

metastatic disease. Boxplots

show variation within

subgroups. Differences

between groups were

analyzed using Mann-

Whitney U-test * - p<0.050;

** - p<0.001.

Inter- and intra- heterogeneity in DNA methylation profiles

(Paper II).

As shown above, methylation levels between samples in the same group (i.e. M0

and M1) showed high variation (Figure 8). Using principal component analysis

(PCA) based on methylation levels of promoter-associated CpGs, similarities

between groups of samples were analyzed. TF-samples showed a homogenous

methylation pattern and the same for T-samples belonging to the same cluster

(i.e. cluster A/B). On the other hand distribution of T-samples belonging to the

same group (i.e. M0-PF, M0-P, and M1) showed overlapping methylation

patterns (Figure 9A).

In order to evaluate DNA methylation within tumors, two or three samples from

the same tumor in six patients were analyzed using the 450K arrays. PCA analysis

of these samples showed a homogenous methylation pattern (Figure 9B).

29

Figure 9. Principal component analysis (PCA) on (A) 115 ccRCC and 12 tumor-free (TF) tissue

samples. Samples are highlighted as Tumor-Free tissue (blue) non-metastatic tumors that without

progress (M0-PF, green) and with progress (M0-P, orange), and ccRCC with metastasis (M1, red),

and (B) two or three samples from the same tumor were collected from five patients. Each patient

is represented by an individual color and each sample with a dot. PCA was performed using the two

first principal components of the average methylation of promoter-associated CpGs.

Methylation and genetic aberration and their correlation with

biological and mitotic age (Papers I and II)

The biological and mitotic age of patient tissue samples were estimated by

different epigenetically based age predictors (60,61). In paper I the method

described by Horvath (2013) was used to calculate biological age (60) and mitotic

age was calculated by using the epiTOC model described by Yang et al., (2016) in

paper II (61).

In paper I, the biological age of blood from patients with ccRCC, TF and T tissue

were calculated and compared with the chronological age of the individual

patient. There was a significant correlation between biological and chronological

age in blood and T-tissue. However, the estimated biological age in T-tissue was

estimated significantly higher than the chronological age of the patient.

In paper II, the mitotic age of T tissue was calculated and related to the number

of genetic and epigenetic alterations. There was a significant correlation between

the number of genomic aberrations and of hypermethylated CpGs, that were also

correlated to predicted mitotic age. In contrast, there was no correlation between

either the number of hypomethylated CpGs or the number of genomic aberrations

to mitotic age.

30

New prognostic classifiers in ccRCC (Papers II-III).

Whereas methylation alterations showed correlation to tumor progression we

aimed to define differences in methylation patterns that could identify non-

metastatic tumors at diagnosis that later progress (M0-P).

Promoter Methylation Classifier (PMC) panel (Paper II).

The heterogonous methylation

pattern seen between tumors within

the same group (i.e. M0-PF, M0-P,

and M1) made it difficult to use an

average methylation level for each

group to identify CpGs of importance

for tumor progression. Instead, DM-

CpGs were identified at individual

level as described in the material and

method section. DM-CpGs in at least

70% of samples in each group were

selected for further analysis (Figure

10). Combining DM-CpGs from each

group in a Venn-diagram showed that

M0 tumors had a similar methylation

pattern as M1 with only 12% and 17%

unique DM-CpGs for M0-PF and M0-

P respectively whereas M1 tumors had

56% unique DM-CpGs.

A Promoter Methylation Classifier

(PMC) panel was defined based on the

172 DM-CpGs common for M0-P and

M1 (Figure 10). A receiver operating

characteristic (ROC) curve was used to

set the PMC high and PMC low threshold.

Figure 10. Workflow showing the identification of the Promoter Methylator Classifier (PMC)

panel.

Non-metastatic tumors were classified as PMC low or high based on the cut-off

derived from the ROC curves. The PMC status was correlated to tumor

progression, where the PMC high patients showed a significantly higher

frequency of tumor progression (PMC low pCIP5yr 8% vs. PMC high pCIP5yr 38%,

p=0.001). This result was verified in an independent cohort from the TCGA-KIRC

data set (PMC low pCIP5yr 16% vs. PMC high pCIP5yr 39%, p<0.001) (Figure 11).

ccRCC cohort n = 115

M0-PF n = 64

M0-P n = 23

M1 n = 28

DM-CpGs n = 364

DM-CpGs n = 522

DM-CpGs n = 1,058

46

596

231

172 91

M0-P

M0-PF

M1

28 59

PMC Panel

31

Figure 11. Kaplan-Meier cumulative incidence of progress (CIP) for the promoter methylation

classification (PMC) panel (A) in our cohort involving 87 non-metastatic ccRCC and (B) in the TCGA-

KIRC validation cohort including 230 non-metastatic ccRCC. The difference in CIP was between

groups was analyzed using the Log Rank test.

Genes represented in the PMC panel

The 172 CpGs included in the PMC panel represented 160 unique genes of which

109 genes were hypermethylated and 51 hypomethylated in relation to TF tissue.

The hypermethylated genes were enriched for gene ontology terms associated

with SMAD protein complex assembly, RNA polymerase II regulation and

response to pH, whereas hypomethylated genes were associated with several

defense mechanisms, immune response, and response to external stimuli.

In total, 17 genes were represented by two or more DM-CpGs in the PMC panel.

One of these genes (SMAD family member 6 (SMAD6)) showed significantly

higher average methylation in the M0-P group compared to M0-PF whereas

suppressor of cytokine signaling 3 (SOCS3) and MX dynamin-like GTPase 2

(MX2) showed a significantly lower level of methylation. To evaluate if

methylation in these gene promoters was associated with mRNA gene expression

levels data were extracted from 28 analyzed patients on the Illumina Beadchip

gene expression arrays. There were significantly higher mRNA levels for SOCS3

and MX2 expressions in both M0-P and M1 compared with M0-PF whereas no

significant difference was seen for SMAD6.

32

Triple Classifier (Paper III).

In the clinic, the Mayo scoring system is currently used to stratify patients into

risk groups for tumor progression (8). The risk for tumor progression differs

between Mayo score based risk groups and these risk groups are followed up by

different procedures. In our study, the Mayo risk score was set as ‘golden

standard’ when creating new classifiers for tumor progression. Since Mayo

Intermediate and High-risk groups have identical follow up protocols, we pooled

these patients together to be able to calculate sensitivity and specificity for this

classifier. In our cohort, the Mayo scoring system had 85% sensitivity and 50%

specificity for 5-year risk for tumor progression.

A combination of clinicopathological variables and methylation profiles were

used to build risk classifiers for tumor progression in non-metastatic ccRCC. In

total, twelve clinicopathological variables; age, gender, morphological grade,

TNM, tumor diameter and blood analysis (albumin, alkaline phosphatase,

calcium, creatinine, gamma glutamyltransferase, hemoglobin, thrombocyte

particle count) were included, as well as 64 previously identified CpG sites (PI-

CpGs) of relevance in ccRCC and 40 variables identified by Directed Cluster

Analysis (DCA).

The categories described above, i.e. clinical, PI-CpGs and DCA were used

individually or in combination to build six different classifiers; clinical, PI-CpGs,

DCA, clinical+PI-CpGs, clinical+DCA and clinical+PI-CpGs+DCA (denoted

“triple classifier”). The classifiers were built by logistic regression with the five

first principal components were used, to be able to compare the new classifiers to

the clinically used Mayo scoring system the sensitivity was set at 85% and the

specificity was compared.

