Lipid profiling in nontuberculous mycobacterial pulmonary disease based on geospatial locations and disease severity

Abstract

Background

The role of lipid profiles in the pathogenesis of nontuberculous mycobacterial pulmonary disease (NTM-PD) remains largely unexplored. This study aimed to identify lipid profile variations across geospatial lung lesions, their reflection in serum relative to disease severity and their diagnostic discriminative ability using lipidomic analysis.

Methods

Lipidomics was performed using hydrophilic interaction liquid chromatography–tandem mass spectrometry on lung tissues and serum samples. 960 lipid species were analysed across geospatial lung lesions (cavity wall, centre of cavity, granuloma, bronchiectasis and normal lungs) and assessed in serum according to disease severity. Python-based machine-learning models in PyCaret were used to classify NTM-PD based on lipidomic signatures.

Results

This study included 23 lung specimens from seven patients with NTM-PD and 332 serum samples comprising 134 patients with NTM-PD, 136 with non-NTM bronchiectasis and 62 healthy controls. Triacylglycerol (TG) levels were elevated in lung lesions affected by NTM-PD, particularly in the centre of the cavity. In serum, TG levels were higher in patients with NTM-PD than controls but decreased in patients with more severe disease, including those with acid-fast bacilli smear positivity, cavitation or higher BACES (body mass index, age, cavity, erythrocyte sedimentation rate and sex) scores. The top five models, developed using lipid species characteristically altered in NTM-PD, effectively discriminated patients with NTM-PD from healthy controls.

Conclusion

TG levels were elevated in lung lesions affected by NTM-PD but decreased in serum as disease severity increased, suggesting TG accumulation in lung tissues. These findings highlighted the role of lipid metabolism in the pathogenesis of NTM-PD.

Shareable abstract

Our findings indicate that lipid dysregulation, including TG alterations, varies by lesion characteristics and disease severity in NTM-PD. The selected lipids in this study show potential as biomarkers for NTM-PD. https://bit.ly/3IOmdjC

Introduction

Nontuberculous mycobacteria (NTM), comprising over 190 species but excluding Mycobacterium tuberculosis and Mycobacterium leprae, are environmental organisms found in soil, air, dust and municipal water [1]. These organisms can cause chronic infections in humans, and their epidemiological significance has steadily increased. In South Korea, the prevalence of NTM infections rose from 11.4 cases per 100 000 in 2010 to 56.7 cases per 100 000, surpassing the prevalence of tuberculosis [2].

Most NTM infections manifest as pulmonary diseases (PD), with their clinical progression often influenced by histological and radiographical findings. For instance, cavitation is linked to faster disease progression and poorer treatment outcomes [3], whereas nearly half of noncavitary bronchiectasis (BE) cases resolve spontaneously [4]. Within a single host, BE and cavitation may coexist, exhibiting geospatial variability across the lungs and contributing to the complexity of NTM-PD. Understanding the mechanisms underlying this geospatial variability is crucial for elucidating the pathophysiology of NTM-PD.

Lipidomics, a branch of metabolomics, examines disease-specific lipid profiles, providing insights into the underlying pathophysiology [5]. Lipidomics has been widely applied to various respiratory diseases, including mycobacterial infections such as tuberculosis. Lipidomic studies have elucidated mechanisms driving the increased virulence of M. tuberculosis [6], identified biomarkers for treatment monitoring [7] and characterised lipid profiles predictive of treatment responses [8]. However, its application to NTM-PD remains largely unexplored.

In this study, we conducted lipidomic analyses of NTM-PD using lung tissues and serum. First, we aimed to identify characteristic lipid profiles associated with diverse geospatial lesions, enabling correlations with specific pathological findings. Next, we examined how these lesion-specific lipidomic profiles are reflected in serum relative to disease severity. Through these analyses, we sought to deepen our understanding of the role of lipid profiles in the pathophysiology of NTM-PD.

