Sample Stability and Protein Composition of Saliva: Implications for Its Use as a Diagnostic Fluid

Saliva is an easy accessible plasma ultra-filtrate. Therefore, saliva can be an attractive alternative to blood for measurement of diagnostic protein markers. Our aim was to determine stability and protein composition of saliva. Protein stability at room temperature was examined by incubating fresh whole saliva with and without inhibitors of proteases and bacterial metabolism followed by Surface Enhanced Laser Desorption/Ionization (SELDI) analyses. Protein composition was determined by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) fractionation of saliva proteins followed by digestion of excised bands and identification by liquid chromatography tandem mass spectrometry (LC-MS/MS). Results show that rapid protein degradation occurs within 30 minutes after sample collection. Degradation starts already during collection. Protease inhibitors partly prevented degradation while inhibition of bacterial metabolism did not affect degradation. Three stable degradation products of 2937 Da, 3370 Da and 4132 Da were discovered which can be used as markers to monitor sample quality. Saliva proteome analyses revealed 218 proteins of which 84 can also be found in blood plasma. Based on a comparison with seven other proteomics studies on whole saliva we identified 83 new saliva proteins. We conclude that saliva is a promising diagnostic fluid when precautions are taken towards protein breakdown.


Introduction
Saliva is a plasma ultra-fi ltrate that includes specifi c salivary proteins produced by three major salivary glands (parotid, sub-mandibular and sub-lingual) and other smaller glands (Baum, 1993). Salivary glands produce around 750 ml of fl uid each day (Chicharro, 1998). After secretion in the mouth cavity, the fl uid is mixed with bacteria, lining cells, nasal secretions and bronchial secretions and is termed whole saliva (Kaufman, 2000;Kaufman, 2002). Whole saliva is easy to collect in a non-invasive way. This makes saliva an attractive alternative to blood testing (Kaufman, 2002;Lawrence, 2002). Compared to blood sampling, whole saliva collection requires no specially trained personnel, can reduce discomfort and anxiety and may simplify serial sample collection. Saliva tests are also safer than blood tests regarding the risk for hepatitis and HIV. As a diagnostic fl uid, saliva has been studied in pilot experiments for several pathological conditions, such as celiac disease (Lanander-Lumikari, 2000), rheumatoid arthritis (Helenius, 2005), HIV (Holmstrom, 1990;Malamud, 1992;Matsuda, 1993;Frerichs, 1994), diabetes mellitus (Belazi, 1998;Lopez, 2003), preterm birth (Heine, 2000;Ramsey, 2003), breast cancer (Streckfus, 2005;Streckfus, 2006), sjögren's syndrome (Ryu, 2006) and for evaluation of hematopoietic stem cell transplantation (Imanguli, 2007). Saliva composition is infl uenced by several factors, e.g. circadian rhythms, oral health status and exercise (Dawes, 1993;Chicharro, 1998) but also micro organisms and proteases may have a considerable effect on sample stability/protein degradation. Before saliva can be used as a diagnostic fl uid for protein markers in the clinic, its stability should be determined. At present there are only three studies on protein stability in saliva samples (Morris, 2002;Ng, 2003;Schipper, 2007). Two of the studies report on the stability of specifi c proteins i.e. IgA, Lysozyme (Ng, 2003) and IgG (Morris, 2002). One recent study determined overall protein stability of saliva samples stored on ice, at −20 °C and at −80 °C (Schipper, 2007). In the current study we evaluated in detail the overall protein stability of saliva at room temperature over the fi rst four hours after sample collection since this is a critical period where protein breakdown could be expected. The effect of sample handling, inhibition of bacterial growth and inhibition of protease activity on saliva protein stability was examined by comparative profi ling with Surface Enhanced Laser Desorption/Ionization Time of Flight Mass Spectrometry (SELDI-TOF-MS) (Merchant, 2000). In addition we studied whole saliva composition. Whole saliva protein composition has been studied using different proteomic strategies (Huang, 2004;Vitorino, 2004;Wilmarth, 2004;Hu, 2005;Guo, 2006;Walz, 2006). Xie et al. (Xie, 2005) identifi ed 437 proteins in saliva using free fl ow electrophoreses. Guo et al. (Guo, 2006) could identify 1381 proteins employing a capillary electrophoresis approach. However, because of the complexity of whole saliva, each proteomics strategy leads to partial overlapping subsets of saliva proteins (Guo, 2006). Therefore, different proteomics strategies contribute to a comprehensive view of the whole saliva proteome. We analyzed whole saliva composition by fractionating saliva proteins on SDS-PAGE followed by LC-MS/MS analyses of digests from cut-out sections of the gel lane. This proteomics approach has not been applied to saliva before. The results of this approach are compared to previous proteomics studies on whole saliva and discussed in terms of protein origin and function.

