Effect of metformin on the high‐density lipoprotein proteome in youth with type 1 diabetes

Abstract Background Youth with type 1 diabetes (T1D) have normal or elevated High‐Density Lipoprotein Cholesterol (HDL‐C), however, the function of HDL, partly mediated by the HDL proteome, may be impaired. Metformin can be used as an adjunct therapy in youth with T1D, but its effects on the HDL proteome are unknown. Objective To determine the effect of metformin on the HDL proteome. Subjects Youth (12–20 years old) with T1D who had a BMI > 90th percentile, HbA1c > 8.0% and Tanner stage 5. Methods Double‐blinded, placebo‐controlled randomized sub‐study. We examined the effects of metformin (n = 25) or placebo (n = 10) after 6 months on HDL proteome. Changes in HDL proteins were measured by data‐independent acquisition (DIA) mass spectrometry and compared between treatment groups. As a secondary outcome, associations between proteins of interest and the most studied function of HDL, the cholesterol efflux capacity (CEC), was examined. Results The relative abundance of 84 HDL‐associated proteins were measured. Two proteins were significantly affected by metformin treatment, peptidoglycan recognition protein 2 (PGRP2; +23.4%, p = .0058) and alpha‐2‐macroglobulin (A2MG; +29.8%, p = .049). Metformin did not significantly affect CEC. Changes in affected HDL proteins did not correlate with CEC. Conclusions Despite having little effect on HDL‐C, metformin increased PGRP2 and A2MG protein on HDL in youth with T1D, but had no significant effect on CEC. Further studies are needed to understand the impact of PGRP2 and A2MG on other HDL functions.


| MATERIAL S AND ME THODS
This project utilized blood samples that were collected during the "T1D Exchange Clinic Network Metformin RCT Study Group" trial, a double-blind, placebo-controlled clinical trial assessing metformin usage as an adjunct therapy in obese adolescents with T1D. The design and the results of the main clinical trial have been previously described. 11 In summary, this trial was conducted in 26 clinical sites of the T1D Exchange Clinic Network. Eligibility criteria for inclusion in this trial included: age 12-20 years old, diagnosis of T1D, on treatment with insulin for at least 1 year, BMI at or above the 85th percentile based on CDC growth charts, total daily insulin dose of at least 0.8 units/kg/day, HbA1c values at enrollment of 7.5%-10.0% and checking blood glucose at least three times a day. A total of 140 adolescents (aged 12.1-19.6 years old) were enrolled in this study.

| Participants
For the current proteomic project, our inclusion criteria included participants with T1D who participated in the "T1D Exchange Clinic Network Metformin RCT Study Group" trial and had a baseline BMI > 90th percentile, HbA1c > 8.0% and were Tanner stage 5. Those criteria were selected in an effort to the minimize potential effects of different stages of pubertal development on protein cargo of HDL. We selected heavier participants and those with suboptimal glycemic control, given the expected lower insulin sensitivity and potentially enhanced benefit from metformin treatment. Of the eligible participants, the core lab randomly selected 25 participants with T1D who were treated with metformin for 6 months and matched them for age and sex with 10 participants with T1D treated with placebo for 6 months.
The T1D Exchange Biobank studies are governed by individual site institutional review boards. Qualifying participants provided consent or assent as age-appropriate and parents provided consent for participants <18 years old. scans were done to assess % body fat. BMI percentile was calculated using the childs package in r, using the 2000 CDC growth charts as a reference. 12 Further studies are needed to understand the impact of PGRP2 and A2MG on other HDL functions.

