Abstract

   Background: Mass spectrometry imaging (MSI) lipidomics is a subset of
   spatially resolved techniques wherein lipids are detected by mass
   spectrometry, allowing their multiplexed detection and acquiring
   position-correlated spectra along a tissue section. Rapid advances in
   the field provide solid evidence demonstrating how specific and
   regulated lipid distribution is in any biological context. Objectives:
   Herein, we describe the MSI, particularly matrix-assisted laser
   desorption/ionization (MALDI-MSI), challenges and advantages in
   defining human lung pathophysiology, particularly in lung cancer and
   chronic obstructive pulmonary disease, leading causes of death.
   Methods: MALDI-MSI analysis of lung tissue sections at 25 μm of lateral
   resolution allowed associating specific lipid profiles with the main
   tissues present and independently assessing the impact on lipid
   composition of smoking, chronic inflammation, and lung cancer. Results:
   Consistent with MALDI-MSI studies in tumor epithelia, arachidonic
   acid-containing phospholipids increased, agreeing with its role as a
   precursor of numerous bioactive molecules participating in cell
   differentiation and malignization. Next, a gene expression dataset of
   epithelial human non-small cell lung cancer samples was analyzed using
   system biology approaches, revealing that, consistent with the most
   relevant changes in lipid profiles, the network dominated by the
   tumor-associated module included genes tightly involved in
   phosphatidylinositol and sphingolipid metabolism. Hence, despite the
   intrinsic difficulties entailed by lung tissue handling, the results
   strongly encourage future analysis at higher lateral resolutions so
   that the lipidome changes associated with each lung cellular type, even
   subtype, could be fully mapped. Therefore, MALDI-MSI lipidomics
   definitively broadens the options, some still rather unexplored, to
   delve into pathophysiology at the cell-type level.

1. Introduction

   Single-cell technologies are finally making it possible to capture the
   cellular complexity of any tissue. However, transcriptomic and
   metabolomic programs are highly dependent on both cell-to-cell and
   cell-to-environment interactions. Thus, it seems that spatially
   resolved molecular profiles will be critical for identifying molecular
   drivers of cellular interactions in both physiological and pathological
   conditions [[40]1]. In this context, mass spectrometry imaging (MSI)
   techniques are a subset of spatially resolved techniques wherein the
   analytes are detected based on mass spectrometry, which is the best in
   terms of high sensitivity and specificity, high throughput, and high
   accuracy [[41]2]. MSI enables the multiplexed detection and
   localization of biological molecules by acquiring position-correlated
   spectra along a tissue section [[42]3,[43]4]. Once spectra are
   acquired, specialized software transforms the variable “peak intensity”
   of a particular compound into a pixel colored according to a color
   scale, one per coordinate, generating a final image of its distribution
   ([44]Figure 1). MSI methodology can be applied to analyze proteins,
   metabolites, or lipids ([45]Supplementary Figure S1), providing
   high-quality molecular data and connecting both pathology and mass
   spectrometric analysis to tissue-based research. There are several MSI
   techniques, but MALDI (matrix-assisted laser desorption/ionization),
   which uses soft ionization to extract the analytes from the tissue, is
   currently the most widely used to analyze lipid profiles [[46]5,[47]6].
   One of the key parameters in MSI analysis is the lateral (or spatial)
   resolution achieved, which is determined by laser spot size and the
   distance the laser moves between one analysis zone and the next. Most
   current MALDI-MSI equipment allows reaching 50 to 10 µm of lateral
   resolution; consequently, the molecular information gathered from the
   tissue is close to single-cell.

Figure 1.

   [48]Figure 1
   [49]Open in a new tab

   Typical MALDI-MSI workflow in a biomedical context. A standard MALDI
   mass spectrometry imaging (MALDI-MSI) protocol in biomedical
   applications involves three key phases: Sample Preparation: Thin tissue
   sections from human biopsies are cryopreserved and coated with a
   crystalline organic matrix; Spectral Acquisition: The prepared samples
   are systematically scanned using a high-resolution mass spectrometer to
   generate spatial molecular profiles; and Multimodal Integration:
   Computational alignment of MSI segmentation maps with histochemical
   stains (e.g., H&E) and targeted protein localization data from
   immunohistochemistry/immunofluorescence enables precise demarcation of
   both morphological landmarks and disease-specific zones (adapted from
   Maimó-Barceló et al. [[50]7]).

