Aqueous humor metabolomic profiles in association with diabetic mellitus

Abstract

Diabetic mellitus (DM), commonly referred to diabetes, is a worldwide metabolic disorder, which usually causes high morbidity and mortality rates. Especially, DM may result in serious macrovascular problems including cataract. To investigate the underlying molecular mechanism, here we for the first time employed gas chromatography-time-of-flight mass spectrometry (GC-TOF MS) for an untargeted metabolomics study. Totally 263 metabolites were determined in aqueous humor (AH) samples from 30 patients: 15 for the controls and 15 with DM. Both the heat map and principal component analysis (PCA) plot showed a significantly distinct metabolomics profiles between patients with DM and the controls. Moreover, 20 metabolites were determined to be significantly altered (P ≤ 0.05) in DM patients, some of which were associated with oxidative stress. Metabolic pathway analysis of these significantly different metabolites identified ten most relevant pathways in the group of DM patients when compared with the control group. Among them, three pathways including fatty acid biosynthesis, fatty acid metabolism, and linoleic acid metabolism were the three most significantly influenced pathways (P ≤ 0.05), which probably play key roles in the formation of DM and its complication, cataracts. Altogether, this work not only indicated a distinct AH metabolomic profile in association with DM, but presented novel insights into the molecular mechanisms of DM formation, as well as formation of cataracts.

Introduction

Diabetic mellitus (DM), commonly referred to diabetes, is a chronic disease that affects millions of people worldwide [1]. The latest 2016 data from the WHO reported that an estimated 422 million adults are living with DM. Notably, DM usually causes complications such as kidney function loss, eye problems, heart problems, or other serious problems [2]. More importantly, DM remains a main cause of blindness. Especially in the long term, DM may result in serious macrovascular problems including vitreous hemorrhage, rubeosis, temporary blurring of vision, glaucoma, diabetic retinopathy (DR), and cataracts [3]. Among them, retinopathy is responsible for most of the sight-threatening complications of DM, while cataract is another major secondary complication [4].

So far, there have been already a great number of researches focusing on the underlying mechanism of DM and its complications, which aims to find out potential prevention and possible treatment strategies. For example, to determine the role of the cytokines, Demircan et al (2006) measured interleukin-1 beta and tumor necrosis factor-alpha in both serum and vitreous humor samples from patients with proliferative DR, which were attributed to the role of interleukins in the development of this disease [5]. So far, there have been several genome-wide association studies on the complications of diabetes including DR and cataract [6,7]. Burdon et al (2015) employed a genome-wide association approach to determine novel contributors of sight-threatening DR, showing a strong connection to genetic variation close to the GRB2 gene [6]. Similarly, Chang et al (2016) revealed genetic factors for diabetic cataract by using a genome-wide association method, indicating that the CACNA1C gene is connected to diabetic cataracts [7]. Moreover, another useful approach, proteomics analysis, has been employed in various ocular diseases associated with DM and its complications. For example, Chiang et al (2012) conducted a comparative study on proteomics between the controls and DM patients with the development of DR, which finally identified potential AH biomarkers, as well as susceptibility factors for predicting DR development [8]. Furthermore, by using two-dimensional differential in gel electrophoresis connected to MS, Su et al (2014) determined differential changes on proteomics and metabolomics between “slow” type 2 and “fast” type 1 diabetic cataracts in rats, which was helpful for identifying the shared and differential mechanisms [9].

Besides being an emerging and potentially powerful tool, metabolomics has been employed on studies of various diseases including DM and ophthalmology researches. It allows the simultaneous determination of numerous endogenous compounds including amino acids, organic acids, lipids, and nucleic acids in specific cells/tissues at a special time. Generally, two main analytical platforms for metabolomics studies includes nuclear magnetic resonance (NMR), and MS based metabolomic methods such as GC-MS, liquid chromatography connected to MS (LC-MS), and capillary electrophoresis connected to MS (CE-MS). By using high-resolution 1H NMR, Mayordomo-Febrer et al (2015) profiled the AH in corresponding controls and in glaucoma-induced eyes, which showed that after a series of sodium hyaluronate injections, levels of certain metabolites were significantly different [10]. The metabolomic data played a very important role in glaucoma pathogenesis. Furthermore, Barbas-Bernardos et al (2016) employed both LC-MS and CE-MS to compare patients with various severities of myopia, which not only showed metabolic variation among various severities of myopia, but provided potential biomarkers and new targets [11].

