Comparative metabolomic analysis of human lung slices (hu-PCLS) exposed to either standard or FLASH protons: a pilot study

Abstract

Background

Recent advances in radiation biology and preclinical research have identified that high doses of radiation at ultra-high dose rate can lead to sparing of normal tissue, while maintaining tumor control. This has been termed the FLASH effect and has been extended from electrons to protons, heavy ions and photons. Lung cancer treatments, despite the advancements in radiotherapy with precise protons, are still associated with significant damage to the normal tissue. FLASH proton exposures have not been characterized yet with regards to lung tissue injury or sparing.

Methods

In this pilot study we used precision cut lung slices from human healthy non-smoker donors, two male and one female, and exposed this ex vivo model to a single dose of 12 Gy of standard or FLASH protons and analyzed them with untargeted metabolomics at 24 h after exposure.

Results

We identified 22 metabolites of interest, with primary enrichment in β-alanine metabolism, sphingolipid metabolism, primary bile acid biosynthesis, and one carbon pool by folate, while chemical classes showed that sphingoid bases and eicosanoids were two of the most enriched chemical classes after radiation exposure. Classification analysis with receiver operating characteristic curves based on the 22 metabolites indicated the presence of individual variability to responses, that warrant future studies with a larger cohort.

Conclusions

Although this is a pilot study, we show the utility of ex vivo models in radiation research and particularly with standard and FLASH protons, and that further research in this area should include both male and female subjects. This ex vivo human model should therefore be investigated further to identify early responses to proton exposures.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13014-025-02714-8.

Background

Novel radiation treatments for lung cancer using proton beams decrease the dose to the organs outside the target volume while maintaining the dose to the tumor target. Proton radiation therapy (PRT) can improve spatial dose delivery compared with photons or electrons. Despite the sophisticated beam delivery systems, standard proton radiotherapy for the treatment of lung cancer is still associated with damage to normal tissue parenchyma that manifests as radiation pneumonitis with clinical symptoms such as dyspnea and cough. The clinical incidence with newer treatment regimens has been greatly reduced [32, 38] and the lung sparing capacity of protons could be better suited for patients with initially decreased lung function [34].

Recent studies suggest that FLASH electron RT (60–100 Gy/s) induces less normal tissue damage while maintaining tumor response compared with standard electron RT (conventional dose rate) [11, 23, 31]. Similar data are emerging with FLASH proton therapy in preclinical models, demonstrating sparing of normal tissue (salivary glands, gastrointestinal tissues, skin, and bone) [5, 40, 41]. In a nonrandomized clinical trial with 10 patients, it was shown for the first time that clinical proton FLASH irradiation was feasible [29].

Preclinical models provide significant mechanistic underpinnings of such exposures with the ability to study long term responses, both in survival and in normal and tumor tissue effects. However, generating models utilizing human tissues will be invaluable in understanding and comparing animal models to humans, and if possible reducing the use of animals in research studies. In this pilot study we employed human precision cut lung slices (hu-PCLS) from non-smoking donors. Hu-PCLS are representative of the in vivo tissue that can be kept viable in culture, as confirmed by the beating cilia of airway epithelial cells lining bronchioles [39]. This useful tool has been used in studies on asthma [17, 18, 26], in idiopathic pulmonary fibrosis [2], viral infections such as pneumonic plague [4], ventilator associated lung injury [9], inhalation of particulates [22, 30], screening for chemical toxicity [42] and even to cigarette smoke extract [43].

Previous work on this model has shown changes in gene expression with protons, demonstrating its utility in these studies [39]. We now extend our studies to determine metabolic effects between sham, standard proton radiation treated (SPRT) and FLASH proton radiation treated (FPRT) hu-PCLS from three different donors, two male and one female origin. At 24 h after exposure, global assessment of metabolic differences was conducted with untargeted metabolomics. While this is a pilot study, results revealed promising metabolic pathway enrichment when comparing sham to either type of exposure or comparing SPRT to FRPT, indicating differences in responses compared to sham but also between the two radiation modalities. Metabolic perturbations have been reported in studies with electron FLASH, with reduced lipid peroxidation as a mechanism of interest to spare normal tissue [13] and distinct lipid chain lengths [28]. This is the first study to our knowledge indicating metabolic changes in human living lung tissue with FPRT and results warrant further research in this area.

