5.  Discussion

The aim of this study was to establish a reference 2D-electrophoresis map of the human mitochondrial proteome that could be used in subsequent studies to identify patients with mitochondrial disease. I chose immortalized lymphoblastoid cells as the material to investigate. Compared to other possible materials like muscle cells and fibroblasts, lymphoblastoid cells can be more easily obtained from patients. Routinely, 5-10 ml heparinized whole blood are sufficient to establish a permanent cell culture. This procedure is more acceptable to parents and children as compared to muscle or skin biopsies. Lymphocytes can be cultured in vitrowhich allows the yield of enough material for my investigation. Since the cells can grow in suspension, high cell density can be obtained without too much expenditure of work and cost. Lymphoblastoid cells are transformed with Epstein-Barr Virus (EBV) for immortalization. EBV-transformation as compared to SV40-transformation has the advantage of higher chromosomal stability during subsequent passages [Jha et al., 1998]. Another advantage of lymphoblastoid cells is that they express a respiratory chain defect – in contrast to fibroblasts – more frequently [Bourgeron et al., 1993]. Although the number of mitochondria in lymphocytes (about 15-20/cell) is far lower than that in liver cells (2,200/cell [Rohr et al., 1976]), the easier access to lymphoblastoid cells compared to liver cells counterbalances this shortcoming.

In recent years, several studies of the human mitochondrial proteome have been carried out on other materials like placental cells, transmitochondrial cybrids and neuroblastoma cell lines [Rabilloud et al., 1998 and 2002; Lopez et al., 2002; Fountoulakis et al., 2003]. However, for diagnostic purposes these cells are generally not available. In contrast, lymphoblastoid cells can be obtained from any patients at any age without anaesthesia. This also allows the study of mitochondrial proteins in normal control individuals at different age-groups in order to find out age-related changes. Additionally it also allows investigation of the genetic variability in a larger number of control patients at any given age.

Isolation of mitochondria is the first step to investigate mitochondrial function in vitro and to analyze their proteome. Subfractionation of cellular components can intensify the low-abundance proteins and make their detection easier. The effect of enrichment of mitochondrial proteins is illustrated in Fig. 4-3. The number of spots in the mitochondrial gel is distinctly reduced compared to that of the gel from whole lymphocyte proteins. The background noise and the complexity of the sample are thus reduced. This simplifies and focusses the study on the mitochondrial proteome. I have introduced several modifications to standard protocols in my experiment in order to solve some basic problems of mitochondrial isolation. First of all, in order to disrupt the cell membrane more easily, the use of digitonin is necessary. However, this membrane destabilizing agent disrupts membranes indiscriminately, i.e. the membrane of subcellular organelles can be attacked as well. Bronfman et al. (1998) found out that subcellular organelles are only affected if the concentration of digitonin is higher than 0.5-1.0 mg/ml. I therefore used only low concentrations of digitonin (0.1 mg/ml). The electron microscopic photographs of the mitochondrial pellet demonstrate that most of the mitochondria have intact inner and outer membranes. The slight swelling of the mitochondria may be due to the lower osmotic pressure of the used buffer. Secondly, I repeated the mechanical homogenizing cycle three times to insure the disruption of all cells. After the third round of homogenization about [page 55↓]97% cells were disrupted. The number of repetitive homogenizing cycles depends not only on the type of biological material but also on the type of pestle and on the rotating speed of the homogenizer. Thirdly, I separated mitochondria from other subcellular organelles by centrifugation on a discontinuous hybrid density gradient. Since the densities of mitochondria and lysosomes are very close to each other (1.11 g/ml of mitochondria versus 1.07 g/ml of lysosomes), the separation of these two fractions is of special importance. For this task standard protocols suggest the use of a continuous Percoll gradient which forms during centrifugation. However, this method did not lead to a satisfactory result. Since the density-gradient was continuous, more than five bands were formed after centrifugation. Three of them were the mitochondrial bands and lay very close to each other. In addition the mitochondrial band was too broad. Therefore, it was difficult to assure the purity of the product, especially when the original material was limited. So I opted for a discontinuous gradient (35% and 17% metrizamide and 6% Percoll, with the densities 1.1304 g/ml, 1.1029 g/ml and 1.0331 g/ml), in order to separate the mitochondria from the lysosomes sharply at two different interfaces. The electron microscopic photographs of the products from each interface confirmed that they contained the expected subcellular fractions. Disadvantage of the discontinuous gradient is the high cost of the metrizamide and the additional work to prepare the gradient.