Each patient was classified as low risk for progress (LRP) or high risk for progress

(HRP) by each classifier and specificity was evaluated at 85% sensitivity. The

specificity for the six classifiers ranged from 43% to 64%, with the lowest

specificity was seen for DCA alone and the highest for the triple classifier.

The triple classifier showed higher specificity (64%) compared to Mayo (50%) at

85% sensitivity in our ccRCC cohort. Both Mayo and the triple classifier showed

a significant difference in the cumulative incidence of progress within five years

(pCIP5yr p=0.005 and <0.001, respectively). The Triple classifier was better to

prognosticate progress (CIP5yr low riskMayo 9,4% vs 7.5% LRPtriple classifier and CIP5yr

intermediate/high riskMayo 37% vs 45% HRPtriple classifier) (Figure 12).

33

Figure 12. Cumulative incidence of progress (pCIP5yr). 87 non-metastatic ccRCC tumors were

classified according to (A) the Mayo Scoring System and (B) the triple classifier (clinical+PI-

CpGs+DCA) at diagnosis. The pCIP5yr was compared between risk groups and Log Rank p-values

are presented.

2.0

0.8

0.6

0.4

0.2

0.0 0 50 100 150 200

p = 0.005

Intermediate/High Risk (n = 32)

Low Risk (n = 46)

Mayo Scoring System

0 50 100 150 200 p < 0.001

HRP (n = 35)

LRP (n = 43)

Triple Classifier

Months from diagnosis

Cu

mu

lati

ve In

cid

en

ce o

f P

rogr

ss

B A

34

Discussion

There is evidence that histological normal tissue taken from a tumor-bearing

kidney contains differently methylated CpGs compared to tissue from a tumor-

free kidney (69). It has also been shown that DNA methylation is an early event

in several different cancer types and therefore, made us investigate the role of

DNA methylation in ccRCCs at diagnosis and to evaluate if there is a correlation

to tumor progression and clinical outcome.

This project started with a cohort of 45 ccRCCs that were analyzed for

methylation using the 27K arrays with a promoter-directed approach. The

conclusion drawn from paper I that, methylation classification has a potential

role as a prognostic marker in ccRCC led us to expand the study. In paper II and

III, the study cohort included 115 patients, and methylation was analyzed using

the 450K arrays. The 27K array includes 27,758 CpG dinucleotides located in gene

promoter regions. The 450K array is an evolvement of the 27K array including

485,577 CpGs scattered throughout the whole genome. The decision to use the

450K arrays in paper II and III became natural when the updated array not only

analyzed promoter-associated CpGs, but also CpGs within the whole genome as

well as intergenic regions. The total number of CpGs associated with each gene

promoter also increased from the 27K arrays to 450K arrays. The distribution of

CpGs per gene is biased on both types of methylation arrays, more

“interesting/well-characterized genes” are represented by a higher number of

CpGs. The decision to focus this thesis on promoter-associated CpGs was a result

of the conclusions drawn in paper I concerning methylation alterations. We

wanted to verify these patterns in a larger more population-based cohort.

The 450K arrays can not only be used for methylation analysis but also to analyze

genetic gains and/or losses. Therefore, the twelve genetic aberrations previously

identified with SNP arrays and described as common in a subset of the ccRCC

cohort were analyzed using methylation arrays as well and the agreement

between both methods was good (22). Using a predefined region when analyzing

gains and losses with the 450K arrays might have led to an underestimation of

genetic aberrations in the cohort. Other regions that might be of interest were not

analyzed and important information might have been missed, ccRCC biomarkers

are summarized by Gulati et al., (2014) (17).

35

Previously identified genetic aberrations show heterogeneity between studies,

and their correlation to clinical outcome might be cohort dependent due to the

diversity in identified regions between different studies. Chromosomal losses on

4p, 9p, 9q, and 14q are correlated to survival time in several studies indicating

that these aberrations are of relevance for ccRCC tumor progression (17). In our

study, the correlation between genetic aberrations and clinical groups of patients

(M0-PF, M0-P, and M1) was analyzed. Out of the twelve regions, gain of

chromosome arm 7p and loss of chromosomal arms 9p and 9q showed a

significant difference when all three groups were compared. Only the loss of

chromosome arm 9q was significantly more common in M0-P compared to M0-

PF, whereas loss of chromosome arms 9p, 9q, and 14q were significantly more

common in metastatic disease compared to M0-PF. A gain of both chromosome

arms 7p and 7q were significantly more common in metastatic disease compared

to M0-P. Interestingly, the region lost on 9q was enriched for DM-CpGs

indicating an epigenetic regulation in this region. This contributes to the previous

finding that chromosome 7, 9 and chromosome arm 14q constitute genomic

regions of importance for ccRCC tumor progression.

In both papers I and II, overall methylation alterations were analyzed. As paper

II was an expansion of the study in paper I some analyzes are identical for the two

studies. In paper II, the number of patients was 2.5 times higher than in paper I

(115 in paper II compared to 45 in paper I), the cohort was more selectively chosen

in paper I compared to a more population-based cohort in paper II (58% M1 in

paper I compared to 24% M1 in paper II), and the number of analyzed CpGs, in

paper II, was more than seven times higher.

Hierarchical cluster analysis on promoter-associated CpGs in both studies

divided the tumors into two distinct clusters, denoted A and B. Cluster B was

associated with a higher level of average methylation and a higher number of

hypermethylated CpGs. In both studies patients belonging to cluster B had poorer

prognosis and shorter cancer-specific survival. When the analysis was restricted

to non-metastatic tumors only (paper II) there was a significant difference in the

cumulative incidence of tumor progression. This indicates that when analyzing a

high number of CpGs, the ccRCCs can be divided into groups correlated to clinical

outcome of patients. Our studies on ccRCC have identified similar DNA

methylation patterns as described in previous studies, confirming that general

hypermethylation correlated to poor prognosis (69,94-96).

36

The TF-tissue samples clustered together with tumors in cluster A in paper II and

in close proximity to cluster A in paper I. This indicates that the methylation

pattern is altered even though the TF-tissue is histologically normal. This is in

line with previous results showing different methylation patterns in normal tissue

from a tumor-free organ, compared to histological normal tissue from a tumor-

bearing organ in both kidney and lung (69,71), and what has been presented as

TINT (tumor-indicating normal tissue) in prostate cancer (97). In colorectal

cancer, the difference in methylation between cancer adjacent polyps (CAP) and

cancer-free polyps (CFP) has been analyzed (70). A significantly higher average

methylation level in analyzed CpGs was seen in CAP compared to CFP, which

might be due to the close proximity to a tumor in CAP and that close proximity to

a tumor can affect methylation patterns (70).

Today, patients with small ccRCCs undergo partial nephrectomy to spare kidney

function post-operatively. When comparing outcome in T1 tumors depending on

radical (RN) or partial nephrectomy (PN) no difference in time to progression or

overall survival could be established. One important benefit of PN compared to

RN is the reduced risk of chronic kidney disease post-operative. Therefore, larger

tumors have also been enrolled to be treated with PN, the risk of tumor

progression in this group is yet to be evaluated (98).