Materials and methods

Chemicals

All chemicals and solvents were of the highest purity for liquid chromatography mass spectrometry (LC-MS) analysis. 4-Nitrobenzoic acid, chlorpropamide, debrisoquine sulfate, spermine, folic acid, palmitoylcarnitine, and phytosphingosine were obtained from Sigma Aldrich (St. Louis, MO). Lysosphingomyelin, prostaglandin E3, ursodeoxycholic acid, malonyl coenzyme A, D-erythro-sphinganine, and thromboxane B2 were obtained from Cayman Chemicals (Ann Arbor, MI). Decanoylcarnitine was obtained from Tocris Bioscience (Minneapolis, MN).

Donor lung samples

The normal human samples used in this study were from de-identified non-used lungs donated for organ transplantation following an established protocol (PROPEL, approved by University of Pennsylvania Institutional Review Board) with informed consent in accordance with institutional and NIH procedures. Consent was provided by next of kin or healthcare proxy. The institutional review board of the University of Pennsylvania approved this study, and all patient information was removed before use. While this tissue collection protocol does not meet the current NIH definition of human subject research, all institutional procedures required for human subject research at the University of Pennsylvania were followed throughout the reported experiments.

Human precision cut lung slices (hu-PCLS) were prepared according to previously published reports [6, 21, 39]. In short, whole human lungs from non-asthma otherwise healthy donors (males and females) were obtained through the National Disease Research Interchange (Philadelphia, PA, USA). Samples from two male [designated as D1 (70-years-old) and D2 (42-years-old)] and one female (designated as D3, 74-years-old) donors were used in this study. All donors were Caucasian and non-smokers. The lungs were dissected and inflated with 2% (w·v^− 1) low melting point agarose. Cores of 8 mm diameter were made from sectioned lungs and further sliced into 350 μm sections. Slices were randomized as to avoid lung location bias and frozen in -80 °C.

Irradiations

Lung tissue samples were maintained in DMEM F12 medium in petri dishes, at 37 °C in a humidified air: CO₂ (ratio of 95:5) incubator. Prior to irradiation, the sections were carefully transferred to a 50 ml tube with fresh culture medium. The sections were irradiated in the tube to a dose of 12 Gy using a 230 MeV (range ~ 32.0 g/cm²) proton beam generated by an IBA Proteus Plus Cyclotron (Louvain-La-Neuve, Belgium) to a fixed research beam line [10]. Lung tissue was placed in the center of a 26 mm diameter circular proton field using the entrance region of the high energy beam. Field uniformity was verified prior to each experiment with radiographic film (EBT3, Ashland Advanced Materials, Bridgewater, NJ, USA). Dose was measured with a calibrated NIST-traceable Advanced Markus Chamber (PTW, Freiburg, Germany). A thin window Bragg peak chamber (Type 34070, PTW, Freiburg, Germany) was cross-calibrated daily to monitor the delivered charge to each sample. FLASH and standard irradiated samples were exposed to average dose rates of 95.1 ± 23 Gy/s and 0.58 ± 0.16 Gy/s, respectively.

Sample Preparation and metabolomic profiling

Sample preparation (n = 5 per group were used for metabolomics) and extraction of small molecules were conducted in the same manner as previously published reports [8, 25]. Briefly, each tissue samples was homogenized in 50%:50% methanol: water with internal standards (final concentrations of 4 µM debrisoquine, 30 µM 4-nitrobenzoic acid, 5 µM chlorpropamide). Following centrifugation, the supernatant was transferred to a clear tube and the pellet was re-extracted with 50%:50% acetonitrile: water. Following incubation on ice and centrifugation, the supernatant was combined with the previously collected fraction and vacuum dried. The sample was resuspended in 50%:50% methanol: water and filtered (0.2 μm) to remove impurities.