I used carrier ampholytes for isoelectric focussing instead of commercially available immobilized pH-gradients. In Tab. 5-1 the advantages and disadvantages of these two methods are compared. Several advantages for my special task made me to use carrier ampholytes to run IEF-separations. A very good protein resolution can be expected by using long

Tab. 5-1: Comparison of carrier ampholytes and immobilized pH-gradients for first dimension isoelectric focussing.

gradients of carrier ampholytes [Klose, 1999b; Lopez, 1999]. Secondly, variable pH ranges can be prepared by the blending of different ampholytes. In addition, the preparation of a tube gel is much easier than that of immobilized pH-gradients and does not require complex gradient casting equipment. However, a variable batch-to-batch reproducibility, the effect of protein concentration on the shape of the pH-gradient, and the cathodic drift of the pH-gradient with time are some drawbacks.

Since the aim of my study was to detect as many mitochondrial protein spots as possible, optimum resolution was paramount. I thus chose carrier ampholytes to run my IEF-separations. On the other hand, the disadvantages of carrier ampholytes were minimized since I blended a large volume of gel stock-solution and aliquoted it into small portions so that all experiments were carried out with the same batch of ampholyte mixture. The cathodic drift of the pH-gradient with time was controlled by using a programmable power supply. This added to the standardization of IEF-running conditions. The effect of the protein sample on the pH-gradient was also investigated and is discussed below.

The reproducibility of 2D-separations can be assessed by three aspects:

1. The presence/absence of spots : the reproducibility of protein resolution,
2. The position of spots : the reproducibility of protein spot location on the gel and
3. The quantity of spots : the reproducibility of protein abundance.

In order to test the reproducibility of my gel-runs, I isolated mitochondrial proteins from the same cell line and run them on four different gels.

1. The gel-runs achieved a satisfactory reproducibility of protein resolution since each spot on one gel was reproduced on the other test-gels.
   
2. The reproducibility of position was only satisfactory when spots were compared to one another in a small perimeter of 3-5 cm from a certain anchor-point, i.e. the corresponding spots were only congruent when the gels were compared subsection by subsection. These running differences between two large gels might be due to minor inconsistencies of running conditions (temperature, buffer-composition), gel quality or protein-concentration.
   
3. Most of the protein spots achieved satisfactory quantitative reproducibility. However, the varying reproducibility of certain protein spots might be due to additional interfering factors. Spots 11 and 12, which correspond to the protein actin-beta, are examples. Since the test-samples of four gels were isolated from four different aliquots of the same cell line, the reproducibility of the mitochondrial isolation method was tested indirectly as well. The tight affiliation of actin-beta with the outer membrane of the mitochondria makes its removal difficult. The degree of actin-beta removal most probably depends on the cell fractionation process, which is the most difficult step to keep completely uniform.

Several other factors can also impair the reproducibility. For example if the protein samples are thawed more than once not only the resolution of the protein spots but also their quantity is reduced (Fig. 4-8). This phenomenon is most probably due to the degradation of proteins [page 57↓] during the freezing and thawing process. As expected the effect is most pronounced in low abundance proteins.

In order to resolve as many protein spots as possible I used the large gel 2D-electrophoresis method with a length of 40 cm for isoelectric focussing [Klose, 1999b]. This way Iseparated 420 protein spots on a silver stained 2D-electrophoresis gel. I do not know how far this number is from the actual mitochondrial protein content. Until now, a total of 525different proteins have been registered in a special database of mitochondrial proteins (Human Mitochondrial Proteins Database). This database comprises information from “SwissProt”, “LocusLink”, “Protein Data Bank” (PDB), “GenBank”, “Genome Database” (GDB), “Online Mendelian Inheritance in Man” (OMIM), “Human Mitochondrial Genome Database” (mtDB), MITOMAP, and “Neuromuscular Disease Center and Mendelian Inheritance and the Mitochondrion” (MitoDat).

Additionally, the existence of protein-isoforms increases the spot-number considerably. The number of isoforms of a certain proteinvaries between two to ten [Rabilloud et al., 1998; Jung et al., 2000; Fountoulakis et al., 2003]. Although not all spots of putative isoforms were analyzed in my study, from a representative sample of heat shock 60 kDa proteins (spot 9 and 10) one can assume that the horizontal arrays of spots are isoforms of the same protein. In my sample, the number of isoforms of a protein ranged from two to six.