Materials and methods

Study population and data collection

This cross-sectional study followed the principles of the Declaration of Helsinki and received approval from the Institutional Review Board of Seoul National University Hospital (number 2403-058-1518). Informed consent was obtained from all participants.

For lung tissue analysis, we included patients who underwent surgical resection for NTM-PD at Seoul National University Hospital between 1 June 2022 and 31 December 2023. These patients had received guideline-based treatment for NTM-PD [1] but required surgical resection due to 1) worsening radiographical findings, 2) persistent cavities or 3) intolerance to antibiotic treatment. For serum analysis, we included patients diagnosed with NTM-PD between 1 July 2011 and 31 December 2020, registered in the prospective NTM-PD cohort of Seoul National University Hospital (NCT01616745) [3, 9, 10]. These patients had not received anti-mycobacterial treatment prior to enrolment. Additionally, we included patients with noncavitary nodular BE, matched for age and sex (registered in a prospective BE cohort, NCT01616745) and healthy controls (registered in the Controls for Respiratory Disease Study, NCT03120481). The absence of NTM infection was confirmed by at least two negative acid-fast bacilli (AFB) cultures obtained at intervals of at least 1 month [11]. Many patients and healthy controls participated in our previous work [12].

Clinical, laboratory, and radiographic data were collected during surgical resection (for patients providing lung tissue) or at cohort enrolment (for patients providing serum). Radiographical findings were obtained from chest computed tomography scans performed within 3 months of surgery or cohort enrolment. Sputum AFB smear results and the corresponding causative species were also recorded. The severity of NTM-PD was classified as mild (score 0–1), moderate (score 2–3) or severe (score 4–5) based on the BACES (body mass index, age, cavity, erythrocyte sedimentation rate and sex) score [10].

Sample preparation

A lung pathologist with over 10 years of experience dissected the lung tissue. After palpating the lung tissue, the dissection was initiated from the visceral pleura. Cavity, BE and granuloma were identified, and the central necrotic material within the cavity was separately sampled. Tissue fragments measuring 5×5 mm were obtained from each area using a sterile blade. Additionally, tissue from areas adjacent to the lesion, which appeared normal, was collected. Each sample was immediately stored at −80 °C [13].

Serum was extracted from blood samples obtained when patients participated in the cohorts. The collected blood samples were incubated at 25 °C for 15–30 min, then centrifuged at 13 756×g for 20 min. The isolated serum (500 μL) was stored at −80 °C until further analysis [12].

Lipid nomenclature

The lipid nomenclature was annotated according to the guidelines provided by LIPID MAPS consortium. Abbreviations of the lipid classes are as follows: FA, fatty acid; PC, phosphatidylcholine; PC-O, alkylphosphatidylcholine; PC-P, alkenylphosphatidylcholine; LPC, lysophosphatidylcholine; LPC-O, lysoalkylphosphatidylcholine; LPC-P, lysoalkenylphosphatidylcholine; PE, phosphatidylethanolamine; PE-O; alkylphosphatidylethanolamine; LPE, lysophosphatidylethanolamine; PG, phosphatidylglycerol; PS, phosphatidylserine; PI, phosphatidylinositol; DG, diacylglyceride; TG, triacylglyceride; Cer, ceramide; DhCer, dihydroceramide; HexCer, hexosylceramide; LacCer, latosylceramide; SM, sphingomyelin; AC, acylcarnitine; and CE, cholesteryl ester.

Lipid extraction from sera and tissues

For lung tissue homogenisation, 80% isopropyl alcohol (IPA) was added to the tissue at a ratio of 1 mg:5 µL (tissue, 80% IPA) and homogenised using a BioMasher II homogeniser (Nippi, Tokyo, Japan). The homogenates were centrifuged for 5 min at 18 341×g and 4 °C. After centrifugation, the supernatants were collected into a new 1.5-mL Eppendorf tube. For lipid extraction from serum and lung tissue, the internal standard mixture (at the concentrations listed in supplementary table S1), was added to 20 µL of serum and tissue extracts. Lipids were further extracted using a solution of water:IPA: methyl tert-butyl ether in a 240:160:600 (v:v:v, µL) ratio, followed by vortexing and centrifugation for 5 min at 18 341×g and 4 °C. The upper phase was collected into a new 1.5-mL Eppendorf tube. The collected upper phases were mixed and dried under a nitrogen (N₂) gas stream. The dried residue was reconstituted with 0.2 mL of 95% acetonitrile (ACN) containing 10 mM ammonium acetate and transferred into an autosampler vial. Pooled sera and lung tissue extracts were used as quality control samples for signal drift correction.

