Integrative Raman Spectroscopy and Multi-Omics Analysis of Lipid and Matrix Protein Heterogeneity in Triple-Negative Breast Cancer.

Introduction
Breast cancer (BC) is one of the most
prevalent types of cancer
among women and is the top cause of cancer-related deaths worldwide.
According to the World Health Organization (WHO), in 2022, 2.3 million
incidences and 670,000 deaths were reported worldwide. Breast cancer is widely recognized as a heterogeneous disease,
characterized by a broad range of clinical, pathological, and molecular
properties. Based on immunohistochemical
features, it is classified into three main subtypes according to intra-
and extracellular domain receptors: estrogen receptor (ER), progesterone
receptor (PR), and human epidermal growth factor receptor 2 (HER2).
Triple-negative breast cancer (TNBC) is characterized by a lack of
ER/PR expression and the absence of HER2 amplification, accounting for ∼10–20% of breast
cancer cases. TNBC is characterized by
high metastasis rates, resistance to most adjuvant and neo-adjuvant
therapies, and a lack of targetable receptors for traditional therapies. One issue in accurately targeting TNBC is its
heterogeneity, both within and between tumors. TNBC heterogeneity
can arise from interactions of cancer cells with the tumor microenvironment
(TME), which includes diverse cellular populations, metabolites, lipids,
and the extracellular matrix (ECM). Recently, both lipids and ECM
have gained interest as novel regulators of tumor heterogeneity and
mediators of breast cancer progression to a more aggressive phenotype.
−

Several types of cancers are associated with alterations in
the
biochemical characteristics of cells. For example, compared to nontumor cells, lipid metabolism in cancer
cells is elevated to fulfill the biogenesis, bioenergetics, and metabolic
demands of tumors, thereby promoting cell proliferation and survival. The association between altered metabolism and
cancer is generally described by glycolysis, generally known as the
Warburg effect.
−

However, recent research has shown that other pathways,
especially those related to lipoprotein metabolism and fatty acid
production, may contribute to cell proliferation by providing the
energy and membrane components necessary for rapid growth.
,−

Lipid biosynthesis has been reported in cancerous tissues
to support
the rapid proliferation of cancer cells.
,
However, fatty acid (FA) and cholesterol biosynthesis primarily
occur in the liver, adipose tissue, and lactating breast tissues.
−

Studies have shown that cancer cells undergo a transition from lipid
uptake to de novo lipogenesis.
,
This could alter the
cell membrane lipid saturation,
,
leading to modifications
in the levels of saturated and monounsaturated phospholipids,
−

which may protect cancer cells from oxidative damage by decreasing
lipid peroxidation.
,
Aggressive breast cancer has
shown elevated levels of saturated FAs,
,
indicating
that decreased membrane fluidity may be a characteristic of advanced
stages of the disease. A rise in polyunsaturated
fatty acids (PUFAs) has been correlated with aggressiveness in cancer
cell lines and the enhancement of tumor development. In addition to lipid metabolism, the TME plays a significant
role in cancer progression.

The
microenvironment comprises different cell types, including
fibroblasts, immune cells, adipocytes, and endothelial cells, all
of which are supported by the ECM. Components
of the ECM provide cells with biomechanical and biochemical cues and
play a crucial biological role in breast cancer progression and metastasis.
,
The ECM comprises various proteins including laminins, fibronectin,
collagens, proteoglycans, glycosaminoglycans, matricellular proteins,
and ECM remodeling enzymes.
,
There are distinct
differences between normal and cancerous ECM compositions because
of cancer progression. Mammary gland
involution, a tissue remodeling procedure, is characterized by major
changes in the ECM, which involve a pronounced increase in fibrillar
collagens, fibronectin, matricellular proteins, and numerous ECM remodeling
enzymes. Changes in the ECM can be recognized
as a cancer hallmark.