   MALDI-MSI has provided a new level of understanding by clearly showing
   the specific spatial distribution of lipid species in tissues
   ([51]Figure 1) and also that lipid profiles are tightly linked to
   cell-type identities and, even more importantly, their precise
   pathophysiological states [[52]8,[53]9,[54]10]. While this precision
   prompted the lipid analysis of a wide variety of tissues by spatially
   resolved techniques, few studies used human lungs [[55]11,[56]12]. This
   is notable because the lungs are affected by two of the deadliest
   diseases: lung cancer and chronic obstructive pulmonary disease (COPD)
   [[57]13,[58]14]. Thus, lung cancer remains the leading cause of
   cancer-related deaths, accounting for over 1.8 million deaths and 2.1
   million new cases in 2020 alone [[59]14], while COPD is the third
   leading cause of death worldwide [[60]13]. In addition, COPD increases
   the risk of lung cancer, and around 40 to 70% of lung cancer patients
   are diagnosed with COPD [[61]15,[62]16]. Herein, we focused on these
   two diseases, analyzing lung tissue sections at a lateral resolution of
   25 μm and comparing them to the non-affected counterpart and healthy
   tissue. At this resolution, we were able to associate specific lipid
   profiles with the main tissue types and independently assess the impact
   of smoking, chronic inflammation, and tissue malignization. Consistent
   with numerous studies analyzing tumor epithelia, the most robust change
   was the increase in arachidonic acid (AA)-containing phospholipids, in
   agreement with the role of this fatty acid in cell differentiation as
   well as cell malignization [[63]8,[64]9]. Despite the intrinsic
   difficulties that the handling of lung tissues entails, the results
   strongly encourage the analysis of the tissue at higher lateral
   resolutions to fully map the changes in lipid composition associated
   specifically with each lung cellular type and subtype. There is no
   doubt that the rapid and constant advances in MSI techniques offer a
   unique opportunity to reach this ambitious aim.

2. Materials and Methods

2.1. Study Design

   Consecutive smoker and non-smoker patients undergoing bronchoscopy for
   diagnosis purposes were invited to participate. Written informed
   consent was obtained for each patient recruited at Hospital
   Universitari Son Espases (Palma, Balearic Islands, Spain). Patients
   underwent conscious sedation during fiber-optic bronchoscopy, which was
   performed for diagnosis purposes in either the right or left lung.
   Bronchial biopsies were performed with alligator forceps in the second
   or third sub-segmental division of the right or left lower lobes in the
   COPD and non-smoker control (hereafter, Control) groups. In patients
   with endobronchial lesions, later confirmed as neoplasms, a biopsy was
   taken from the tumoral lesion (labeled as tumor) and another biopsy
   from the opposite lung to the neoplasm (labeled as non-tumor). Biopsies
   were immediately snap-frozen in liquid nitrogen and saved at −80 °C
   until analysis.

   All patients underwent lung function assessment by means of forced
   spirometry. Patients were then classified based on the results of
   bronchoscopy, chest computed tomography, and spirometry as COPD, lung
   cancer, smoker, or control.

   Tissue sections approximately 10 μm thick were prepared in a cryostat
   (Leica CM3050S, Leica Biosystems, Wetzlar, Germany) at −20 °C without
   the use of cryoprotective substances or embedding material and mounted
   on regular glass microscope slides. To facilitate structural
   identification, the consecutive section was stained with hematoxylin
   and eosin (HE, SigmaAldrich, Madrid, Spain).

2.2. Sample Preparation for MALDI-Imaging, Measurement Conditions, and Data
Analysis

   Tissue sections were coated with 1,5-diaminonaphthalene (DAN, 97%,
   Sigma-Aldrich Chemie, Steinheim, Germany), a suitable matrix for
   negative-ion detection [[65]17]. To help structure identification, a
   consecutive section was stained with HE. Leucine enkephalin acetate
   hydrate, ammonium acetate, and sodium hydroxide solution were purchased
   from Sigma Aldrich Chemie (Steinheim, Germany).

   Garate et al. describes in detail MALDI imaging and data analysis
   [[66]18,[67]19]. Briefly, a total of 23 sections were scanned in
   negative-ion mode using a MALDI-LTQ-Orbitrap XL (Thermo Fisher
   Scientific, San Jose, CA, USA), an orbitrap analyzer equipped with an
   N[2] laser (LTB, Berlin, model MNL 100), 100 µJ max power, an
   elliptical spot, and a 60 Hz repetition rate. The mass resolution used
   to record the data was 60,000 at m/z = 400 Da. The scanning range was
   550–1000 Da. Within this m/z range, we focused on glycerophospholipids
   and sphingolipids detected more consistently in the negative mode,
   namely PE, ether PE, PI, PS, and SM. For each pixel, two microscans of
   10 laser shots were recorded, and the raster size used was 25 μm. We
   used in-house programs developed in MATLAB (MathWorks, Natick, MA, USA)
   to align and analyze spectra.

   The MSI data processing protocol may be found in ref. [[68]19]. In
   brief, spectra were normalized using a total ion current algorithm and
   aligned using the parsing stage the Xiong method [[69]20]. To
   streamline analysis, peaks with intensity values below 0.5% of the
   intensity of the strongest peak were filtered, reducing the number of
   m/z. Lipid assignments were made by matching experimental m/z values
   and the species in the software’s database (<33,000 lipid species plus
   adducts) and in the LIPID MAPS^® database ([70]www.lipidmaps.org (last
   accessed on 17 March 2025)). Mass accuracy was established <9 ppm and
   typically <3 ppm.