The correlation between higher levels of methylation and tumor development in

several types of cancers has led to the development of DNA-demethylating agents

for treatment. The demethylation agents used today, i.e. 5-azacytidine and 5-aza-

2’-deoxycytidine, have been approved for the treatment of myelodysplastic

syndrome (MDS) and acute myeloid leukemia (AML) and have improved the

outcome in these diseases. These agents have not been approved or found suitable

for treatment of solid tumors, that depend on the instability of the molecule, its

short half-life and above all, the lack of clinical response (99).

In paper I, genes represented by DM-CpGs were analyzed in relation to a list of

1,893 genes presented as polycomb repressor complex 2 (PRC2) target by Lee et

al., (2006) and one-fourth of the DM genes was a target for PRC2 (93) indicating

epigenetic regulation of these genes. The PRC2 complex can both attract DNMTs

to induce DNA methylation (100) and the PRC2 subunit AEBP2 favors binding

methylated DNA and induce H3K27 methylation (101). This indicates that DNA

methylation of PRC2 target genes can both be induced by PRC2 binding as well

as induce PRC2 recruitment. EZH2 is a catalytic factor in the PRC2 complex that

methylates H3K27, which in turn induces heterochromatin formation and gene

silencing. EZH2 overexpression is associated with prostate, bladder, breast, and

gastric cancer as well as hepatocellular and renal cell carcinoma (41).

37

Degerman et al., (2014) identified an accumulation of hypermethylated CpGs

with the number of cell divisions in immortalized T-cells (102). This made us

hypothesize that metastasized tumors were biologically older than non-

metastatic tumors. The method described by Horvath (2013) to calculate

biological age was created for several different tissues and when analyzing

different tumors they were predicted both older and younger than normal tissue

(60,103). Interestingly, in paper I there was a trend towards lower biological age

in tumors belonging to cluster B compared to cluster A even though the number

of hypermethylated CpGs was significantly higher. In paper II, the EpiTOC model

was used to calculate the mitotic age in tumor tissues (61). There was an

accumulation of both number hypermethylated CpGs and the total number of

genetic aberrations associated with increased mitotic age (61). Metastatic tumors

harbor significantly higher numbers of hypermethylated CpGs and genetic

aberrations compared to non-metastatic tumors and also have a significantly

higher mitotic age. This made us speculate that metastatic tumors have a longer

and/or more accelerated proliferative history than non-metastatic tumors.

One challenge in ccRCC is to identify the 30% of non-metastatic tumors that later

will progress. In Sweden today, non-metastatic tumors are scored using the Mayo

scoring system with different follow-up strategies depending on the risk group,

with the goal of early detection of tumor progression. None of the identified

genetic and/or epigenetic biomarkers for ccRCC are established in clinical use as

the reproducibility between different cohorts is scarce. Joosten et al., (2017) drew

the conclusion that a combination of several CpGs and/or genes might be

beneficial in creating biomarkers for ccRCC (85). This made us investigate

different approaches to predict outcome in non-metastatic ccRCC.

The knowledge that hierarchical cluster analysis on promoter-associated CpGs

could separate tumors with significant different methylation patterns and

outcome was the starting point for building our prognostic classifiers. We wanted

to create a prognostic classifier that could predict outcome in a single ccRCC

sample after the classifier was built. Even though a centroid point for each cluster

can be calculated and a new sample can be related to that we decided against that

method. Instead, we analyzed the data in three different ways and identified two

classifiers for ccRCC tumor progression.

Firstly, in paper II, principal component analysis (PCA) was used to investigate

if prognostic groups showed different methylation patterns. Tumors belonging to

the same cluster showed similar methylation patterns whereas prognostic groups

had overlapping methylation patterns and the hypothesis was therefore

discarded.

38

Secondly, in paper II, CpGs identified as DM in M0-P and M1, but not in M0-PF,

were used to create a promoter methylator classifier (PMC). The PMC panel

divided non-metastatic tumors into PMC high and low with a strong association

with the risk of tumor progression. The clinical relevance remained significant in

a multivariate analysis including PMC status, TNM, morphological grade, age,

and gender and was also validated using the TCGA-KIRC cohort. The prognostic

classifier presented by Wei et al., (2015) predicting overall survival was also

applied to our cohort to elucidate if it could predict progress in non-metastatic

tumors (84). Almost 50% of M0-P tumors were classified as low risk and there

was no significant difference in CIP between low- and high-risk tumors indicating

that the Wei Risk score can not be used as a prognostic classifier for progress in

non-metastatic ccRCC.

Thirdly, in paper III, an unsupervised approach combining clinicopathological

variables with previously identified CpGs of prognostic relevance and DCA

clusters was used. In total, six classifiers were built with a fixed sensitivity at 85%

and compared to the clinically used Mayo scoring system (8). The high fixed

sensitivity might lead to a reduced specificity. The requirement of the specificity

for a prognostic marker depends on the treatment program for high-risk patients,

i.e. close monitoring or if to enroll them in adjuvant treatment. Adjuvant

treatment in non-metastatic ccRCC has been reported as beneficial in the S-TRAC

study, with longer progression-free survival time in the sunitinib arm, but no data

are yet available for overall survival (10). There was no survival advantage in none

of the other adjuvant TKI studies in non-metastatic ccRCC, e.g., but a high

proportion of adverse side-effects and adjuvant treatment for non-metastatic

tumors prior to progression is not recommended (9). New immunotherapies have

been proven beneficial in the treatment of metastatic ccRCC. These therapies

block the inhibitory T-cell receptor RD-1 or cytotoxic T-lymphocyte associated

antigen 4 (CTLA-4) signaling to restore tumor-specific T-cell immunity and are

shown to beneficial for patients with metastatic disease (13). Due to the high risk

of adverse side-effects with adjuvant treatment, there is a need for high specificity

of risk classifier markers for tumor progression before enrolment of patients with

non-metastatic disease in adjuvant treatment program (9,13,104).

The classifier with the best specificity, the triple classifier, combined all included

variables and had a specificity of 64%. The PMC panel had a specificity of 55%

which was lower than the triple classifier but still higher compared to the Mayo

scoring system which had a specificity of 50% in our cohort. It has previously been

shown that the combination of methylation and gene-expression profiles with

clinical variables can predict overall survival in ccRCC (105). This knowledge

supports our result that a combination of molecular biomarkers and clinical

variables improves risk-classification in non-metastatic ccRCC.

39

Previous analysis concerning heterogeneity in ccRCC have identified differences

in the mutational rate and metabolic patterns from different samples from the

same tumor (18,106). ClearCode34 is a ccRCC classifier that uses gene expression

levels of 34 genes to separate non-metastatic tumors into ccA or ccB with poorer

outcome in the ccB subgroup (107). ClearCode34 is a “slimmed-down version” of

the 110 gene classifier presented by Brannon et al., (2010) (108). Heterogeneity

analysis with several samples from the same tumor, and the 110 gene panel,

identified eight out of ten tumors classified as both ccA and ccB (17). Even though,

there was a difference in methylation patterns between tumors in the same

prognostic group we could identify a homogenous methylation pattern within the

individual tumor. This is in line with the analysis presented by Wei et al., (2015)

whose risk classifier could correctly classify over 90% of tumor samples analyzed

for intra-tumoral heterogeneity (84). This indicates that methylation alterations

are more stable within an individual tumor and therefore might be better suited

in developing prognostic markers for ccRCC. Another advantage with a DNA

based marker is that DNA is more stable than, eg. RNA, and is routinely extracted

in most clinics.