Chemical
imaging using Raman spectroscopy provides a label-free,
spatially resolved analysis of lipid composition and distribution,
offering insights into dynamic metabolic changes that immunohistochemistry
(IHC) cannot capture due to its reliance on predefined markers. By
directly mapping lipid heterogeneity in the TME, Raman-based spatial
lipidomics can uncover the metabolic signatures of cancer progression,
resistance mechanisms, and functional diversity, advancing our understanding
of cancer biology beyond static protein markers. Confocal Raman microspectroscopy
is a nondestructive, quantitative, and label-free method that has
been applied in many biological studies for analyzing chemical composition.
,
Raman spectroscopy provides vital biological information and chemical
changes associated with a disease, as each molecule displays a distinct
pattern of vibrations, which can be used as a Raman biomarker.
,
This nondestructive method provides chemical information and the
spatial distribution of components, such as proteins and lipids, in
tissues. Moreover, in comparison to other common methods (e.g., liquid
chromatography–mass spectrometry (LC–MS) and proteomics
analysis), the samples do not require preparation and are not destroyed
during analysis.
This study aims to leverage Raman microspectroscopy
to investigate
FA and ECM alterations in the TME of triple-negative human breast
cancer compared with matched breast adipose tissue. Raman spectra
and Raman images were collected from both cancerous and matched breast
tissues, providing a basis for detailed biochemical comparisons. Ratio
analysis of the Raman spectra revealed higher levels of unsaturated
fatty acids in cancerous tissues compared to their matched nontumor
counterparts. The spatial distribution of FAs and collagens within
tissues was visualized using the DCLS (direct classical least-squares)
method applied to Raman images.
To further elucidate the biochemical
landscape, DCLS analysis was
performed on Raman spectra to investigate changes in ω-3 and
ω-6 FAs, which were subsequently correlated with the LC–MS
results for cross-validation. Protein changes between the two tissue
groups were explored by combining the DCLS analysis of Raman spectra
with proteomics data for comprehensive validation. This integrative
approach highlights the potential of Raman microspectroscopy not only
for studying FA and protein alterations in cancerous tissues but also
for identifying novel biomarkers and gaining insights into TME dynamics.
Additionally, the methodology could be extended to study other tumor
types and explore spatial ECM and metabolic changes under diverse
pathological conditions.

Experimental Methods

Materials
α-Linolenic acid (CAS #822-18-4), docosahexaenoic
acid (CAS #6217-54-5), docosapentaenoic acid (CAS #24880-45-3), eicosapentaenoic
acid (CAS # 10417-94-4), stearidonic acid (CAS # 20290-75-9), arachidonic
acid (CAS #506-32-1), linoleic acid (CAS #60-33-3), oleic acid (CAS
#112-80-1), and cholesterol (CAS #57-88-5), cholesteryl palmitate
(CAS #601-34-3), and 1,2-dipalmitoyl-sn-glycero-3-phosphate
sodium salt (CAS #169051-60-9) were obtained from Cayman Chemical,
and glyceryl tripalmitate (CAS #555-44-2) was purchased from Sigma-Aldrich.
Collagen I (CAT 354249), fibronectin (CAT 354008), and laminin (CAT
354232) were purchased from Corning Life Sciences. Collagen III (CAT
#5019-10MG) and collagen IV (CAT #5022-5MG) were purchased from Advanced
BioMatrix. Collagen V (CAT #1270-01S) was purchased from Southern
Biotech.

Preparation of Triple-Negative Breast Cancer (TNBC) Tissue
Frozen tissue sections from three patients were purchased from
OriGene Technologies and stored at −80 °C until further
use. For each patient, tumor and matched breast adipose tissue sections
(5 × 5 μm2) collected from the same patient
were mounted onto slides with an optimal cutting temperature (OCT)
embedding medium. After thawing, the OCT was removed by washing the
tissues twice with 70% cold ethanol, each for 30 min. Following this,
the tissues were shaken in 1× PBS for 10–15 min to eliminate
any residual ethanol. Tumor donor demographics are presented in Table
.

Histology
Hematoxylin and eosin (H&E) dyes were
used to stain tissue sections for histological examination. Staining
was performed on a Leica ST5020/CV5030 autostainer (Leica Biosystems)
using Eosin Phloxine 515, SelecTech Hematoxylin 560, bluing, and define
solutions (Leica Biosystems). The stained tissues were mounted using
Surgipath Sub-X mounting medium and scanned with a Hamamatsu NanoZoomer
HT whole slide scanner, which features an Olympus 20x/0.8 NA dry objective.

Proteomics
A comprehensive explanation of the proteomics
method employed in this study can be found in the referenced paper . The details of the procedure
can be found in the Supporting Information.

Lipidomics
A comprehensive explanation of the lipidomics
method employed in this study can be found in the referenced paper . The details of the procedure
can be found in the Supporting Information.

