Glycomics based biomarkers of the rate of aging :
development and applications of high-throughput N-glycan analysis
Ruhaak, L.R.
Citation
Ruhaak, L. R. (2011, March 24). Glycomics based biomarkers of the rate of aging : development and applications of high-throughput N-glycan analysis.
Retrieved from https://hdl.handle.net/1887/16559
Version: Corrected Publisher’s Version License:
Licence agreement concerning inclusion of doctoral thesis in the Institutional Repository of the University of Leiden
Downloaded from: https://hdl.handle.net/1887/16559
Note: To cite this publication please use the final published version (if applicable).
page 165
8
Chapter 8.
General discussion
page 166
The aim of the work described in this thesis was to search for changes in glycosyla- tion that are associated with familial longevity within the Leiden Longevity Study. As techniques for large scale glycan analysis were largely lacking, a part of the studies in this thesis have been devoted to the development of techniques for high-through- put glycan analysis –especially at the level of sample preparation- to allow glycan profiling of large study cohorts, such as the LLS. Thus, firstly the progress that was reached with method development for large scale glycan analysis, as described in Chapters 2, 3 and 4 is discussed, while the second part of the discussion will focus on the results obtained with these techniques from the glycan analysis of the Leiden Longevity Study (Chapters 5, 6 and 7).
Advances in large scale N-glycan analysis
Over the years, several methods have been developed for the analysis of N-glycans from glycoproteins, whether at the levels of the ‘total glycoproteome’ in body fluids or at the level of individual, purified proteins. However, at the start of this project, such methods were time-consuming and had only been applied in studies with sample sizes of up to 150 samples. Recently, the potential of glycosylation patterns as bio- logical features marking changes in physiological states associated with e.g. disea- ses has been recognized [9]. To allow the conduction of further studies aimed at biomarker discovery and unraveling the biological processes responsible for these changes in glycosylation, there is a need for sample preparation techniques that allow high-throughput N-glycan analysis in cohort sizes as typically encountered in epidemiolocial and clinical studies. In the meantime several other groups have also developed procedures for high-throughput N-glycan analysis.
Protein glycosylation analysis can be performed at three different levels: at the level of the intact the glycoproteins, glycopeptides or released glycans may be studied.
While the first two approaches are highly suitable for purified glycoproteins, the latter strategy allows the in-depth analysis of N-glycan pools from multiple glycoproteins in one analysis. Moreover, this strategy is more universal and can, once developed, be applied to all types of glycoprotein samples, whether in a mixture or from specific
page 167
8
glycoproteins. A drawback is that information regarding the glycosylation site (and in mixtures therefore also the protein of attachment) is lost.
In Chapters 2 and 4 the development of sample preparation techniques for N-glycan analysis of released glycans using HILIC-HPLC-FL combined with MALDI-TOF and multiplexed CGE-LIF are described respectively. The procedures have been optimi- zed for the analysis of N-glycans from human plasma. The first advantage of these methods is that they have been optimized to be performed at the 96-well microti- tration plate level, thus allowing the simultaneous preparation of 96 samples in a standardized format. A drawback of previous methods was the large amount of time- consuming steps necessary to obtain samples ready for analysis. In the procedures described in this thesis, glycans are labeled directly after enzymatic release, wit- hout prior purification, decreasing the time needed for analysis. However, this one- pot sample preparation limits the possibilities for stationary phases for subsequent SPE purification, due to the large amounts of lipids, salts and other plasma matrix constituents, present in the samples. As a reverse phase purification step will not purify the samples from salts, and carbon-SPE material will bind almost everything and therefore clog when plasma samples are applied, HILIC-SPE material, to which glycans bind on the basis of their hydrophilic properties [134], is most suitable for subsequent purification. Lipids and salts show hardly any retention on this material, while most proteins are also too hydrophobic to bind. The interaction of glycans with this material is based on hydrogen bonds with the OH groups of the sugar moieties.
To allow fluorescence detection of the glycans, the reducing end of the oligosaccha- ride has to be derivatized by reductive amination. A large disadvantage of the use of reductive amination for oligosaccharide labeling is the use of NaBH3CN, which, upon contact with water, releases the toxic HCN gas. Several reducing agents have been described in other processes (e.g. [101]); however, no really successful substituent had yet been described. We have shown (Chapters 3 and 4) that 2-picoline-borane, which was previously reported for the reductive amination of several synthetic keto- nes/aldehydes [179], is an excellent alternative for NaBH3CN, as it performs equally
page 168
well, or even better than 2-picoline-borane. We expect that 2-picoline-borane may ultimately substitute NaBH3CN in most reductive aminations of oligosaccharides.
