Journal of Pharmaceutical and Biomedical Analysis 240 (2024) 115966

Available online 8 January 2024

Untargeted metabolomics of bladder tissue using liquid chromatography

and quadrupole time-of-ight mass spectrometry for cancer

biomarker detection

Joanna Nizioł

, Krzysztof Ossoli´

nski

, Aneta Płaza-Altamer

, Artur Kołodziej

Anna Ossoli´

nska

, Tadeusz Ossoli´

nski

, Zuzanna Krupa

, Tomasz Ruman

Rzesz´

ow University of Technology, Faculty of Chemistry, 6 Powsta´

nc´

ow Warszawy Ave., 35-959 Rzesz´

ow, Poland

Department of Urology, John Paul II Hospital, Grunwaldzka 4 St., 36-100 Kolbuszowa, Poland

Doctoral School of Engineering and Technical Sciences at the Rzesz´

ow University of Technology, 8 Powsta´

nc´

ow Warszawy Ave., 35-959 Rzesz´

ow, Poland

ARTICLE INFO

Keywords:

Bladder cancer

Biomarkers

Human tissue

Metabolomics

LC-MS

UHPLC-UHRMS

ABSTRACT

Bladder cancer (BC) ranks among the most common cancers globally, with an increasing occurrence, particularly

in developed nations. Utilizing tissue metabolomics presents a promising strategy for identifying potential bio-

markers for cancer detection. In this study, we utilized ultra-high-performance liquid chromatography coupled

with ultra-high-resolution mass spectrometry (UHPLC-UHRMS), incorporating both C18-silica and HILIC col-

umns, to comprehensively analyze both polar and non-polar metabolite proles in tissue samples from 99 pa-

tients with bladder cancer. By utilizing an untargeted approach with external validation, we identied twenty-

ve tissue metabolites that hold promise as potential indicators of BC. Furthermore, twenty-ve characteristic

tissue metabolites that exhibit discriminatory potential across bladder cancer tumor grades, as well as thirty-nine

metabolites that display correlations with tumor stages were presented. Receiver operating characteristics

analysis demonstrated high predictive power for all types of metabolomics data, with area under the curve (AUC)

values exceeding 0.966. Notably, this study represents the rst report in which human bladder normal tissues

adjacent to cancerous tissues were analyzed using UHPLC-UHRMS. These ndings suggest that the metabolite

markers identied in this investigation could serve as valuable tools for the detection and monitoring of bladder

cancer stages and grades.

1. Introduction

Bladder cancer (BC), also known as urological or urinary bladder

cancer, ranks as the tenth most common and thirteenth most deadly

cancer worldwide. According to the latest GLOBOCAN data, bladder

cancer represents around 3% of all cancer cases globally. Its occurrence

is on the rise, especially in developed countries, with approximately

550,000 new cases being diagnosed each year [1]. Environmental and

occupational factors play a signicant role in the development of

bladder cancer, with tobacco smoke being the most signicant risk

factor, responsible for nearly 50% of bladder tumors. Smokers face a

2.5-fold higher risk compared to nonsmokers. Approximately 7% of

bladder cancer cases are attributed to hereditary genetic predisposition

[2].

BC is a diverse condition with a signicant risk of morbidity and

recurrence. BC manifests in both primary and recurrent forms, with

stages ranging from T1 to T4, including Ta [3]. Currently, BC diagnosis

primarily relies on urinary cytology, cystoscopy, and radiological im-

aging. Cystoscopy, though, is invasive, painful, expensive, and less

effective in detecting high-grade supercial tumors, often causing

considerable psychological stress, especially when combined with a bi-

opsy. Urinary cytology, while noninvasive and highly specic, lacks

sensitivity [4].

Recent advancements in BC research have led to the development of

several biomarkers primarily based on protein detection. However, none

of these biomarkers have been incorporated into clinical practice

guidelines due to their insufcient diagnostic performance at the time of

evaluation, high identication costs, and the need for sophisticated

instrumentation, which is not commonly accessible to many clinicians.

Consequently, there are currently no clinically viable biomarkers for the

* Corresponding author.

E-mail address: jniziol@prz.edu.pl (J. Nizioł).

Contents lists available at ScienceDirect

Journal of Pharmaceutical and Biomedical Analysis

journal homepage: www.journals.elsevier.com/journal-of-pharmaceutical-and-biomedical-analysis

https://doi.org/10.1016/j.jpba.2024.115966

Received 27 September 2023; Received in revised form 18 December 2023; Accepted 4 January 2024

Journal of Pharmaceutical and Biomedical Analysis 240 (2024) 115966

early detection, diagnosis, or prognosis of BC [5].

Metabolomics stands out as a potent method for detecting low mo-

lecular compounds in biological samples, playing a crucial role in un-

derstanding disease processes [6]. It focuses on the changes in

metabolites, which are often the result of disruptions in biochemical

pathways caused by diseases. These alterations in metabolites have

gained increasing recognition as vital indicators in clinical diagnostics.

As a result, metabolomics has become a leading technique in the iden-

tication of biomarkers, early disease detection, and the exploration of

related biochemical pathways. By analyzing metabolites present in

biological samples such as urine, serum, and tissue, metabolomics en-

ables the monitoring of the metabolic response of living systems to

diseases or drug toxicity. A key advantage of metabolomics is its ca-

pacity to simultaneously detect hundreds of metabolites, offering a

comprehensive view of biochemical changes [7].

In the past two decades, two main analytical platforms, mass spec-

trometry (MS) and nuclear magnetic resonance spectroscopy (NMR),

have been predominantly utilized for metabolomic analysis of various

bladder cancer samples [8]. Each of these techniques has its advantages

and disadvantages. NMR stands out for its minimal preparation re-

quirements, speed, non-destructive nature, and highly consistent results.

Its advantages also include the ability to analyze a wide array of sample

types such as liquid, gas, solid, and tissue, enhancing its versatility.

However, NMR faces limitations in sensitivity, posing a challenge for

future advancements in this area. NMR often requires a larger amount of

sample compared to other techniques like MS, which can be a limitation

when sample availability is restricted. Moreover, for samples with a

large number of different metabolites, the spectral resolution of NMR

might not be sufcient to distinguish between closely related com-

pounds. In contrast, MS demonstrates superior specicity and sensi-

tivity, making it a more effective technique. MS can analyze a wider

range of compounds, including those that are non-volatile or thermally

unstable, which NMR might not effectively analyze. MS generally re-

quires a smaller amount of sample for analysis compared to NMR, which

is advantageous when sample availability is limited. MS is generally less

affected by the complexity of the sample matrix, which can be a chal-

lenge in NMR analysis. While both NMR and MS can be used for quan-

titative analysis, MS is often preferred for its ability to handle complex

mixtures and provide highly accurate quantitative data. Moreover, MS

can be coupled with various separation techniques like Liquid Chro-

matography (LC) or Gas Chromatography (GC), enhancing its ability to

analyze complex mixtures by separating compounds before detection

[9].

The data derived from NMR and MS experiments in metabolomics

are often intricate, offering both qualitative and quantitative insights

into numerous metabolites. Metabolomics data, known for their high

variability and susceptibility to noise, are inuenced by factors like

environment, diet, age, lifestyle, medications, microbiota, and sample

handling. They also exhibit a higher degree of missing data, compli-

cating multivariate analysis and classication. Therefore, it’s crucial to

apply suitable statistical methods to avoid losing important information

or identifying incorrect trends and models. In metabolomics, multivar-

iate statistical analysis is preferred for its comprehensive approach to

studying system-level changes in biological systems. This analysis in-

cludes both unsupervised methods like Principal Component Analysis

(PCA), mainly used for quality control, and supervised methods like

Partial Least Square Discriminant Analysis (PLS-DA) and its advanced

form, Orthogonal PLS-DA (OPLS-DA), which are crucial for biomarker

discovery. These techniques help in effectively breaking down and

analyzing predictor variables, with OPLS-DA providing clearer models

by distinguishing predictive variance from noise [10].

However, the literature on human bladder tissue analysis in the

context of metabolomics remains limited. The initial exploration of

metabolomic proling in bladder tissues was conducted in 2011 by

Putluri et al. using LC-MS. Their analysis of 58 tissues unveiled signi-

cant alterations in the levels of 35 mass spectral features within the

bladder tissues [11]. In 2013, Tripathi et al. employed the

high-resolution magic angle spinning (MAS) NMR technique to conduct

further research. They identied 22 distinct metabolites associated with

different stages of bladder cancer. These ndings were cross-validated

using targeted GC-MS analysis, although normal, unaffected tissues

were not included as controls [12]. Another study, published in 2017 by

Piyarathna et al., examined 165 bladder tissues, comprising 126 bladder

cancer tissues and 39 benign or normal adjacent tissues. Through

UHPLC-HRMS analysis, they discovered 570 lipids associated with the

survival and diverse clinical stages of bladder cancer [13]. In addition,

our recent study employed nanoparticle-based laser desorption/ioniza-

tion mass spectrometry imaging (LDI-MSI) to investigate BC tissues. This

innovative approach allowed us to identify 10 potential biomarkers

associated with this disease by analyzing the distinct distribution pat-

terns of these compounds on the surface of the imaged tissue fragment

[14].

To discern specic cellular markers that can differentiate between

different grades and stages of BC, numerous metabolomics studies have

been conducted on urine and blood samples from BC patients [7].