Raman Spectroscopy and Microscopy Experiments
Raman
spectra from TNBC tissues mounted on glass slides were collected by
a Renishaw inVia Reflex spectrometer using a 532 nm laser, 50XL air
objective, 1800 grating at 50% power (15 mW), 10 s exposure, and 100–3200
cm–1 spectral range. In each group of collected
tissues (n = 3), for both normal and cancerous tissues,
a total of 180 Raman spectra were collected. Raman images of OCT tissues
at a spatial resolution of 100 μm × 100 μm were acquired
using StreamHR mode, a 532 nm laser, 50XL air objective, 1800 grating
at 100% power (30 mW), and 1 s exposure time at 1000 cm–1 center. Raman spectra of lipid and protein standards in the form
of liquid or powder were acquired using a 785 nm laser, 50XL air objective,
1200 grating, 10 s exposure time, and 10% power (18 mW) for lipids
and 50% power (90 mW) for proteins. Thirty Raman spectra were acquired
from each tissue type.

Statistical Analysis and Data Processing
To perform
statistical analysis, first, the collected Raman spectra were baseline-subtracted
using Renishaw’s WiRE 4.4 software’s “Intelligent
Fitting”. OriginLab software (OriginLab, Northampton, MA) was
used for plotting the spectra, data processing, and principal component
analysis. DCLS was utilized for generating the mapping data. For this
purpose, unprocessed Raman data were modified to eliminate the sudden
and intense spikes caused by the cosmic rays. After cosmic ray removal,
data were smoothed using a Savitzky–Golay process with a window
size of 7 and a polynomial order of 2. Finally, corrected data were
baseline-subtracted using Renishaw’s WiRE 4.4 software’s
“Intelligent Fitting”. A comprehensive explanation of
the DCLS analysis employed in this study can be found in the referenced
papers
,
.