Saliva collection
Whole saliva was collected from healthy subjects, four male and three female, between 08:00 a.m. and 10:00 a.m. after overnight fasting, to minimize the infl uence of circadian rhythms and food debris. Subjects were asked to rinse their mouths with water and discard this before sample collection. Saliva was allowed to accumulate in the fl oor of the mouth. The accumulated saliva was then spit into a polypropylene test tube and this was repeated until enough saliva was collected (Navazesh, 1993). During the collection process the sample tubes were kept on ice.

Sample pretreatment
Samples were processed according to Hu et al. (Hu, 2005). Briefl y, samples were centrifuged for 5 minutes at 1300 g at 4 °C. The pellet was discarded (debris) and the supernatant was centrifuged for 15 minutes at 14000 g at 4 °C. After centrifugation, the supernatant was stored at −20 °C until analysis. Samples were analyzed the same day.

Sample stability studies
In the fi rst experiment, sample stability was determined in saliva obtained from seven healthy volunteers (four male, three female). Freshly collected saliva samples were either directly processed (time point 0) or left at room temperature for four hrs before processing. Aliquots of the 0 and 4 hrs time points were then analyzed by SELDI-TOF-MS in duplicate (see below).
In a second experiment a unique saliva sample freshly collected from one healthy male volunteer was divided in 3 aliquots of 1.2 ml and incubated for 0, 0.5, 1 and 4 hrs at room temperature with either a) 40 μl of 100 mM sodium azide to inhibit bacterial activity, b) protease inhibitors: 60 μl of 2 mg/ml PMSF in 2-propanol, 1.2 μl of 1 mg/ml leupeptin in water, 12 μl of 100 mM EDTA, or c) no additives (control). Water was added to a fi nal volume of 1.273 ml for all 3 conditions. At each time point an aliquot was taken and treated as described above (sample pre-treatment section). Thereafter, samples were analyzed in triplicate by SELDI-TOF-MS for protein profi ling.