However, to the best of our knowledge, only a few studies have exam-

ined the associations between alterations in metabolite levels in tissues

from BC patients and the specic grades and/or stages of tumor pro-

gression [4,12,15].

The identication of BC tissue biomarkers could potentially be

crucial for understanding cancer mechanisms at a molecular level and

aiding in the development of targeted therapies. These biomarkers

might be more specic to certain cancer types than blood biomarkers

and could be invaluable in identifying bladder cancer subtypes for

personalized treatment and prognosis. They might also be essential in

monitoring treatment responses and detecting recurrences, offering in-

sights not always available from serum biomarkers [16]. In the context

of surgical interventions, particularly in procedures like transurethral

resection of bladder tumor (TURBT) or partial cystectomy, BC tissue

biomarkers could play a signicant role. Surgeons might aim to remove

the tumor while conserving as much healthy tissue as possible, but

visually differentiating between cancerous and normal tissue can be

challenging, especially with small tumors or unclear margins. Bio-

markers specic to bladder cancer could provide molecular information

that might help identify tumor boundaries, potentially enhancing the

precision of these surgeries. Furthermore, tissue biomarkers could

potentially be used during tumor removal surgery to accurately delin-

eate tumor margins. This precision would be crucial in surgeries aimed

at preserving organs and in operations near critical structures. Tech-

niques like uorescence-guided surgery might employ biomarkers

linked to uorescent dyes, possibly illuminating cancer cells under

special lighting conditions during surgery [17]. This could aid surgeons

in real-time differentiation between cancerous and non-cancerous tis-

sues, a feature especially benecial in minimally invasive surgeries such

as laparoscopic or robotic procedures. Additionally, these biomarkers

might assist in preserving bladder function by ensuring that only

cancerous tissues are removed, which could be critical for maintaining

quality of life. They could guide targeted biopsies and intraoperative

frozen section analysis, potentially helping conrm that surgical mar-

gins are free of cancer cells.

In this study, UHPLC-UHRMS was employed to investigate metabolic

changes in a substantial sample size of 198 human tissue samples

derived from 99 bladder cancer cases. The primary objective of this

research was to identify the metabolites that distinguish cancerous tis-

sues from normal tissues, while also enabling differentiation between

different stages and grades of bladder cancer.

2. Materials and methods

2.1. Materials and equipment

All solvents were of at least LC-MS grade and were obtained from

J. Nizioł et al.

Journal of Pharmaceutical and Biomedical Analysis 240 (2024) 115966

Merck Life Science Sp.z.o.o (Darmstadt, Germany). Formic acid in LC-

MS grade was purchased from Fisher Scientic GmbH Im (Schwerte,

Germany). High-purity deionized water (>18 MΩ

cm) was produced

locally.

2.2. Collection of human tissue samples

Tissue samples were collected from 99 individuals with bladder

cancer (79 males, 20 females, average age 72) undergoing surgical

therapy at John Paul II Hospital in Kolbuszowa (Poland). The research

was approved by the local Bioethics Committee at the University of

Rzeszow (Poland, permit number 2018/04/10) and followed all current

rules and regulations. All of the patients in this study were Caucasian.

Patients’ specimens and clinical data were acquired with their informed

consent. The ndings of all laboratory tests (complete blood count,

bleeding prole, kidney function tests, and CRP) were all within normal

levels. Tissues for the metabolomic investigation were obtained during a

bladder tumor transurethral resection. We obtained about cubic frag-

ments of 2–4 mm in size of the malignant tumor and a fragment of

normal bladder mucosa for the metabolomic research. Both of these

fragments were sliced in half, with one component examined and the

other sent for histological testing to conrm the diagnosis. The samples

were promptly frozen and stored at −60◦C until future use. Table 1

shows the pathological and clinical characteristics of the individuals.

2.3. Preparation of tissue metabolite extracts

As previously described, medium-to-high polarity metabolites were

recovered from tissue samples [18]. In summary, a weighted amount of

sectioned tissue (0.3–19 mg) was put into a 2 ml centrifuge tube, and a

1:2 MeOH/CHCl

(1:2, v/v, 90

l/mg of tissue) solution was added and

homogenized with borosilicate glass beads for 60 s. After that, each tube

was homogenized twice for 60 s with a 5-minute pause interval in cold

O (12

l/mg of tissue). The samples were incubated at −20 degrees

Celsius for 20 min before being centrifuged at 14,000xg for 10 min at 4

degrees Celsius to remove cells and other precipitated material. The

polar (upper) phase was moved to a new 1.5 ml microcentrifuge tube,

and the non-polar (lower) phase was similarly transferred. Finally, 20%

of the volumes of the polar phase and the whole volume of non-polar

phase samples were taken and used for UHPLC-UHRMS analysis with

C18 and HILIC columns respectively. Using a SpeedVac-type vacuum

concentrator (at 0.9 mbar) all samples were lyophilized to complete

dryness. Due to the relatively big spread of tissue weights, dried polar

extracts were dissolved in constant volume - 1100

l of methanol, and

the tissue weight normalization was performed in the MetaboScape

program (version 2022b). Afterward, the mixtures were incubated at a

temperature of 4 ◦C for 15 min. To separate any remaining particles, the

solutions underwent centrifugation at 4 ◦C and 12,000 x g for 10 min.

From the resulting solutions, volumes of 900

l of polar extracts were

transferred to HPLC vials. Vacuum-dried lower phases (low polar ex-

tracts) were dissolved in weight-normalized volumes of methanol (33.3

l/mg of tissue). The vial, containing the extracted sample, was then

positioned in the autosampler and introduced for analysis. All proced-

ures involving human tissue samples were conducted in compliance

with applicable standards under appropriate biosafety conditions.

Personnel were equipped with standard PPE, including gloves and eye

protection, and worked within designated biosafety cabinets. Chloro-

form was handled in a well-ventilated fume hood with appropriate PPE

to prevent inhalation and skin contact, in accordance with applicable

standards. All waste materials were disposed of following established

protocols for biohazardous and chemical waste.

2.4. Analysis of tissue samples

Tissue extracts were analyzed using a Bruker Elute UHPLC system

with Hystar 3.3 software and an ultra-high-resolution mass spectrometer

Bruker Impact II (60,000 +resolution version; Bruker Daltonik GmbH)

ESI QTOF-MS with Data Analysis 4.2 (Bruker Daltonik GmbH) and

Metaboscape (ver. 2022b) as described in our previous publication [18].

A Waters UPLC column ACQUITY BEH (C18 silica, 1.7 µm particles, 50

×2.1 mm) was used for separation and analysis of polar tissue extracts,

while the HILIC column (1.7 µm particles 50 ×2.1 mm) was used for

non-polar extract of tissues. Suitable column guards for all studies were

employed for all analyses. RP C18 column was used with mobile phases:

A =water with 0.1% formic acid, B =acetonitrile with 0.1% formic acid

(v/v). Samples in the autosampler were thermostated at 4 ◦C, and a

volume of 2

l (5

l for HILIC) of the extract was loaded onto the column

at a ow rate of 200

l/min, with 1% B. The percentage of B varied with

time as follows: 0 min – 1%, 0.56 min - 1% B, 4.72 min - 99%, 5.56 min -

99%, 5.6 min – 1%, 9.45 min – 1%. The solvent ow was set at 450

l/min, and the column was maintained at 40 ◦C. HILIC column was

used with mobile phases: A =95:5 acetonitrile: water (v/v) with 10 mM

Table 1

Participant characteristics.

BC patients

Training set Validation set

Number

General 70 29

Male 56 23

Female 14 6

Age (mean/SD) 73(10) 71(10)

Grade

High grade 27 12

Low grade 39 16

LG (70%) and HG (30%) 0 1

LG (85%) and HG (15%) 1 -

PUNLMP 3 -

Stage

pT1 13 5

pT2 7 3

pTa 50 21

Type of surgery

TURBT 68 29

Cystectomy 2 0

Tumor origin

Primary 40 18

Recurrent 30 11

Hematuria

At diagnosis 69 29

At sampling 51 20

Tumor size [cm]

<1.9 51 9

2-3 3 10

>3 15 6

Multifocal/at 0 4

Multifocality

0 1 0

1 50 21

2-3 7 5

>3 12 3

Previous treatment

BCG 10 4

neoadjuvant chemotherapy (cisplatin based) 2 0

Tumor histology

Papillary 69 27

Flat, CIS 1 0

Solid, non-papillary 1 2

Tobacco smoking

Non smoking 53 21

Currently smoking 9 3

Previous smoking 7 5

Tumors were classied according World Health Organization (WHO)/In-

ternational Society of Urological Pathology (ISUP) classication criteria; BC –

bladder cancer; BCG - bacillus calmette guerin; CIS - carcinoma in situ; LG – low-

grade; HG – high-grade; PUNLMP - papillary urothelial neoplasm of low ma-

lignant potential; pT1 and pTa – high risk non-muscle invasive bladder cancer;

pT2 – muscle invasive bladder cancer; pT- the stage has been based on patho-

logical or microscopic ndings; SD: standard deviation.