Recently taking advantage of GC-TOF MS, Ji et al (2017) reported metabolic characterization of human AH referred to high myopia, showing significant variation not only in metabolite abundances but also in metabolite-metabolite correlations [12]. Likewise in the present study, we also employed GC-TOF MS to profile 30 AH samples including 15 for controls, and 15 with DM. We believed that our work may provide potential AH biomarkers for clinical diagnosis and monitoring DM. More importantly, it may present novel insights into the molecular mechanism of DM formation and its complication of cataracts.

Materials and methods

Subjects

Thirty subjects were recruited in the present study as shown in Table 1: 15 patients with DM and 15 for the controls. All of them met the inclusion criteria as the previous study [12]. Moreover, the mean age for DM patients was more or less 60, while the mean age in the control group was nearly 65. The statistical analysis showed no significance for both age and sex between these two groups. Additionally, other characters including axial length were also shown in Table 1. The Ethics Committee of Baoshan District Traditional Chinese and Western Medicine Hospital (Shanghai, China) has reviewed and approved the study protocol.

Table 1.

Data of human AH samples

Group	Patient No.	Gender	Age (Years old)	Axial length	LOCSIII
Controls	A1_1	Female	50	21.91	C4N3P1
	A2_1	Female	66	21.75	C4N3P2
	A3_1	Female	68	23.88	C2N4P2
	A4_1	Female	63	24.3	C3N3P2
	A5_1	Female	68	23.54	C3N3P1
	A6_1	Male	70	23.63	C5N4P2
	A7_1	Male	66	24.01	C4N5
	A8_1	Female	65	23.94	C3N4P2
	A9_1	Female	53	24.63	C3N4P2
	A10_1	Male	62	23.2	C3N2P4
	A11_1	Male	76	21.48	C2N2P3
	A12_1	Female	57	22.59	C3N3P2
	A13_1	Male	78	23.19	C4N5P2
	A14_1	Female	53	24.75	C2N2P5
	A15_1	Male	78	22.97	C3N3P4
Patients with diabetic mellitus	D1_1	Male	70	25.9	C4N4P4
	D2_1	Female	48	21.98	C2N2P5
	D3_1	Female	63	24.47	C3N3P4
	D4_1	Male	68	24.95	C3N3P3
	D5_1	Female	61	26.22	C3N2P2
	D6_1	Male	40	22.96	C3N2P5
	D7_1	Male	50	23.51	C2N2P3
	D8_1	Male	58	22.86	C3N3P3
	D9_1	Male	60	24.8	C3N3P4
	D10_1	Female	76	24.17	C5N3P2
	D11_1	Female	55	27.45	C4N5
	D12_1	Female	57	22.56	C3N3P4
	D13_1	Male	68	23.29	C4N4P4
	D14_1	Female	52	21.75	C2N2P3
	D15_1	Female	72	23.11	C4N2P2

Sample collection and GC MS analysis

Sample collection and preparation of AH were done as per the previous report [12]. The supernatant after final centrifuging was immediately transferred in liquid nitrogen until GC MS analysis. Metabolic profiling of all the AH samples was performed similarly to that described in the previous study [12]. After the process of metabolite extraction, the supernatant (400 μL from the samples) was then collected and dried in a vacuum concentrator, followed by derivatization and injection into the GC system for metabolomic analysis as described before [12].

Metabolites identification and metabolomic data analysis

The mass spectrometry data for each sample were mapped to the databases for metabolites identification as previously reported [12]. After data normalization, the metabolomic data were input to Mev (MultiExperiment Viewer) 4.8 for hierarchical cluster analysis. And meanwhile, SIMCA-P 13.0 software (Umetrics, Malmö, Sweden) was employed for PCA and partial least squares discrimination analysis (PLS-DA), together with which independent t-tests were conducted for identifying the distinct metabolomics profiles and determining significant differences between the controls and the patients with DM [12].