High-throughput analysis, off course, not only depends on fast, robust and large sca- le sample preparation, but also needs similarly performing separation and detection methods. As stated in the introduction, several strategies have previously been used in small scale glycan analysis. Such strategies usually comprise chromatographic or electromigrative separation techniques followed by mass spectrometric or fluo- rescence detection. It has been postulated that fluorescence detection of glycans is better for quantitation than mass spectrometric detection. Alternatively mass spec- trometry may be applied directly. Separation techniques used for carbohydrates are usually slow (at least 30 min.), and faster methods that are commercially available (e.g. the labchip GCII microchip-CE from Caliper lifesciences, Hopkinton, MA, http://
www.caliperls.com) so far appear to have a lower separation power. Alternatively, high-throughput may be obtained by using multiplexed systems. In Chapters 2 and 5 this was attempted using a dual LC-system, but the time needed for the measure- ment of the full LLS (approximately 2500 samples) was still six months. In Chapters 4 and 7, a 48 capillary CGE separation was proposed, allowing the analysis of the same LLS cohort in four days, thus significantly outperforming HPLC separations. A disadvantage of the use of multiplexed CGE is its non-compatibility with mass spec- trometric detection, making peak annotation more difficult.
While glycan separation is often combined with detection on the basis of a fluores- cent tag, which results in good quantitation, an alternative technique for fast glycan analysis is direct mass spectrometry. We have applied such methods in Chapters 2 and 4. In Chapter 2, direct MALDI-TOF-MS was applied in addition to HILIC-HPLC- FL, and it was indeed observed that the quantitative power of MALDI-MS was less than that of HILIC-HPLC-FL. In Chapter 4, the use of MALDI-FTICR-MS is descri- bed. This technique, developed for detection of IgG-Fc glycopeptides by intermedi- ate pressure MALDI-FTICR-MS has recently been described by our group, allowing detection with high-resolution and relative quantitation of both sialylated and non-
page 169
8
sialylated glycopeptides [58]. We could observe a similarly improved detection of sialylated glycans in MALDI-FTICR-MS over MALDI-TOF-MS, however, the quan- titation-results still showed a larger variability compared to fluorescence detection.
When analyzing glycan patterns, it is obviously important to know the composition of the N-glycan pool, and to be able to annotate the signals to glycan structures. Using mass spectrometric detection, this is relatively easy, as the obtained mass is indica- tive of the glycan composition in terms of hexoses, deoxyhexoses, N-acetylhexosa- mines and sialic acids. However, linkage information is more difficult to obtain. When applying fluorescence detection, peak annotation is more straining, as hardly any standards are available, and therefore no direct procedure is available. For such analyses it is thus important to generate libraries, similar to the GlycoBase, which was developed in the group of Prof. Rudd [135]. In this database, the standardized elution positions of 2-AB-labeled glycans in HILIC with fluorescence detection are used for structural assignment [80]. In chapters 2 and 4, a first attempt was made to generate peak annotations for 2-AA-labeled N-glycans found in HILIC-FL and CGE- LIF analyses, respectively. The annotations were obtained by glycan fractionation on HILIC material and subsequent mass spectrometric detection. However, using this strategy, we could only annotate the most abundant glycans. The application of glycosidase treatments and further fractionation strategies will be necessary for the annotation of additional glycans (including linkage isomers), but as such techniques are very time consuming this was beyond the scope of this thesis.
Several other aspects that are associated with large scale glycan analysis have to be considered when executing such analyses. Most large scale analyses have been performed on plasma samples. Glycans are then released from the protein pool, and subsequently profiled. The N-glycosylation pattern thus reflects N-glycans at the level of the total glycoprotein pool. Altered glycosylation patterns may be due to changes in concentrations of plasma glycoproteins, but may also be caused by alte- red glycosylation of one or more glycoproteins. It would thus appear to be of benefit to generate plasma protein profiles in parallel, to identify altered protein expression.
page 170
Alternatively, the glycosylation pattern of a specific glycoprotein may be monitored, as we described in chapters 6 and 7.
Another aspect of total plasma N-glycan profiling is that glycans from the high- abundant proteins dominate the glycan profiles, and therefore only changes in the glycosylation pattern of these high-abundant proteins will be observed using such strategies. This issue is similar to the problem faced in human plasma proteomics analyses: high abundant proteins are present in plasma at concentrations of mg/
ml, while low abundant proteins are present at pg/ml levels, which implies a 1010 to 1012 magnitude difference in plasma concentrations (e.g. [16;212;213]). Human plasma proteins may roughly be divided in two groups [15;16]: 1. the classical plas- ma proteins, whose function depends on their presence in the plasma (e.g. albumin, transferrin and immunoglobulins) and 2. tissue leakage proteins and other secreted proteins, whose function is as far as known not depending on their presence in plasma. While the first group mainly comprises the high-abundant proteins, which represent at least 95% of the plasma protein mass, the second group comprises many more protein species at lower abundances and most protein-derived biomar- kers have been discovered in the second group of low abundant plasma proteins.