J. Nizioł et al.

Journal of Pharmaceutical and Biomedical Analysis 240 (2024) 115966

ammonium acetate, B =1:1 acetonitrile:water (v/v) with 10 mM

ammonium acetate. Percentage of B varied with time as follows: 0 min –

0%, 6 to 7 min – 20%, 7.1 to 8 min – 0%. Internal calibration using 10

mM sodium formate (water: isopropanol 1:1 v/v) ions was automatically

performed in Metaboscape, utilizing a syringe pump at an infusion ow

rate of 0.12 ml/h, in high precision calibration (HPC) mode. Analyses in

positive autoMSMS mode were conducted with specic parameters,

including m/z range of 50–1200, capillary voltage of 4.5 kV, nebulizer

pressure of 2.7 bar, dry gas ow of 12 L/min, drying gas temperature of

220 ◦C, hexapole voltage of 50 Vpp, funnel 1 and funnel 2 voltages of

200 Vpp, pre-pulse storage time of 5

s, and transfer time of 60

Collision-Induced Dissociation (CID) was employed with predetermined

settings: absolute threshold (per 100 sum): 200 cts; absolute threshold

88 cts; active exclusion 3 spectra; release after 0.3 min, isolation mass:

for m/z =100, the width was 3, for 500 width was 4, for 1000 was 6 and

for 1300 was 8); collision energy value was 30 eV. MS frequency was 20

Hz and MSMS from 5 to 30 Hz. Untargeted annotations were carried out

in Metaboscape (ver. 2022b) with specic criteria for mass deviation

(Δm/z) under 2 ppm and mSigma value under 20 to ensure accurate

compound identication. For identication and molecular formula

generation, the exact mass of parent ions was matched with <3 ppm

error and mSigma value <50 in most cases. All the molecular formulas

were obtained using the Smart Formula tool and the C, H, N, O, P, S, Cl,

Br, I, and F elements. MSMS spectra were automatically matched against

MSMS libraries, such as Bruker HMDB 2.0 library, MassBank of North

America (MoNA) library [19], and NIST ver. 2020 MSMS library [20], to

aid in compound identication. Quality control (QC) samples were

regularly prepared from 10 different tissue extracts and measured every

10 samples throughout the analytical run to assess method stability and

repeatability. All measurements were performed in technical triplicates

to ensure data reliability and consistency. Data for replicates was

managed by Bruker Metaboscape ver. 2022b. This program uses the

Time aligned Region complete eXtraction (T-ReX) algorithm that ex-

tracts all relevant information, automatically and in a “region complete”

manner. In cohorts of LC-QTOF-MS/MS data sets ions belonging to the

same compound are combined, aligned across all samples, and auto-

matically re-extracted in individual samples if initially below the peak

picking threshold therefore addressing the missing value problem in

statistics. For each m/z value in each sample, we calculated the mean

value from the triplicate measurements. In cases where one or more of

the triplicate measurements were missing or fell below the detection

limit, we calculated the mean using the available data from the tripli-

cates for that specic m/z value.

2.5. Statistical analysis

All metabolite datasets were analyzed using MetaboAnalyst version

5.0 online software [21]. The multivariate statistical analysis employed

in this study is similar to that detailed in our recent papers [22,23]. In

brief, the metabolite data from each analytical technique was

log-transformed and auto-scaled. Unsupervised Principal Component

Analysis (PCA) and Orthogonal Partial Least Squares Discriminant

Analysis (OPLS-DA) were performed on the obtained metabolite pro-

les. Metabolites with Variable Importance in Projection (VIP) values

more than 1.0, as determined by the OPLS-DA modeling, were regarded

as potentially important discriminators of BC from AN tissue. Validation

and accuracy of the OPLS-DA models were assessed using permutation

tests with 2000 steps and 5-fold cross-validation. To assess the statistical

signicance of tissue metabolite level differences, a paired

non-parametric t-test with Mann-Whitney and Bonferroni correction

and paired fold change (FC) analysis was used. P-values and false dis-

covery rates (FDR) less than 0.05, as well as FC >2.0 or <0.5, were

considered statistically signicant. The P-value and FDR corrections

were applied across the entire list of identied metabolic features.

Furthermore, receiver operating characteristic curve (ROC) evaluations

using random forest modeling were done to validate the OPLS-DA

models and assess the diagnostic value of the metabolites. Metabolite

variables having an AUC (area under the curve) larger than 0.75 were

judged important for distinguishing the BC versus AN tissue metab-

olome. Independent multivariate statistical analyses were performed on

the training and validation metabolite datasets. External validation was

performed using two independent datasets (hence referred to as the

training and validation datasets) to evaluate the performance of the

OPLS-DA models. Both the training and validation datasets were sub-

jected to predened statistical criteria. To identify metabolic pathways

that are most likely impacted by bladder cancer, a metabolic pathway

impact analysis was performed using MetaboAnalyst 5.0 [21] with the

Kyoto Encyclopedia of Genes and Genomes [24] for Homo sapiens and

quantitative pathway enrichment analysis with the Small Molecule

Pathway Database (SMPD), We used a one-way analysis of variance

(ANOVA) with Tukey’s post-hoc testing to see if there were any signif-

icant differences in average math test results between different stages

and grades of BC.

3. Results

In this study, we characterized the metabolic prole of cancer and

normal tissue extracts from 99 patients diagnosed with BC to identify the

compounds that most differentiate the bladder cancer tissue (BCT) from

the adjacent tissue (ANT). Following the extraction of tissue samples, we

obtained two distinct fractions – polar and low-polar one. To analyze

these fractions effectively, we employed a C18 column for the polar

fraction and a HILIC column for well-controlled retention of the low

polar compounds of the low polar fraction. From both measurements,

two distinct datasets were created: a training set consisting of 70% of all

samples and a validation set containing the remaining 30% of samples.

Subsequently, tissue metabolic proling was conducted separately on

these two datasets. The training dataset was utilized to identify tissue

markers that differentiate between ANT and BCT and also both high-

grade and low-grade BCT. The validation dataset was employed to

independently validate the diagnostic performance of tissue metabolite

biomarkers. However, when analyzing samples from different stages of

cancer, as well as different age and gender groups, the number of

available samples was insufcient to carry out a reliable statistical

analysis when dividing them into two independent sets. Consequently,

the study was performed using the entire dataset as a whole. Clustering

of QC samples prepared from bladder tissue extracts measured on HILIC

and C18 columns is presented as PCA plots in the supplementary data in

Fig. S1.

3.1. Distinguishing between normal tissue and bladder cancer

A total of 1999 m/z features were identied using a HILIC column,

while 659 m/z features were detected using a C18 column in each tissue

sample. These features were required to be present in at least nine

samples corresponding to the smallest group of cancer subtypes. The

dataset obtained from the HILIC and C18 columns was partitioned into

separate training and validation sets. Subsequently, multivariate statis-

tical analysis was performed on these datasets using the MetaboAnalyst

v 5.0 platform.

For the tissue extract data obtained from the C18 column, the un-

supervised PCA analysis revealed a partial separation between BCT and

ANT samples. The rst two principal components (PC1 and PC2)

accounted for 75.0% and 4.6% of the total variance, respectively,

demonstrating a signicant degree of group separation in the training

set. Outliers were sparsely distributed within the central 95% of the eld

of view, as depicted in Fig. 1A. Furthermore, in the validation set, PC1

(66.8%) and PC2 (7.3%) exhibited the most prominent separation be-

tween BCT and ANT samples, as shown in Fig. S2A of the supplementary

information.

To elucidate the metabolic distinctions between the BCT and ANT

groups, a supervised OPLS-DA analysis was employed. The score plot

J. Nizioł et al.

Journal of Pharmaceutical and Biomedical Analysis 240 (2024) 115966

generated from the training set exhibited a clear demarcation between

the two groups, as illustrated in Fig. 1B. To validate the OPLS-DA model,

5-fold cross-validation and 2000 permutation tests were conducted

(refer to Table S1 in the supplementary information). Notably, the re-

sults indicated robust discrimination between the groups, with a Q

value of 0.921, an R

Y value of 0.954, and an extremely low P-value of

5E-04 (0/2000). These ndings underscore the presence of signicant

metabolic variations between BCT and ANT samples. The OPLS-DA

model demonstrated high interpretability and predictability, as evi-

denced by its elevated R

Y and Q

values. Consistently, the validation

set’s OPLS-DA model exhibited a similar ability to discriminate between

BCT and ANT, with excellent permutation test outcomes (Q

=0.901,

Y=0.989, p-value 5E-04 (0/2000)). The VIP plot, derived from the

OPLS-DA model, served as a valuable tool for identifying potential tissue

biomarkers associated with BC. To evaluate the statistical signicance of

metabolite level differences, a comprehensive analysis was conducted,

employing paired fold change analysis, paired non-parametric t-tests

(Mann-Whitney), and Bonferroni correction. Subsequently, univariate

ROC analysis was performed on both the training and validation sets to

assess the diagnostic capability of the models, as depicted in Fig. 1C. The

area under the ROC curve (AUC), a reliable indicator of model perfor-

mance, was utilized as a metric to evaluate the sensitivity and specicity

Fig. 1. Analysis of BCT and ANT metabolite proles in the training set. (A) 2D PCA and (B) OPLS-DA scores plots and (C) ROC curves of the BCT (violet) and ANT

(orange) samples obtained using column C18; (D) 2D PCA and (E) OPLS-DA scores plots and (F) ROC curves of the BCT (violet) and ANT (orange) samples obtained

using column HILIC; (G-K) Box-whisker plots of selected metabolites levels in tissue samples from BCT and ANT.

J. Nizioł et al.

Journal of Pharmaceutical and Biomedical Analysis 240 (2024) 115966

of the identied biomarkers. By integrating VIP scores (>1.0), t-tests

(FDR and P-values <0.05), fold change (FC >2.0 or <0.5), and AUC (>

0.75) analyses from both the training and validation sets, a total of nine

metabolites were identied as signicant discriminators between BCT

and ANT tissue. Among the m/z values that met the statistical criteria in

both the training and validation sets, our focus was solely on those that

corresponded to specic compounds with conrmed structures.