Samples were analyzed with an Ultra Performance Liquid Chromatography (UPLC) coupled to the Xevo G2S time-of-flight mass spectrometer (Waters Corp., Milford, MA). LC and MS conditions of the instrument, including gradient used for chromatography, are shown in Supplementary Table 1. Quality controls (QCs), consisting of the pooled samples within each donor, were also run to monitor for chromatographic integrity and mass accuracy.

Moleculyzer pipeline

Moleculyzer, an in-house LC-MS metabolomics data processing pipeline that conducts chromatographic deconvolution, alignment, and deleterious feature reduction, was utilized to generate an MS1 peak table and putative chemical identifications from the raw chromatographic data. MolecuLyzer evolved from the Disparate Metabolomics Data Reassembler algorithm (DIMEDR) [27]. Moleculyzer attempts to identify and minimize the potentially deleterious impact of in-source fragments (ISF), which results in numerous extraneous and confounding features in the final peak table. The NIST20 Tandem Mass Spectral Library is used by the program for inferencing potential in-source fragments of over 28,000 compounds. Candidate parent ions are initially identified in the experimental chromatographic data via exact mass matching to the NIST library, and ions with similar retention times to the parent are compared to the associated library fragments (Fig. 1). This process also yields putative ISF based identifications of the parent ions, and can also be applied to MSE data for additional IDs.

Fig. 1.

Moleculyzer identifies in-source fragments in the MS1 chromatographic data, removing them as independent features and incorporating them into their parent ion. In doing so, the potentially deleterious impact of in-source fragmentation is mitigated, and the number of “unknown features” is significantly reduced in the final data matrix. These fragments also enable the assignment of putative identities for parent ions

Biomarker identification, validation, and data analysis

Data extracted from MolecuLyzer were manually mined and reduced based on the following criteria: concordance of putative identifications between in source fragmentation identifications and MSE identifications, same directionality of changes in 2/3 of the donors, and putative identifications should have biological relevance. Chemical classes were used for further reduction of the data.

Each donor was analyzed separately with MetaboAnalyst 5.0 (https://www.metaboanalyst.ca/MetaboAnalyst/home.xhtml). Missing values > 70% were removed from the analysis and remaining missing values were replaced by LoDs (1/5 of the minimum positive value of each variable). Data were normalized by the sum and presented in the form of principal component analysis (PCA) scores plots, PLS-DA scores plots, and heatmaps with Euclidean distance and Ward clustering, averaged within each group. Pathway enrichment and chemical structures analysis were also conducted through MetaboAnalyst 5.0 using KEGG and a minimum of 4 entries per metabolite set.

Validation of select metabolites from each chemical class was conducted with tandem mass spectrometry by fragmenting QC samples with the m/z of interest and comparing the fragmentation patterns to those of pure standards and further through the NIST MS database v2.3. MSE fragmentation patterns and database matching were also visualized through the software Progenesis QI (NonLinear Dynamics, Newcastle, UK) with matches through the databases LipidBlast [20] or empirical METLIN [15], with a ppm error set to 10. Levels 1 and 2 were used for the identification of metabolites, with positively identified metabolites denoted with a # in Table 1 [12, 36].

Table 1.