40

General conclusions and future perspectives

In this thesis, DNA methylation classification was identified as a prognostic

marker for tumor progression in ccRCC. We identified a stepwise increase in

average methylation and the number of hypermethylated CpGs at increasing

tumor stages. A high number of methylation alterations at diagnosis correlated

to shorter cancer-specific survival and tumors more prone to progress. We have

created two different classifiers that used DNA methylation profiles, with or

without combination with clinical variables to predict tumor progression in non-

metastatic tumors.

Our analysis was solely focused on CpGs located within gene promoter regions.

The decision not to analyze DNA methylation in non-coding regions, in samples

analyzed with the 450K arrays, might have led to a miss in information correlated

to tumor progression. Demethylation of non-coding regions can contribute to

genomic instability and it would be interesting to correlate genetic aberrations to

DM-CpGs in intrinsic regions and correlate them, if possible, to ccRCC tumor

progression. However, this analysis would preferably be performed on whole

genome bisulfite-sequenced DNA.

Even though the methylation analysis developed by Illumina Inc. easily can be

implemented in the laboratory work, there are still some challenges to over-build.

At many diagnostic hospitals in Sweden, including Norrlands

Universitetssjukhus, Umeå, analysis of genetic aberrations using the SNP arrays

from Illunina Inc. are already implemented. However, the amount of information

derived from the methylation arrays demands bioinformatic knowledge to pre-

process and adapt the data before applying it to an available script to calculate

the risk for tumor progression. Although bioinformatics pipelines have been set

up at many hospitals, this is still a limiting factor for the integration of our

predicted DNA methylation-based risk classifier for ccRCC.

The epigenetic regulation of gene expression is a complex machinery, with the

interaction between the different regulatory mechanisms. The consequence of

specific DNA methylation alterations for gene expression has not been analyzed

in detail in this thesis. It would be of outmost interest to further investigate the

functional consequence of specific epigenetic alterations for gene expression.

Functional analysis might contribute to deepened knowledge of the role of DNA

methylation in ccRCC progression and could be beneficial in identifying targets

for treatment. Methylation analysis can also be used to select patients for

adjuvant treatments. Hopefully, patients identified as high risk for recurrence by

DNA methylation might be helped by the new immunotherapies at hand.

41

The long term goal of this project is to further evolve the prognostic marker to

include the best combination of methylation markers and clinical variables, to

verify it in additional ccRCC cohorts and lastly implement it in clinical use.

42

Acknowledgment

I never thought this day would come, for me to defend my thesis. Since acceptance as a

Ph.D. student, I have had two wonderful kids, medical school, internship and now finally

this day. It would not have been possible without my supervisors, collaborators, friends,

and family.

To my supervisor Sofie Degerman, we started off as fellow Ph.D.-students and now a

lector and (hopefully soon) a Ph.D. A role model in laboratory work and organization, a

hard-working and dedicated supervisor that can keep a lot of projects and work in her head

at once. Always ready to lend a helping hand, all day of the week. Through thick and thin

(and a lot of chocolate cakes), we made it!

To my co-supervisor Göran Roos for sticking by my side through the years, first as a

supervisor and later as a co-supervisor. Thank you for giving me the opportunity to become

a part of your research group. It took a bit longer than expected but kids and medical school

came in between. If some reason I do have an urge to answer with a “schooo” every time

you ask me something.

Börje Ljungberg, my co-supervisor since the beginning. Your clinical knowledge has

been so important to this thesis and a big help along the way.

Magnus Hultdin, my newest co-supervisor. Intelligent, funny and always with a big

smile. Your contributions to me, and the group as a whole, have made me want to work a

bit harder each day. We will see, maybe I one day will become a pathologist as well.

“The Group”: Zahra my Pakistani friend and co-worker. Intelligent and hard-working,

you are a true inspiration and your Swedish skills are top-notch. Always up for “äventyr”!

Mattias the statistician that has had to endure all my million questions concerning

everything. All emails with the subject “Heeeeeeej”, without you this thesis had been

nothing. Oliva the new PhD-student, funny and open-minded, you will be great! Pia both

former and current member of the group. Your technical skills are amazing. Former

members; Ulrika went from study mates at the beginning of the 21st century to Ph.D.

students in the same group. You inspire me with your intelligence and your sewing skills.

Magnus B, you are one of the smartest and funniest men I know, thanks for the years as

co-working PhD-students. Linda, you took care of me when I started as a “stipendiat” in

Irina’s lab. Bright and cool, I’ve learned a lot from you. Helene, we never worked with the

same project but you are a fun and inspiring woman in so many ways. Ravi, the other

kidney cancer Ph.D. in the group, thanks for inspiring discussions through the beginning

of my Ph.D. studies.

Thanks to the collaborating group of Patrik Rydén, finally we finished the last

manuscript. A special thanks to Linda Vidman that did not give up on me and all my

questions about the statistical stuff that “MatStat” contributed to the last manuscript and

this thesis.

43

Fikamänniskorna, the once that drinks coffee (and tea) and have lunch at 6M 2nd floor.

The once that like chocolate and have taken care of me when I have been in and out from

the lab. Smart, bright and funny (and good looking) folks that make even the hard days a

bit funnier.

Ida, Terry, and Carina the administrative staff that have made this journey easier for

me. Percentage of employment, registration of courses, salary, parental leave, and

everything in between.

And to all other employees at 6M for making all these years as a Ph.D. fun, interesting

and worthwhile with interesting seminars and discussions.

My dear friends/Mina kära vänner:

Pumorna, ni som funnits där genom alla år. Ni som alltid har tid att ses när vi är på

samma breddgrader. Ni som känner mig inifrån och ut oavsett hur lång tid det går mellan

gångerna vi ses och hörs är allt alltid ”som vanligt”. Ni som älskar mig innerst inne OCH

ytterst ute, jag älskar er på samma sätt!

Ida, du får stå här också. Så många gånger du har fått mig att skratta så jag fått

träningsvärk, så många gånger jag suttit på ditt skrivbord. Min eldfluga på 6M.

Angelica tack för alla felspringningar och kaffedejter. Så himla skönt att bara kunna dyka

upp hemma hos dig och din familj, dricka kaffe och kolla på barn som leker och känna

lugnet.

Heidi, jag vet inte vad jag ska säga om dig. Genom år av äventyr, häng, kids, kaffe, studier,

jobb och allt annat. Att gå ut och dansa med dig är helt underbart. Vi vet EXAKT vart vi

har varandra och vi kan snacka om allt, alltid. Du som dricker kaffe med mycket mjölk,

min extrasyster.

Emma, du som gått från bekant till nära vän. Du som alltid har tid för en kaffe och lite

snack på balkongen, i soffan eller vid köksbordet. Du som lurat med mig att bada

#kallabad. Du som är klok som en bok och har underbara kramar. Tack för att DU är min

vän.

Elin, utan dig (och din familj) vet jag inte hur mitt liv skulle se ut. Mina barn har fått en

storasyster och en lille-/storebror. Du är en trygghet att ha som vän, du finns alltid där,

ställer alltid upp och du är bland de bästa som finns. Att bara kunna andas hemma hos

dig, lyssna på musik eller dra på äventyr. You see the whole of the moon.