Pathway analysis

All 20 differential metabolites between the controls and the patients with DM were imported into the website for pathway analysis (http://www.metaboanalyst.ca/). The pathway library of Homo sapiens was chosen, while hypergeometric test and relative-betweenness centrality were selected in the algorithms, respectively. Moreover, the reference metabolome was “used all compounds in the selected pathways”.

Results

Metabolites profiling for human AH

To fully uncover human AH metabolome, we took advantage of GC-TOF MS for untargeted metabolites profiling. Totally 263 metabolites were determined in all 30 samples including 15 controls and 15 with DM (Supplementary Table 1). Moreover, these 263 metabolites covered the major and central metabolism pathways, which included 35 amino acids, 50 carbohydrates, 15 lipids, 5 nucleotides, and other 158 compounds (Supplementary Table 1). Among those 158 compounds, 39 biochemicals were named, while 59 biochemicals were identified as analytes and the left were defined to be unknown.

Distinct metabolomics profiles between the controls and patients with DM

To snapshot metabolic characterization between the controls and patients with DM, we inputed all the metabolic profiles into Mev software for hierarchical cluster analysis (Figure 1). The results indicated very distinct metabolomic profiles in these two groups. The 15 samples from patients with DM clustered together, which were clearly apart from the other 15 samples from the controls. We further performed PCA on all the 30 samples to provide an overview of the information hidden in the metabolomic data (Figure 2). Likewise, the 15 samples from patients with DM were clearly separated from those of the controls, which re-confirmed that the metabolomic profile in patients with DM was significantly different from that in the control group.

Figure 1.

Hierarchical cluster analysis of 263 metabolites between controls and patients with DM.

Figure 2.

PCA model of patients with DM and controls. The red circle dots represent samples from patients with DM, while the blue ones represent samples from the controls.

Metabolic changes in patients with DM

We further employed both the supervised statistical method PLS-DA and t tests for determining significant metabolites responsible for the identified metabolic separations. The result showed that 20 metabolites were found to be significantly altered (P ≤ 0.05), which included 18 up-regulated and 2 down-regulated biochemicals participated in 4 super pathways (Table 2 and Figure 3). The only two down-regulated biochemicals were beta-mannosylglycerate 2 and lactose 2, whose fold-change between patients with DM and controls were respectively 0.21 and 0.09. The 18 signicantly up-regulated biochemicals included 2,6-diaminopimelic acid, 1,6-phosphogluconic acid, leucrose 1, linoleic acid methyl ester, palmitic acid, and stearic acid, ranged from 2.61 to 29.73 folds.

Table 2.

Significantly different metabolites betweencontrols and patients with DM

Super pathway	Compound name	Retention time (minutes)	Ratio (DM/ARC)*	p value
Amino acid	2,6-Diaminopimelic acid 1	12.65	7.39	2.30E-02
Carbohydrate	beta-Mannosylglycerate 2	11.96	0.21	3.00E-03
	6-phosphogluconic acid	14.72	6.55	0.00E+00
	Lactose 2	16.00	0.09	1.20E-02
	Leucrose 1	16.67	3.73	9.00E-03
Lipids	Linoleic acid methyl ester	13.23	13.09	8.00E-03
	Palmitic acid	12.96	3.63	1.20E-02
	Stearic acid	13.86	3.21	3.40E-02
Others	3-Hydroxypyridine	7.50	2.93	2.70E-02
	Analyte 290	10.91	12.58	0.00E+00
	Analyte 411	12.86	5.93	1.70E-02
	Analyte 449	14.04	6.44	0.00E+00
	Analyte 467	14.64	9.4	1.00E-03
	Conduritol b epoxide 2	12.51	4.79	3.70E-02
	Indole-3-acetamide 4	13.98	5.83	0.00E+00
	Trans-3,5-Dimethoxy-4-hydroxycinnamaldehyde 1	13.32	29.73	0.00E+00
	Unknown 058	13.93	4.52	0.00E+00
	Unknown 059	14.00	6.24	0.00E+00
	Unknown 060	14.03	5.78	0.00E+00
	Urea	8.21	2.61	6.00E-03

^*DM represents patients with diabetes mellitus while ARC represents the controls.