Selected putative and positively identified metabolites with ISF and MSE concordance or with MS/MS

	Putative metabolites	m/z experimental				m/z calculated		ppm error	Ret. Time			Adduct
1	N-Octanoylsphingosine-1-phosphate	488.3641				488.3505		27	8.26			[M + H-H2O]+
2	N-Myristoylsphinganine	512.5034				512.5042		1.6	8.64			[M + H]+
3	N, N-Dimethylsphingosine	310.3109				310.311		0.32	8.87			[M + H-H2O]+
4	15-Ketoprostaglandin E1	335.2304				335.2222		24	5.53			[M + H-H2O]+
5 #	D-erythro-Sphinganine	302.3055				302.3058		0.99	6.91			[M + H]+
6 #	Spermine	203.2232				203.2235		1.48	0.27			[M + H]+
7	Cholesterol	369.3518				369.3521		0.81	9.31			[M + H-H2O]+
8	Glycodeoxycholic acid	432.3189				432.3113		17.58	7.25			[M + H-H2O]+
9 #	Folic acid	442.1476				442.1474		0.45	1.55			[M + H]+
10	Prostaglandin F2.alpha. 1,15-lactone	337.2353				337.2378		7.41	6.95			[M + H]+
11	L-Glutarylcarnitine	276.1445				276.1446		0.36	0.41			[M + H]+
12	Lyso-sphingomyelin	465.3347				465.3457		23.6	5.33			[M + H]+
13 #	Decanoylcarnitine	338.2386				338.2307		23.35	6.94			[M + Na]+
14	Palmitoylcarnitine	422.3209				422.3246		8.76	6.62			[M + Na]+
15	Phytosphingosine	300.2899				300.2902		0.99	8.17			[M + H-H2O]+
16	Prostaglandin E3	349.2049				349.2015		9.73	7.8			[M-H]-
17	7.alpha.-Hydroxy-3-oxo-4-cholestenoic acid	429.3004				429.3005		0.23	7.27			[M-H]-
18	Ursodeoxycholic acid	391.2846				391.2848		0.51	7.01			[M-H]-
19	1,11-Undecanedicarboxylic acid	243.1595				243.1596		0.41	6.33			[M-H]-
20	L-Glutamic acid	146.0453				146.0453		0	0.32			[M-H]-
21	Thromboxane B2	369.2275				369.2277		0.54	5.66			[M-H]-
22	Malonyl coenzyme A	852.1217				852.1078		16	4.06			[M-H]-

# signifies validated through MS/MS

Graphing of processed data from MetaboAnalyst 5.0 was conducted through GraphPad Prism v9 and data for select MS/MS validated biomarkers were presented as mean with 95% confidence intervals. Receiver operating characteristic (ROC) curves were constructed with MetaboAnalyst for multi-biomarker panels with Random Forests as classification and ranking methods. The area under the curve (AUC) indicates the classification accuracy, with ~ 0.5 indicating no discernible change in metabolic effects and ~ 1 indicating a good model with substantial metabolic changes.

Spermine Immunofluorescence

Formalin-fixed paraffin-embedded HuPCLS (10 μm-thick) were subjected to sequential steps of a standard protocol of deparaffinization and rehydration, including incubations in xylene, 100% Ethanol, 95% Ethanol, 70% Ethanol, 50% Ethanol and deionized water. After rehydration, the slides were incubated for 10 min in a boiling sodium citrate buffer (10 mM, pH 6.0). After washing the slides with deionized water, the slides were blocked in 8% BSA (in PBS) for 1 h, at room temperature. The sections were stained overnight with a rabbit polyclonal anti-spermine antibody (Ab26975, Abcam, 1:50 dilution in 1% BSA/PBS), at 4° C. Next day, the sections were washed three times for 5 min in PBS-TT (0.05% Tween-20, 0.01% Triton in PBS), and were incubated in Alexa Fluor 488 goat anti-rabbit IgG (Invitrogen, Ref#A11008), for 2 h, at room temperature. The sections were washed three times for 5 min in PBS-TT and the nuclei were counterstained with Hoechst (1:1000 in PBS) for 30 min, at room temperature. The sections were finally washed five times for 5 min in PBS, mounted and cover-slipped and stored at 4 °C, until analysis. N = 5 for sham or no radiation, n = 14 for SPRT, and n = 14 for FPRT.