Team Tallberg ni vet vilka ni är, Göteborgsvarv, plump och underbara släkthelger med

världens bästa Mormor som mittpunkt. Farfar, farbröder, fastrar och kusiner alltid

lika underbart att komma ”hem” och träffa er.

44

”Familjen Evelönn” – Åke och Katrin, Maria och Mikael, Erik och Nicolin, Laila och

Håkan, Sofia och Christian, Anna och Owen plus alla barnen. Jag hade verkligen tur när

jag även fick er som del av min familj. Alltid lika trevligt att umgås med alla er och era

barn.

Övriga vänner, ni som där runt omkring mig. Ni som peppar, stöttar och finns där för

att göra mitt och barnens liv ännu lite bättre.

Sist men inte minst…

Syrran, du som hatar forskning. Läkarstudierna utan dig hade kanske gått men inte lika

bra. Kommer fortfarande ihåg känslan att ”idag ska jag bara springa 21 km”, dags för en

ny utmaning snart! Oavsett vad, jag kunde inte fått en bättre storasyster. Daniel,

fantastisk svåger som lagar underbart god mat. Du som ska lära mina barn göra upp eld,

använda yxa och kniv. Selma och Ida, mosters töser, har älskat er sen den dagen ni

började växa i eran mammas mage.

Mamma och Pappa den stabila punkten i livet. Bara 27 mil bort, alltid redo att ställa upp

närhelst och vadhelst jag har behövt och behöver. Med er i min ringhörna är allt möjligt,

även om ni inte alltid förstår vad jag håller på med. Även om jag nog aldrig kommer flytta

hem igen åker jag alltid hem till er, NI är hemma för mig.

Johan, pappa till mina barn. Utan dig hade jag inte stått här. Tryggheten genom tuffa

småbarnsår kombinerade med studier. Du som tvingar mig att tänka till en gång extra,

som ställer upp och stöttar och försöker sätta perspektiv på saker. Jag är glad att ha dig

som del i mitt liv.

Det bästa jag gjort i livet: Hugo och Hanna, ni gör livet värt att leva. Älskar er av hela

mitt hjärta och varje dag med er gör mig lycklig.

45

References

1. Ferlay J, Colombet M, Soerjomataram I, Mathers C, Parkin DM, Pineros M, et al. Estimating the global cancer incidence and mortality in 2018: GLOBOCAN sources and methods. Int J Cancer 2019;144(8):1941-53 doi 10.1002/ijc.31937.

2. Regionalt Cancercentrum. Nationellt vårdprogram njurcancer. 2019. 3. Scelo G, Larose TL. Epidemiology and Risk Factors for Kidney Cancer. J Clin Oncol

2018:Jco2018791905 doi 10.1200/jco.2018.79.1905. 4. Haas NB, Nathanson KL. Hereditary kidney cancer syndromes. Adv Chronic Kidney

Dis 2014;21(1):81-90 doi 10.1053/j.ackd.2013.10.001. 5. Kim HS, Kim JH, Jang HJ, Han B, Zang DY. Clinicopathologic Significance of VHL

Gene Alteration in Clear-Cell Renal Cell Carcinoma: An Updated Meta-Analysis and Review. Int J Mol Sci 2018;19(9) doi 10.3390/ijms19092529.

6. Sobin LH GM, Wittekind C. TNM classification of malignant tumors. UICC International Union Against Cancer.: Wiley-Blackwell; 2009.

7. Fuhrman SA, Lasky LC, Limas C. Prognostic significance of morphologic parameters in renal cell carcinoma. Am J Surg Pathol 1982;6(7):655-63.

8. Leibovich BC, Blute ML, Cheville JC, Lohse CM, Frank I, Kwon ED, et al. Prediction of progression after radical nephrectomy for patients with clear cell renal cell carcinoma: a stratification tool for prospective clinical trials. Cancer 2003;97(7):1663-71 doi 10.1002/cncr.11234.

9. Bai Y, Li S, Jia Z, Ding Y, Gu C, Yang J. Adjuvant therapy for locally advanced renal cell carcinoma: A meta-analysis and systematic review. Urol Oncol 2018;36(2):79.e1-.e10 doi 10.1016/j.urolonc.2017.10.001.

10. Bex A, Albiges L, Ljungberg B, Bensalah K, Dabestani S, Giles RH, et al. Updated European Association of Urology Guidelines Regarding Adjuvant Therapy for Renal Cell Carcinoma. Eur Urol 2017;71(5):719-22 doi 10.1016/j.eururo.2016.11.034.

11. Escudier B, Sharma P, McDermott DF, George S, Hammers HJ, Srinivas S, et al. CheckMate 025 Randomized Phase 3 Study: Outcomes by Key Baseline Factors and Prior Therapy for Nivolumab Versus Everolimus in Advanced Renal Cell Carcinoma. Eur Urol 2017;72(6):962-71 doi 10.1016/j.eururo.2017.02.010.

12. Mejean A, Ravaud A, Thezenas S, Colas S, Beauval JB, Bensalah K, et al. Sunitinib Alone or after Nephrectomy in Metastatic Renal-Cell Carcinoma. N Engl J Med 2018;379(5):417-27 doi 10.1056/NEJMoa1803675.

13. Singla N, Freifeld Y, Ghandour RA, Hammers HJ. Rational Approaches to Treatment Duration with Immunotherapy in Metastatic Renal Cell Carcinoma. Eur Urol Focus 2020;6(1):31-3 doi 10.1016/j.euf.2019.10.010.

14. Comprehensive molecular characterization of clear cell renal cell carcinoma. Nature 2013;499(7456):43-9 doi 10.1038/nature12222.

15. Ogawa O, Kakehi Y, Ogawa K, Koshiba M, Sugiyama T, Yoshida O. Allelic loss at chromosome 3p characterizes clear cell phenotype of renal cell carcinoma. Cancer Res 1991;51(3):949-53.

16. D'Avella C, Abbosh P, Pal SK, Geynisman DM. Mutations in renal cell carcinoma. Urol Oncol 2018 doi 10.1016/j.urolonc.2018.10.027.

17. Gulati S, Martinez P, Joshi T, Birkbak NJ, Santos CR, Rowan AJ, et al. Systematic evaluation of the prognostic impact and intratumour heterogeneity of clear cell renal cell carcinoma biomarkers. Eur Urol 2014;66(5):936-48 doi 10.1016/j.eururo.2014.06.053.

18. Gerlinger M, Rowan AJ, Horswell S, Math M, Larkin J, Endesfelder D, et al. Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. N Engl J Med 2012;366(10):883-92 doi 10.1056/NEJMoa1113205.

19. Kaelin WG, Jr. The von Hippel-Lindau tumor suppressor gene and kidney cancer. Clin Cancer Res 2004;10(18 Pt 2):6290s-5s doi 10.1158/1078-0432.CCR-sup-040025.

46

20. Arai E, Ushijima S, Tsuda H, Fujimoto H, Hosoda F, Shibata T, et al. Genetic clustering of clear cell renal cell carcinoma based on array-comparative genomic hybridization: its association with DNA methylation alteration and patient outcome. Clin Cancer Res 2008;14(17):5531-9 doi 10.1158/1078-0432.ccr-08-0443.

21. Sanjmyatav J, Junker K, Matthes S, Muehr M, Sava D, Sternal M, et al. Identification of genomic alterations associated with metastasis and cancer specific survival in clear cell renal cell carcinoma. J Urol 2011;186(5):2078-83 doi 10.1016/j.juro.2011.06.050.