CM10 weak cation exchange proteinchip arrays (Ciphergen biosystems, Fremont, CA, U.S.A.) were assembled in a 96 well bioprocessor and the spots were washed two times with 200 μl binding buffer (100 mM NH 4 Ac pH 4.0, 0.05% Triton X-100) for 5 minutes with vigorous shaking. After removing the buffer from the wells, 90 μl binding buffer and 10 μl saliva sample were randomly applied to the spots (in duplicate or triplicate as detailed above). Samples were allowed to incubate for 30 minutes with continuous shaking. Then, they were removed and spots were washed 3 times with 200 μl binding buffer for 5 minutes and once with 200 μl de-ionized water for 5 minutes. The chips were removed from the bioprocessor and air-dried for 15 minutes, followed by two additions (1 μl each) of a 20% solution of CHCA prepared in 50% ACN and 0.5% TFA. Spots were analyzed using the ProteinChip Reader (model PBS II, Ciphergen Biosystems). The mass spectrometer was calibrated using the All-in-One peptide calibration kit (Ciphergen Biosystems) with a focus mass of 6000 Da. Spectra from the saliva samples were collected with the proteinchip software 3.1 (Ciphergen Biosystems) in the mass range 1-20 kDa. Laser intensity was 190, ionsource voltage 20000 V and detector voltage 2150 V. Cluster analysis was performed by Ciphergen Express 3.0 software (Ciphergen Biosystems): a) between samples collected at 0 hr and 4 hr, combining spectra of all seven volunteers measured in duplicate (28 spectra in total) and b) between different time points (0 hrs, 0.5 hrs, 1 hr and 4 hrs) for every condition (control, in presence of azide and in presence of protease inhibitors) measured in triplicate. Before cluster analyses, spectra to be compared were selected, the baseline was subtracted and profi les were normalized using total ion current. Peaks with a signal to noise ratio higher than 5 were selected and were clustered with peaks with similar masses (mass deviation 0.3%) in other profi les with signal to noise ratios higher than 2. The percentage of spectra in which a peak must appear in order to form a cluster was set to 20%. Signifi cant differences (p Ͻ 0.05) in peak height of particular masses were calculated.

Saliva protein composition
A saliva sample freshly obtained from a healthy volunteer was processed immediately after collection as described in the sample pretreatment section. 10 μl of processed saliva were mixed with NuPAGE LDS sample buffer (Invitrogen, Carlsbad, CA, U.S.A.) according to standard protocol from the manufacturer. SDS-PAGE was then performed on a NuPAGE 12% Bis-Tris gel (Invitrogen) run at 200 V for 50 minutes with MES buffer (Invitrogen). Proteins were visualized with Coomassie staining. For protein identifi cation, the whole lane was excised in 30 bands. Each band was cut into small pieces and stored at −20 °C until analysis. Then, they were washed in water and dehydrated in ACN. Reduction was performed by covering gel pieces with 10 mM DTT in 100 mM ammonium bicarbonate for 1 hr at 60 °C. DTT solution was then replaced by 55 mM iodoacetamide in 100 mM ammonium bicarbonate and gel pieces were incubated at room temperature in the dark for 45 minutes. After washing in water and dehydration in ACN, 0.1 μg of trypsin (in 50 mM ammonium bicarbonate) was added and gel pieces were allowed to rehydrate on ice for 20 minutes. Digestion was carried out overnight at 37 °C. Peptides were extracted by treating the gel pieces with 0.1%FA for 30 minutes with continuous shaking. Peptide mixtures were then stored at −20 °C until LC-MS/MS analysis was performed.
Separation of the resulting tryptic peptide mixtures was performed by nano-scale reversed-phase LC-MS/MS. The Agilent 1100 nanofl ow/capillary LC system (Agilent, Paolo Alto, CA, U.S.A.) was equipped with a trapping column (5 × 0.3 mm C 18 RP) (Dionex/LC Packings, Amsterdam, The Netherlands) and a nanocolumn (150 × 0.075 mm, C 18 Pepmap) (Dionex/LC Packings). Peptides mixtures were injected into the trapping column at a fl ow rate of 10 μl/min (3%ACN/0.1%TFA). After 10 minutes the trapping column was switched into the nanofl ow system and the trapped peptides were separated using the nanocolumn at a fl ow rate of 0.3 μl/min in a linear gradient elution from 95%A (3%ACN/0.1%TFA) to 50%B (97%ACN/ 0.1%TFA) in 70 minutes, followed by an increase up to 80%B in 5 minutes. The eluting peptides were on-line electrosprayed into the QStar XL Hybrid ESI Quadrupole time-of-flight tandem mass spectrometer, ESI-QTOF-MS/MS (Applied Biosystems, Framingham, MA; MDS Sciex, Concord, Ontario, Canada) provided with a nanospray source equipped with a New Objective ESI needle (10 μm tip diameter). Typical values for needle voltage were 2 kV in positive ion mode. The mass spectrometer was set to perform data acquisition in the positive ion mode, typically with a selected mass range of 300-1500 m/z. Peptides with +2 to +4 charge states were selected for tandem mass spectrometry, and the time of summation of MS/ MS events was set to be 2 s. The three most abundant charged peptides above a 30 count threshold were selected for MS/MS and dynamically excluded for 60 s with 50 amu mass tolerance.