Moreover, only metabolites naturally occurring in the human body were

considered for further analysis and interpretation. These signicant

metabolites are listed in Table 2.

A similar analysis was performed on data obtained using the HILIC

column. The PCA analysis revealed a clear separation of BCT and ANT.

The rst two principal components (PC1 and PC2) accounted for 36.7%

and 9% of the total variance, indicating signicant group separation in

Table 2

Differential metabolites for discrimination between BCT and ANT (P-value and FDR <0.001; VIP >1; FC <0.5 and >2; AUC >0.8).

No. Name Column

type

Formula m/z

[s]

VIP

P-

value

FDR

AUC

Spec.

[%]

Sens.

[%]

1 Heptadecanoic acid

f,h,i

HILIC C

288.2894 46 1.63 0.0 2.5E-12 1.3E-

1.000 100 100

2 Isobutyrylcarnitine

f,h,i

HILIC C

232.1542 404 1.61 16.6 2.5E-12 1.3E-

1.000 100 100

3 PC(34:4)

f,h,i

HILIC C

P 754.5356 238 1.40 84.1 2.5E-12 1.3E-

1.000 100 100

4 SM(34:1)

f,h,i

HILIC C

P 703.5744 307 1.67 5.7 2.5E-12 1.3E-

1.000 100 100

5 Valerylcarnitine

f,h,i

HILIC C

246.1698 384 1.67 5.7 2.5E-12 1.3E-

1.000 100 100

6 Hypoxanthine

f,h,i

HILIC C

O 137.0456 67 1.41 0.0 2.5E-12 1.3E-

0.998 65 75

7 C17 Sphinganine

f,h,i

C18 C

288.2895 214 1.16 0.4 2.7E-09 9.8E-

0.815 79 80

8 C17 Sphingosine

f,h,i

HILIC C

286.2739 28 1.34 0.0 2.5E-12 1.3E-

0.995 91 99

9 PC(38:3)

f,h,i

HILIC C

P 812.6133 227 1.32 113.7 3.6E-12 1.5E-

0.987 92 92

10 Ortophosphate

f,i

HILIC H

P 98.9839 26 1.42 0.0 2.5E-12 1.3E-

0.986 95 95

C18 H

P 98.9840 28 1.42 0.3 4.8E-12 7.0E-

0.886 80 90

11 Inosine

f,g,h,i

HILIC C

269.0879 81 1.35 30.9 2.7E-12 1.3E-

0.979 97 92

12 Iminodiacetic acid

f,i

HILIC C

134.0446 165 1.51 0.0 1.1E-11 3.6E-

0.969 100 94

13 Uracil

f,i

HILIC C

113.0344 33 1.19 0.1 1.8E-11 5.2E-

0.962 99 89

14 PC(36:5)

f,h,i

HILIC C

P 780.5531 227 1.36 58.0 5.2E-12 1.9E-

0.950 99 92

15 8,11,14-

Eicosatrienoylethanolamide

f,h,i

HILIC C

350.3050 29 1.38 0.0 1.0E-11 3.3E-

0.931 89 97

16 PC(16:1/16:1)

f,h,i

HILIC C

P 730.5379 237 1.29 32.3 8.2E-12 2.7E-

0.923 99 91

17 PC(34:1)

f,h,i

HILIC C

P 760.5846 234 1.01 0.1 3.5E-09 7.1E-

0.904 94 85

18 Glycerophosphocholine

f,g,h,i

C18 C

P 258.1100 21 1.87 67.9 2.7E-11 2.5E-

0.893 83 93

19 PC(40:6)

f,h,i

HILIC C

P 834.6004 221 1.20 47.4 6.1E-11 1.6E-

0.887 88 80

20 LPC(18:2)

f,h,i

HILIC C

P 520.3393 321 1.17 28.8 1.7E-10 4.1E-

0.877 99 82

21 5-Dodecenoic acid

f,g,h,i

C18 C

199.1692 234 1.35 0.4 1.5E-11 1.6E-

0.871 86 81

22 PC(18:0e)

f,h,i

HILIC C

P 524.3701 331 1.05 6.4 1.6E-10 4.0E-

0.865 94 83

23 Palmitoleoylethanolamide

f,h,i

C18 C

298.2738 236 1.32 0.4 5.6E-11 4.4E-

0.850 83 80

24 Hydrocinnamic acid

f,g,h,i

C18 C

151.0752 132 1.21 0.4 3.3E-10 1.7E-

0.833 80 79

25 Sorbose

f,g,h

C18 C

203.0526 22 1.20 0.4 3.4E-09 1.2E-

0.823 86 81

SM: sphingomyelin; VIP: variable inuence on projection

Experimental monoisotopic mass of ion;

VIP scores derived from OPLS-DA model;

fold change between cancer and control tissue extract calculated from the abundance mean values for each group – cancer-to-normal ratio;

Derived from paired non-parametric t-test;

ROC curve analysis for individual biomarkers;

the metabolites identied by high precursor mass accuracy;

the metabolites identied by matching retention time;

the metabolites identied by matching isotopic pattern;

the metabolites identied by matching MS/MS fragment spectra; AUC: area under the curve; FC: fold change; FDR: false discovery rate; LPC: lysophosphati-

dylcholine; m/z: mass-to-charge ratio; PC: phosphatidylcholine; RT: retention time; Sens.: Sensitivity; Spec.: Specicity;

J. Nizioł et al.

Journal of Pharmaceutical and Biomedical Analysis 240 (2024) 115966

the training set. As seen in Fig. 1D, outliers were sparsely dispersed in-

side the middle 95% of the eld of vision. Furthermore, as shown in

Fig. S2D of the supplementary information, PC1 (37.4%) and PC2

(8.1%) demonstrated the most noticeable separation between BCT and

ANT samples in the validation set. The OPLS-DA score plots generated

for both the training and validation sets demonstrated exceptional dif-

ferentiation between the two groups, as depicted in Fig. 1B and

Figure 1SB (supplementary material). The 5-fold cross-validation and

permutation test were performed to validate the OPLS-DA model (refer

to Table S2 in the supplemental information). Notably, the results

showed strong group discrimination, with a Q

value of 0.997, an R

value of 0.998, and a low P-value of 5E-04 (0/2000). These data high-

light the existence of signicant metabolic differences between BCT and

ANT samples. A comprehensive analysis was conducted to determine the

statistical signicance of metabolite level differences. This analysis

included, additionally, paired fold change analysis, paired non-

parametric t-tests (Mann-Whitney), and Bonferroni correction and uni-

variate ROC analysis (Fig. 1F, S1F) was performed on both the training

and validation sets to evaluate the statistical signicance of metabolite

level differences and the diagnostic capacity of the models. Using the

same statistical criteria as before (VIP >1.0), FDR and P-values <0.05,

FC >2.0 or <0.5, and AUC >0.75) from both the training and vali-

dation sets, a total of twenty-ve metabolites were identied as signif-

icant discriminators between BCT and ANT tissue using HILIC column.

These signicant metabolites are listed in Table 2.

These ndings suggest that in total, twenty-ve metabolites may

have enhanced diagnostic potential and could be valuable indicators of

malignant versus normal tissues of patients with bladder cancer when

evaluated together.

3.2. Distinguishing between grades of bladder cancer

To evaluate the potential of tissue extract metabolite proles in

distinguishing between different grades of BCT and ANT, we conducted

a multivariate statistical analysis on separate training and validation

datasets. The analysis involved 96 tissue samples obtained from patients

diagnosed with high-grade (HG) and low-grade (LG) cancer, excluding

three samples from patients with papillary urothelial neoplasm of low

malignant potential (PUNLMP). We used the training dataset,

comprising 26 HG BCT and HG ANT samples, along with 39 LG BCT and

LG ANT samples. To verify the reliability and consistency of the distinct

group clustering observed in the PCA model, we employed the valida-

tion dataset comprising 12 HG BCT and ANT samples and 16 LG BCT and

ANT samples.

In both the training and validation datasets obtained using the HILIC

and C18 column, the PCA and OPLS-DA scores plots demonstrated clear

differentiation between HG BCT and ANT groups, as depicted in Fig. S3

A-D of the supplementary material. The OPLS-DA models were validated

using a 5-fold cross-validation and permutation test, demonstrating

robust group discrimination. The results exhibited strong performance,

with a Q

value exceeding 0.960, an R

Y value surpassing 0.865, and a

very low P-value of 5E-04 (0/2000) (Table S1, supplementary material).

Additional analyses such as ROC analysis, paired no-parametrical t-test,

and fold change analysis were conducted. These assessments, along with

careful consideration of statistical parameters (VIP >1, P-value, FDR <

0.05, FC <0.5 or >2.0, AUC >0.70) led to the identication of 26

specic compounds. These compounds exhibited the highest potential

for distinguishing cancerous tissue from normal tissue in patients diag-

nosed with HG BC (Table S2, supplementary data).

Similar analyses were conducted on tissue extracts collected from

patients diagnosed with LG BC. The PCA and OPLS-DA scores plots in

both the training and validation datasets produced using the HILIC and

C18 column exhibited clear differences between the LG BCT and LG ANT

groups, as shown in Fig. S3 E-H of the supplemental material. A vali-

dation of the OPLS-DA models, revealing robust group discrimination.