Image analysis

Fluorescent signal from antibody staining was visualized by a Zeiss Observer.Z1 inverted microscope and Zen 3.0 software (Zeiss) using the same exposure settings for all conditions. Quantification of the positively-stained areas of spermine (Sham, number of fields: 5; SPRT and FPRT, number of fields: 14) was performed with Fiji [Fiji, RRID: SCR_002285]. Briefly, background for the channel of interest was subtracted based on negative control, and then each channel was automatically thresholded. Area of all thresholded regions were measured. Quantification is presented as spermine positive area normalized to total nuclei area.

Results

The overall graphical representation of the hu-PCLS processing and downstream analysis is shown in Fig. 2. Briefly, hu-PCLS were prepared from donated lungs, maintained in DMEM F12 medium, irradiated in 50 mL conical tubes, and at 24 h after exposure flash frozen and subsequently subjected to untargeted metabolomics, with candidate metabolites further validated through MS/MS.

Fig. 2.

Graphical representation of the experimental approach. Slices from healthy donor lungs were irradiated with Standard protons or FLASH protons using a single dose of 12 Gy and metabolomic analysis conducted at 24 h after exposure. Each donor served as their own control

hu-PCLS viability through TUNEL assay showed no statistically significant differences between exposed and unexposed at 24 h after exposure

Viability of hu-PCLS has been extended successfully in culture for several days after embedding [3]. Increased cell death after exposure was evaluated in slices from Donor 1 with the TUNEL assay, and found higher than no irradiation, although the variability was high and there was no significant difference between all three groups (data not shown).

Moleculyzer used to reduce and eliminate in-source fragments leading to data reduction

A graph summarizing the results from the chromatographic deconvolution is shown in Fig. 3. Raw spectral data from all three donors underwent chromatographic deconvolution and alignment using Moleculyzer, an in-house liquid chromatography mass spectrometry (LC-MS) data processing pipeline that leverages the NIST Mass Spectral Library to produce high quality annotated MS1 spectral feature peak tables for downstream analysis. Moleculyzer significantly decreases the needless complexity endemic to metabolomics datasets by identifying and eliminating ISFs. This ISF data were also leveraged to provide putative compound identifications, as well as MSE based identifications.

Fig. 3.

A graph summarizing the results from the chromatographic deconvolution, alignment, and deleterious feature reduction of the spectral data generated from all three donors. In-source fragmentation is a major source of extraneous spectral features that result in unnecessarily complex metabolomics datasets. Using the NIST Tandem Mass Spectral Library, the impact of many of these deleterious features can be reduced (blue dots), and in many cases entire features can be removed (green hued dots)

Untargeted metabolomics: the ex vivo hu-PLCS model shows biologically relevant responses to SPRT and FPRT

Deconvoluted data for ESI + and ESI- were manually curated with the following criteria: MSE identification and in source fragmentation identification concordance, biologically relevant, and same directionality of changes in 2/3 of the donors in exposed groups compared to Sham. Twenty-two potential metabolites were identified in ESI + and seven in ESI- (Table 1 and Supplementary Table 2 presenting normalized data, with average values and fold changes). Further positive identification of select putative metabolites was performed with tandem mass spectrometry to a metabolomics standard initiative level 1 [36]. Enrichment analysis through MetaboAnalyst 5.0 for metabolic pathways and metabolic classes, taking into account the totality of metabolites from Table 1, as shown in Fig. 4 revealed significant enrichment in β-alanine, sphingolipid, primary bile acid metabolism, one carbon pool by folate and pyruvate metabolism as the top 5 most enriched metabolic pathways (Fig. 4A). The top 5 classes included sphingoid bases, eicosanoids, bile acids, fatty esters, and amines (Fig. 4B).

Fig. 4.