K E Y W O R D S
cholesterol efflux, high-density lipoprotein, metformin, proteomics, type 1 diabetes

| Proteomics experiments
The preparation of serum samples for the proteomics experiments was performed as previously described. 13,14 In brief, (1) purification of HDL from serum using size-exclusion chromatography; (2) pooling of all the HDL containing fractions; (3) application of lipid-binding resin to pooled HDL; (4) washing of HDL on the resin in order to remove contaminating proteins; (5) digestion of resin-bound HDL with trypsin (overnight 37°C); (6) washing of the resin in order to collect tryptic peptides; (7) reduction and carbamidomethylation (DTT and iodoacetamide, respectively); and (8) desalting of samples using ZipTips. The samples were dried for proteomics analysis by using a nano Acquity UPLC coupled with a TripleTOF 6600 mass spectrometer (MS).
As a first step, we established a library of all HDL proteins detectable by MS for the analysis of a sample pool from all the participants (with an equal amount of proteins combined). The pooled sample was analysed in data-dependent acquisition (DDA) mode.
Then, we ran each sample individually via label-free Sequential The mass spectra were recorded with Analyst TF 1.7 software in the DDA mode. Each cycle consisted of a full scan (m/z 400-1800) and fifty (IDAs; m/z 100-1800) in the high sensitivity mode with a 2+ to 5+ charge state. Rolling collision energy was used. For the SWATH acquisition, each of the samples was injected individually into the same NanoUPLC-MS/MS system but acquired by repeatedly cycling through 32 consecutive 25-Da precursor isolation windows, generating time-resolved fragment ion spectra for all the analytes detectable within the 400-1200 m/z precursor range.

| Proteomics data analysis
In order to generate the spectra library, raw mass spectra files after

| Cholesterol efflux assay
In order to determine whether changes in the HDL-bound proteins of interest are associated with changes in the HDL function, we measured the CEC of HDL as previously described by our group. 3 In summary, the HDL-CEC assays were performed using the murine macrophage cell line, J774, as per published methods. [17][18][19] Briefly, 3 x 105 J774 cells/well were plated and radiolabeled for

| Statistical analysis
The distributions of all variables were examined prior to analysis.
Descriptive statistics reported include frequencies and percentages for categorical variables, means and standard deviations for normally distributed continuous variables, and medians and percentiles for non-normally distributed continuous variables. To compare groups, either the chi-square test or Fisher's exact test was used for categorical variables, and either the Mann-Whitney test or t-tests were used for continuous variables. For the initial identification of any proteins that changed significantly after 6 months of treatment (among a total of 84 HDL proteins), we used the Mann-Whitney test to compare the % change in each protein in the metformin vs. the placebo group. For further analysis of the selected proteins, we performed T-tests on ion intensity values test to confirm our results.
Finally, partial least squares discriminant analysis (PLS-DA) with leave-one-out cross-validation was used to examine whether participants in the two treatment groups could be discriminated on the basis of changes in the proteins. All analyses were performed using R version 3.5.1 or GraphPad Prism version 8.3.0.

| Study population
The participants in the two treatment groups did not differ in age, sex, or race and all were Tanner stage V at baseline (Table 1). Baseline (Table 2) and after 6 months (Table 3) HbA1c, plasma lipids, % body fat and BMI were not significantly different between groups. In a univariate analysis, the insulin dose was significantly different between patients with T1D and HC (Tables 2 and 3).

| Effect of metformin on the HDL proteome
Using a label-free SWATH mass spectrometry analysis, an HDL proteome quantification library was generated using a subset of HDL samples. This library allowed for reliable detection of the relative abundance of 84 different HDL-associated proteins. The Mann-Whitney test was used to screen measured percent changes in protein abundance among treatment groups to identify likely affected proteins (Table S1). This initial screen-detected differences in average percent change between placebo and metformin groups for and the ROC AUC for this model was 0.6, only slightly better than chance. In this analysis, A2MG contributed most to discrimination between treatment groups (Table S2). In our analysis, we had identified one outlier with a protein that could skew the data. We also performed sensitivity analysis which indicated that removal of the outlier point did not impact the ultimate conclusion of the statistical comparison.