The results were strong, with a Q

value greater than 0.959, an R

value greater than 0.821, and an extremely low P-value of 5E-04 (0/

2000) (Table S1, supplemental material). Following that, additional

analyses were performed, including ROC analysis, paired no-

parametrical t-test, and FC analysis. These evaluations, combined with

careful consideration of statistical characteristics (VIP >1, P-value, FDR

0.05, FC 0.5 or >2.0, AUC >0.70), resulted in the discovery of 24

distinct compounds. In individuals with LG BC, these metabolites have

the greatest potential for differentiating malignant cells from normal

tissue. Unfortunately, comparing the three cancer grade groups (HG BCT

versus LG BCT) revealed no statistically signicant differences (data not

shown).

3.3. Distinguishing between different stages of bladder cancer

A metabolomics analysis of tissue samples was also performed to see

if distinct metabolite patterns can help distinguish between stages of BC.

On a total of 198 tissue extracts from 71 patients with pTa BC, 18 pa-

tients with pTa BC, and 10 patients with pT2 BC, we performed PCA,

OPLS-DA, ROC analysis, non-parametric paired t-test, and FC analysis.

However, due to the limited availability of patients with pT1 and pT2

malignancy, the analysis was carried out without stratication into

training and validation sets. The obtained datasets from both the HILIC

and C18 columns were subjected to analysis. In both cases, the PCA and

OPLS-DA scores plots clearly exhibited distinct separation between the

pTa, pT1, and pT2 stages of cancer tissues and their corresponding

adjacent normal tissues. These ndings are visually represented in

Fig. S4 of the supplementary material. A validation of these OPLS-DA

models, demonstrates strong predictive power, accurately distinguish-

ing between the different groups with a Q

value greater than 0.976 and

a high level of accuracy in capturing the variance in the dataset with, an

Y value greater than 0.894 (Table S1, supplementary material). Based

on the cut-off criteria (FC >2 or <0.5, VIP >1; AUC >0.75, P-value and

FDR <0.05), nally, 26, 30, and 33 metabolites appeared to be most

relevant for sample distinction between pTa BCT vs. pTa ANT, pT1 BCT

vs. pT1 ANT and pT2 BCT vs. pT2 ANT, respectively (Table S3, sup-

plementary material). Unfortunately, comparing the three cancer stage

groups (pT1 BCT versus pTa BCT versus pT2 BCT) revealed no statisti-

cally signicant differences (data not shown). LC-MS data for all iden-

tied compounds are provided in Table S4 of supplementary data.

3.4. Pathway analysis of potential biomarkers

MetaboAnalyst 5.0 was utilized to conduct a metabolic pathway

impact analysis, with the goal of identifying the key pathways impli-

cated in the observed alterations in tissue metabolite levels. Pathway

and quantitative pathway enrichment analyses were performed on a set

of 41 compounds that played a crucial role in distinguishing BCT and

ANT. Notably, these analyses also considered the compounds that

exhibited signicant differences across various stages and grades of BC.

Out of the 41 compounds examined, a total of 40 compounds were found

to be relevant to human metabolism. Leveraging the KEGG (Kyoto

Encyclopedia of Genes and Genomes) database, the pathway analysis

revealed 13 pathways specically associated with BC. Among these

identied pathways, four were found to be particularly signicant:

glycerophospholipid metabolism, purine metabolism, pyrimidine

metabolism, and linoleic acid metabolism. These ndings are illustrated

in Fig. 2A and detailed in Table S5 of the supplementary information.

To broaden the investigation of BC-related pathways in our metab-

olomic study, we performed a quantitative enrichment analysis using

the pathway enrichment module of MetaboAnalyst 5.0, utilizing the

Small Molecule Pathway Database (SMPDB). Through this analysis, a

comprehensive set of 59 pathways associated with BC was identied.

Among these pathways, two additional pathways emerged as particu-

larly signicant: lactose synthesis and glycolysis. These ndings, visu-

ally represented in Fig. 2B and detailed in Table S6 of the supplementary

information, provide crucial insights into the metabolic processes

J. Nizioł et al.

Journal of Pharmaceutical and Biomedical Analysis 240 (2024) 115966

underlying bladder cancer development and progression.

4. Discussion

In this study, our primary objective was to employ untargeted

metabolic proling to investigate tissue extracts obtained from 99 pa-

tients diagnosed with BC. Our aim was to uncover distinctive metabolic

signatures that could serve as potential biomarkers for early and accu-

rate detection of BC, utilizing the UHPLC-UHRMS technique. To

accomplish this, we conducted an untargeted analysis of both ANT and

BCT, seeking to identify a panel of metabolites that exhibited differential

abundance between the two tissue types. Through our analysis, we

successfully identied 25 metabolites. Notably, 12 of these compounds

were found to be present at higher levels in cancer tissue compared to

adjacent normal tissue, while 13 compounds were present at lower

concentrations. The elevated levels of these 12 metabolites could

potentially indicate increased synthesis of tumor-related metabolites,

which may be secreted by cancer cells or result from alterations in the

composition of non-cancerous tissues due to tumor inltration through

the epithelial barrier. Moreover, the presence of tumors could trigger

inammatory responses, contributing to the observed elevation of

certain metabolites.

Our research utilized the UHPLC-HRMS method to compare urine

and serum proling results in bladder cancer (BC) patients [22,25]. In

this study, we focused on investigating tissue biomarkers to complement

information obtained from serum and urine biomarkers. Our ndings

reveal that changes in tumor tissues can signicantly affect the

composition of metabolites in the blood and urine. However, these

correlations are often complex and not always direct, with some bio-

markers only detectable in tissue, particularly in early cancer stages.

One key discovery was the markedly elevated level of valer-

ylcarnitine in cancerous tissue extracts, showing 100% specicity and

sensitivity, a nding corroborated in blood serum [25] and contrasted

by decreased levels in urine [22]. The same class of compounds also

includes isobutyrylcarnitine, which also turned out to be an excellent

tissue marker with 100% specicity and sensitivity. Valerylcarnitine and

isobutyrylcarnitine, both acylcarnitines, play crucial roles in fatty acid

metabolism and energy production, and their altered levels are indica-

tive of the cancer cells’ aberrant metabolism, commonly referred to as

the Warburg effect [26]. This effect leads to an accumulation of specic

metabolites, like valerylcarnitine, in cancer cells and consequently in

the tissues, with potential implications for cancer diagnosis and treat-

ment [27]. The elevated levels of valerylcarnitine in serum could be due

to the release of this metabolite from the tumor tissues into the blood-

stream. As tumor cells proliferate and metabolize, they may release

more of certain metabolites as a byproduct of their increased metabolic

activities. Previous research has indicated elevated levels of carnitines in

bladder cancer tumor tissues. Several studies have investigated the

metabolomic proles of BC and identied dysregulated metabolites,

including carnitines. For instance, a study conducted by Putluri et al.

[11], analyzed the metabolomic proles of BC tissues using liquid

chromatography-mass spectrometry (LC-MS). The ndings of this study

revealed that carnitine levels were signicantly elevated in bladder

cancer tumor tissues when compared to adjacent benign and normal

Fig. 2. Analysis of the topology of selected statistically signicant tissue metabolites in BC. (A) Pathway analysis based on KEGG, with bubble area corresponding to

the impact of each pathway and color representing signicance from red to white, from greatest to least. (1) Glycerophospholipid metabolism; (2) Purine metabolism;

(3) Pyrimidine metabolism; (4) Linoleic acid metabolism; (B) Quantitative enrichment analysis based on SMPDB. KEGG: Kyoto Encyclopedia of Genes and Genomes;

SMPDB: The Small Molecule Pathway Database.

Fig. 3. Venn diagrams illustrate the number of tissue metabolites common and

unique in different (A) grade and (B) stages of BC.

J. Nizioł et al.

Journal of Pharmaceutical and Biomedical Analysis 240 (2024) 115966

tissues. Similarly, subsequent examination of the metabolic prole of

urine samples from BC patients indicated elevated levels of carnitines

such as acetylcarnitine and isovalerylcarnitine in the urine of BC pa-

tients compared to controls [28–30]. These ndings suggest that alter-

ations in carnitine metabolism are associated with bladder cancer

development and progression.

Another compound that was related to our previous results was the

lipid PC(18:0) a specic type of phosphatidylcholine (PC), which is a

major class of phospholipids and a crucial component of cell mem-

branes. Our results indicated that PC(18:0) was present in signicantly

higher amounts in tumor tissues compared to adjacent normal tissues

(FC =6.4), and in lower amounts in serum from the same BC patients

compared to the control group (FC =0.69) [25]. Apart from PC(18:0),

11 other lipids belonging to the PC class were found to be signicantly

associated with cancer tissues, of which, interestingly, as many as 6 were

signicantly changed only in patients with pT2 stage of bladder cancer

(Table 3). Among these compounds, PC(34:4) had the highest diagnostic

ability with 100% specicity and sensitivity (Table 2). Phosphatidyl-

cholines, including PC(18:0), are integral in maintaining membrane

integrity, and uidity, and in facilitating signal transduction processes.