LC–MS analysis

Lipidome measurements were performed using a Waters ACQUITY ultra-performance liquid chromatography (LC) (Waters Corporation, Milford, MA, USA) coupled to an AB Sciex 5500 triple-quadrupole mass spectrometer (AB Sciex, Concord, ON, Canada), as previously described with modifications [14]. An ACQUITY Premier BEH amide column (1.7 µm, 100 mm×2.1 mm, Waters, Milford, MA, USA) was used for lipid class separation during the LC analysis. The mobile phase consisted of 95% ACN for phase A and 50% ACN for phase B, with both phases containing 10 mM of ammonium acetate. The LC gradient was as follows: 0% phase B at 0 min, held at 0% phase B for 2 min, increased to 80% phase B by 5 min, reduced to 0% phase B at 5.1 min and equilibrated at 0% until 8 min. The column temperature was maintained at 45 °C, and the sample injection volume was 2 µL. The electrospray ionisation source parameters were set to a voltage of 4500 V in the positive mode and −3500 V in the negative mode. Due to the large number of TG species in human serum, the TG species with corresponding IS was analysed in separated batches at a 2- min runtime with 0% phase B. Multiple reaction monitoring (MRM) mode was used to acquire mass spectrometry (MS) data. During lipid acquisition in MRM mode, 2046 lipid species were screened using pooled lung extracts. Of these, 656 signals from pooled serum extracts and 960 signals from lung extracts were identified, each with a peak intensity exceeding 1000 and a relative sd <30%. A scheduled MRM mode was applied to measure the identified 656 and 960 lipid species with corresponding retention times, as detailed in supplementary table S1, using deuterated lipid standards and a retention time window of 1.5 min.

Data analysis

The peak areas of lipid species were integrated using Analyst software (AB Sciex, Concord, Canada) and subsequently analysed with the Shiny app of LICAR to correct isotopic overlap within the same lipid class [15]. The corrected peak areas were normalised using class-specific lipid standards. Additionally, signal drift in the large batch for the sera cohort was corrected using a quality control robust LOESS signal correction (QC-RLSC) from the statTarget package [16].

A hierarchically clustered heatmap was generated to depict the distribution patterns of lipid metabolites across the groups. A t-test and one-way ANOVA, followed by Tukey's post hoc test, were applied to identify significantly altered features between groups. Paired t-tests were used to identify lipid species significantly altered in lung lesions compared with normal lesions, with significance determined by raw p-values. A significance threshold of p<0.05 was used for all analyses. An upset plot was generated to visualise shared lipid alterations across different lesions. The classification model was constructed using serum lipid levels from NTM-PD lung tissue to distinguish NTM-PD from healthy controls (see supplementary methods S1 for further details). All statistical analyses and graphical outputs were conducted using MetaboAnalyst 6.0 [17], GraphPad Prism 7 (GraphPad Software, CA, USA), Python (Python Software Foundation, Fredericksburg, VA, USA) and R software v.4.3.1 (R Software Foundation, Vienna, Austria).

Results

Baseline characteristics of study population

Seven patients (six female and one male) who provided lung tissues were included in the analysis (table 1). The most common reason for surgical resection was the persistent presence of cavities (five patients). The causative organisms included M. avium (three patients), M. intracellulare (three patients) and mixed infection of M. intracellulare and M. abscessus subspecies abscessus (one patient). Detailed characteristics are provided in supplementary table S2. For the serum analysis, 332 patients (134 NTM-PD, 136 BE and 62 controls) were included. Among the 134 patients with NTM-PD, 26 (19.4%) had cavities and 31 (23.1%) had AFB smear positivity.