   For each experiment, the spectra of each tissue segment was obtained
   using hierarchical divisive-rank compete analysis (HD-RCA) as an MSI
   segmentation algorithm [[71]21]. The specific nature of the bronchial
   tissue type associated with each of the MSI segments was identified and
   confirmed by at least two expert pathologists working at the hospital
   where the patients underwent the bronchoscopic procedure.

   The percentage of each lipid species was calculated by dividing its
   intensity by the total intensity of all species within the same lipid
   class (e.g., for PE species: PE[i]% = I[PEi]/ΣI[PE]·100). This method
   enables averaging the lipid profiles across sections within the same
   study group or comparisons between multiple study groups. To be able to
   delve into the potential biological meaning of the results and the
   potential fatty acid moieties involved, we compared the phospholipid
   profiles obtained with those previously published
   [[72]22,[73]23,[74]24]. Hence, we will describe species as “highly
   enriched” in a particular fatty acid only when this characterization is
   supported by published data. Finally, the compounds identified as
   plasmalogens of PE (PE P) have the same m/z values as their
   corresponding alkyl-ether analogs (PE O) with an additional double
   bond. In the graphs we have assigned them to PE P for simplicity, but
   they are included in the [75]Supplementary Tables S1–S3. This decision
   is based on the following rationale. On the one hand, in most of the
   tissues analyzed in the literature, the content of alkenyl-PE is
   substantially higher than that of alkyl-PE. On the other hand, the
   resultant species profile fits with the published literature.
   Nevertheless, the presence of additional isobaric lipid species sharing
   the m/z cannot be ruled out.

2.3. Gene Set Enrichment Analysis

   The [76]GSE168466 dataset from Gene Expression Omnibus (GEO) was used
   to explore the difference in gene expression level between healthy and
   tumor lung epithelial cells. The dataset contains RNA-seq information
   on human epithelial cells from non-small cell lung cancer tissue (n =
   49) and its healthy counterpart (n = 49). Weighted correlation network
   analysis (WGCNA) was used to identify correlation patterns of gene
   expression across samples using the CEMiTool package, which provides a
   comprehensive set of functions for WGCNA [[77]25]. The parameters were
   set as follows: (1) the most variant genes were selected according to a
   p-value cutoff of 0.1 and (2) Pearson’s correlation was used to
   identify the modules of co-expressed genes, using a soft-thresholding
   power automatically set to ensure scale-free topology, with a β-value
   of 7 (scale-free R^2 = 0.81). The minimum module size was set at 20
   genes, and similar modules were merged. Pathway enrichment was
   calculated with EnrichR [[78]26] using the Reactome database (2022).
   The top 10 pathways were ranked and plotted according to p-value
   enrichment and considered representative of the modules included. Lipid
   genes listed in co-expressed modules were analyzed with the STRING tool
   [[79]27]. The default combined string score represents node–strength
   interaction, and enrichment analysis was calculated to illustrate node
   contribution to the lipid pathway. Cytoscape v3.10.2 was used to color
   nodes according to co-expression modules, calculate network enrichment
   using the String enrichment app, and generate the final network layout.

3. Results

3.1. MALDI-MSI Lipid Segments Established Accurately Described the Tissue
Architecture Found in Bronchoscopic Biopsies

   Twenty subjects were recruited, and samples were classified into the
   following groups: patients with lung cancer (tumor samples (n = 5) and
   non-tumor samples (n = 5)), patients with COPD (n = 4), smokers (n =
   5), and non-smokers with normal lung function (n = 4, control group).
   The characteristics of the cohort are summarized in [80]Table S1.

   Bronchoscopy biopsies usually consist of respiratory mucosa, including
   epithelium and stroma, and, occasionally, cartilage ([81]Figure 2a). In
   turn, bronchi have cartilaginous reinforcement and a pseudostratified
   columnar epithelium, which is highly heterogeneous in composition and
   function, containing ciliated, goblet, club, and underlying basal cells
   [[82]28]. The heterogeneity described at the structural and cellular
   level posed a real challenge during the MALDI-MSI image interpretation.

Figure 2.

   [83]Figure 2
   [84]Open in a new tab

   Differences in membrane lipid profile between lung epithelium,
   cellular, and fibrous stroma. (a) First row, HE image of the
   consecutive section obtained from bronchoscopy biopsies. Scale bar =
   150 μm. Second and third rows, segmentation images generated using
   non-supervised HD-RCA segmentation establishing 5 and 10 segments.
   Fourth and fifth rows, distribution of selected lipid species. (b)
   HE-stained section with the tissue types identified (E: epithelium, CS:
   cellular stroma, FS: fibrous stroma), segmentation images associated
   with E (light red cluster), CS (purple cluster), and FS (white
   cluster), and images of selected m/z. (c) Bar diagrams showing the
   comparison of the lipid profiles of the three main clusters. Values are
   expressed as % of total membrane lipids (mean ± SEM), n = 3.
   Statistical differences were assessed by multiple unpaired t-tests; aa,
   p ≤ 0.01; aaa, p ≤ 0.001.