Results
To distinguish regions of cancer cells from
the surrounding stroma
on the TNBC tumor slides, we used H&E-stained images (Figure
). H&E staining
of the patient tissue slides indicated varied levels of neoplasia
and changes compared to the matched breast tissues.
For example, Figure
A shows an increased presence of tumor cells toward
the lower portion
of the tissue (purple ellipse) and arranged in nests and anastomosing
cords. Additionally, there was a large blood vessel with a mass of
tumor cells localized in the lumen (yellow arrow, bottom right of
the tissue). Compared to the matched breast tissue from the same patients
(Figure
D), there
are only a few ducts with neoplastic epithelium (yellow arrows). The
other two tumor slides had similar features with regions of neoplastic
duct epithelium toward the bottom of the tissue (yellow arrows, Figure
B) and increased
tumor cell density on the right side of the tissue (purple ellipse, Figure
C). Corresponding
matched breast tissues exhibited much lower/smaller neoplastic duct
epithelium regions (purple ellipses). In addition, calculation of
the nucleus/cell area ratio revealed higher levels for the TNBC tissues
compared to the matched breast tissues (Figure
G–I).
To understand the chemical composition within breast adipose tissue
and cancerous tissues, we collected Raman spectra from both TNBC and
matched breast tissues; Figure
A shows average Raman spectra for all groups (n = 3, 30 spectra/patient). The first column of Figure S1A–C shows average Raman spectra for each group.
In both groups of normal and cancerous tissues, bands at around 1003
cm–1 (Phenylalanine; CH3 rocking coupled
with C–C stretching of carotenoids), 1097 cm–1 (O–P–O (stretching PO2 symmetric (phosphate
II) of phosphodiesters), 1156 cm–1 (C–C and
C–N stretching of proteins and carotenoids), 1245 cm–1 (amide III), 1451 cm–1 CH2 deformation
(lipids and proteins), 1520 cm–1 (β-carotene;
CC stretching mode), 1665 cm–1 (lipids (CC
stretching)), 2852 cm–1 (symmetric CH2 stretching of lipids), 2889 cm–1 (asymmetric CH2 stretching of lipids), 2938 cm–1 (C–H
stretching, aromatic and aliphatic amino acids, charged amino acids,
proline, threonine, histidine, lipids, proteins), and 3063 cm–1 (nucleic acid to protein ratio; C–H aromatic
ring vibration) were identified.
,
The exact
peak positions of the average Raman spectra plotted in Figure
A are summarized in Table S2. In this study, to achieve better classification
with higher accuracy, we utilized a machine learning approach, partial
least-squares discriminant analysis (PLS-DA), to analyze and visualize
Raman spectra. PLS-DA is considered a supervised form of principal
component analysis (PCA) while reducing the number of latent variables
to a lower-dimensional space that minimizes error. Figure
B presents the PLS-DA results,
which revealed distinct clustering between TNBC and normal tissues. Figure
C presents the PCA
biplot, created by using the first and second principal components.
In this plot, pink dots representing TNBC samples are predominantly
clustered on the right side, while blue dots representing normal tissues
are more concentrated on the left side. Together, the first two principal
components account for 28.7% of the total variability. Additionally,
arrows pointing to the right correspond to Raman peaks at 748, 755,
1206, 1517, and 1519 cm–1, highlighting their stronger
contribution to distinguishing TNBC data. The results show that there
is a large heterogeneity among tumors collected from patients.
The alterations in cancerous and matched breast
tissues were further
evaluated using PCA analysis of the Raman spectra collected from each
tissue type. This involved calculating the percentage of variables
in the Raman bands, reflecting changes in the content of components
such as lipids and proteins because of cancer. For this purpose, Raman
spectra of all replicates in the range of 100–3200 cm–1 for both groups of cancerous and matched breast tissues from all
three groups were categorized as the first and second principal components
(PC1 and PC2). The second column of Figure S1A–C displays the PCA analysis of Raman spectra for each group. In all
groups, the first PCs captured most of the variances within the data
set, accounting for 92.8% in group 1, 83.6% in group 2, and 84.3%
in group 3. The second PC, which distinguishes between matched breast
tissue and cancerous groups, accounted for 3.5% in group 1, 7.2% in
group 2, and 7.0% in group 3. Together, the first two PCs explained
a total variance of 96.3% for group 1, 90.8% for group 2, and 91.3%
for group 3. Peaks at around 1445 and 1660 cm–1 were
observed in both types of cancerous tissues and matched breast tissues
of all three groups, which are associated with lipids. The intensity
of the area under these peaks was measured and utilized to determine
the level of lipid unsaturation by calculating the ratio of unsaturated
lipid (CC at 1660 cm–1) to saturated lipid
(CH2 at 1445 cm–1) bands. The calculated
ratio can be seen in Figure
D, which indicates an increase in lipid unsaturation (P < 0.0001, Mann–Whitney two-tailed test) following
breast cancer based on the mean ratio for all three groups. While
the Raman spectra of cancerous and matched breast tissues appear similar,
notable differences exist due to biochemical changes related to proteins,
lipids, and other components. To explore these variations between
cancerous and matched breast tissues, we employed analyses such as
volcano plots, heat maps, and variable importance in projection (VIP)
scores. The data was generated by analyzing the intensity of the Raman
spectra, representing the expression levels of the biomolecular constituents
within the tissues, as shown in Figure
. The volcano plot in Figure
A highlights the Raman peak positions that
are upregulated (shown in red) and downregulated (shown in blue) in
TNBC tissues compared with matched breast tissues. Specifically, the
peaks at 748, 1517, and 1523 cm–1 are upregulated,
while the peaks at 2780, 2851, and 2881 cm–1 are
downregulated in TNBC tissues. Figure
B illustrates the hierarchical heatmap clustering analysis,
showing the upregulated peaks in red and the downregulated peaks in
blue for each peak position. The results indicate that the peaks at
278, 748, 958 cm–1 (carotenoid/cholesterol), 994,
1002 cm–1 (phenylalanine/proteins), 1206 cm–1 (tyrosine), 1155 cm–1 (proteins/cholesterol),
1175 cm–1 (tyrosine), 1517 cm–1 (carotenoid), 1519 cm–1 (β-carotene), 1523
cm–1 (β-carotene), 1555 cm–1 (proteins), 1558 cm–1 (proteins), 1583, 1586,
1605 cm–1 (tyrosine/phenylalanine), 1608 cm–1 (tyrosine/phenylalanine), and 2332 cm–1 are upregulated, whereas the peaks at 2849 cm–1 (lipids), 2851 cm–1 (lipids), 2877 cm–1 (lipids), 2878 cm–1 (lipids), 2881 cm–1 (lipids), 2886, and 2812 cm–1 are downregulated
in TNBC tissues. The VIP score analysis (Figure
C) highlights the distinct Raman peak positions
between the TNBC and normal tissue samples for each group. Notably,
the upregulated peaks at 278, 295, 748, 755, 923, 994, 1002, 1203,
1206, 1513, 1517, 1519, 1523, 1555, 1558, 1586, and 2332 cm–1, along with the downregulated peaks at 388, 875 cm–1, and 2780 (lipids) cm–1, effectively differentiate
between the two tissue types. Peaks associated with proteins at 1002,
1155, 1555, and 1558 cm–1 are prominent in cancerous
tissue, whereas Raman bands linked to lipids, such as 2849, 2851,
2877, 2878, and 2881 cm–1, dominate in matched breast
tissues. Figure
D
presents a violin plot comparing the mean intensities of the 2780
cm–1 peak, associated with lipids, and the 1517
cm–1 peak, linked to carotenoids. The results indicate
a lower mean intensity for the lipid-associated peak at 2780 cm–1 and a higher mean intensity for the carotenoid-associated
peak at 1517 cm–1 in TNBC tissues compared to that
in normal tissues. The elevated β-carotene levels in TNBC compared
to matched breast tissue may stem from tumor-driven metabolic reprogramming,
including increased lipid uptake, storage, and altered retinoid metabolism.
β-carotene serves as a precursor for retinoic acid, which regulates
cell differentiation and proliferation. TNBC may have an altered retinoid metabolism,
,
leading to an accumulation of β-carotene due to inefficient
conversion to retinoic acid. TNBC cells, under high oxidative stress,
may accumulate β-carotene as an adaptive antioxidant response.
−