Figure 3.

PLS-DA model of patients with DM and controls. A. Score plot. The red boxes represent samples from patients with DM, while the blue ones represent samples from the controls. B. S plot. Metabolites marked with red triangles play key roles for separation.

Metabolic pathway analysis

To further facilitate the biological interpretation, all 20 differential metabolites between patients with DM and the controls were imported into the MetaboAnalyst web server. The results indicated that nine metabolites mapped to HMDB/PubChem/KEGG were involved in ten most relevant pathways (Figure 4 and Supplementary Table 2) including fatty acid biosynthesis, fatty acid metabolism, linoleic acid metabolism, fatty acid elongation in mitochondria, pentose phosphate pathway, and alpha-linolenic acid metabolism. Especially, fatty acid biosynthesis, fatty acid metabolism, and linoleic acid metabolism were the most significantly influenced pathways (P < 0.05) in DM patients.

Figure 4.

A systemic view of disordered metabolic pathways in association with DM.

Discussion

There already have been many metabolomic studies focusing on DM and DM-related diseases including DR and diabetic kidney disease [13-15]. Most of these studies were conducted for serum samples and only one recent study was performed for aqueous and vitreous humors samples by metabolomic profiling of reactive persulfides and polysulfides. Here, we took advantage of GC-TOF MS for untargeted metabolite profiling for 30 AH samples from 15 control patients and 15 patients with DM. More importantly, the identified metabolites discovered here, for the first time disclosea much broader AH metabolome in association with diabetic mellitus [12-17].

Increasing evidence revealed oxidative stress plays critical roles in the formation of both types of DM and cataracts [4,18]. Here the level of linoleic acid methyl ester was greatly increased in DM patients, which was reported to increase oxidative stress in patients with DM, further triggering and modulating the process of apoptosis [19,20]. Likewise, high levels of stearic acid and palmitic acid in patients with DM may induce apoptosis and finally lead to DM and cataract [21]. The other 17 metabolites may also play critical roles and be involved in regulatory pathways in relation to DM. For example, a direct precursor of indole-3-acetic acid, indole-3-acetamide 4, triggered an increased tolerance to different toxic compounds and several stress conditions [22]. Moreover, both mannosylglycerate 2 and lactose in carbohydrate metabolism were reported to be involved in protecting functional protein (such as photoreceptor proteins) activities from denaturation [23,24]. Another metabolite in carbohydrate metabolism, leucrose 1, was found to exhibit a very strong hydrophobic effect, which may also play pathophysiologic roles in DM and its complication cataract [25].

DM is a chronic progressive metabolic disorder that remains a growing and major global health problem. It is characterized by impaired carbohydrate (especially glucose) metabolism with hyperglycemia, mainly due to deficiency of insulin. Recent studies have identified that tissue lipid accumulation and dysregulated fatty acid metabolism both participate in the formation of insulin resistance and DM. Likewise, fatty acids metabolism are reported to be contributed to cataractogenesis [26]. For example, some fatty acids including linoleic and linolenic acid are involved in the development of higher risk of nuclear cataract [12,27]. In the current study, the levels of palmitic, stearic, and linoleic acid methyl ester were all significantly elevated in patients with DM, which was consistent with a previous study [28]. The metabolic pathway analysis here suggested pathways including fatty acid biosynthesis, fatty acid metabolism, and linoleic acid metabolism may be critical in the formation of DM and diabetic cataracts.

In summary, by using a non-targeted technology, GC-TOF MS, we comprehensively revealed AH metabolomic profiles from a serious of 30 patients (including 15 for controls and 15 with DM). Significantly different metabolomics were observed in association with diabetic mellitus. More importantly, significantly changed metabolites related to oxidative stress and their corresponding pathways including fatty acid biosynthesis, fatty acid metabolism, and linoleic acid metabolism may play key roles in the formation of DM and diabetic cataracts. Our effort may present potential biomarkers for predicting DM in AH, and also broaden our understanding of the underlying molecular mechanisms.

Disclosure of conflict of interest

None.