22. Kohn L, Svenson U, Ljungberg B, Roos G. Specific Genomic Aberrations Predict Survival, But Low Mutation Rate in Cancer Hot Spots, in Clear Cell Renal Cell Carcinoma. Appl Immunohistochem Mol Morphol 2014 doi 10.1097/pai.0000000000000087.

23. Monzon FA, Alvarez K, Peterson L, Truong L, Amato RJ, Hernandez-McClain J, et al. Chromosome 14q loss defines a molecular subtype of clear-cell renal cell carcinoma associated with poor prognosis. Mod Pathol 2011;24(11):1470-9 doi 10.1038/modpathol.2011.107.

24. Klatte T, Rao PN, de Martino M, LaRochelle J, Shuch B, Zomorodian N, et al. Cytogenetic profile predicts prognosis of patients with clear cell renal cell carcinoma. J Clin Oncol 2009;27(5):746-53 doi 10.1200/jco.2007.15.8345.

25. Antonelli A, Arrighi N, Tardanico R, Balzarini P, Zanotelli T, Corti S, et al. Prognostic value of cytogenetic analysis in clear cell renal carcinoma: a study on 131 patients with long-term follow-up. Anticancer Res 2010;30(11):4705-9.

26. Mitchell TJ, Rossi SH, Klatte T, Stewart GD. Genomics and clinical correlates of renal cell carcinoma. World J Urol 2018;36(12):1899-911 doi 10.1007/s00345-018-2429-x.

27. Nabi S, Kessler ER, Bernard B, Flaig TW, Lam ET. Renal cell carcinoma: a review of biology and pathophysiology. F1000Res 2018;7:307 doi 10.12688/f1000research.13179.1.

28. Li F, Mao G, Tong D, Huang J, Gu L, Yang W, et al. The histone mark H3K36me3 regulates human DNA mismatch repair through its interaction with MutSalpha. Cell 2013;153(3):590-600 doi 10.1016/j.cell.2013.03.025.

29. Hakimi AA, Ostrovnaya I, Reva B, Schultz N, Chen YB, Gonen M, et al. Adverse outcomes in clear cell renal cell carcinoma with mutations of 3p21 epigenetic regulators BAP1 and SETD2: a report by MSKCC and the KIRC TCGA research network. Clin Cancer Res 2013;19(12):3259-67 doi 10.1158/1078-0432.ccr-12-3886.

30. Svenson U, Ljungberg B, Roos G. Telomere length in peripheral blood predicts survival in clear cell renal cell carcinoma. Cancer Res 2009;69(7):2896-901 doi 10.1158/0008-5472.can-08-3513.

31. Nicoglou A, Merlin F. Epigenetics: A way to bridge the gap between biological fields. Stud Hist Philos Biol Biomed Sci 2017;66:73-82 doi 10.1016/j.shpsc.2017.10.002.

32. Herceg Z, Ushijima T. Introduction: epigenetics and cancer. Adv Genet 2010;70:1-23 doi 10.1016/b978-0-12-380866-0.60001-0.

33. Shen H, Laird PW. Interplay between the cancer genome and epigenome. Cell 2013;153(1):38-55 doi 10.1016/j.cell.2013.03.008.

34. Audia JE, Campbell RM. Histone Modifications and Cancer. Cold Spring Harb Perspect Biol 2016;8(4):a019521 doi 10.1101/cshperspect.a019521.

35. MacDonald VE, Howe LJ. Histone acetylation: where to go and how to get there. Epigenetics 2009;4(3):139-43 doi 10.4161/epi.4.3.8484.

36. Sawicka A, Seiser C. Sensing core histone phosphorylation - a matter of perfect timing. Biochim Biophys Acta 2014;1839(8):711-8 doi 10.1016/j.bbagrm.2014.04.013.

37. Huang H, Weng H, Chen J. The Biogenesis and Precise Control of RNA m(6)A Methylation. Trends Genet 2020;36(1):44-52 doi 10.1016/j.tig.2019.10.011.

47

38. Yang PK, Kuroda MI. Noncoding RNAs and intranuclear positioning in monoallelic gene expression. Cell 2007;128(4):777-86 doi 10.1016/j.cell.2007.01.032.

39. Wei JW, Huang K, Yang C, Kang CS. Non-coding RNAs as regulators in epigenetics (Review). Oncol Rep 2017;37(1):3-9 doi 10.3892/or.2016.5236.

40. Denis H, Ndlovu MN, Fuks F. Regulation of mammalian DNA methyltransferases: a route to new mechanisms. EMBO Rep 2011;12(7):647-56 doi 10.1038/embor.2011.110.

41. Yan KS, Lin CY, Liao TW, Peng CM, Lee SC, Liu YJ, et al. EZH2 in Cancer Progression and Potential Application in Cancer Therapy: A Friend or Foe? Int J Mol Sci 2017;18(6) doi 10.3390/ijms18061172.

42. Cedar H, Bergman Y. Linking DNA methylation and histone modification: patterns and paradigms. Nat Rev Genet 2009;10(5):295-304 doi 10.1038/nrg2540.

43. Avery OT, Macleod CM, McCarty M. STUDIES ON THE CHEMICAL NATURE OF THE SUBSTANCE INDUCING TRANSFORMATION OF PNEUMOCOCCAL TYPES : INDUCTION OF TRANSFORMATION BY A DESOXYRIBONUCLEIC ACID FRACTION ISOLATED FROM PNEUMOCOCCUS TYPE III. J Exp Med 1944;79(2):137-58 doi 10.1084/jem.79.2.137.

44. Robertson KD. DNA methylation and chromatin - unraveling the tangled web. Oncogene 2002;21(35):5361-79 doi 10.1038/sj.onc.1205609.

45. Lister R, Pelizzola M, Dowen RH, Hawkins RD, Hon G, Tonti-Filippini J, et al. Human DNA methylomes at base resolution show widespread epigenomic differences. Nature 2009;462(7271):315-22 doi 10.1038/nature08514.

46. Saxonov S, Berg P, Brutlag DL. A genome-wide analysis of CpG dinucleotides in the human genome distinguishes two distinct classes of promoters. Proc Natl Acad Sci U S A 2006;103(5):1412-7 doi 10.1073/pnas.0510310103.

47. Kulis M, Esteller M. DNA methylation and cancer. Adv Genet 2010;70:27-56 doi 10.1016/b978-0-12-380866-0.60002-2.

48. Ooi SK, Qiu C, Bernstein E, Li K, Jia D, Yang Z, et al. DNMT3L connects unmethylated lysine 4 of histone H3 to de novo methylation of DNA. Nature 2007;448(7154):714-7 doi 10.1038/nature05987.

49. Edwards JR, Yarychkivska O, Boulard M, Bestor TH. DNA methylation and DNA methyltransferases. Epigenetics Chromatin 2017;10:23 doi 10.1186/s13072-017-0130-8.

50. Wu X, Zhang Y. TET-mediated active DNA demethylation: mechanism, function and beyond. Nat Rev Genet 2017;18(9):517-34 doi 10.1038/nrg.2017.33.

51. Cedar H, Bergman Y. Programming of DNA methylation patterns. Annu Rev Biochem 2012;81:97-117 doi 10.1146/annurev-biochem-052610-091920.