TABLE 1.

Clinical characteristics of study participants

Groups	Lung tissue	Serum
	NTM-PD	HC	NTM-PD	BE	p-value
Clinical characteristics
N	7	62	134	136
Sex, female/male	6/1	39/23	88/46	31/19
Age, years	63.6±7.8	59.2±9.9	62.5±10	58.9±10.1	0.09
BMI, kg·m−2	21.7±1.3	23.3±2.6	20.8±2.5	22.6±2.8	8.3×10−10
Laboratory indices
ESR, mm·h−1
Female	27.8±16.1	NA	36±16.3	28.7±21	0.07
Male	5#	NA	21.7±20.9	18.5±15.4	0.20
Creatinine, mg·dL−1	0.78±0.13	NA	0.8±0.2	0.8±0.2	0.07
Albumin, g·d−1	4.27±0.24	NA	4.3±0.3	4.3±0.3	0.30
Bilirubin, mg·dL−1	0.49±0.17	NA	0.7±0.3	0.6±0.2	0.06
Presence of cavity	5 (71.4)	NA	26 (19.4)	NA
AFB smear positivity	0 (0)	NA	31 (23.1)	NA

Data are presented as n, mean±sd or n (%). HC: healthy control; NTM: nontuberculous mycobacterial disease; BMI: body mass index; BE: bronchiectasis; ESR: erythrocyte sedimentation rate; NA: not available; AFB: acid-fast bacilli. ^#: ESR was measured in only one male patient; thus, the absolute value is presented.

Lipid profiles in geospatial lesions

A comprehensive lipidomic analysis was conducted on lung tissues (cavity wall, centre of cavity, granuloma, BE and normal lung) and serum (figure 1a). At the lipid-class level, lesions affected by NTM-PD showed increased levels of TG, DG and phospholipids, whereas FA levels were decreased when compared with unaffected tissue. These changes were particularly pronounced at the centre of the cavity, which indicates more severe disease (figure 1b). Consistent with these findings, TG and phospholipid levels were elevated, whereas FA levels were decreased in the serum of patients with NTM-PD compared with healthy controls (figure 1c and supplementary table S3).

FIGURE 1.

Systemic and localised lipid profiles of participants and their nontuberculous mycobacterial pulmonary disease (NTM-PD)-specific alterations. a) The lipidomic analysis scheme in this study. b) Heatmap comparing the average lipid-class levels among healthy control (HC), NTM-PD and bronchiectasis (BE) groups and c) among the lung lesions of patients with NTM-PD. d) An upset plot of paired t-test results between healthy and diseased tissues showing the number of common lipids with significant changes. The upset plot shows alterations in the lung tissue of patients with NTM-PD. TG: triacylglycerol; DG: diacylglyceride; PC-P: alkenylphosphatidylcholine; PC-O: alkylphosphatidylcholine; LPC: lysophosphatidylcholine; PC: phosphatidylcholine; FA: fatty acid; PE: phosphatidylethanolamine; PG: phosphatidylglycerol; PE-O: alkylphosphatidylethanolamine; PS: phosphatidylserine; Cer: ceramide; DhCer: dihydroceramide; HexCer: hexosylceramide; LPE: lysophosphatidylethanolamine; LPC-O: lysophosphatidylcholine ether; LPC-P: lysoalkenylphosphatidylcholine; PI: phosphatidylinositol; SM: sphingomyline.

At a more detailed species level, 37 lipids were consistently dysregulated across three or more lesions (figure 1d). Lipids with a fold change >1.5 and a p-value <0.05 are listed in table 2. Notably, lipids related to TG, lysophospholipids and sphingolipids (SM and HexCer) were generally upregulated at the centre of the cavity, whereas these changes were less pronounced or not significant in BE. Moreover, phospholipids (PC and PG) were typically downregulated at the centre of the cavity (figure 2).