Additionally, enhanced lipid droplet formation and dysregulated
lipid transport proteins could contribute to their sequestration within
the tumor. These findings highlight a potential metabolic vulnerability
in TNBC that can be leveraged for therapeutic targeting.
After applying Raman spectroscopy, which is an
untargeted method,
we employed Raman microscopy to target specific components and profile
lipid and protein distribution within cancerous and matched breast
tissues. To generate spatial mapping of the components and show their
distribution in the tissues, we analyzed Raman images of the samples
using the DCLS method and generated images of ω-3, ω-6,
and ω-9 FAs, like α-LA, SDA, EPA, DPA, DHA, LA, AA, OA,
and TG, and matrix proteins, like COL I, COL III, COL IV, COL V, FN,
and LAM. An example of the generated Raman mapping image for both
cancerous and matched breast tissues is shown in Figure
A,B for COL I, LAM, and COL
III for patients nos. 1, 2, and 3, respectively. Figure
C quantitatively compares the
protein score extracted from their corresponding Raman mapping images,
in which there is a significant (p < 0.01) decrease
in the level of COL I and COL III and a significant increase (p < 0.01) in the level of LAM within cancerous tissues
in comparison to matched normal tissues. We analyzed comprehensive
Raman mapping images for all fatty acid components to identify their
distribution, and the generated images can be seen in Figures S2A,B and S3A,B for patient 1, Figures S5A,B and S6A,B for patient 2, and Figures S8A and S9A,B for patient 3. In these
mapping images, red indicates a higher concentration of the components,
while blue signifies a lower concentration of the same fatty acids.
Moreover, Raman images of the samples were analyzed to generate spatial
mapping images of different proteins such as COL I, COL III, COL IV,
COL V, FN, and LAM, which can be seen in Figure S4A,B for patient 1, Figure S7A,B for patient 2, and Figure S10A,B for
patient 3.
In addition to Raman mapping, which shows the distribution
of components
within tissues, we also performed DCLS analysis on the Raman spectra
to quantitatively measure the lipid and protein changes in tissues. Figure
A–H presents
the mean of the lipid scores extracted by DCLS analysis of Raman spectra
for all three patients for the respective fatty acids. To achieve
this, Raman spectra collected from lipid standards of various FAs
provided by Corning Life Sciences were used for DCLS analysis. Each
patient’s sample spectrum was compared against specific lipid
standardsα-LA, EPA, DHA, AA, and LAto generate
a lipid score for each component. For further validation, we also
performed LC–MS analysis, the result of which can be seen in Figure
I–P.
Based on the data, the level of all FAs (except
α-LA) was
elevated in the cancerous tissues compared with that in the matched
breast tissues. For the α-LA FA, while LC–MS analysis
showed a decreasing trend in cancer tissue in comparison to that in
breast tissue, the Raman analysis indicated no changes between the
two groups. After comparing each component, we calculated the levels
of ω-3 (α-LA, EPA, and DHA) and ω-6 (AA and LA)
FAs. According to these two methods, the cancerous tissues indicated
higher levels for both ω-3 and ω-6 FAs. Moreover, due
to the importance of LA and α-LA, which are the main components
of ω-3 and ω-6 FA families, respectively, we calculated
the ratio of LA/α-LA using both methods, revealing that a higher
level of LA/α-LA was observed in cancer tissues compared to
the matched breast tissues.
For further quantitative analysis
of Raman data with proteins,
we analyzed collected Raman spectra using the DCLS method to generate
a specific protein score for each component within the Raman spectra.
The protein scores generated through Raman spectroscopy and the DCLS
method were plotted and are shown in Figure
A–F. In addition
to Raman microspectroscopy, several methods can be applied to examine
the expression of proteins, including proteomics and immunohistochemistry
(IHC), which is semiquantitative. In our study, to compare our Raman
results for protein expression, we employed a proteomics technique,
which is a destructive method and damages the structure of the tissue,
and the results of the analysis are shown in Figure
G–L. Our results
showed that there is a positive correlation between the two methods.
According to these results, TNBC tissues indicated lower expression
levels of COL I, COL III, COL IV, and LAM and higher expression levels
of FN and COL V compared with the breast tissue based on both Raman
and proteomics techniques.
TNBC samples were compared with matched breast
adipose tissue for
differences in protein profiles. Results demonstrated that the matched
breast adipose tissues were distinguished from TNBC by a multivariate
analysis of proteomics data. The PC1 axis (explaining 38.3% variance)
and PC2 axis (explaining 19.1% variance) revealed clear separation
between the TNBC tumor and control groups, indicating significant
proteomic differences between these groups, as shown in Figure