ProID software (Applied Biosystems) was used to identify proteins from the mass spectrometric datasets according to UniProt database (May 2005, 181,000 entries). Mass tolerance was set to 0.15 Da (MS) and 0.1 Da (MS/MS) and carboxamidomethylation and methionine oxidation were chosen as modifi cations for database search.

Sample stability
The stability of saliva at room temperature after sample collection from four male and three female volunteers was evaluated. Freshly collected saliva samples were either directly processed (time point 0) or left at room temperature for 4 hrs and processed as described in the sample pretreatment section. Aliquots taken at the two time points were then spotted in duplicate on CM10 weak cation exchange chips and protein profi les were obtained by SELDI-TOF-MS. Representative spectra obtained from the 1 to 10 kDa range are shown in Figure 1. In this mass range degradation products of larger proteins can be expected. When comparing the protein profi les of fresh samples (0 hrs) of the seven volunteers (A-G) it is evident that there is already considerable variation, especially in the mass range of 1 to 5 kDa between the different individuals. This may be due to biological variation and/or may indicate different degrees of protein degradation between individuals. Also over the period of 4 hrs at room temperature many changes in the spectra can be observed (Fig. 1). To fi nd common peaks that were changed over the 4 hrs period in all seven samples we performed a cluster analyses on the acquired spectra. In total 11 differences were detected and listed in Table 1 together with their fold change in peak intensity between the two conditions. Most peptides are decreased in abundance at 4 hrs, probably because they are further degraded into single amino acids during this period. However, 3 peptides with masses 2937 Da, 3370 Da and 4132 Da are increasing in time which indicates that they are relatively stable breakdown products of larger proteins. Although SELDI technology allows rapid comparison of sample composition, protein/peptide identification is troublesome because it involves purifi cation of each degradation product. Therefore, we attempted to identify the 2937 Da breakdown product by direct SELDI-MS/MS which is possible for peptides with masses below 3000 Da. However, the 2937 Da peptide could not be fragmented by MS/MS even with the highest energy settings and argon as collision gas. This indicates that it is very stable, possibly due to a high degree of post-translational modifi cations such as glycosylation which may also explain its stability in vivo. As Figure 2 indicates, the three marker peptides can already Da. Protein profi les were generated using CM10 proteinchips and 100 mM NH 4 Ac pH 4.0 as binding and washing buffer. CHCA was used as matrix.
be detected in "fresh" samples (0 hrs), although with lower intensities. This indicates that breakdown already starts during sample collection and suggests that these markers may be useful indicators of protein breakdown in saliva samples. We examined the protein degradation in saliva in detail to obtain more knowledge on the time frame of the degradation process and whether it was possible to inhibit degradation. Protein degradation could be caused by bacteria in the mouth cavity and/or by proteases present in saliva. Therefore, we studied the infl uence of sodium azide, an inhibitor of bacterial energy metabolism, and of a protease inhibitor cocktail consisting of PMSF, leupeptin (both serine and cysteine protease inhibitors) and EDTA, an inhibitor of metallo-proteases. Saliva samples were incubated for 0, 0.5, 1 and 4 hrs in the presence and absence of the above mentioned inhibitors. Subsequently, protein profi les were generated (Fig. 3) and compared for differences by cluster analyses as described above. In Figure 4 the number of signifi cant differences is depicted for the different conditions and the different time points compared to 0 hrs (control). For saliva without inhibitors already 19 differences were observed in the fi rst 30 minutes incubation. Thereafter, the number of differences stabilizes. This can be explained by assuming that equilibrium has been reached at this point between the formation of peptides from the breakdown of larger proteins and the degradation of these peptides into single amino acids which are in the mass range of the matrix peaks and therefore are not detected. At 4 hrs 26 differences were observed compared to 0 hrs. This indicates that protein degradation in saliva is a relatively rapid process. It is also clear from Figure 4 that protein breakdown is almost not affected by the addition of azide to the samples, indicating that bacterial metabolism is not contributing much to the protein degradation process, at least for the fi rst hour after collection. The protease inhibitor cocktail is more effective in slowing down the degradation process ( Fig. 4). At 4 hrs, 19 differences were observed in the presence of protease inhibitors compared to 26 differences in the control sample. Nevertheless, protein degradation is still substantial with the inhibitor cocktail used. Different inhibitors or combinations of inhibitors need to be evaluated to determine their effectiveness.