Metabolic enrichment based on the metabolites from Table 1. Panel A: Metabolic pathway enrichment. Panel B: Chemical classes of metabolites. In both panels, the darker color signifies higher enrichment of the pathway or class

Multivariate data analysis shows responses to radiation in all three donors with inter- and intra-variability

Multivariate data analysis was performed for each donor separately, as shown in Fig. 5, based on the panel of 22 metabolites. Principal component analysis (PCA) of the first two components showed variable separation across component 1 in the three donors (55.9 to 71.3%), with less separation along component 2 (13.6 to 19.1%). Interestingly, the partial least squares discriminant analysis (PLS-DA) models showed a more robust validated model for Donors 1 and 2, both male, than Donor 3, a female (Supplementary Fig. 1), with permutation p-value of p < 0.05 for Donors 1 and 2. Nonetheless, all three donors showed responses to radiation, on the metabolic level at 24 h after exposure. All samples from each donor and each condition were also presented together in a PCA and PLS-DA model, highlighting the inter- and intra-variability (Supplementary Fig. 2). Heatmaps of the averaged values within each group for the individual metabolites clearly showed the existence of differential responses in select metabolites between Donors 1 and 2 with Donor 3, particularly in inflammation intermediates such as prostaglandins E3 and A1 and thromboxane B2.

Fig. 5.

PCA scores plots for each donor and heatmaps of averaged values within each group, based on the 22 metabolites from Table 1. Responses in Donors 1 and 2 are more similar compared to Donor 3, indicating the potential of sex differences in initial responses to the different radiation modalities

Select biomarkers that were further validated through tandem MS with pure chemicals were graphed using normalized abundance levels to highlight the intra- and inter-variability in responses. As shown in Fig. 6, inter-donor differences were evident in select metabolite responses, particularly between male and female responses, such as in folic acid, thromboxane B2, and malonyl coenzyme A. However, spermine was decreased in all three donors with both radiation modalities compared to their baseline levels. For validation of this finding, immunohistochemistry was performed on hu-PCLs from Donor 2 (Supplementary Fig. 3) to validate the patterns identified by MS. Similarly to MS, spermine levels were decreased with both radiation types, although the effect was less pronounced with FPRT compared to SPRT.

Fig. 6.

Select metabolites from different chemical classes that were further validated through MS/MS. N = 5 per group. Data are presented as mean with 95% confidence intervals for each of the donor samples

ROC curves indicate inter-individual variability in response to FPRT

In order to investigate the degree of metabolic changes compared to baseline and between radiation types, receiver operating characteristic (ROC) curves were constructed with Random Forests to show the sensitivity and specificity levels of the metabolic profiles in classifying the degree of changes between two groups. Figure 7 shows the ROC curves for 6 different models for each ROC curve, with a combination of 2 (red), 3 (green), 5 (blue), 10 (light blue), 20 (purple) or the full 22 (yellow) list of metabolites used to construct each ROC curve. Values for confidence intervals and AUC in bold letters shown for the full 22 list panel, with most AUC’s approaching 1. However, Donor 3 showed essentially no radiation response after FPRT compared to untreated levels, indicative of the potential of heterogeneous responses that could be specific to FPRT.

Fig. 7.

ROC curves constructed with a combination of 2 (red), 3 (green), 5 (blue), 10 (light blue), 20 (purple) or the full 22 (yellow) list of metabolites from Table 1, with Random Forests as the classification and ranking methods. An AUC of ~ 1 is considered a good to excellent classification model, while an AUC ~ 0.5 is considered a poor to none classification model. Comparisons were conducted for each Donor for Sham vs. SPRT, Sham vs. FPRT, and SPRT vs. FPRT

Discussion

Sparing of normal tissue during radiotherapy and decrease of radiotoxicity that can lead to improved quality of life have been at the forefront of developing new clinical trials and testing new radiation protocols, in addition to effectively treating solid tumors. Preclinical research has identified the FLASH effect, where high doses of radiation are delivered at an ultra-high dose rate, generally at multiple Gy/s in the order of > 40 Gy/s, where normal tissue is spared from radiation responses that are generally observed with standard dose rates. Extensive research in the characterization of electron FLASH has been underway in the past few years, with oxygen depletion and immune system effects presented as some of the unique aspects affecting responses. Comparisons between electron and proton FLASH (FPRT) beams have shown preservation of neurocognitive abilities in comparison to standard exposures [1], although some differences in generation of reactive oxygen species (ROS) have been measured between the two [37]. More work in the area of ROS is warranted, as computational models have already determined some hypotheses to be tested that include oxygen availability and capillary tension [31].