TABLE 2.

Selected lipid species with common significant changes across the lung lesions

Lipid class	Lipids	Cavity wall		Granuloma		Centre of cavity		Bronchiectasis
		p-value	log2(FC)	p-value	log2(FC)	p-value	log2(FC)	p-value	log2(FC)
Triacyl glycerols	TG50:2_18:2	1.1×10−3	1.69	1.2×10−2	1.15	2.3×10−2	1.59	NA	NA
	TG50:4_20:4	3.2×10−3	2.61	1.7×10−2	2.79	1.3×10−2	3.33	NA	NA
	TG50:5_20:4	6.4×10−3	2.40	4.8×10−2	2.73	4.9×10−2	4.03	NA	NA
	TG50:6_20:4	3.0×10−3	2.92	NA	NA	4.4×10−2	4.91	3.0×10−2	2.25
	TG51:4_20:4	2.5×10−3	3.63	4.7×10−2	3.22	1.8×10−2	3.75	NA	NA
	TG52:3_20:3	2.1×10−4	2.83	4.7×10−2	2.48	3.1×10−2	2.69	NA	NA
	TG52:4_20:4	3.7×10−4	3.00	4.3×10−2	2.84	5.5×10−3	4.30	NA	NA
	TG52:4_22:4	1.5×10−2	2.32	NA	NA	2.9×10−2	3.26	8.6×10−3	1.60
	TG53:4_20:4	3.7×10−4	3.54	3.9×10−2	3.18	1.4×10−2	3.78	NA	NA
	TG54:4_22:4	2.2×10−4	3.44	3.9×10−2	2.93	8.6×10−3	3.80	NA	NA
	TG54:5_20:3	5.6×10−3	1.77	2.5×10−2	1.61	3.9×10−2	2.03	NA	NA
	TG54:5_22:4	1.5×10−3	2.45	4.0×10−2	2.41	1.5×10−2	2.82	1.9×10−2	1.15
	TG54:5_22:5	2.8×10−4	2.61	3.7×10−2	2.22	3.5×10−2	2.35	NA	NA
	TG54:6_20:4	5.4×10−4	2.51	4.0×10−2	2.31	1.7×10−2	3.69	NA	NA
	TG56:4_22:4	4.9×10−5	3.97	3.8×10−2	3.13	4.4×10−3	4.18	NA	NA
	TG56:5_18:2	2.7×10−4	1.92	3.8×10−2	1.63	2.4×10−2	2.21	NA	NA
	TG56:5_20:2	6.2×10−4	1.52	2.0×10−2	1.27	5.2×10−3	2.19	4.3×10−2	0.78
	TG56:6_18:2	8.7×10−4	2.25	4.7×10−2	1.92	7.4×10−3	2.78	NA	NA
	TG56:7_20:3	6.5×10−3	2.01	4.1×10−2	2.16	8.8×10−3	3.18	NA	NA
	TG58:10_18:2	7.4×10−4	2.31	3.5×10−2	1.72	5.9×10−3	2.78	NA	NA
	TG58:10_20:4	1.4×10−3	2.47	3.8×10−2	2.44	1.2×10−2	3.66	NA	NA
	TG58:6_22:4	1.8×10−3	2.54	4.5×10−2	2.29	5.0×10−3	3.09	NA	NA
	TG58:9_22:5	1.6×10−3	2.18	4.9×10−2	1.92	1.5×10−2	2.66	NA	NA
	TG60:10_22:6	1.6×10−3	2.66	4.8×10−2	2.54	3.4×10−2	2.71	NA	NA
Phospholipids	PC16:0_20:5	2.2×10−3	−0.61	NA	NA	5.8×10−3	−1.39	1.2×10−3	−0.69
	PG16:1_18:0	2.9×10−2	−1.17	4.0×10−3	−0.87	NA	NA	4.4×10−2	−2.11
	PG18:0_20:1	NA	NA	7.1×10−4	−0.86	1.1×10−2	−1.11	6.9×10−3	−0.80
	PG18:1_22:5	3.9×10−2	−1.42	3.3×10−2	−1.22	NA	NA	4.1×10−2	−2.06
	LPC20:2	1.1×10−2	1.33	9.6×10−3	1.65	7.2×10−4	3.00	NA	NA
	LPC-O24:0	1.8×10−3	1.62	3.1×10−2	1.62	9.8×10−4	3.42	NA	NA
	LPC-O24:2	8.3×10−3	2.39	4.3×10−2	2.62	7.8×10−3	5.51	NA	NA
Sphingolipids	SM32:0	6.1×10−2	1.15	3.2×10−2	1.13	3.3×10−2	1.67	NA	NA
	SM32:1	1.5×10−3	0.80	6.4×10−3	0.63	2.0×10−2	1.34	NA	NA
	SM44:3	3.2×10−2	0.70	NA	NA	1.5×10−2	1.40	1.3×10−2	0.74
	HexCer 16:1_18:0	6.7×10−3	0.69	1.5×10−2	0.94	4.4×10−2	0.28	NA	NA
	HexCer 18:1_14:0	4.4×10−4	1.70	9.6×10−3	1.58	1.8×10−2	3.31	3.6×10−2	0.77
	HexCer 18:1_24:1	3.3×10−2	0.85	2.0×10−2	0.94	3.8×10−2	1.80	NA	NA