   In [85]Figure 2a, a series of MALDI-MSI images of a representative
   section of each study group was compiled. We used a segmentation
   algorithm (HD-RCA) to sort and group the pixels with the most similar
   lipid profile which, in the end, constitutes a lipid segment or
   cluster. It is important to bear in mind that segment generation is a
   mathematical process, and the number of segments can range from one to
   virtually the number of pixels the image is composed of. So, as a rule
   of thumb, we established as many segments as biological significance we
   can ascribe. Due to a lack of previous knowledge on lung epithelial
   lipidome, we took a prudent approach while establishing the epithelial
   segment and used just one. Nevertheless, considering that the segment
   can be further divided, it is highly plausible that analysis at higher
   lateral resolution would help identify additional segments accounting
   for the presence of epithelial subtypes, similar to that observed in
   colon epithelial cells during differentiation [[86]8,[87]9].

   In bronchoscopy biopsies of the control group, we were able to identify
   three segments that matched the epithelium, the cell stroma, and the
   fibrous stroma ([88]Figure 2b). Interestingly, the lung stroma
   preserved one of the characteristics of colon stroma: the content of
   AA-containing lipid species in stromal phosphatidylinositol (PI) and
   phosphatidylethanolamine (PE) was at least 2-fold higher than in the
   epithelium (PI 38:4 and PE 38:4, [89]Figure 2c) [[90]8]. Conversely,
   stromal clusters contained lower levels of species highly likely to be
   enriched in docosahexaenoic acid (DHA, 22:6n-3) according to the
   literature, such as PE 38:6, PE 40:6, PE P 38:6, PE P 40:6, and PS 40:6
   [[91]22,[92]23,[93]24], which, depending on the species, decreased by
   30 to 90%.

3.2. Identification of a Tissue-Type Dependent Response to Lung Malignization
at the Lipid Profile Level

   It is well established that tissue neoplastic transformation induces
   profound changes in cells at both gene and metabolic levels. Numerous
   studies have demonstrated the impact of the malignant transformation on
   the lipidome in a wide range of cancers, including lung cancer
   [[94]11,[95]29,[96]30]. We used MALDI-MSI to study the impact of cancer
   at the tissue-specific level, avoiding the bias induced by tissue or
   cell-type isolation protocols. Thus, we first compared the lipid
   profiles of the malignant and non-malignant lung tissues obtained from
   patients with lung cancer ([97]Figure 3).

Figure 3.

   [98]Figure 3
   [99]Open in a new tab

   Impact of lung cancer on lung tissue membrane phospholipid profile. Bar
   diagrams show the comparison of non-tumor and tumor lung tissues: PI,
   PE, PEP, SM, and PS composition in each of the clusters identified
   (from left to right): (a) epithelium, (b) cell stroma, and (c) fibrous
   stroma. Values are expressed as % of total membrane lipids (mean ± SEM)
   in non-tumor (lighter colors, n = 4–5 E; n = 3–4 CS; n = 3–4 FS) and
   tumor tissue (darker colors, n = 4–5 E; n = 5–6 CS; n = 2 FS).
   Statistical differences were assessed by multiple unpaired t-tests. a,
   p ≤ 0.05; aa, p ≤ 0.01; aaa, p ≤ 0.001. Detailed results are included
   in [100]Table S3.

   Globally, the most consistent change in the lung epithelium was a
   robust increase in all the lipid classes detected in species highly
   likely to be enriched in AA [[101]22,[102]23,[103]24]. The greater
   impact was in diacylglycerophospholipids (PI, PE, and PS lipid classes,
   approx. 60%) than in PE plasmalogens (PE P, approx. 18%). These
   observations were consistent with the literature and our previous
   results in the colon epithelium [[104]8,[105]9]. Thus, the lung PE
   profile was enriched in 36:1, 36:2, 38:2, and 38:4 species, each
   accounting for 10–20% of total PE, which were the most altered species
   in the tumor epithelium. While 38:4 and 36:4 content drastically
   increased by 91.3% and 64.4% in the malignant tissue compared to
   non-malignant tissue (14.5 vs. 27.7% and 1.3 vs. 2.2%, respectively),
   PE 36:1 and PE 36:2 content tend to decrease by 16.4% and 14.4%,
   respectively (31.0 vs. 25.9% and 18.8 vs. 16.1%). PE plasmalogen
   species showed a similar trend, that is, an increase in AA- and a
   decrease in MUFA-containing species, but reached significance only for
   PE P 36:1, which decreased by 49.5%. The decrease observed in
   DHA-containing species, PE P 38:6 and PE P 40:6, was also particularly
   interesting, albeit not significant.