Loss of proteins is doubtless taking place during the whole process from protein preparation until the staining of the gel. In order to minimize this loss I used a highly standardized sample preparation protocol introduced by Klose (1999a). One key point of this protocol is to solubilize hydrophobic proteins and keep them soluble. This is achieved by using CHAPS and high concentrations of urea to break up non-covalent interactions and by using DTT to break up disulfide-bridges. Another key point is to reduce protein degradation as good as possible. This is achieved by adding a cocktail of protein inhibitors and by performing all preparations in the cold at 4°C.

In my experiments the silver stain was much more sensitive than the Coomassie stain. This has already been demonstrated in other studies [Switzer et al., 1979; Rabilloud, 1990; Shevchenko et al., 1996a]. On the other hand, for the purpose of protein identification by MALDI-TOF mass spectrometry the Coomassie stain seemed to be superior. None of my silver stained spots could be identified. However, this disappointing result might not be due to the silver stain itself but due to the fact that I only cut out those silver spots, which could not be identified on the Coomassie stained gel or which had been very weak. Therefore, the reason of the zero identification rate of my silver-stained spots was more likely to be caused by low protein abundance than by a strong influence of silver ions on the proteins. Other authors like Shevchenko (1996a) and Rabilloud (1998) have demonstrated that silver-stained gels can indeed be used for satisfactory mass spectrometric identification of proteins.

The proteins were identified mainly by MALDI-TOF mass spectrometry on the basis of their peptide mass fingerprints and in some cases by sequencing a peptide tag with MALDI-QTOF tandem mass spectrometry.

According to the annotations in SWISS-PROT and NCBI, 78% of the identified proteins were human mitochondrial proteins. The other 22% identified proteins included 12% non-[page 58↓]mitochondrial proteins (cytoplasmic, endoplasmic reticulum or B-cell-associated) and 10% unknown proteins (most of them were only present as expressed sequence tags (EST)). The majority of my spectra was of good quality which allowed rather easy protein identification. The high accuracy in mass determination of the peptide fragments is made possible by MALDI-TOF mass spectrometry with “delayed extraction”. The delayed extraction method allows peptide fragments of the same weight to be better focused. This leads to sharp and accurate spectra which allow the reliable identification of proteins even with limited sequence coverage [Jensen et al., 1996]. Therefore, my results were reliable even with low sequence coverage of around 12-17%. The array of identified proteins contained eleven proteins from other cell compartments. This “contamination” cannot simply be attributed to the purification method of the mitochondria. Such proteins as the B-cell associated protein (spot 103) are very abundant and/or specialized for the lymphoblastoid cells. These proteins are difficult to remove completely and are absent when mitochondria from other tissues were investigated (e.g. placenta [Rabilloud et al., 1998] or neuroblastoma cells [Fountoulakis et al., 2003]). However, I cannot rule out that these eleven proteins might interact with the mitochondria as well, since some other non-mitochondrial proteins are known to have very tight links to the mitochondria. They either bind to the outer membrane of mitochondria or transport material to the mitochondria. An example for these proteins is actin-beta, which is attached to the outer membrane of the mitochondrion and is involved in the movement of mitochondria within the cell.

There are 25 spots in the list, which could not be identified. This might be due to the fact that the spectrum was not good enough or that the spectrum could not be matched with a protein from the databases. Besides the low abundance of certain proteins in the sample, several other reasons could also account for this. If two spots are very close to each other on the gel or overlapping each other, the proteins are mixed during excision of the spot. The spectra from those overlapping spots may confuse the database search. Peptide loss during the tryptic digest or the failure to extract the peptides from the gel pieces for MALDI-TOF mass spectrometry may be another reason for low amplitude mass spectra.

Notably, a large part (72%) of the unidentified spots comprises small proteins whose molecular weights are below 17.5 kDa. Some of these proteins are fragments of larger proteins. Spot 114 is an example for that. This spot is the smallest protein among the identified ones. Its molecular weight is around 5-10 kDa and it could only be identified by protein ladder sequencing to be a small fragment of actin-beta.

I established a reference map of the mitochondrial proteome of human lymphoblastoid cells. This map will be used as a tool for further investigation of mitochondrial proteins of patients with mitochondrial diseases. A total of 95 proteins, 74 of them of sure mitochondrial origin, have been identified on this reference map (Fig. 4-12). The sub-mitochondrial location and the function of each identified mitochondrial protein were looked up in various databases and original publications.

Similar to previous analyses of the mitochondrial proteome [Jung et al., 2000; Lopez et al., 2000; Fountoulakis et al., 2003], most of my identified proteins are hydrophilic or easy to solubilize proteins. Although 18 proteins of the list were annotated to be membrane-associated proteins, only four of them were clearly transmembrane proteins with at least one transmembrane helix.