FC: fold change; HexCer: hexosylceramide; LPC: lysophosphatidylcholine; LPC-O: lysophosphatidylcholine ether; NA: not available; PC: phosphatidylcholine; PG: phosphatidylglycerol; SM: sphingomyline; TG: triacylglycerol.

FIGURE 2.

Relative lipid species abundance across lung lesions in patients with nontuberculous mycobacterial pulmonary disease. The plots depict relative abundance levels of lipid species in normal lung tissue and four distinct lesion types: cavity wall, granuloma, centre of cavity and bronchiectasis. Data points represent mean±sd.

Lipid profiles by disease severity

Next, we analysed dysregulated lipid levels in lesions affected by NTM-PD and evaluated their variations by disease severity using serum. TG levels were decreased in patients with cavities and similarly reduced in those with AFB smear positivity (figure 3). Likewise, TG levels exhibited a decreasing trend from mild to severe disease severity, as classified by the BACES score (figure 4). In contrast, HexCer levels demonstrated the opposite pattern. HexCer levels were generally elevated in patients with cavities and AFB smear positivity, and they tended to rise with increasing BACES scores.

FIGURE 3.

Significantly altered serum lipid levels in patients with nontuberculous mycobacterial pulmonary disease (NTM-PD) stratified by the presence or absence of cavity and smear. The box plot illustrates lipid species with significant differences (p<0.05) in serum levels among patients with NTM-PD, grouped by a) the presence or absence of cavity and b) smear positivity.

FIGURE 4.

Serum lipid levels across nontuberculous mycobacterial pulmonary disease (NTM-PD) severity groups based on BACES (body mass index, age, cavity, erythrocyte sedimentation rate and sex) scores. The box plots represent alterations in serum lipid levels across NTM-PD severity groups. The NTM-PD severity groups were divided into three categories based on BACES scores: mild (0–1), moderate (2–3) and severe (4–5). A significant linear trend in lipid levels between groups (p<0.05) is presented in the plot.

Classification of NTM-PD using selected dysregulated lipids

Of the 37 significantly altered lipids identified in lung tissue, 21 corresponding serum lipid levels were used to assess the top five machine-learning models using PyCaret's automated model comparison framework for distinguishing patients with NTM-PD from healthy controls (supplementary table S4). Based on these selected models, receiver operating characteristic (ROC) curves obtained through 10-fold cross-validation are presented in figure 5. The CatBoost Classifier achieved the highest area under the curve (AUC) (0.8341), followed by Linear Discriminant Analysis (0.8243), Ridge Classifier (0.8243), Support Vector Machine (SVM), Linear Kerne (0.8106) and Ada Boost Classifier (0.7692). All evaluated models achieved AUC values >0.77, indicating that both ensemble and linear classifiers can effectively distinguish patients with NTM-PD from healthy controls.