   Generally, PI 38:4 is considered the most abundant PI species
   [[106]31]. However, similar to the colon epithelium, the lung
   epithelium PI repertoire was enriched in more species, such as PI 34:1,
   PI 34:2, PI 36:2, PI 36:4, and PI 38:3, each of which accounted for 5
   to 10% of total PI. The results indicated that lung cancer induced a
   profound remodeling in PI species in the non-tumor epithelium. Thus, in
   the tumor epithelium, PI 38:4 content increased by 62.1% compared to
   the control epithelium (41.5 vs. 67.3%), while PI 36:2, PI 34:2, PI
   34:1, and PI 32:0 content decreased by 57.5, 69.2, 72.1, and 81.6%,
   respectively (12.5 vs. 5.3%, 5.9 vs. 1.8%, 13.7 vs. 3.8%, and 1.8 vs.
   0.3%). Consistent with changes in PE and PI, PS 38:4 content tended to
   increase (49.1%) in the tumor epithelium compared with NT, while PS
   40:6 (enriched in DHA) significantly decreased by 50.8%. Finally, SM
   d34:1, the most abundant SM species, decreased by 17.3% in tumor tissue
   (70.7 vs. 58.5%).

   Next, we explored the differences in the remaining tissues. In most of
   the samples, the pathologists identified two types of stroma, cellular
   and fibrous stroma, which differ in the cell content. However, only PS
   showed a robust increase in PS 38:4 and a drastic decrease in PE 36:1,
   which was sharper in fibrous than in cellular stroma. PE 38:4 also
   increased in cellular but not in fibrous stroma, most probably due to
   the lower content of immune cells in the latter.

   Altogether, these results indicated that the lung epithelium was the
   most affected tissue in terms of lipid composition, showing a solid
   increase in AA-containing species throughout all the lipid families
   detected.

3.3. Impact of Chronic Inflammation on Healthy Lung Tissue

   Chronic inflammation is associated with a higher risk of developing
   cancer [[107]15]. For this reason, we were interested in assessing the
   differences in lipid composition of tissues obtained from patients with
   COPD and lung cancer and comparing them to the healthy control group
   ([108]Figure 4). Interestingly, the most affected lipid classes were PI
   and, to a lesser degree, PE, while in PE P, SM, and PS, changes were
   scattered. Compared to the control group, COPD showed lower levels of
   PI 36:2, PE 36:2, PE 38:2, and PS 38:4 (32.4, 23.3, 26.0, and 44.9%,
   respectively) and higher content of PI 38:3 and PE 38:4 (99.0 and
   40.3%, respectively). Consistently, PE 38:4, which is also an increase
   in tumor tissue (82.3%), was one of the species that clearly increased
   in the colon epithelium of patients with chronic inflammation [[109]10]
   and adenomatous polyps [[110]8]. The comparisons of the cellular and
   fibrous stroma lipid profile can be found in [111]Table S3.

Figure 4.

   [112]Figure 4
   [113]Open in a new tab

   Impact of chronic inflammation (COPD) and lung cancer on the lung
   epithelium membrane lipid profile. Bar diagrams show the comparison of
   epithelial (a) PI, (b) PE, (c) PEP, (d) SM, and (e) PS composition in
   the control (C), tumor (T), and COPD groups. Values are expressed as %
   of total membrane lipids (mean ± SEM) in C (n = 4), T (n = 5), and COPD
   (n = 4) groups. Statistical differences were assessed by multiple
   unpaired t-tests: a, C vs. NT; b, C vs. T; c, NT vs. COPD. Depending on
   the figure one, two, and three symbols may represent, p ≤ 0.05, p ≤
   0.01, and p ≤ 0.001, respectively. Detailed results are included in
   [114]Table S3.

3.4. Changes in Lipid Composition in Non-Malignant Tissues

   Finally, we investigated the changes in the lipid composition in
   tissues not affected by chronic inflammation or cancer, wherein
   macroscopic changes, if any, are subtle, by comparing the control,
   smoker, and non-tumor groups ([115]Figure 5). The most consistent
   changes occurred in the epithelium, particularly in PI, PE, and PE P.
   Conversely, the changes followed similar patterns in cellular and
   fibrous clusters and differed from those in the epithelium ([116]Table
   S3).

Figure 5.

   [117]Figure 5
   [118]Open in a new tab

   Non-tumor lung epithelial membrane phospholipid profile comparison. Bar
   diagrams show the comparison of epithelial (a) PI, (b) PE, (c) PEP, (d)
   SM, and (e) PS composition in the control (C), smokers (Sm), and
   non-tumor (NT) groups. Values are expressed as % of total membrane
   lipids (mean ± SEM) in C (n = 4), Sm (n = 5), and NT (n = 5).
   Statistical differences were assessed by multiple unpaired t-tests: a,
   C vs. Sm; b, C vs. NT; c, and Sm vs. NT. One, two, and three symbols
   represent, p ≤ 0.05, p ≤ 0.01, and p ≤ 0.001, respectively. Detailed
   results are included in [119]Table S3.