A. Comparing TNBC tumor samples
to matched breast tissue, the volcano plot analysis found that 463
total proteins significantly altered between the two groups, as shown
in Figure
B. Pathway
analysis in Figure
C shows enrichment of protein changes associated with focal adhesion,
ECM–receptor interactions, and FA breakdown, which is in accordance
with the observed changes in our Raman and LC–MS analyses.
Analysis of Gene Ontology (GO) biological processes demonstrated significant
changes in the ECM and structure organization, cellular detoxification,
and chromatin assembly, as shown in Figure
D. Significantly enhanced categories in terms
of cellular components included blood microparticles, ECM including
collagen, focal adhesion sites, and nucleosome structures, as shown
in Figure
E. ECM structural
elements, antioxidant activity, protein heterodimerization activity,
and lipase inhibitor activity were significantly changed by molecular
function enrichment (Figure
F). These results, taken together, show clear proteome reprogramming
associated with the ECM and the extracellular TME in TNBC.
We built a protein–protein interaction network
(Figure
A) based on
differentially
expressed proteins to identify the important functional modules linked
with TNBC. Three main connected clusters were identified by the network
analysis: enzymes including alcohol dehydrogenases (ADH1A, ADH1B,
and ADH5), acyl-CoA thioesterases (ACOT1, ACOT2), and hydroxyacyl-CoA
dehydrogenase (HADHA, HADHB) are connected to lipid metabolism-related
proteins (green nodes). Collagens (COL6A3, COL6A2), integrins (ITGAV,
ITGA7), laminins (LAMA3, LAMB2), matrix adhesion proteins (SDC1, THBS2),
and ECM-related proteins (pink nodes) constituted the largest cluster.
The clustering of tumor-specific proteins (blue nodes), including
poly­(ADP-ribose) polymerase 1 (PARP1) and vimentin (VIM), suggested
roles as either targets or biomarkers unique to TNBC. The heatmap
of the proteins in the interaction network is shown in Figure
B. The heatmap of the top 75
differentially expressed proteins found in the study is shown in Figure
C.
To better understand how the lipid profile is remodeled
in TNBC,
we summarized the data in a circos plot (Figure
A). Lipid species were labeled by subclass
and annotated with their total acyl chain length and degree of unsaturation.
Within each subclass, species were organized from fully saturated
(number of double bonds, n = 0) to highly polyunsaturated
(n ≥ 3) and, within each group, ranked by
increasing chain length. The TNBC lipidome showed a consistent shift
toward polyunsaturated fatty acid (PUFA) species. Phosphatidylethanolamine
(PE) and phosphatidylinositol (PI) contained highly enriched polyunsaturated
lipids, such as PE (42:8), PI (42:9), and PI (44:12). Phosphatidylglycerol
(PG) species like PG (38:7) also showed the same trend. In contrast,
shorter and less unsaturated phosphatidylcholine (PC) and lysophosphatidylcholine
(LPC) species, including PC (30:0), PC (32:0), and LPC (20:4), were
reduced in TNBC compared to normal samples. The inner track of the
circos plot, which reflects chain elongation, revealed a clear increase
in the level of elongation among polyunsaturated PE and PI species.
Together, these data indicate that TNBC tumors are characterized by
a loss of short, saturated PCs and LPCs, alongside an enrichment of
long-chain, highly unsaturated PEs, PIs, and PGs. This remodeling
suggests metabolic reprogramming that promotes both elongation and
desaturation, contributing to a lipid environment.
The BioPAN lipid pathway analysis (Figure
C) indicated unique metabolic
alterations
in the TNBC samples. Reactions involving LPE → PE (−0.117)
showed decreased PE biosynthesis, while suppressed reactions (PC →
PS, −1.058; PI → LPI, −0.859; and LPC →
PC, −0.584) showed decreased flux toward PS and PC synthesis
pathways. These changes indicate considerable metabolic reprogramming,
as evidenced by suppressed LPE to PE conversion and altered glycerophospholipid
balance, which may support membrane remodeling and tumor growth.
The targeted lipidomic study showed that TNBC samples were significantly
different from breast adipose control samples, as shown in Figure
. The volcano plot
in Figure
A showed
that some lipid species were significantly differentially expressed
(P < 0.05). These included PI (40:7), PG (36:7),
and PG (36:4) that were upregulated, and LPC (20:5) and LPC (20:4)
that were downregulated. The OPLS-DA (Figure
B) clearly separated the lipid profiles
of TNBC and the control groups, showing strong metabolic differences.
The main and orthogonal components accounted for 18.4 and 33.3% of
the variance, respectively. Heatmap analysis in Figure
C showed that some polyunsaturated
PE and PI lipid species were found to be more abundant in TNBC. Individual
lipid species abundances, as shown in Figure
D with boxplots, confirmed these lipidomic
changes, showing that polyunsaturated PI and PG lipids increased significantly,
while LPC species decreased in TNBC samples.