52. Guenther MG, Levine SS, Boyer LA, Jaenisch R, Young RA. A chromatin landmark and transcription initiation at most promoters in human cells. Cell 2007;130(1):77-88 doi 10.1016/j.cell.2007.05.042.

53. Lehnertz B, Ueda Y, Derijck AA, Braunschweig U, Perez-Burgos L, Kubicek S, et al. Suv39h-mediated histone H3 lysine 9 methylation directs DNA methylation to major satellite repeats at pericentric heterochromatin. Curr Biol 2003;13(14):1192-200 doi 10.1016/s0960-9822(03)00432-9.

54. Du J, Johnson LM, Jacobsen SE, Patel DJ. DNA methylation pathways and their crosstalk with histone methylation. Nat Rev Mol Cell Biol 2015;16(9):519-32 doi 10.1038/nrm4043.

55. Breitling LP, Yang R, Korn B, Burwinkel B, Brenner H. Tobacco-smoking-related differential DNA methylation: 27K discovery and replication. Am J Hum Genet 2011;88(4):450-7 doi 10.1016/j.ajhg.2011.03.003.

56. Fraga MF, Ballestar E, Paz MF, Ropero S, Setien F, Ballestar ML, et al. Epigenetic differences arise during the lifetime of monozygotic twins. Proc Natl Acad Sci U S A 2005;102(30):10604-9 doi 10.1073/pnas.0500398102.

48

57. Bell JT, Pai AA, Pickrell JK, Gaffney DJ, Pique-Regi R, Degner JF, et al. DNA methylation patterns associate with genetic and gene expression variation in HapMap cell lines. Genome Biol 2011;12(1):R10 doi 10.1186/gb-2011-12-1-r10.

58. Relton CL, Davey Smith G. Epigenetic epidemiology of common complex disease: prospects for prediction, prevention, and treatment. PLoS Med 2010;7(10):e1000356 doi 10.1371/journal.pmed.1000356.

59. Marioni RE, Shah S, McRae AF, Ritchie SJ, Muniz-Terrera G, Harris SE, et al. The epigenetic clock is correlated with physical and cognitive fitness in the Lothian Birth Cohort 1936. Int J Epidemiol 2015 doi 10.1093/ije/dyu277.

60. Horvath S. DNA methylation age of human tissues and cell types. Genome Biol. Volume 14. England2013. p R115.

61. Yang Z, Wong A, Kuh D, Paul DS, Rakyan VK, Leslie RD, et al. Correlation of an epigenetic mitotic clock with cancer risk. Genome Biol 2016;17(1):205 doi 10.1186/s13059-016-1064-3.

62. Gaunt TR, Shihab HA, Hemani G, Min JL, Woodward G, Lyttleton O, et al. Systematic identification of genetic influences on methylation across the human life course. Genome Biol 2016;17:61 doi 10.1186/s13059-016-0926-z.

63. Heyn H, Moran S, Hernando-Herraez I, Sayols S, Gomez A, Sandoval J, et al. DNA methylation contributes to natural human variation. Genome Res 2013;23(9):1363-72 doi 10.1101/gr.154187.112.

64. Feinberg AP, Vogelstein B. Hypomethylation distinguishes genes of some human cancers from their normal counterparts. Nature 1983;301(5895):89-92 doi 10.1038/301089a0.

65. Feinberg AP, Tycko B. The history of cancer epigenetics. Nat Rev Cancer. Volume 4. England2004. p 143-53.

66. Jones PA, Baylin SB. The fundamental role of epigenetic events in cancer. Nat Rev Genet 2002;3(6):415-28 doi 10.1038/nrg816.

67. Reis AH, Vargas FR, Lemos B. Biomarkers of genome instability and cancer epigenetics. Tumour Biol 2016;37(10):13029-38 doi 10.1007/s13277-016-5278-5.

68. Kako K, Kim JD, Fukamizu A. Emerging impacts of biological methylation on genetic information. J Biochem 2019;165(1):9-18 doi 10.1093/jb/mvy075.

69. Arai E, Chiku S, Mori T, Gotoh M, Nakagawa T, Fujimoto H, et al. Single-CpG-resolution methylome analysis identifies clinicopathologically aggressive CpG island methylator phenotype clear cell renal cell carcinomas. Carcinogenesis 2012;33(8):1487-93 doi 10.1093/carcin/bgs177.

70. Druliner BR, Wang P, Bae T, Baheti S, Slettedahl S, Mahoney D, et al. Molecular characterization of colorectal adenomas with and without malignancy reveals distinguishing genome, transcriptome and methylome alterations. Sci Rep 2018;8(1):3161 doi 10.1038/s41598-018-21525-4.

71. Sato T, Arai E, Kohno T, Tsuta K, Watanabe S, Soejima K, et al. DNA methylation profiles at precancerous stages associated with recurrence of lung adenocarcinoma. PLoS One 2013;8(3):e59444 doi 10.1371/journal.pone.0059444.

72. Sandoval J, Esteller M. Cancer epigenomics: beyond genomics. Curr Opin Genet Dev 2012;22(1):50-5 doi 10.1016/j.gde.2012.02.008.

73. Locke WJ, Guanzon D, Ma C, Liew YJ, Duesing KR, Fung KYC, et al. DNA Methylation Cancer Biomarkers: Translation to the Clinic. Front Genet 2019;10:1150 doi 10.3389/fgene.2019.01150.

74. Jaunmuktane Z, Capper D, Jones DTW, Schrimpf D, Sill M, Dutt M, et al. Methylation array profiling of adult brain tumours: diagnostic outcomes in a large, single centre. Acta Neuropathol Commun 2019;7(1):24 doi 10.1186/s40478-019-0668-8.

75. Capper D, Stichel D, Sahm F, Jones DTW, Schrimpf D, Sill M, et al. Practical implementation of DNA methylation and copy-number-based CNS tumor

49

diagnostics: the Heidelberg experience. Acta Neuropathol 2018;136(2):181-210 doi 10.1007/s00401-018-1879-y.

76. Toyota M, Ahuja N, Ohe-Toyota M, Herman JG, Baylin SB, Issa JP. CpG island methylator phenotype in colorectal cancer. Proc Natl Acad Sci U S A 1999;96(15):8681-6.

77. Teodoridis JM, Hardie C, Brown R. CpG island methylator phenotype (CIMP) in cancer: causes and implications. Cancer Lett 2008;268(2):177-86 doi 10.1016/j.canlet.2008.03.022.

78. Hughes LA, Melotte V, de Schrijver J, de Maat M, Smit VT, Bovee JV, et al. The CpG island methylator phenotype: what's in a name? Cancer Res 2013;73(19):5858-68 doi 10.1158/0008-5472.can-12-4306.

79. Ricketts CJ, Hill VK, Linehan WM. Tumor-specific hypermethylation of epigenetic biomarkers, including SFRP1, predicts for poorer survival in patients from the TCGA Kidney Renal Clear Cell Carcinoma (KIRC) project. PLoS One 2014;9(1):e85621 doi 10.1371/journal.pone.0085621.

80. Joosten SC, Smits KM, Aarts MJ, Melotte V, Koch A, Tjan-Heijnen VC, et al. Epigenetics in renal cell cancer: mechanisms and clinical applications. Nat Rev Urol 2018;15(7):430-51 doi 10.1038/s41585-018-0023-z.

81. Zhang Y, Weinberg RA. Epithelial-to-mesenchymal transition in cancer: complexity and opportunities. Front Med 2018;12(4):361-73 doi 10.1007/s11684-018-0656-6.