Intense work with FPRT has shown preservation of normal tissue in mice, such as intestinal crypts, skin, bone, salivary glands, and cardiac function [5, 19, 41], leading to the first FLASH protons nonrandomized clinical trial in the United States (FAST-01) [29]. However, clinical trials are not always feasible and animal models to test radiation toxicity of normal tissues have certain limitations, and ethical concerns do not permit to biopsy human tissues to evaluate toxicity at the cellular and molecular level. Cell cultures are isolated models that are inadequate in representing a “live” organ while rodent respiratory systems differ from human. In addition, reduction of animal use in research and development of alternate models is now considered a necessity for future research innovations. Therefore, a transition from animal and in vitro and animal testing methods to a reproducible screening platform that recreates the human lung is necessitated.

For this, hu-PCLs were used from healthy donors that have been shown previously to have gene expression changes involved in radiation induced inflammatory responses, senescence, and oxidative damage [39]. We used this model to assess early metabolic differences and compare FPRT to SPRT, as a pilot study into insights that could lead to development of lung injury, pneumonitis, and fibrosis. Metabolomic responses showed that Donors 1 and 2, both male, had similar changes in directionality of key metabolites involved in pathways such as eicosanoids (e.g. prostaglandin A1 and E3), while metabolites such as acylcarnitines, certain lipids (e.g. cholesterol, octanoylsphingosine), and spermine had similar responses between all three donors (Figs. 5 and 6). Interestingly, spermine levels were decreased in all three donors with FPRT but remained higher than SPRT, and were further validated with immunohistochemistry in Donor 2. Spermine is a downstream product of putrescine and spermidine and an endogenous polyamine. Importantly, spermine has been associated with protection from ROS [33]. Therefore, this metabolite and others in the pathway such as spermidine, which were not investigated in this study, could be a potential mechanism for protection of normal tissue that should be further evaluated.

Another pathway of interest based on the enrichment pathway analysis was primary bile acid biosynthesis. Although bile acids are primarily considered associated with gut dysbiosis, reports are now showing changes in bile acids in lung in relation to radiation exposure but also connecting them to normal pulmonary function [14, 16]. As an understudied area in the pulmonary function and injury, particularly in connection to microbial dysbiosis, it remains to be seen whether such metabolites can lead to perpetuation of radiation induced injury or provide a potential protective mechanism. Finally, two potential metabolites of interest include thromboxane B2 and folic acid. While thromboxane B2 can be a marker of platelet activation and increased tissue damage [35], folic acid is a vitamin and a cofactor that can protect against radiation induced DNA breaks [7]. The dynamics between age, sex and time post exposure require further elucidation and we are only now beginning to dissect the biological responses to FLASH radiotherapy, and particularly FPRT.

Differences in lipid species and particularly in intermediates associated with pro-inflammatory responses (such as prostaglandins and thromboxanes) are also of special interest. These lipids and lipid products have been identified previously as circulating biomarkers of radiation exposure in total body irradiated patients [24] and in irradiated lung tissue as early as 1982 [35] and can act as signaling molecules, in addition to indicating an imbalance in a pro- and anti-inflammatory response. Although the pathways have not been fully elucidated in this study, future directions should employ a targeted approach with MS in a larger cohort to identify whether there are clear differences between SPRT and FPRT on normal lung tissue. As stated earlier, electron FLASH has shown an inverse effect on lipid peroxidation [13] and therefore more research is needed, as it is unclear at the moment whether there are tissue specific effects, i.e. skin vs. lung. A certain limitation of this study remains the fact that this is an in vitro system, future studies however will be conducted under reduced oxygen concentrations to better reproduce an in vivo environment.