Several strategies for protein depletion and enrichment have been developed for use in proteomics analyses, and such techniques have been reviewed extensively (e.g. [15;213;214]). All techniques are based on protein binding: while depletion of the high-abundant proteins is often performed using commercially available im- munoaffinity kits (except for immunoglobulin G, which may be captured with Prot A/G), enrichment is often performed using peptide libraries (e.g. the commercially available ProteoMiner beads) [214], RNA or DNA libraries (so-called aptamers) [215]
or lectin affinity strategies [205]. Recently, the effects of abundant protein depletion on plasma N-glycosylation profiles were evaluated for the first time [216]. A commer- cially available MARS 6 column was used to remove the six most abundant proteins (albumin, immunoglobulin G, transferrin, α1-anti-trypsin, haptoglobin and immuno- globulin A) from human plasma. Glycans were released both from total plasma as
page 171
8
well as from depleted plasma and it was observed that relative levels of low-molecu- lar-weight glycans carrying a core fucose were lower in the depleted samples [216].
Such glycans were previously observed to be highly abundant on immunoglobulin G (e.g. [165]).
While depletion is a powerful tool, only the top 1 to 20 plasma proteins are generally removed. Using peptide or aptamer libraries for enrichment, the protein levels of both high- and lower abundant proteins are standardized. When glycans are re- leased after this standardization, a much more representative glycan profile can be obtained, which is not affected by altered protein concentrations. Application of such approaches would greatly enhance the possibility to find glycosylation-based markers. Preliminary studies in our lab using ProteoMiner beads have generated promising results.
Using the strategy described in Chapter 4, combining glycan release, APTS labe- ling using 2-PB, HILIC-SPE and CGE-LIF detection we now have a high-throughput and robust analysis platform available, which can be applied for the evaluation of changes in N-glycan profiles in large scale sample sets derived from cohorts with se- veral diseases, such as the LLS, cancer cohorts, individuals with diabetes mellitus or Alzheimer’s disease. This should significantly contribute to study the association of glycosylation patterns with several disease states, and thus help to unravel the bio- logical processes regulating protein glycosylation in normal vs. diseased conditions.
Glycan based biomarker discovery of the rate of human aging
To allow the discovery of novel biomarkers that mark the rate of human aging, the first question that needs to be answered is: What exactly is a biomarker? One de- finition has been proposed by the NIH Biomarkers Definitions working group: “ A biomarker is a characteristic that is objectively measured and evaluated as an indi- cator of normal biological processes, pathogenic processes, or pharmaceutical re- sponses to a therapeutic intervention” [217]. While biomarkers have previously been identified based on clear medical or biological insight, recent interest in biomarker
page 172
discovery is mainly due to new high-throughput analytical techniques that promise to find relevant markers rapidly, without detailed insight into mechanisms of disease (a holistic approach). The full process from first measurement to final biomarker consists of six essential components: candidate discovery, qualification, verification, research assay optimization, biomarker validation and, ultimately, commercialization [218]. All studies performed in this thesis were at the level of candidate discovery, and therefore, only candidate markers have been identified.
The majority of studies conducted so far in the field of aging research have been fo- cused on markers that change as a function of calendar age. For the identification of biomarkers that reflect biological age, however, such studies are insufficient. Since familial longevity is hypothesized to represent a delayed process of biological aging, this thesis was partially aimed at the identification of markers that reflect familial longevity.
To identify glycan based candidate markers for human longevity, the offspring of long-lived individuals can be compared to their partners in the LLS. In such a com- parison, the offspring is regarded to represent individuals with a more healthy and slowly aging phenotype and a higher ‘susceptibility’ to become long-lived, while their partners, representing the general population, serve as controls. Glycosylation patterns originating from total human plasma, plasma-derived immunoglobulin G, alpha-1-antitrypsin and immunoglobulin A from the offspring and their partners have been generated and analyzed in this thesis.
Several changes in N-glycosylation patterns associated with familial longevity were observed in this thesis. In chapter 5, two glycan features (LC-7 and LC-8) origina- ting from total plasma proteins could be associated with familial longevity (Table 5 3). Three IgG glycoforms, all non-or mono-sialylated and containing a bisecting GlcNAc, were found to reflect familial longecity in chapter 6 (Table 6 3). As the pre- dictive value of these markers is rather low (the area under the ROC curve of the logistic model including both IgG glycosylation markers and plasma features LC-7
page 173
8
and LC-8 is only 0.564) it has to be concluded that multiple markers are needed for the prediction of complex phenotypes like familial longevity.