Aberrant membrane choline phospholipid metabolism (MCPM) is a

hallmark of cancer cells, leading to altered levels of choline-containing

metabolites like choline, phosphocholine, glycerophosphocholine, and

phosphatidylcholine. These alterations are driven by the activation of

biosynthetic enzymes such as choline kinase, and catabolic enzymes like

phospholipases A, C, and D [31]. In bladder cancer, particularly, the

phospholipase C (PLC) pathway, among others, is notably affected,

contributing to abnormal gene regulation and tumor development. This

dysregulation is implicated in the malignant transformation in various

cancers, making these enzymes potential targets for anticancer chemo-

therapies [32]. Our ndings align with the research by Tripathi et al.

[12], which revealed elevated levels of phosphatidylcholine and other

choline-containing compounds in tumor tissues compared to benign

tissues. Similar dysregulation in phosphatidylcholines has been

observed in other studies, further conrming the disrupted MCPM in

bladder cancer [33]. This dysregulation may be reected in the altered

lipid metabolism of cancer cells. In cancerous tissues, there is an

increased activity of lipid synthesis to support the rapid proliferation of

cancer cells, potentially leading to the accumulation of compounds like

PC(18:0). Conversely, the exchange of phospholipids such as PC(18:0)

between cell membranes and blood may be impaired in cancer cells,

leading to lower levels of PC(18:0) in the serum. Additionally, bladder

cancer can inuence local blood ow and substance distribution,

causing variations in metabolite concentrations between tumor sites and

the bloodstream. This complex interplay underscores the signicance of

PC(18:0) as a potential biomarker and a participant in the

Table 3

Differential metabolites for discrimination in all groups (P-value and FDR <0.001; AUC >0.8).

No. Name BCT vs. ANT LG BCT vs. ANT HG BCT vs. ANT pTa BC vs. ANT pT1 BC vs. ANT pT2 BC vs. ANT

VIP FC VIP FC VIP FC VIP FC VIP FC VIP FC

1 3-Methylhistidine - - - - - - - - - - 1.06 0.117

2 5-Dodecenoic acid 1.35 0.379 1.35 0.377 1.32 0.379 1.36 0.370 1.26 0.295 - -

3 8,11,14-Eicosatrienoylethanolamidef 1.38 0.033 1.35 0.036 1.35 0.048 1.33 0.041 1.53 0.052 1.06 0.048

4 Acetylcarnitine - - 1.07 0.055 - - 1.01 0.071 - - 1.09 0.115

5 Adenine - - 1.55 119.690 1.56 207.540 1.55 126.760 1.54 171.430 1.53 138.940

6 C17-Sphinganine 1.16 0.442 - - 1.06 0.461 - - 1.17 0.330 - -

7 C17-Sphingosine 1.34 0.031 1.33 0.028 1.29 0.040 1.34 0.033 1.32 0.039 1.36 0.024

8 Glycerophosphocholine 1.87 67.856 1.74 9.577 1.57 27.991 1.81 9.318 1.32 20.114 - -

9 Heptadecanoic acid 1.63 0.003 1.63 0.003 1.59 0.003 1.63 0.003 1.61 0.003 1.55 0.003

10 Hydrocinnamic acid 1.21 0.432 1.21 0.423 1.2 0.430 1.22 0.418 1.19 0.341 - -

11 Hypoxanthine 1.41 0.013 1.41 0.010 1.37 0.022 1.4 0.014 1.43 0.008 1.33 0.029

12 Iminodiacetic acid 1.51 0.020 1.47 0.020 1.51 0.021 1.46 0.021 1.38 0.034 1.48 0.049

13 Inosine 1.35 30.883 1.34 32.208 1.33 25.262 1.32 29.105 1.26 13.671 1.34 32.877

14 Isobutyrylcarnitine 1.61 16.593 1.6 16.010 1.56 16.636 1.59 15.049 1.56 17.967 1.51 11.719

15 LPC(18:2) 1.17 28.845 1.09 16.960 1.29 51.270 1.19 22.118 1.09 20.026 1.07 19.064

16 Oleamide - - - - - - - - - - 1.03 2.910

17 Ortophosphate 1.42 0.015 1.41 0.016 1.39 0.334 1.42 0.013 1.52 0.010 1.26 0.023

18 Palmitamide - - - - - - - - 1.04 0.158 - -

19 Palmitoleoylethanolamide 1.32 0.380 1.32 0.381 1.26 0.386 1.35 0.364 1.17 0.326 1.08 8.400

20 Palmitoylcarnitine - - - - - - - - - - 1.08 8.353

21 PC(16:0/16:0) - - - - - - - - - - 1.09 3.435

22 PC(16:1/16:1) 1.29 32.284 1.26 29.925 1.33 34.979 1.29 34.072 1.21 13.062 1.32 30.346

23 PC(18:0e) 1.05 6.393 - - 1.15 6.764 1.04 6.916 - - 1.26 6.715

24 PC(30:0) - - - - - - - - - - 1.04 4.811

25 PC(34:1) 1.01 0.057 - - - - - - 1.06 0.096 1.15 0.172

26 PC(34:4) 1.4 84.147 1.38 79.115 1.42 79.586 1.4 89.443 1.4 41.755 1.41 65.550

27 PC(35:2) - - - - - - - - - - 1.26 32.427

28 PC(36:3) - - - - - - - - - - 1.1 10.819

29 PC(36:5) 1.36 57.985 1.31 50.443 1.39 75.985 1.34 53.697 1.26 45.363 1.36 60.950

30 PC(38:3) 1.32 113.740 1.29 73.122 1.38 162.680 1.3 100.260 1.3 64.000 1.36 138.300

31 PC(40:6) 1.2 47.412 1.08 30.398 1.36 89.348 1.16 40.833 1.24 20.093 1.26 54.511

32 PC(O-34:1) - - - - - - - - - - 1.03 7.846

33 PC(O-36:5) - - - - - - - - - - 1.26 51.256

34 Propionylcarnitine - - - - - - - - 1.3 0.000 - -

35 SM(34:1) 1.67 5.710 1.66 5.687 1.62 5.508 1.66 5.657 1.64 5.633 1.57 5.500

36 Sorbose 1.2 0.429 1.18 0.410 1.41 0.469 1.18 0.419 1.26 0.337 1.45 0.476

37 Stearoylcarnitine - - - - - - - - - - 1.14 14.060

38 Uracil 1.19 0.071 - - - - - - 1.31 0.068 1.16 0.089

39 Valerylcarnitine 1.67 5.669 1.66 5.630 1.62 5.682 1.66 5.733 1.64 5.520 1.57 5.498

VIP scores derived from the OPLS-DA model;

fold change between cancer and control tissue extract calculated from the abundance mean values for each group – cancer-to-normal ratio; AUC: area under the curve;

FC: fold change; FDR: false discovery rate; LPC: lysophosphatidylcholine; m/z: mass-to-charge ratio; PC: phosphatidylcholine; SM: sphingomyelin; VIP: variable in-

uence on projection

J. Nizioł et al.

Journal of Pharmaceutical and Biomedical Analysis 240 (2024) 115966

pathophysiology of bladder cancer.

Another compound that was linked to our earlier research on the

same group of patients was palmitoleoylethanolamide (PEA). PEA was

signicantly lower in extracts from cancerous tissues compared to

adjacent healthy tissues, and conversely, its levels were signicantly

higher in the serum of BC patients than in the control group. This pattern

can be attributed to the distinct metabolic behaviors of cancer cells.

Unlike normal cells, cancer cells often have altered lipid metabolism,

which could lead to the depletion of PEA in tumor tissues due to its

higher consumption by rapidly growing cancer cells.

Our research has shown that, in addition to isobutyrylcarnitine and

valerylcarnitine, the excellent ability to distinguish healthy tissue from

the adjacent normal tissue with an AUC value of 1.0, indicating 100%

specicity and sensitivity is demonstrated by three other lipids: PC

(34:4), SM(34:1) and heptadecanoic acid. SM(34:1) is a sphingomyelin

molecule involved in cell signaling and membrane uidity. It’s linked to

diseases like cancer and has been associated with drug resistance and

cancer progression, including cell proliferation and apoptosis. Several

prior studies have consistently reported increased levels of various

sphingomyelin species in both BC cell lines [34] and the urine of in-

dividuals diagnosed with bladder cancer [29]. Heptadecanoic acid, a

saturated fatty acid with 17 carbon atoms, was found in lower amounts

in cancerous tissues compared to healthy ones. Naturally present in

dairy, meats, and certain oils, it serves as an energy source and is a

component of lipids like triglycerides and phospholipids in cell mem-

branes. Heptadecanoic acid has been previously identied as a potential

cancer biomarker. For example, ndings from the analysis of tissue

samples from 66 patients with invasive ovarian carcinomas and 9 pa-

tients with borderline tumors of the ovary indicated that saturated fatty

acids, including heptadecanoic acid, exhibited signicant reductions in

invasive ovarian carcinomas compared to ovarian borderline tumors

[35]. In relation to the BC a study conducted by Pasikanti et al. [36]

using two-dimensional gas chromatography time-of-ight mass spec-

trometry (GC×GC−ToFMS) revealed elevated levels of heptadecanoic

acid in the urine of BC patients compared to controls.

A comprehensive analysis identied 23 differential metabolites that

hold promise as potential markers for distinguishing between LG BCT

and ANT (Table 3, S2, supplementary data). Among these metabolites,

adenine, heptadecanoic acid, hypoxanthine, isobutyrylcarnitine, PC

(34:4), SM(34:1), and valerylcarnitine exhibited the highest levels of

specicity and sensitivity, making them the most distinguishing com-

pounds in this model. Apart from heptadecanoic acid and hypoxanthine

other compounds were identied in signicantly higher levels in the

BCT compared to ANT. Also, twenty-four differential metabolites were

identied as a potential marker for discriminating between HG BCT and

ANT. Among these metabolites, 5 had tremendous discriminant signif-

icance with the highest AUC, including heptadecanoic acid, iso-

butyrylcarnitine, PC(34:4), SM(34:1), and valerylcarnitine.