FIGURE 5.

Receiver operating characteristic (ROC) curves of the top five classifiers trained on nontuberculous mycobacterial pulmonary disease (NTM-PD)-associated lipid species. ROC curves for the top five machine-learning models selected via PyCaret were evaluated using 10-fold cross-validation based on 21 selected lipid species to distinguish patients with NTM-PD from healthy controls. The mean true-positive rate was computed across 10 cross-validation folds and the average area under the curve (AUC) for each model is displayed in the legend. The diagonal dashed line represents the random classifier baseline (AUC=0.50). SVM: Support Vector Machine.

Discussion

In this study, a comprehensive lipidomic analysis was performed using lung tissues and serum from patients with NTM-PD. TG levels were elevated in lung tissues affected by NTM-PD, exhibiting notable variation based on the geospatial location of the lesions. The most significant increase in TG was observed in the centre of cavities, which are sites of severe inflammation, whereas no significant changes were detected in regions of BE. Interestingly, serum TG levels exhibited an opposite trend in disease severities, decreasing as disease severity increased, as indicated by cavities, smear positivity and higher BACES scores. These findings indicate that lipid dysregulation, including TG alterations, varies according to lesion characteristics and disease severity. Also, the machine-learning-based classifier model demonstrates outstanding discrimination ability between patients with NTM-PD and healthy controls. This highlights the potential role of lipid metabolism in the pathogenesis and progression of NTM-PD.

Mycobacteria, including NTM, modulate macrophage lipid metabolism during infection, inducing the formation of lipid-laden foam cells [18–20]. These foamy macrophages accumulate lipids, particularly TG, within granulomas, facilitating metabolic adaptations and enhancing mycobacterial survival [21]. As granulomas mature, lipid-rich caseum forms at their core, creating a more protective niche that facilitates mycobacterial survival [21, 22]. Within the lipid-rich caseum core, mycobacteria exploit host lipids as an energy source while evading immune responses [23]. This mechanism explains the significant elevation of TG levels in lesions, particularly at the centre of cavities, as these structures typically represent more advanced and severe diseases than BE [10, 24].

A notable finding of this study was the contrasting trends in TG levels between lung lesions and serum with increasing disease severity. TG levels were elevated in the cavity centre, whereas serum TG levels decreased with advanced disease indicators, such as AFB smear positivity, cavity formation and higher BACES scores. This discrepancy is likely due to the sequestration of host lipids within lesions. During mycobacterial infection, the expression of CD36, a gene involved in lipid uptake and efflux, including TG, is upregulated in macrophages and lung tissue [25–27]. Under these conditions, TG accumulation in lesions primarily results from incorporating FAs released from host TG [28]. Consequently, TG accumulates in tissue lesions of higher severity, whereas its systemic levels in the bloodstream decline. Interestingly 14 of the 21 lipid species (TG50:2_18:2, TG50:4_20:4, TG52:4_20:4, TG54:5_20:3, TG54:5_22:4, TG54:5_22:5, TG54:6_20:4, TG56:5_18:2, TG56:5_20:2, TG56:6_18:2, TG56:7_20:3, TG58:10_18:2, TG58:10_20:4, and TG58:6_22:4) used in the machine-learning-based classification model were TGs. This underscores the importance of TG dysregulation in the pathophysiology of NTM-PD, with serum TG levels contributing to the classifier's performance. Consequently, TGs may serve as potential biomarkers, reflecting both localised lipid sequestration in advanced lesions and systemic shifts in lipid metabolism.