   In the epithelium, the PI profile in the control and smoker groups was
   quite identical, while statistical changes were mainly identified when
   compared with the non-tumor group. The largest change occurred in PI
   38:4, decreasing by 12.8 and 25.7%, while PI 34:1 increased by 51.9 and
   90.6% (control vs. non-tumor and smoker vs. non-tumor, respectively).
   Conversely, PE composition in the control and smoker groups showed some
   changes, such as increased PE 38:4, despite it reaching significance
   only for PE 38:2 ([120]Figure 5, [121]Table S3).

3.5. Lipid Metabolism Gene Expression Regulation in Non-Small Cell Lung
Cancer

   To delve into the biological significance associated with the changes
   observed in malignant epithelia, we used the gene expression dataset
   [122]GSE168466 containing healthy and tumor epithelial human non-small
   cell lung cancer samples of 98 patients. These samples were classified
   by the authors of the data based on the WHO basis for histological
   morphology and molecular typing, based on labeling by
   immunohistochemical P63/Napsin A and TTF-1/CK7 double staining, and
   were divided into the following five groups according to tissue origin:
   bronchoalveolar carcinoma, alveolar epithelial carcinoma, bronchial
   epithelial carcinoma, secretory adenocarcinoma, and mucinous
   adenocarcinoma.

   First, we conducted a Weighted Gene Co-expression Network Analysis
   (WGCNA) to identify modules of co-expressed genes associated with
   epithelial malignization [[123]32]. Such a strategy is useful for
   uncovering unexpected patterns and connections that can be easily
   overlooked using more targeted approaches. WGCNA application of omic
   and multiomic data integration to phenotypic information has been
   applied in fields covering the discovery of regulatory programs, target
   genes, proteins, and metabolites involved or associated with specific
   phenotypes [[124]33,[125]34]. The WGCNA was calculated using the
   CEMiTool package [[126]25], which returned 20 modules using automatized
   determination of beta value (β = 7, r^2 = 0.81) to ensure network
   scale-free topology. In [127]Figure 6a, the tSNE plot summarizes gene
   expression variance according to the profile of co-expression modules
   for each sample (module eigengene). Generally, healthy and tumor
   samples show several expression modules, demonstrating an opposite
   distribution (green vs. red). The sample distribution also conveyed
   lung region and tissue of origin (right side of the tSNE plot).
   Similarly, the hierarchical-clustering heat map of the eigengene
   modules arranged the samples according to pathological state
   ([128]Figure 6b).

Figure 6.

   [129]Figure 6
   [130]Open in a new tab

   Lipid metabolism network associated with lung epithelial malignization.
   WGCNA was used to identify co-expressed genes representative of lung
   epithelium malignization based on the gene dataset [131]GSE168466.
   (a,b) tSNE plot summarizing sample distribution according to (a) their
   co-expression module profile and pathological state (healthy group,
   green dots; Tumor group, red dots), and (b) lung region and epithelial
   tissue of origin; (c) hierarchical-clustering heatmap of module
   eigengene highlighting the association of the module to epithelial
   malignancy; (d) modules more differentially expressed in healthy or
   tumor samples were selected based on ranked ANOVA score, representing
   profiles that better described epithelial transformation; (e) pathway
   enrichment analysis computed using EnrichR for healthy (green plot) and
   tumor modules (red plot) describing the top 10 p-value ranked Reactome
   pathways; (f) STRING network representing the overall interaction of
   genes involved in lipid metabolism.

   Then, we ranked the modules according to the difference between healthy
   and tumor samples and selected those with the best scores. Thus, nine
   up- and down-regulated modules were identified in the healthy vs. tumor
   comparison ([132]Figure 6c). Modules 4, 7, and 18 were more correlated
   to healthy samples, which showed significant enrichment for processes
   related to the cilium assembly and termination of O-glycan
   biosynthesis. Conversely, tumor samples showed a positive correlation
   for modules 1, 9, 10, 11, and 15. The pathway enrichment analysis
   described protein metabolism processes such as GTP hydrolysis and
   joining of the 60S ribosomal subunit and SRP-dependent co-translational
   protein targeting to the membrane, among others (Reactome 2023.1.Hs;
   Enrichr) ([133]Figure 6d) [[134]26].