Interestingly, most metabolites that distinguished cancerous from

healthy bladder tissue were common to both LG and HG bladder cancer

patients. Acetylcarnitine uniquely differentiated cancerous from healthy

tissues only in LG patients, being notably lower in cancerous tissues.

Only in HG bladder cancer patients, C17-sphinganine was identied in

signicantly lower amounts, while PC(18:0e) was found in substantially

higher levels in tumor tissues compared to adjacent healthy tissues.

Our study demonstrates that a tissue-based metabolite prole can

effectively distinguish between different stages of bladder cancer (pTa,

pT1, and pT2), as well as normal tissue (Table S3, supplementary data).

Specically, we identied 24 differentiating compounds in the tissue of

pTa stage patients, 27 compounds in pT1 stage patients, and 32 com-

pounds in pT2 stage patients, all exhibiting an area under the curve

(AUC) greater than 0.804. These ndings highlight the potential of these

metabolites as robust indicators for accurately characterizing the

various stages of bladder cancer. In the analysis of different stages of BC,

most metabolites distinguishing tumor tissue from adjacent healthy

tissue were common across stages pTa, pT1, and pT2. However, a

notable nding emerged in pT1 BC patients, where three metabolites,

including C17-sphinganine, palmitamide, and propionylcarnitine, were

signicantly lower in tumor tissues compared to adjacent healthy tis-

sues. In patients with stage pT2 BC, nine compounds were identied in

signicantly higher quantities in tumor tissues compared to adjacent

healthy ones. These included oleamide, palmitoylcarnitine, PC(16:0/

16:0), PC(30:0), PC(35:2), PC(36:3), PC(O-34:1), PC(O-36:5), and

stearoylcarnitine, alongside 3-methylhistidine, which was found in

signicantly lower amounts in tumor tissues compared to healthy

adjacent tissues.

5. Conclusion

In summary, our study demonstrates the efcacy of ultra-high-

performance liquid chromatography and ultra-high-resolution mass

spectrometry in identifying tissue metabolome changes in patients with

BC. We have identied thirty-nine metabolites that hold promise for

distinguishing cancerous tissue from adjacent normal tissue, considering

the differentiation across various grades and stages of BC. Notably, our

research represents a novel contribution as there is currently no pub-

lished evidence highlighting these specic combinations of metabolites

as potential crucial markers for early detection of BC. The identied

tissue biomarkers of the BC potentially can be used during tumor

removal surgery. They might enable surgeons to accurately delineate

tumor margins, ensuring the complete removal of cancerous tissue while

conserving as much healthy tissue as possible. This precision is partic-

ularly crucial in surgeries aimed at preserving organs and in operations

near critical structures. This could aid surgeons in distinguishing be-

tween cancerous and non-cancerous tissues in real time, a feature that

would be especially benecial in minimally invasive surgeries, such as

laparoscopic or robotic procedures, where direct visualization is more

challenging. During surgery, they could guide targeted biopsies and

intraoperative frozen section analysis, helping to conrm that surgical

margins are free of cancer cells. Moreover, we emphasize the importance

of considering factors such as stage and grade of malignancy, as they add

complexity to the identication of relevant metabolomic signatures in

BC.

Ethical approval

The study protocol was approved by the local Bioethics Committee at

the University of Rzeszow (Poland) (permission no. 2018/04/10). The

patients provided written consent to participate in research. The pa-

tients provided written informed consent for the publication of any

associated data.

Funding

The research was supported by National Science Centre (Poland),

research project SONATA Number UMO-2018/31/D/ST4/00109.

CRediT authorship contribution statement

Ossoli´

nski Tadeusz: Resources. Krupa Zuzanna: Writing – original

draft. Ruman Tomasz: Writing – review & editing, Supervision, Re-

sources, Methodology, Investigation, Data curation. Nizioł Joanna:

Writing – review & editing, Writing – original draft, Visualization, Su-

pervision, Project administration, Methodology, Investigation, Funding

acquisition, Formal analysis, Data curation, Conceptualization. Osso-

li´

nski Krzysztof: Writing – original draft, Resources, Methodology,

Investigation. Płaza-Altamer Aneta: Investigation, Data curation.

Kołodziej Artur: Investigation, Data curation. Ossoli´

nska Anna:

Resources.

J. Nizioł et al.

Journal of Pharmaceutical and Biomedical Analysis 240 (2024) 115966

Declaration of generative AI and AI-assisted technologies in the

writing process

During the preparation of this work, the author(s) used ChatGPT3

(OpenAI) in order to improve language and readability. After using this

tool/service, the author(s) reviewed and edited the content as needed

and take(s) full responsibility for the content of the publication.

Declaration of Competing Interest

The authors declare that they have no known competing nancial

interests or personal relationships that could have appeared to inuence

the work reported in this paper.

Appendix A. Supporting information

Supplementary data associated with this article can be found in the

online version at doi:10.1016/j.jpba.2024.115966.

References

[1] H. Sung, J. Ferlay, R.L. Siegel, M. Laversanne, I. Soerjomataram, A. Jemal, F. Bray,

Global cancer statistics 2020: globocan estimates of incidence and mortality

worldwide for 36 cancers in 185 countries, CA Cancer J. Clin. 71 (2021) 209–249,

https://doi.org/10.3322/CAAC.21660.

[2] M.C.S. Wong, F.D.H. Fung, C. Leung, W.W.L. Cheung, W.B. Goggins, C.F. Ng, The

global epidemiology of bladder cancer: a joinpoint regression analysis of its

incidence and mortality trends and projection, 8, Sci. Rep. 2018 8 (1) (2018) 1–12,

https://doi.org/10.1038/s41598-018-19199-z.

[3] J. Mushtaq, R. Thurairaja, R. Nair, Bladder cancer, Surg. (Oxf. ) 37 (2019)

529–537, https://doi.org/10.1016/J.MPSUR.2019.07.003.

[4] D. Sahu, Y. Lotan, B. Wittmann, B. Neri, D.E. Hansel, Metabolomics analysis reveals

distinct proles of nonmuscle-invasive and muscle-invasive bladder cancer, Cancer

Med 6 (2017) 2106–2120, https://doi.org/10.1002/CAM4.1109.

[5] S. Tabaei, M.R. Haghshenas, T.J. Webster, A. Ghaderi, Proteomics strategies for

urothelial bladder cancer diagnosis, prognosis and treatment: Trends for tumor

biomarker sources, Anal. Biochem 666 (2023) 115074, https://doi.org/10.1016/J.

AB.2023.115074.

[6] X.W. Zhang, Q.H. Li, Z. Di Xu, J.J. Dou, Mass spectrometry-based metabolomics in

health and medical science: a systematic review, RSC Adv. 10 (2020) 3092–3104,

https://doi.org/10.1039/C9RA08985C.

[7] N.A. Di Meo, D. Loizzo, S.D. Pandolfo, R. Autorino, M. Ferro, C. Porta, A. Stella,

C. Bizzoca, L. Vincenti, F. Crocetto, O.S. Tataru, M. Rutigliano, M. Battaglia,

P. Ditonno, G. Lucarelli, Metabolomic approaches for detection and identication

of biomarkers and altered pathways in bladder cancer, Int J. Mol. Sci. 23 (2022)

4173, https://doi.org/10.3390/IJMS23084173/S1.

[8] H.A. García-Perdomo, A.M. D´

avila-Raigoza, F. Korkes, Metabolomics for the

diagnosis of bladder cancer: a systematic review, Asian J. Urol. (2023), https://doi.

org/10.1016/J.AJUR.2022.11.005.

[9] P.K. Cheung, M.H. Ma, H.F. Tse, K.F. Yeung, H.F. Tsang, M.K.M. Chu, C.M. Kan, W.

C.S. Cho, L.B.W. Ng, L.W.C. Chan, S.C.C. Wong, The applications of metabolomics

in the molecular diagnostics of cancer, Expert Rev. Mol. Diagn. 19 (2019) 785–793,

https://doi.org/10.1080/14737159.2019.1656530.

[10] N.R. Anwardeen, I. Diboun, Y. Mokrab, A.A. Althani, M.A. Elrayess, Statistical

methods and resources for biomarker discovery using metabolomics, 24, BMC

Bioinforma. 2023 24 (1) (2023) 1–18, https://doi.org/10.1186/S12859-023-

05383-0.

[11] N. Putluri, A. Shojaie, V.T. Vasu, S.K. Vareed, S. Nalluri, V. Putluri, G.S. Thangjam,

K. Panzitt, C.T. Tallman, C. Butler, T.R. Sana, S.M. Fischer, G. Sica, D.J. Brat,

H. Shi, G.S. Palapattu, Y. Lotan, A.Z. Weizer, M.K. Terris, S.F. Shariat,

G. Michailidis, A. Sreekumar, Metabolomic proling reveals potential markers and

bioprocesses altered in bladder cancer progression, Cancer Res 71 (2011)

7376–7386, https://doi.org/10.1158/0008-5472.CAN-11-1154.

[12] P. Tripathi, B.S. Somashekar, M. Ponnusamy, A. Gursky, S. Dailey, P. Kunju, C.