Discussion
Abnormalities within breast tissue are inherently
heterogeneous
in both morphology and molecular composition.
−

In this study, we applied histological
images to demonstrate structural variations within cancerous and normal
breast tissues. Additionally, variations in lipid and protein levels
within the tissues were evaluated by using Raman microspectroscopy
combined with lipidomic and proteomic analyses. In all techniques,
tissues from both normal and cancerous regions of the breast were
extracted and examined from three patients. Raman spectroscopy analysis
revealed several significant differences, including variations in
the intensity of bands associated with amino acids, nucleic acids,
proteins, carbohydrates, and lipids. For example, the ratio analysis
of lipid unsaturation over saturated lipid bands indicated an increase
in lipid unsaturation in breast cancer tissues compared to normal
tissues (Figure
C).
Moreover, Raman spectra collected from whole tissue slides of normal
and cancerous samples were analyzed by using the DCLS method to identify
lipid and protein scores of each specific component within tissues.
The results of lipid scores extracted from DCLS indicated no change
in the level of α-LA and an increase in the levels of EPA, DHA,
AA, and LA within cancerous tissues compared to normal tissues. To
validate these results, we applied the lipidomics technique, which
also demonstrated a positive correlation with our DCLS results of
Raman spectra, except for α-LA, which showed a decreasing trend
in cancer tissues compared to the matched breast adipose tissues.
While both Raman and LC–MS/MS analyses revealed elevated levels
of ω-3 and ω-6 fatty acids in cancer tissues compared
to the normal ones based on the LC–MS/MS results, the difference
of ω-6 is much higher (about 8767 (pmol/mg)) than the difference
of ω-3 (about 788 (pmol/mg)) in cancer tissues than the matched
normal tissues.
Lipid remodeling is an established metabolic
cancer hallmark. TNBC cells take up exogenous
FAs from the TME
and can also switch to de novo FA synthesis as an alternative source
of energy required for increased cell proliferation.
,,
Metabolic profiling of TNBC shows
that FA biosynthesis upregulates cellular pathways like the phosphoinositide
3-kinase (PI3K)/Akt signaling pathway, a central cancer signaling
pathway actively involved in cell proliferation and metabolism.
−

Consistent with our findings, studies have shown an altered
composition
of PUFA in normal versus tumor tissues.
,
PUFAs also
play a role in the lipid composition of the breast tissue TME, favoring
tumor progression. You et al. showed the elevation of PUFA composition
in TME of both proximal and distal tissues obtained from breast cancer
patients with invasive ductal carcinoma and rat mammary tumor models
compared to their matched normal. Elevated levels of lipids have also
been reported in breast cancer cells.
,
Literature
data on the lipid profiling of MDA-MB-231 (a TNBC cell line) also
showed an increased expression of PUFA compared to MCF-7, a hormone
receptor-positive breast cancer cell line.