   Module composition was further examined using the Reactome database,
   identifying the lipid-related genes included as well as the most
   representative pathways according to the enrichment analysis
   ([135]Figure 6e). The network was computed using the subtracted genes
   as input in the STRING database [[136]27] and the combined string score
   was used to define the strength of node interaction. Overall, the
   network is dominated by the tumor-associated module 1. This module
   includes a set of lipid-related genes tightly involved in phospholipid
   metabolism (CSNK2A1, CDIPT, PI4KA, GDE1, LPCAT3, LPIN2, PIK3C3, LPCAT2,
   CSNK2A2, PIK3C2A, RAB5A, MBOAT2, LCLAT1, STARD10, STARD7, PIP5K1A,
   PLEKHA1, RAB14, CSNK2B, INPP5D, CDS2, PIK3R1, PTDSS1, ARF1, PLEKHA2,
   DGAT1, and PITPNB) and, more specifically, in PI and in sphingolipid
   metabolism (ORMDL2, SPTSSA, SGPP2, DEGS1, VAPA, PRKD3, CERS6, ORMDL3,
   PRKD1, and VAPB). The alteration in these pathways would be consistent
   with the most relevant changes in lipid profiles.

4. Discussion

   In this study, we reinforce the current evidence demonstrating the
   great potential of MALDI-MSI as a powerful tool for detecting and
   localizing specific lipid profiles within biological tissues, making
   MALDI-MSI a valuable approach to providing new insights into complex
   biological systems, such as the pathophysiology of the respiratory
   system. While these techniques generate comprehensive information about
   tissue composition, drawing the correct functional conclusions about
   tissue or organ behavior is delicate due to the high heterogeneity
   coexisting at the cell compositional level
   [[137]28,[138]35,[139]36,[140]37]. Further, MSI techniques are
   challenging due to the vast and specific amount of data generated,
   demanding a reinterpretation of lipid metabolism and highlighting the
   gap in knowledge regarding the molecular mechanisms tailoring the
   precise lipid profile [[141]38].

   MSI clearly establishes that the lipid composition is highly cell
   type-specific but also very sensitive to the cell physiopathology
   state. Spatially resolved techniques have been used to analyze an
   extensive list of human tissue lipidomes, including brain
   [[142]39,[143]40], skin [[144]41,[145]42,[146]43], kidney [[147]44],
   colon [[148]8,[149]9], or breast tissue [[150]45,[151]46], all
   revealing the tissue-dependent distribution of lipid profiles.
   Curiously, information about human lung composition at this level is
   scarce, and only a few studies [[152]30,[153]47], including studies on
   humans [[154]11,[155]12], can be found. The underlying reasons for the
   absence of data could be related to certain difficulties while working
   with lung tissue, such as its spongy nature [[156]47], heterogeneity in
   anatomical and functional structures (bronchia, bronchioles, alveoli),
   and cell type variety, with more than forty different existing cell
   types [[157]28,[158]37] that interact tightly to provide the right
   tissue architecture and functional capacity for effective gas exchange
   [[159]28,[160]37].

   Current single-cell technologies allow for detangling tissue cell
   heterogeneity and analyzing its composition at the RNA level. However,
   because of the low quantity of cells obtained, establishing the lipid
   composition is highly challenging. MALDI-MSI offers a unique
   opportunity because, on the one hand, it is highly sensitive to lipid
   species, and, on the other hand, it provides a lateral resolution
   equivalent to the size of one to two cells. Following the single-cell
   approach, Wang et al. [[161]48] isolated nine cell lineages from
   healthy lung tissue and established lineage-dependent differences in
   genes involved in fatty acid and glycerophospholipid metabolism. They
   showed that AA metabolism is downregulated, while alpha-linolenic acid
   (an essential fatty acid and precursor of DHA) is upregulated in the
   epithelial lineage compared to fibroblast and myeloid lineages
   [[162]48], which is consistent with stromal AA enrichment and a lower
   DHA epithelial content ([163]Figure 2c). Interestingly, AA- and
   DHA-containing phospholipids localize at the edges of airways in murine
   lungs, and although the specific tissue origin was not fully uncovered,
   the evidence pointed to the epithelial lining of airways [[164]47].

   Tissue malignization involves a profound metabolic reprogramming
   wherein lipid metabolism plays a prominent role, although the specific
   role of membrane phospholipid species is less established [[165]49].
   Tumor cells acquire the ability to thrive in a nutrient-poor
   environment by modifying their metabolic fluxes in response to a great
   demand for growth and proliferation. These changes extend to all cells
   in the tumor microenvironment that will need to compete with tumor
   cells for survival. Lipid metabolism exerts a pivotal role in this
   metabolic adaptation, and alterations of tumor lipid metabolism at the
   transcriptional level leading to the de novo lipogenesis activation
   were described [[166]49,[167]50]. Still, the specific role of membrane
   phospholipid species is less established.