The accumulation of TG has significant clinical implications for NTM-PD. TG accumulation, induced by NTM, facilitates their proliferation [21]. Upregulated CD36-mediated lipid uptake has been shown to increase the bacterial burden of M. avium in mice [25]. Furthermore, TG accumulation enables NTM to enter a dormant state [28], characterised by slow growth and reduced metabolism, allowing the bacteria to evade antibiotic treatment. This TG-driven dormancy complicates the eradication of NTM [21]. Additionally, TG accumulation exacerbates treatment challenges by promoting the development of drug resistance in NTM [29]. These findings highlight the pivotal role of TG metabolism in the pathogenesis and persistence of NTM-PD.

Another notable finding in this study is the increased presence of ceramide-related metabolites [30], specifically HexCer, in lung tissue affected by NTM-PD. Unlike TG, HexCer showed an increasing trend in serum in relation to disease severity. Ceramide is crucial in maintaining cell wall integrity and regulating cytokines and immune responses [31–33]. However, whether ceramide plays a protective role in infections remains unclear. A recent study reported increased ceramide levels in sputum following NTM infection [34]. Therefore, HexCer could serve as a biomarker for both the presence and severity of NTM-PD.

Nevertheless, caution is needed when interpreting our results, given the potential influence of antibiotics on the lipid profile. Lung tissues were obtained from patients who had received at least nine months of antibiotic therapy, whereas serum samples were collected at the time of initial diagnosis. This temporal mismatch may have introduced lipidomic changes related to antibiotic exposure. Antibiotics can alter the lipid profile of host cells and also affect bacterial lipid composition [35, 36]. In mycobacteria, changes to lipids happen as the bacteria adjust to being inactive due to antibiotic treatment [36]. Although it is practically impossible to obtain lung tissue from antibiotic-naive patients, this point should be considered when interpreting our findings.

In addition to these considerations, this study has certain limitations. First, as the patients providing serum and tissue samples belonged to separate cohorts, direct comparison of the lipid profile between serum and tissue was limited. Nevertheless, in the tissue analysis, we measured lipid concentrations in multiple lesion areas within each patient's lung tissue. Although the number of patients was limited, similar patterns were observed across these areas. In the serum analysis, we included samples from over 130 patients, which supports the generalisability of our findings. Second, we were unable to adjust for the potential effects of systemic nutritional status or inflammatory responses. However, although these factors can influence the lipid profile, they may also represent downstream consequences of NTM-PD [10].

Despite these limitations, our findings are anticipated to contribute to developing novel therapies for NTM-PD. Current antibiotic treatments demonstrate limited efficacy [37, 38], highlighting the need for alternative strategies. Host-directed therapies informed by lipidomic findings could offer promising new treatment options. For instance, novel agents targeting cholesterol metabolism are under investigation as potential treatments for tuberculosis [39, 40]. Similarly, we anticipate that future therapies targeting NTM's TG metabolism could provide a novel approach to treating this challenging disease.

Conclusion

In conclusion, lipid metabolites, particularly TG, in NTM-PD show distinct changes in tissue and serum, influenced by the spatial distribution of lesions and disease severity. These findings offer valuable insights into the pathophysiology of NTM-PD and may guide the development of novel therapeutic approaches.

Take-home points

• Study question: how do lipidomic profiles vary across different geospatial lesions in nontuberculous mycobacterial pulmonary disease (NTM-PD), and what is their relationship to disease severity?

• Results: at the lipid-class level, lesions affected by NTM-PD showed increased levels of triacylglycerol (TG), whereas fatty acid (FA) levels were decreased when compared with unaffected tissue. Also, the machine-learning-based classification model built by the serum levels of NTM-PD-associated lipid species effectively distinguishes patients with NTM-PD from healthy controls.

• Interpretation: these findings indicate that lipid dysregulation, including TG alterations, varies according to lesion characteristics and disease severity.

Data availability

The datasets generated during and/or analysed during the current study are available in the Metabolomics Workbench repository (https://dx.doi.org/10.21228/M8B848) and have also been deposited in the Korea BioData Station (K-BDS) under accession number KAP241471.

Supplementary material

Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.

Supplementary material

00275-2025.SUPPLEMENT