Unfortunately none of the mtDNA-encoded proteins could be identified in my study. The mitochondrion is rich in membrane-associated proteins. Most of the 13 mtDNA-encoded proteins are components of multi-protein-complexes and are located at the inner mitochondrial membrane. Analysis of putative transmembrane domains in these 13 proteins with the SOSUI-algorithm revealed that all of them are transmembrane proteins. If the proteome-map is to be used to elucidate mitochondrial diseases, the representation of membrane-proteins is paramount and has to be further worked on.

The difficulty of the detection of membrane proteins seems to be connected to the principle of 2D-electrophoresis. This method has certain limitations to detect four kinds of proteins [Gygi et al., 2000; Nordhoff et al., 2001]:

1. hydrophobic proteins,
2. proteins of multi-protein-complexes,
3. proteins with a ve ry basic pI and
4. very small proteins

The critical point for sufficient membrane-protein separation is the isoelectric focussing in the first dimension:

1. Since membrane proteins are mostly hydrophobic proteins, they need to be solubilized to make them migrate in an electric field. This can be achieved by detergents (e.g. SDS or CHAPS). These detergents, however, affect the isoelectric point of the protein [Rabilloud, 1996]. Therefore, only urea as a protein denaturant can be added to the isoelectric focussing gel to keep the proteins in solution. CHAPS, a zwitterionic detergent, is added to the sample buffer to facilitate the solubility of hydrophobic proteins during sample preparation. However, its effect subsides as soon as the proteins enter the IEF-gel. It seems that the urea in the IEF-gel can hardly counteract the tendency of the hydrophobic proteins to aggregate. Therefore, only a few membrane-proteins maintain their solubility and migrate towards their isoelectric points. The rest of the membrane proteins seem to aggregate and do not even enter the IEF-gel.
   
2. The proteins that are part of multi-protein-complexes are also denatured and solubilized by the use of CHAPS and urea during sample preparation. When the proteins enter the IEF-gel the effect of CHAPS begins to subside and some of the proteins start to refold and partially reconstitute themselves into multi-protein aggregates that cannot migrate though the pores of the 3.5% polyacrylamide IEF-gel.
   
3. If the proteins have a very basic pI (>10) which is often the case for membrane proteins, they do not focus properly in the IEF-gel. At the basic end of the IEF-gel the pH gradient does not reach its equilibrium [Klose, 1995]. This would cause the loss of the very basic proteins unless they are captured in a higher percentage cap-gel.
   
4. Small proteins with a molecular weight below 10 kDa are difficult to detect [Klose, 1995]. This is due to the fact that the pores of the 15% second dimension SDS-PAGE-gel are too large to focus the small proteins properly [Carroll et al., 2002]. For proteins of this size a polyacrylamide-percentage of 20-22% would be appropriate. This could be achieved by pouring a 5-20% gradient-gel. This procedure is very laborious and difficult to reproduce exactly. It would make different second dimension separations difficult to compare. Beyond that, the small proteins fail to stain with Coomassie blue.
   
   [page 60↓] This is due to their low absolute peptide content despite the fact that they might be present in the same molar range as high molecular weight proteins. Even if they can be detected on Coomassie stained gels they might be difficult to identify via MALDI-TOF mass spectrometry since there are only few peptide fragments available for a peptide fingerprint analysis. In this case one has to resort to the sequence determination of a peptide fragment by MALDI-QTOF tandem mass spectrometry.

The under-representation of transmembrane-proteins on a regular 2D-gel is a severe drawback of the method. Further work has to be done to make these proteins more soluble and accessible for isoelectric focussing. [Rabilloud et al., 1996, 1999; Henningsen et al., 2002; Navarre et al., 2002]

17 proteins in my list are represented by multiple spots (2-3 in average). Most of them (n = 15) are present in pairs and are probably isoforms of the same protein.

The isoforms could be subdivided into:

• Naturally occurring isoforms that are formed by any kind of in vivo post-translational modification of proteins, such as phosphorylation, glycosylation or acetylation
  
• Isoforms caused by artifacts during sample preparation, which might be generated by the reaction of unpolymerized acrylamide with SH-groups or by oxidation of methionin residues during sample preparation.

These modifications do not change the molecular weight of a protein so much that a difference in molecular weight might be detected in the second dimension run. However, they might well cause a shift of the isoelectric points either to the acidic or to the basic side.

I also found proteins that were present in multiple spots and differed considerably in molecular weight and/or isoelectric point: COXA (spots 5 and 31), mitofilin (spots 20 and 21), and actin-beta (spots 11,12, 26 and 114). The large differences in molecular weight could be caused by differential splicing of the same precursor-mRNA. Proteolysis during the whole procedure might be another possible reason.