   Lung cancers usually originate from basal epithelial cells, located in
   the lower parts of the epithelium, where the broadest changes in lipids
   occurred, consistent with the central role of these cells [[168]51]
   ([169]Figure 3). PI and PE were the most affected classes, showing a
   robust increase in PI 38:4 and PE 38:4, which in the colon are related
   to the epithelial undifferentiation process occurring in adenomatous
   polyps [[170]8] and chronic inflammation [[171]10]. PC and PE AA
   species, particularly those of the immune cells, are considered the
   main source of lipid mediators, such as leukotrienes and
   prostaglandins, involved in the impairment of mucociliary clearance of
   particles or the promotion of bronchoconstriction and mucous secretion
   [[172]47]. Furthermore, Marien et al. established lipid changes
   agreeing with ours in bulk lung cancer biopsies, revealing an increase
   in species with more than 36 carbons, which they linked to an altered
   expression in fatty acid elongases [[173]11]. Consistently, Wang et al.
   used single-cell RNA-seq to demonstrate that pathways involved in the
   metabolism of glycerolipids, glycerophospholipids, fatty acid
   biosynthesis, and biosynthesis of unsaturated fatty acids were
   dysregulated in early-stage lung cancer in a cell-type-dependent manner
   [[174]48]. Importantly for this study, Gouw et al. investigated gene
   expression changes in the lung of KRAS-activated mice, finding that, in
   addition to genes involved in de novo fatty acid synthesis, the
   activation also affects sphingolipid and PI metabolism genes [[175]30].
   Finally, we show that lung cancer affects not only the tumor lipid
   profile but also non-tumor tissue composition ([176]Figure 5), which is
   consistent with the systemic impact cancer has on the homeostatic
   status.

   Although 40–70% of patients with lung cancer are diagnosed with COPD
   [[177]15,[178]16], the exact link between them is uncertain, as both
   have similar causes, mainly smoking. COPD and lung cancer present
   oxidative damage, some genetic markers, and methylation patterns, as
   well as the inflammatory component of both diseases [[179]51]. A better
   understanding of the relationship between these two diseases could
   benefit their simultaneous treatment. The most relevant change found
   herein was the increase in PE 38:4 in COPD and also smoker groups,
   which, as mentioned, is a precursor of lipid mediators such as LTB4 and
   PGE2, both raised in sputum of patients with COPD [[180]52,[181]53].

   Finally, this study has two main limitations. First, the samples were
   analyzed exclusively in negative-ion mode; consequently, it was not
   feasible to obtain solid data for establishing a phosphatidylcholine
   (PC) profile, the most abundant membrane phospholipid. Fatty acid
   remodeling, the main cause of changes in lipid species content, is a
   phospholipid-dependent process [[182]54]. In a previous study comparing
   non-tumor and malignant colon epithelia and stroma, the changes in PC
   profile followed similar trends to PI and PE, although less intense
   (except for PC 34:1) [[183]8]. Nevertheless, epithelial and stromal PC
   profiles were specific to the tissue type, reinforcing the relevance of
   spatially resolved techniques. Conversely, we obtained unique
   information on phospholipids that are difficult to characterize by
   other methodologies that, despite being considered minor classes, are
   essential in cell signaling, such as phosphatidylinositol.

   Secondly, there is the limitation in unequivocal lipid species
   identification, which is common to MSI lipidomic techniques and stems
   from the lack of a separation stage before the mass analysis. In the
   present study, all lipid species were carefully assigned using highly
   restrictive parameters. Furthermore, the resulting lipid profiles are
   comparable to those published and established with liquid
   chromatography-mass spectrometry (LC-MS) methodologies. Despite this,
   the presence of additional isobaric or isomeric lipid species cannot be
   ruled out, and consequently, the biological interpretation could be
   slightly affected by their presence. It is worth mentioning that,
   conscious of this constraint, the field is constantly improving the
   analytical processes so that this limitation is overcome [[184]7]. In
   this context, the recent introduction of ion mobility cells into mass
   spectrometers helps the segregation of isobaric species [[185]55].
   Concurrently, Claes et al. introduced a new technique based on an
   ozone-induced dissociation to enable the high-resolution mapping of
   phospholipid isomers throughout relatively large areas [[186]56].
   Alternatively, it is possible to run MS/MS experiments directly from
   the section, although most current mass analyzers are still not capable
   of working in image mode and recording MS/MS experiments
   simultaneously. Finally, some MSI researchers use a hybrid approach,
   combining laser capture microdissection to select specific areas and
   analyze them with LC-MS (in this case, it is important to identify the
   impact of laser burning on phospholipids) [[187]57].

5. Conclusions

   Owing to improvements in MALDI-MSI instruments and protocols [[188]54],
   not only is it becoming possible to reach subcellular resolutions but
   also to run kinetic approaches [[189]55,[190]56] or multiplexed
   immunohistochemistry [[191]57]. Herein, we provide evidence sustaining
   the importance of spatially resolved molecular techniques to fully
   understand cell pathophysiology. Despite the intrinsic difficulties
   that lung tissue handling entails, the results strongly encourage
   tissue analysis at higher lateral resolutions to fully map the changes
   in lipid composition associated specifically with each lung cellular
   type and subtype. Undoubtedly, the rapid, constant advances in MSI
   techniques offer a unique opportunity to achieve this ambitious aim.

Acknowledgments