The saliva proteome
To determine the saliva composition, saliva proteins were fractionated by SDS-PAGE (Fig. 5).
The whole lane was sliced into 30 bands and digested by trypsin. The digests of the bands were subjected to LC-MS/MS for protein identifi cation, as described in detail in the Materials and Methods section. In total we identifi ed 218 proteins, 182 with 99% confi dence and 36 with 95% confi dence. A complete list of identifi ed proteins is shown in Table 2. Proteins were classifi ed into 12 functional categories based on information from Swiss Prot, Source and Human Protein Reference Database. Clusters were determined using S/N Ͼ 5 (fi rst pass) and S/N Ͼ 2 (second pass). Differences were considered signifi cant if p Ͻ 0.05.  Table 1.
For each protein also the functional category is listed. In Figure 6 an overview of the different categories is given. The largest category (19.2%) consists of enzymes involved in metabolism, mainly in carbohydrate metabolism (12.8%). This includes enzymes such as α-amylase, lactate dehydrogenase, malate dehydrogenase and fructosebiphosphate aldolase. Another important category (17.9%) includes proteins that are involved in immune response and defense against bacteria. In this group there is a large cluster of IgG chains besides antibactericidal peptides such as dermcidin and bactericidal permeability-increasing protein. Also many proteins from bacterial origin were identifi ed (11% of total). 10.6% of the proteins are involved in degradation. Six proteases were identifi ed in this group e.g. kallikrein, cathepsin D and lysozyme C. Thirteen protease inhibitors are also part of this category such as cystatins, alpha-2macroglobulin and TIMP-1. The proteases are likely to contribute to the rapid breakdown of saliva proteins that was described above. Also many structural proteins (14.7% of total) were found which are probably derived from cells lining the mouth cavity together with other intracellular proteins that were identifi ed. The transport proteins (8.3% of total) are mainly serum-derived such as albumin, apolipoprotein A-1, transferrin, and ceruloplasmin. Minor categories of proteins are signaling (5.5%), protein modifi cation (4.6%), cell growth and differentiation (2.3%), cell adhesion (3.7%), and proteins involved in maintaining redox status (2.3%). We also compared our results, listed in Table 2, to the HUPO plasma proteome initiative list of 3020 plasma proteins, identifi ed with at least two peptides by LC-MS/MS (www.bioinformatics. med.umich.edu/hupo/ppp). According to the

Number of differences
Time (hrs) Figure 4. Number of signifi cant differences in peak intensity between the different conditions (• control, + azide and ▲protease inhibitors) and the different time points compared to 0 hrs. Differences were calculated from the spectra (Fig. 3) for the mass ranges of 1000 to 20000 Da using cluster analyses of triplicate measurements of the samples. Clusters were defi ned using S/N Ͼ 5 (fi rst pass) and S/N Ͼ 2 (second pass). Differences were considered signifi cant if p Ͻ 0.05.     HUPO list, 84 proteins (38.5%) identifi ed in the current study and indicated in Table 2 are also found in plasma. To determine the relevance of the identifi ed proteins, we compared our results to seven other proteomics studies (Huang, 2004;Wilmarth, 2004;Vitorino, 2004;Hu, 2005;Walz, 2006;Guo, 2006) on whole saliva composition. Figure 7 shows that there is only partial overlap with our study. Based on this analyses 83 new saliva proteins were identifi ed in our study which are indicated in Table 2.