Two groups have established a human mitochondrial proteome database before. One is Rabilloud et al. (1998 and 2002) who investigated mitochondria from human placenta. The other group is Fountoulakis et al. (2003) who analyzed mitochondria from a human neuroblastoma cell line IMR-32. The database of Rabilloud et al. (2002) contains 68 proteins in total. The other database of Fountoulakis et al.(2003) contains 185 different proteins. I re-identified 29 proteins in Rabilloud’s database and 65 in Fountoulakis’ database (Fig. 5-1). Excitingly, in comparison to these two databases, I identified 26 new proteins. 54% of them are of definitive mitochondrial origin. 35% of them were unidentified proteins only present in EST databases. Interestingly, three of my four transmembrane proteins were identified for the first time.

Similar to my experience, Rabilloud’s and Fountoulakis’ databases do not include many transmembrane proteins either. Using the SOSUI-algorithm, I controlled all the proteins in Rabilloud’s database. Only two proteins in this database carry transmembrane domains. A [page 61↓]similar check was done with the proteins from Fountoulakis’ database, which carried a SWISS-PROT accession number. None of these proteins had a transmembrane domain. None of the mtDNA-encoded proteins were identified in both databases either. Since their experiments were carried out with the same methods, the limitation and drawbacks of the 2D-electrophoresis become clear.

	Fig. 5-1 : Comparison of my results to Rabilloud‘s and Fountoulakis‘ mitochondrial proteome databases. Compared to the 95 proteins in my database, the database of Rabilloud contains 68 proteins in total and the other database contains 185 different proteins. The shared proteins between the different databases are depicted in the overlaying parts.

The correlation between experimental and theoretical pI values in my study is not as thight as described by other authors (Bjellqvist et al., 1993; Perrot et al., 1999). However, the correlation between the experimental and the theoretical pI still tends to lie on a linear regression line (Fig. 4-16) with a correlation coefficient of R = 0.873 and a lateral dispersion of R² = 0,762. The proteins with a large difference between experimental and theoretical pI values were studied separately. They fall into the following catagories:

• Most of the mitochondrial proteins are nuclear encoded. They are synthesized in the cytosol as preproteins with a N-terminal transit sequence of about 20-30 amino acids. After import into the mitochondrion this transit-sequence is cleaved and the proteins fold into its mature conformation. Therefore, most mitochondrial proteins in my 2D-map are assumed to be mature proteins. The site of cleavage has not been determined experimentally for most of the proteins and in the databases it is generally only calculated from primary characteristics of mitochondrial import sequences [Claros et al., 1996]. In some proteins of my database (n = 7) the putative cleavage point was impossible to determine by calculation. The calculation of pI of these seven proteins therefore only takes into account the pre-protein sequences. The pI is thus wrongly shifted towards the basic side.
• For seven proteins in my database only ESTs were available in the public databases. Since ESTs might be incomplete or encode a pre-protein, the theoretical MW and pI of the EST derived proteins were unraliable. [page 62↓]
• Variation in protein expression patterns might be due to alternative splicing events on mRNA-level or due to post-translational modification. These modifications may be due to glycolysation, partial hydrolysis or phosphrylation and might cause a substantial difference between the experimental pI value and the theoretical one. This was verified by the fact that eight of those obviously deviating spots (pI-difference >1.00 pH unit) were found to be isoforms of four proteins.
• By direct measurement, the linearity of the pH-gradient in the IEF-gel could only be verified between pH 5-9. The pH-gradient of the IEF-gel is not linear any more close to its basic end. Therefore, the experimental pI values of those proteins focussing at the basic end were unreliable. This results in a pI-difference larger than 1.00 pH unit for almost all the basic proteins whose theoretical pI is larger than pH 9.00.
• Preparation artifacts (e.g. partial hydrolysis by proteases) might cause protein degeneration. The ensuing fragments run differently on the 2D-gel than the intact proteins and might be as well a reason for lager deviations between the theoretical and the experimental pI-values.

The correlation of experimental and theoretical molecular weights is better (correlation coefficient: R = 0.941) than that of the isoelectic points. The values lie on a regression line with a lateral dispersion of R²=0.885. In my study I used a mass reference marker set which only spanned the range betweeb 17.5 kDa and 76 kDa. Therefore, the MW of the proteins could not be determined with certainty when they run below 17.5 kDa or above 76 kDa. It is exactly these proteins which deviated most from the regeression line (Fig. 4-16).