82. Serrano-Gomez SJ, Maziveyi M, Alahari SK. Regulation of epithelial-mesenchymal transition through epigenetic and post-translational modifications. Mol Cancer 2016;15:18 doi 10.1186/s12943-016-0502-x.

83. Tian Y, Arai E, Gotoh M, Komiyama M, Fujimoto H, Kanai Y. Prognostication of patients with clear cell renal cell carcinomas based on quantification of DNA methylation levels of CpG island methylator phenotype marker genes. BMC Cancer 2014;14:772 doi 10.1186/1471-2407-14-772.

84. Wei JH, Haddad A, Wu KJ, Zhao HW, Kapur P, Zhang ZL, et al. A CpG-methylation-based assay to predict survival in clear cell renal cell carcinoma. Nat Commun 2015;6:8699 doi 10.1038/ncomms9699.

85. Joosten SC, Deckers IA, Aarts MJ, Hoeben A, van Roermund JG, Smits KM, et al. Prognostic DNA methylation markers for renal cell carcinoma: a systematic review. Epigenomics 2017;9(9):1243-57 doi 10.2217/epi-2017-0040.

86. van Vlodrop IJ, Joosten SC, de Meyer T, Smits KM, Van Neste L, Melotte V, et al. A four-gene promoter methylation marker panel consisting of GREM1, NEURL, LAD1 and NEFH predicts survival of clear cell renal cell cancer patients. Clin Cancer Res 2016 doi 10.1158/1078-0432.ccr-16-1236.

87. Weisenberger DJ, Campan M, Long TI, Kim M, Woods C, Fiala E, et al. Analysis of repetitive element DNA methylation by MethyLight. Nucleic Acids Res 2005;33(21):6823-36 doi 10.1093/nar/gki987.

88. Teschendorff AE, Marabita F, Lechner M, Bartlett T, Tegner J, Gomez-Cabrero D, et al. A beta-mixture quantile normalization method for correcting probe design bias in Illumina Infinium 450 k DNA methylation data. Bioinformatics 2013;29(2):189-96 doi 10.1093/bioinformatics/bts680.

89. Nordlund J, Backlin CL, Wahlberg P, Busche S, Berglund EC, Eloranta ML, et al. Genome-wide signatures of differential DNA methylation in pediatric acute lymphoblastic leukemia. Genome Biol 2013;14(9):r105 doi 10.1186/gb-2013-14-9-r105.

90. Ward Jr JH. Hierarchical grouping to optimize an objective function. Journal of the American statistical association 1963;58(301):236-44.

91. Wang ZR, Wei JH, Zhou JC, Haddad A, Zhao LY, Kapur P, et al. Validation of DAB2IP methylation and its relative significance in predicting outcome in renal cell carcinoma. Oncotarget 2016;7(21):31508-19 doi 10.18632/oncotarget.8971.

50

92. Wang J, Vasaikar S, Shi Z, Greer M, Zhang B. WebGestalt 2017: a more comprehensive, powerful, flexible and interactive gene set enrichment analysis toolkit. Nucleic Acids Res 2017 doi 10.1093/nar/gkx356.

93. Lee TI, Jenner RG, Boyer LA, Guenther MG, Levine SS, Kumar RM, et al. Control of developmental regulators by Polycomb in human embryonic stem cells. Cell. Volume 125. United States2006. p 301-13.

94. Li J, Jin H, Wang X. Epigenetic biomarkers: potential applications in gastrointestinal cancers. ISRN Gastroenterol 2014;2014:464015 doi 10.1155/2014/464015.

95. Lai RK, Chen Y, Guan X, Nousome D, Sharma C, Canoll P, et al. Genome-wide methylation analyses in glioblastoma multiforme. PLoS One 2014;9(2):e89376 doi 10.1371/journal.pone.0089376.

96. Barrow TM, Michels KB. Epigenetic epidemiology of cancer. Biochem Biophys Res Commun 2014 doi 10.1016/j.bbrc.2014.08.002.

97. Halin S, Hammarsten P, Adamo H, Wikstrom P, Bergh A. Tumor indicating normal tissue could be a new source of diagnostic and prognostic markers for prostate cancer. Expert Opin Med Diagn 2011;5(1):37-47 doi 10.1517/17530059.2011.540009.

98. Forbes CM, Rendon RA, Finelli A, Kapoor A, Moore RB, Breau RH, et al. Disease progression and kidney function after partial vs. radical nephrectomy for T1 renal cancer. Urol Oncol 2016;34(11):486.e17-.e23 doi 10.1016/j.urolonc.2016.05.034.

99. Mehdipour P, Murphy T, De Carvalho DD. The role of DNA-demethylating agents in cancer therapy. Pharmacol Ther 2019:107416 doi 10.1016/j.pharmthera.2019.107416.

100. Vire E, Brenner C, Deplus R, Blanchon L, Fraga M, Didelot C, et al. The Polycomb group protein EZH2 directly controls DNA methylation. Nature 2006;439(7078):871-4 doi 10.1038/nature04431.

101. Wang X, Paucek RD, Gooding AR, Brown ZZ, Ge EJ, Muir TW, et al. Molecular analysis of PRC2 recruitment to DNA in chromatin and its inhibition by RNA. Nat Struct Mol Biol 2017;24(12):1028-38 doi 10.1038/nsmb.3487.

102. Degerman S, Landfors M, Siwicki JK, Revie J, Borssen M, Evelonn E, et al. Immortalization of T-cells is accompanied by gradual changes in CpG methylation resulting in a profile resembling a subset of T-cell leukemias. Neoplasia 2014;16(7):606-15 doi 10.1016/j.neo.2014.07.001.

103. Horvath S. Erratum to: DNA methylation age of human tissues and cell types. Genome Biol. Volume 16. England2015. p 96.

104. Ravaud A, Motzer RJ, Pandha HS, George DJ, Pantuck AJ, Patel A, et al. Adjuvant Sunitinib in High-Risk Renal-Cell Carcinoma after Nephrectomy. N Engl J Med 2016;375(23):2246-54 doi 10.1056/NEJMoa1611406.

105. Thompson JA, Marsit CJ. A methylation-to-expression feature model for generating accurate prognostic risk scores and indentifying disease targets in clear cell kidney cancer. Pac Symp Biocomput 2017;22:509-20 doi 10.1142/9789813207813_0047.

106. Okegawa T, Morimoto M, Nishizawa S, Kitazawa S, Honda K, Araki H, et al. Intratumor Heterogeneity in Primary Kidney Cancer Revealed by Metabolic Profiling of Multiple Spatially Separated Samples within Tumors. EBioMedicine 2017;19:31-8 doi 10.1016/j.ebiom.2017.04.009.

107. Brooks SA, Brannon AR, Parker JS, Fisher JC, Sen O, Kattan MW, et al. ClearCode34: A prognostic risk predictor for localized clear cell renal cell carcinoma. Eur Urol 2014;66(1):77-84 doi 10.1016/j.eururo.2014.02.035.

108. Brannon AR, Reddy A, Seiler M, Arreola A, Moore DT, Pruthi RS, et al. Molecular Stratification of Clear Cell Renal Cell Carcinoma by Consensus Clustering Reveals Distinct Subtypes and Survival Patterns. Genes Cancer 2010;1(2):152-63 doi 10.1177/1947601909359929.