| Effect of metformin on CEC
To evaluate the possible impact of metformin on the most studied function of HDL, CEC was measured in all participants at baseline and 6 months follow-up. CEC of apoB-depleted plasma from each participant was measured using cholesterol-loaded J774 macrophages stimulated with cAMP to upregulate ABCA1 expression.
There was no significant difference between treatment groups at baseline or after 6 months of treatment with metformin or placebo ( Figure 3).

| DISCUSS ION
In this analysis, from age and sex-matched youth with T1D treated with metformin or placebo for 6 months, we demonstrate for the first time that metformin has an increase on the PRGP2 (peptidoglycan recognition protein 2 also known as N-acetylmuramoyl-L-alanine  In the current study, we found that HDL-associated PGRP2 increased significantly after metformin treatment while no change was observed in the placebo group. Our group has previously published that PGRP2 was higher in the HDL of youth with T1D compared to healthy controls. 20 This would suggest that the effect of metformin shifts the level of this protein even further from the level of healthy control participants rather than normalizing it toward the level of healthy controls. We also have shown previously that PGRP2 was not associated with glycemic control in youth with T1D. 20 It is possible that the percent change in PRGP2 may be more closely related to other factors relative to the metformin use rather than glycemic control. In fact, the T1D-Metformin RCT trial did not show any significant impact of metformin on HbA1c. 11 Overall, our findings suggest that metformin increases the PGRP2 and A2MG proteins on HDL and below we discuss the possible effects on metabolic and Peptidoglycan recognition proteins are known to be responsible for keeping a normal gut microbiome environment and protect the host from inflammation and colitis. 23 PGRP2 is present in the serum, the skin and the epithelial cells of the intestines. 24 The expression of PGRP2 is highest in the liver, from where it is secreted in the blood, and PGRP2 expression is also induced by bacteria in the F I G U R E 1 Effect of 6-month placebo or metformin treatment on HDLassociated proteins. Ion intensity values detected by data-independent acquisition (DIA) mass spectrometry analysis of sizeexclusion chromatography purified HDL were compared for subjects receiving placebo or metformin for 6 months.  Table S2 epithelial cells of all segments of the gastrointestinal tract. 23 There is growing evidence about the role of altered gut microbiota to promote CVD via several mechanisms, such as altering lipid metabolism, playing a role in vascular dysfunction and hypertension, increasing systemic inflammation, foam cell formation and insulin resistance. 25 There is recent evidence that treatment with metformin not only alters the gut microbiome but some of the beneficial effects of metformin could be due to the change in the gut microbiome. 26,27 It is also known that the intestine has a significant contribution to the production of HDL-C. 28 Based on the above, it is possible that metformin alterations to the gut microbiome could be associated with changes in the peptidoglycan protein expression of PRGP2 in the intestines and also on PRGP2 concentration on HDL.
We also found that A2MG increased after metformin. Our group has previously found that the A2MG protein of HDL was positively associated with calcified burden of coronary angiography, a marker of increased CVD risk. 29 Another group also found higher A2MG protein of HDL in patients with non-alcoholic fatty liver disease, suggestive of increased fibrinogenesis activity. 30 More research is needed to explore whether metformin changes the anti-thrombotic function of HDL by altering the expression of A2MG protein.
Our study did not find any effects of metformin on HDL efflux capacity, contrary to our hypothesis. A previous study by Matsuki et al 31 showed that glycation of HDL reduced the CEC of HDL and that metformin restored this defect in studies done in vitro, by preventing the glycation of HDL. One possible explanation for the difference with our findings is that the dose used for the in vitro studies was relatively higher and had a more potent effect than the dose that has been approved for use in humans. In addition, in our study, all patients had diabetes with a HbA1c well above target, and we thus assume that a certain degree of glycation of HDL was present in all of them. However, our study was not designed to look at the function. 32 Our group had also previously found that KLKB1 of HDL was associated with lower CEC. 29 However, given that KLKBI and A2AP proteins also increased in some participants in the placebo group ( Figure 1), we believe more research is needed to explore whether metformin changes the anti-thrombotic function of HDL by altering the expression of the above proteins.