Evident through the data, however, is the inherent variability in humans, particularly in baseline levels, and therefore the results of this study should be validated in a larger cohort. This was particularly evident when comparing the three donors, as two out of three showed similar responses both to SPRT and FPRT, whereas Donor 3 differed substantially. As there are specific differential responses of one donor demonstrating differential responses from the other two, future studies should include higher numbers of donors with variable age and sex as these may be important factors in determining one’s response and potential radioprotection of normal tissue with FPRT. ROC curves however shown in Fig. 6 clearly demonstrated that model performance for Donor 3 (female) showed no discernible difference for Sham vs. FPRT, showing that the FPRT responses are in line with Sham exposure, unlike SPRT. This was not evident in the two other donors (male), further increasing the necessity to conduct more extensive studies. In addition, a longitudinal study with consecutive time points beyond 24 h could provide clarity on whether sex is a key variable in metabolic responses with this radiation modality, whether delayed responses in metabolism may be patient dependent, or if FPRT is a better treatment modality for certain individuals with better normal tissue protection. Nonetheless, research into FPRT responses could significantly enhance cancer treatments and provide patients with significantly less adverse responses due to radiotoxicity.

While we have shown for the first time metabolic differences in between SPRT and FPRT acute early responses in living hu-PLCs, our study and model are not without limitations. As an ex vivo model it lacks the immune infiltration post exposure that can modify responses, however a co-culture system could also be developed as a next step. As a pilot study, the number of donors is low, however it is demonstrating metabolic changes in human tissue worth investigating more in depth and integrating them with other -omics studies. Finally, long term responses have not been evaluated, where FPRT could more clearly show the protective effect on normal tissue over SPRT, as it has already been shown in various animal models. Nonetheless, there is much to be investigated with FPRT as the mechanism has not yet been clearly defined.

Conclusions

This is the first pilot study utilizing an ex vivo living human tissue model exposed to SPRT or FPRT from three healthy human donors. Results demonstrated the presence of metabolic differences distinct from Sham in exposed lung slices, with inter-individual variability evident in early responses. Longer term responses are needed to fully assess whether FPRT has a radioprotective effect on normal lung tissue compared to SPRT. Nonetheless, hu-PLCs provides an alternate ex vivo model utilizing human living tissue for direct translation to humans.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization, AV, TDK, CK, MC-S. and ECL.; methodology, AV, TDM, AD, MK, CK, MCS, ECL.; software, TDM.; validation, AV, TDM, and ECL.; formal analysis, AV, TDM, and ECL.; investigation, AV, TDK, AD, MCS, and ECL.; resources, AV, MCS, and ECL.; data curation, TDM and ECL.; writing—original draft preparation, AV, TDK, MK, MCS, and ECL.; writing—review and editing, AV, TDM, AD, MK, CK, MCS, and ECL.; visualization, AV, TDM, and ECL.; supervision, MCS, CK, and ECL.; funding acquisition, CK and MCS. All authors have read and agreed to the published version of the manuscript.

Funding

Studies were partially supported by the National Cancer Institute (NCI) award no. 1P01CA257904-01 (P.I. Constantinos Koumenis). The project described above was also supported by award no. P30 CA051008 (P.I. Louis Weiner) from the NCI. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NCI or the NIH.

Data availability

Due to privacy and ethical considerations, data will be available upon request.

Declarations

Ethics approval and consent to participate

Not applicable.

Institutional review board statement

The Institutional Review Board of the University of Pennsylvania determined that the study does not meet the criteria for human subject research and therefore ongoing IRB oversight is not required. This study does not meet the definition of “human subject”: According to 45 CRF § 46.102(f), “Human subject” means a living individual about whom an investigator (whether professional or student) conducting research obtains: (1) data through intervention or interaction with the individual, or (2) identifiable private information.

Competing interests

The authors declare no competing interests.