T. Lee, A.M. Chinnaiyan, T.M. Rajendiran, A. Ramamoorthy, HR-MAS NMR tissue

metabolomic signatures cross-validated by mass spectrometry distinguish bladder

cancer from benign disease, J. Proteome Res 12 (2013) 3519–3528, https://doi.

org/10.1021/PR4004135/SUPPL_FILE/PR4004135_SI_001.PDF.

[13] D.W.B. Piyarathna, T.M. Rajendiran, V. Putluri, V. Vantaku, T. Soni, F.C. von

Rundstedt, S.R. Donepudi, F. Jin, S. Maity, C.R. Ambati, J. Dong, D. G¨

odde, S. Roth,

S. St¨

orkel, S. Degener, G. Michailidis, S.P. Lerner, S. Pennathur, Y. Lotan, C. Coarfa,

A. Sreekumar, N. Putluri, Distinct lipidomic landscapes associated with clinical

stages of urothelial cancer of the bladder, Eur. Urol. Focus 4 (2018) 907–915,

https://doi.org/10.1016/J.EUF.2017.04.005.

[14] K. Ossoli´

nski, T. Ruman, T. Ossoli´

nski, A. Ossoli´

nska, A. Arendowski, A. Kołodziej,

A. Płaza-Altamer, J. Nizioł, Monoisotopic silver nanoparticles-based mass

spectrometry imaging of human bladder cancer tissue: Biomarker discovery, Adv.

Med Sci. 68 (2023) 38–45, https://doi.org/10.1016/J.ADVMS.2022.12.002.

[15] D.W.B. Piyarathna, T.M. Rajendiran, V. Putluri, V. Vantaku, T. Soni, F.C. von

Rundstedt, S.R. Donepudi, F. Jin, S. Maity, C.R. Ambati, J. Dong, D. G¨

odde, S. Roth,

S. St¨

orkel, S. Degener, G. Michailidis, S.P. Lerner, S. Pennathur, Y. Lotan, C. Coarfa,

A. Sreekumar, N. Putluri, Distinct lipidomic landscapes associated with clinical

stages of urothelial cancer of the bladder, Eur. Urol. Focus 4 (2018) 907–915,

https://doi.org/10.1016/J.EUF.2017.04.005.

[16] R.R. Bhanvadia, Y. Lotan, Progress in the development of tissue-based biomarkers

for urothelial cancer, Expert Rev. Anticancer Ther. 22 (2022) 605–619, https://doi.

org/10.1080/14737140.2022.2070154.

[17] H.L. Stewart, D.J.S. Birch, Fluorescence guided surgery, Methods Appl. Fluor. 9

(2021) 042002, https://doi.org/10.1088/2050-6120/AC1DBB.

[18] J. Nizioł, V. Copi´

e, B.P. Tripet, L.B. Nogueira, K.O.P.C. Nogueira, K. Ossoli´

nski,

A. Arendowski, T. Ruman, Metabolomic and elemental proling of human tissue in

kidney cancer, Metabolomics 17 (2021) 30, https://doi.org/10.1007/S11306-021-

01779-2.

[19] MassBank of North America (MoNA), (n.d.).

[20] Mass Spectrometry Data Center, NIST, (n.d.). 〈https://chemdata.nist.gov/〉

(Accessed 8 June 2022).

[21] Z. Pang, J. Chong, G. Zhou, D.A. De Lima Morais, L. Chang, M. Barrette,

C. Gauthier, P.´

E. Jacques, S. Li, J. Xia, MetaboAnalyst 5.0: narrowing the gap

between raw spectra and functional insights, Nucleic Acids Res 49 (2021)

W388–W396, https://doi.org/10.1093/NAR/GKAB382.

[22] J. Nizioł, K. Ossoli´

nski, A. Płaza-Altamer, A. Kołodziej, A. Ossoli´

nska, T. Ossoli´

nski,

A. Nieczaj, T. Ruman, Untargeted urinary metabolomics for bladder cancer

biomarker screening with ultrahigh-resolution mass spectrometry, 13, Sci. Rep.

2023 13 (1) (2023) 1–15, https://doi.org/10.1038/s41598-023-36874-y.

[23] K. Ossoli´

nski, T. Ruman, V. Copi´

e, B.P. Tripet, A. Kołodziej, A. Płaza-Altamer,

A. Ossoli´

nska, T. Ossoli´

nski, A. Nieczaj, J. Nizioł, Targeted and untargeted urinary

metabolic proling of bladder cancer, J. Pharm. Biomed. Anal. 233 (2023) 115473,

https://doi.org/10.1016/J.JPBA.2023.115473.

[24] S. Okuda, T. Yamada, M. Hamajima, M. Itoh, T. Katayama, P. Bork, S. Goto,

M. Kanehisa, KEGG Atlas mapping for global analysis of metabolic pathways,

Nucleic Acids Res 36 (2008) W423–W426, https://doi.org/10.1093/NAR/

GKN282.

[25] J. Nizioł, K. Ossoli´

nski, A. Płaza-Altamer, A. Kołodziej, A. Ossoli´

nska, T. Ossoli´

nski,

T. Ruman, Untargeted ultra-high-resolution mass spectrometry metabolomic

proling of blood serum in bladder cancer, 12, Sci. Rep. 2022 12 (1) (2022) 1–13,

https://doi.org/10.1038/s41598-022-19576-9.

[26] D.W. Foster, The role of the carnitine system in human metabolism, Ann. N. Y

Acad. Sci. 1033 (2004) 1–16, https://doi.org/10.1196/ANNALS.1320.001.

[27] M.V. Liberti, J.W. Locasale, The Warburg effect: how does it benet cancer cells?

Trends Biochem Sci. 41 (2016) 211–218, https://doi.org/10.1016/J.

TIBS.2015.12.001.

[28] X. Jin, S.J. Yun, P. Jeong, I.Y. Kim, W.J. Kim, S. Park, Diagnosis of bladder cancer

and prediction of survival by urinary metabolomics, Oncotarget 5 (2014) 1635,

https://doi.org/10.18632/ONCOTARGET.1744.

[29] B.M. Wittmann, S.M. Stirdivant, M.W. Mitchell, J.E. Wulff, J.E. McDunn, Z. Li,

A. Dennis-Barrie, B.P. Neri, M.V. Milburn, Y. Lotan, R.L. Wolfert, Bladder cancer

biomarker discovery using global metabolomic proling of urine, PLoS One 9

(2014) e115870, https://doi.org/10.1371/JOURNAL.PONE.0115870.

[30] J. Nizioł, V. Bonifay, K. Ossoli´

nski, T. Ossoli´

nski, A. Ossoli´

nska, J. Sunner, I. Beech,

A. Arendowski, T. Ruman, Metabolomic study of human tissue and urine in clear

cell renal carcinoma by LC-HRMS and PLS-DA, Anal. Bioanal. Chem. 410 (2018)

3859–3869, https://doi.org/10.1007/s00216-018-1059-x.

[31] P. Tripathi, P. Kamarajan, B.S. Somashekar, N. MacKinnon, A.M. Chinnaiyan, Y.

L. Kapila, T.M. Rajendiran, A. Ramamoorthy, Delineating metabolic signatures of

head and neck squamous cell carcinoma: phospholipase A2, a potential therapeutic

target, Int J. Biochem Cell Biol. 44 (2012) 1852–1861, https://doi.org/10.1016/J.

BIOCEL.2012.06.025.

[32] B.S. Cummings, Phospholipase A2 as targets for anti-cancer drugs, Biochem Pharm.

74 (2007) 949–959, https://doi.org/10.1016/J.BCP.2007.04.021.

[33] S.A. Mir, P. Rajagopalan, A.P. Jain, A.A. Khan, K.K. Datta, S.V. Mohan, S.S. Lateef,

N. Sahasrabuddhe, B.L. Somani, T.S. Keshava Prasad, A. Chatterjee, K.V. Veerendra

Kumar, M. VijayaKumar, R.V. Kumar, S. Gundimeda, A. Pandey, H. Gowda,

LC–MS-based serum metabolomic analysis reveals dysregulation of

phosphatidylcholines in esophageal squamous cell carcinoma, J. Proteom. 127

(2015) 96–102, https://doi.org/10.1016/J.JPROT.2015.05.013.

[34] M.Y. Lee, A. Yeon, M. Shahid, E. Cho, V. Sairam, R. Figlin, K.H. Kim, J. Kim,

Reprogrammed lipid metabolism in bladder cancer with cisplatin resistance,

Oncotarget 9 (2018) 13231, https://doi.org/10.18632/ONCOTARGET.24229.

[35] C. Denkert, J. Budczies, T. Kind, W. Weichert, P. Tablack, J. Sehouli, S. Niesporek,

D. K¨

onsgen, M. Dietel, O. Fiehn, Mass spectrometry–based metabolic proling

reveals different metabolite patterns in invasive ovarian carcinomas and ovarian

borderline tumors, Cancer Res 66 (2006) 10795–10804, https://doi.org/10.1158/

0008-5472.CAN-06-0755.

[36] K.K. Pasikanti, K. Esuvaranathan, Y. Hong, P.C. Ho, R. Mahendran, L. Raman Nee

Mani, E. Chiong, E.C.Y. Chan, Urinary metabotyping of bladder cancer using two-

dimensional gas chromatography time-of-ight mass spectrometry, J. Proteome

Res 12 (2013) 3865–3873, https://doi.org/10.1021/pr4000448.

J. Nizioł et al.