Strengths of our study include the fact that we were able to match participants that were treated with metformin or placebo in a double-blinded randomized fashion. All participants were overweight and had similar glycemic control and pubertal stage. We also have detailed HDL proteome data: The two-step size-exclusion chromatography and lipid interaction-based HDL purification and the SWATH-MS allowed us to have a very sensitive and robust label-free proteomic quantitation for multiple clinical samples. Our two-step approach to the proteomics data analysis included an initial screening for metformin affected proteins based on percent change by treatment group and was followed up with direct comparisons of candidate proteins by their detected ion intensities. This approach improved confidence in detected differences and reduced the false discovery rate by filtering out weaker effects. Limitations of this study include our small sample size and the lack of additional HDL function assays or assays directly related to PGRP2 function and the lack of directly measured insulin sensitivity. A limitation of our HDL purification method is that it does not allow for analysis of HDL subfractions which are commonly characterized by density and isolated by ultracentrifugation. 33 One advantage of our approach over ultracentrifugation is that it purifies HDL under physiological g forces and salt conditions resulting in less disruption of protein interactions on lipoproteins caused by extreme conditions experienced during ultracentrifugation. 34,35 The downside is that size-exclusion chromatography by itself does not isolate pure HDL preparations and there is some contamination of non-lipoproteinassociated proteins (e.g. albumin and immunoglobulins). This is overcome by our two-step isolation, which includes washing of lipoproteins on a lipid-binding resin. 13 Alternative approaches to the isolation of native HDL include immunoaffinity adsorption usually by pulldown with anti-apolipoprotein A-I antibodies. 36 was to provide an initial characterization of the effect of metformin on the proteomic composition of HDL and identify associations with changes in cholesterol efflux function. Even though we did not find a significant change in CEC, it is possible this was due to our relatively small sample size or that the treatment did not have an impact on pre-beta HDL, the particle species predominantly responsible for ABCA1-mediated cholesterol efflux. 38 Pre-beta HDL are small discoidal particles composed of apolipoprotein A-I, phospholipid, and free cholesterol, and they do not contain many other proteins. 39 Therefore, effects observed in the HDL proteome are likely not reflective of changes to pre-beta particles. Although it is possible that redistribution of apoA-I or other proteins among HDL subspecies could occur even if no change in the total apoA-I protein was observed. However, this could not be detected without isolation of distinct HDL subfractions. It is also possible that there may be interesting glycation modifications to HDL proteins that could impact particle functions and might be prevented or reversed by metformin, but we were unable to examine these in the present analysis. Future studies can further explore whether enrichment of HDL with PRGP2 could potentially alter HDL function.
In summary, we found a significant increase in the PGRP2 and A2MG content of HDL in youth with T1D treated with metformin compared to placebo by combining size-exclusion chromatographybased HDL purification and SWATH-MS-based label-free proteomic quantitation. Further studies are needed to explore whether this effect of metformin impacts other known functions of HDL such as the anti-inflammatory or anticoagulant activity, and to also explore potential novel functions arising from this interaction. With the recent rapid advancements in proteomics technologies, future studies to investigate the factors that contribute to the glycation of HDL proteins in patients with T1D and how those are impacted by metformin or other diabetes drugs will be of significant interest.

ACK N OWLED G EM ENTS
The authors are grateful to the T1D Exchange Biobank and the participants who provide their time and effort in contributing critical research information. Without participant assistance, this research would not be possible. The T1D Exchange is the source of samples and data for this study, however, the individual authors are solely responsible for the analyses, content and conclusions presented as these have not been reviewed or approved by the T1D Exchange.

CO N FLI C T O F I NTE R E S T
Authors have nothing to disclose.

PATI ENT CO N S ENT S TATEM ENT
Patients gave consent to participate in the T1D-Exchange Metformin RCT.

DATA AVA I L A B I L I T Y S TAT E M E N T
Not applicable.