However, the effects of ω-6 and ω-3 FAs are reported
to be pro-oncogenic and antioncogenic, respectively, through the activation
of different sets of metabolic genes. Therefore, these PUFAs are considered to have a pleiotropic impact
on cancer pathology. For example, linoleic
acid, a major ω-6 PUFA, promotes invasion and migration in MDA-MD-231
cells by activating the PI3K/AKT pathway through the induction of
its upstream proteins, EGFR and PI3K. EPA and DHA, which are ω-3 PUFAs, have been shown to reduce
cell proliferation in breast cancer cell lines and in vivo models,
while ω-6 have been shown to promote cell growth.
,
ω-3 fatty acids also synergistically enhance the effect of
chemotherapy.
,
PUFAs, therefore, provide a new
rationale as a target for therapeutics or to be used as a tool to
enhance existing chemotherapeutic regimens.
,

As a crucial element of tissues in multicellular organisms,
the
ECM consists of various proteins including collagens, laminins, fibronectin,
glycoproteins, and proteoglycans, functioning as a structural framework,
offering the support required to preserve tissue integrity. The results of protein score data extracted
from DCLS data for proteins showed lower expression of COL I, COL
III, COL IV, and LAM and higher expression of COL V and FN, compared
to the normal tissue, which aligns with the results of proteomics.
The basement membrane acts as a barrier and protects the tissue from
the initial invasion of tumor cells. We
and others have previously highlighted the loss of COL I and COL III
mRNA in TNBC compared to matched breast adipose. Lochter et al. reported decreased expression of COL IV
and LAM and increased expression of COL V and FN in carcinoma cells
and breast tumors. Moreover, in some
invasive breast carcinomas, a reduction in the level of COL IV was
detected, indicating basement membrane degradation during breast cancer.
,

Our comprehensive proteomic and lipidomic study demonstrates
significant
changes in TNBC samples relative to those of matched breast adipose
samples. Proteomic data revealed substantial changes in proteins associated
with lipid metabolism and ECM remodeling (Figures
A–F and A–C).
Additionally, the volcano plot in Figure
B revealed several differentially expressed
proteins, including upregulated ATIC and MYO1D and downregulated ADH1A
and CAPNS2. These proteins are linked to key biological processes:
ATIC with metabolic reprogramming, MYO1D with cell migration, ADH1A
with angiogenesis and tumor suppression, and CAPNS2 with the formation
of invadopodiaspecialized structures that help cells degrade
the ECM during migration and invasion. Pathway and network analyses
validate that ECM remodeling and lipid metabolism are essential interrelated
processes supporting tumor growth. These data align with research
highlighting ECM modifications important to cancer metastasis and lipid metabolism as a defining characteristic
of cancer cell proliferation and survival.
,

The results were further confirmed by the lipidomic studies,
which
showed that there were significant changes toward PUFAs and an extension
of fatty acyl chains in the PI, PE, and PG lipid subclasses (Figure
A). According to
Baenke et al. and Röhrig and Schulze, BioPAN analysis found that there was a decrease
in PE biosynthesis (LPE → PE) and also a decrease in PC to
PS conversion pathways (Figure
C). This suggests that there is a change in glycerophospholipid
homeostasis, which is likely to play a role in the membrane remodeling
necessary for tumor growth. Figure
A–D shows that the changes in lipidomics were
confirmed at the species level, with an accumulation of polyunsaturated
PI, LPC, and PC lipids. Altogether, our findings provide support to
the idea that TNBC’s aggressive phenotype and metabolic adaptability
are made possible by coordinated proteome and lipidomic reprogramming,
and they point to possible treatment targets in pathways related to
lipid metabolism and ECM remodeling.

Conclusions
We utilized Raman spectroscopy as a nondestructive
and spatially
resolved analytical platform to map unsaturated fatty acids and extracellular
matrix (ECM) proteins in tumor microenvironments. We validated these
findings through quantitative liquid chromatography–mass spectrometry
(LC–MS) and proteomic analysis. We demonstrate that TNBC tumors
exhibit elevated levels of polyunsaturated fatty acids (EPA, DHA,
AA, LA, ω-3/ω-6) and remodeling of ECM components. These
findings are supported by direct classical least-squares (DCLS) modeling
of Raman spectra and confirmed with biochemical quantification. Furthermore,
pathway and network analyses built using the proteomics results revealed
coordinated reprogramming of lipid metabolism and ECM remodeling,
key hallmarks of TNBC aggressiveness. This comprehensive molecular
mapping underscores the utility of Raman-based spatial lipidomics
and proteomics as powerful, minimally invasive tools for studying
cancer heterogeneity and tissue microenvironments.

Supplementary Material