Discussion
There is growing interest in using saliva as a diagnostic fl uid because of its relatively simple and non-invasive collection procedures. A prerequisite for measuring diagnostic protein markers in saliva is that these proteins are stable in saliva. In this study we show that relatively rapid protein degradation occurs in whole saliva samples at room temperature. We developed a SELDI-based assay to quickly monitor sample integrity. With this assay we show that protein degradation in saliva at room temperature is rapid and starts already during sample collection and handling. Three degradation products with masses of 2937 Da, 3370 Da and 4132 Da were discovered that can be used to monitor the degradation process and to determine the quality of a saliva sample before protein analyses. These markers increase 2 to 7-fold over a period of 4 hrs storage at room temperature, suggesting they are stable breakdown products of larger proteins. The proteome analyses indicates that there are at least six proteases present in saliva (see Table 2) which are probably involved in the observed protein degradation. However, also 13 protease inhibitor proteins were identifi ed which may counteract protease activity. Nevertheless, the overall balance is clearly in favour of degradation. Protein breakdown in saliva could be partly inhibited by a protease inhibitor cocktail targeting serine, cysteine and metallo-proteases. Also in another study, that investigated storage of saliva samples at different temperatures, only partially prevention of degradation was observed with a different inhibitor cocktail (Schipper, 2007). More research is clearly needed to fi nd more effective protease inhibitor cocktails to prevent degradation. A complicating factor in such studies will be that many protease inhibitors are peptides themselves or covalently bind to proteins thereby changing their masses. Both will interfere with proteomics measurements in biomarker discovery studies but may not interfere with ELISA-based measurements of individual proteins. Based on the results of our study we recommend to freeze samples immediately after collection, e.g. in liquid nitrogen, to minimize protein breakdown. Sample processing at 4 °C as well as the use of protease inhibitors can help to reduce degradation. Based on a study by Schipper et al. (Schipper, 2007) long time storage at −80 °C is recommended. Several different strategies have been employed to analyze the saliva proteome such as 2 D gel (Huang, 2004;Wilmarth, 2004;Vitorino, 2004;Hu, 2005;Walz, 2006), capillary iso-electric focusing (Guo, 2006), and free fl ow electrophoresis . Our approach was to fractionate saliva proteins by SDS-PAGE followed by LC-MS/MS for protein identifi cation. Overall 218 proteins were identifi ed by this proteomics strategy, not applied to saliva before. From the identified proteins we deduced the main functions. These are: carbohydrate-breakdown, immune response/ defence against bacteria, and protein degradation. By comparing our results with seven previous proteomics studies (Fig. 7) on whole saliva composition we fi nd only partial overlap with our study. 83 new saliva proteins from our study which were not previously identifi ed are indicated in Table  2. These results show that with each proteomics strategy, partial overlapping subsets of saliva proteins are identified. Therefore, different proteomic approaches will contribute to a more comprehensive view of the saliva proteome. Many of the identifi ed proteins are also found in plasma. Comparison with the HUPO plasma proteome database learned that 38.5% of the identified proteins can also be found in plasma. This relatively high percentage of plasma proteins in saliva illustrate the possibilities for use of saliva as an alternative to blood for diagnosis and biomarker  (Walz, 2006) (E), Huang et al. (Huang, 2004) (F), Wilmarth et al. (Wilmarth, 2004) (G), Guo et al. (Guo, 2006) (H). Only part of the data of the study by Guo et al. is publicly available and was used in this comparison. discovery. However, protein breakdown in saliva samples poses a serious problem for quantitative measurements. We conclude that saliva can be a promising diagnostic fl uid when precautions are taken towards protein breakdown.