Figure 8 1. ROC curve of the logistic model including both IgG glycosylation markers and plasma features LC-7 and LC-8.
Moreover, as replication of these results has not been performed by carrying out a similar study in another cohort yet, the current results can till that moment only be classified as candidate biomarkers for familial longevity. The glycan features (LC-7 and LC-8) marking familial longevity in total plasma tend to change with calendar age (see Table 5 2) and the IgG glycoforms marking familial longevity are also highly as- sociated with calendar age (see Table 6 2). As the offspring of the long-lived siblings tend to have ‘younger’ profiles than their partners, these observations would support
page 174
our hypothesis that familial longevity is in essence a delayed aging phenotype.
To allow the development of interventions aimed at disease-free aging, medical sci- ence would obviously benefit from having markers available that reflect healthy and unhealthy aging. It can be questioned to what extent markers that associated to familial longevity and possibly biological aging also reflect the health condition of an individual. To address this issue, the relation between several physiological pa- rameters and protein N-glycosylation patterns was assessed, as well as the relation of N-glycosylation patterns with the occurrence of the age-related cardiovascular and metabolic diseases such as myocardial infarction (MI), cerebrovascular accident (CVA) and diabetes. LC-7 and LC-8 could be associated with several physiological parameters. Moreover, LC-8 could be associated with the incidence of myocardial infarction (MI) in participants of the LLS (Table 5 5), however, the number of individu- als diagnosed with MI in the study population is still very low (2.8%). It is anticipated that in later follow-up questionnaires, more participants in the LLS will have been diagnosed with these diseases, so that glycosylation features such as LC-8 can be tested for its value in predicting incidence of MI in a follow up of the LLS. Alterna- tively, N-glycosylation patterns may be analyzed in existing prospective studies on the incidence of cardiovascular disease.
In Chapter 7, changes in AAT-derived glycosylation could not be associated with fa- milial longevity, but relations with calendar age (Table 7 3) were observed for several glycosylation features. Moreover, relationships were observed between a number of physiological parameters (mainly levels of triglycerides) and AAT-glycosylation patterns. The occurrence of MI (Table 7 7), was positively related to two glycosyla- tion features, indicating that AAT-glycosylation is possibly influenced by metabolic health, but does not mark familial longevity.
While glycosylation patterns can be associated with several disease states, not much is known regarding the physiological parameters that influence the glycosyla- tion patterns. A recent study revealed that body mass index, plasma lipid profiles, and smoking are associated with changes in plasma N-glycosylation patterns [24].
page 175
8
It might be interesting to perform cellular assays in which cells are stimulated by known signaling molecules, and evaluate the changes in glycosylation patterns of the cells, similarly to strategies used in e.g. [219].
More than once, indications have been described in literature for different regulation of plasma N-glycosylation in females prior to and after menopause (e.g. [23;24;185]).
To further evaluate such observations, a larger-scale population study has to be undertaken to properly evaluate such effects, and to unravel possible physiological pathways that regulate this altered regulation.
The regulation of glycosylation is a very complex cellular process, and the biological pathways involved in longevity and healty aging have only started to be unraveled.
Currently, there is no clear regulator that would link the regulation of familial longe- vity with the regulation of the plasma N-glycan profile. It has been demonstrated that total plasma N-glycosylation patterns are partially heritable [21], and it is known that the plasma N-glycosylation pattern is partially determined by genetic variation: in a recent genome wide association study (GWAS) the regulation of FUT 8 and ESR2 was shown to be related to levels of the non-fucosylated non-galactosylated bianten- nary glycan [23]. To date however, this study has only been performed assessing this one glycan, and the results have not yeat been replicated. A GWAS has recently been finished in all subjects of the LLS, allowing evaluation of the replication of the results. These studies may also reveal whether genes that regulate protein N-glyco- sylation patterns also regulate human longevity. To allow further in-depth analysis, the glycosylation in long-lived mice, such as the growth hormone receptor knock-out (GHRKO) mouse described in [194], could be studied. As mice can be bred under standardized conditions, and additional parameters could be analysed more easily than in humans, such a model would even more facilitate the search for alterations associatied with longevity that could regulate protein N-glycosylation.
In conclusion, the present thesis describes the development of several approaches for high-throughput N-glycosylation analysis at the level of the total plasma proteome
page 176
as well as at the level of individual proteins. These techniques were then applied to a biomedical study involving a cohort size which is typical for such studies, and which indeed needs the throughput which was technologically realized. It is anticipated that – apart from allowing more detailed insight in the processes underlying aging and longevity- the developed methodology will as well be of considerable value to other fields of biomedical research.
