Introduction

Two-dimensional gel electrophoresis with immobilized pH gradients (2-DE) [1, 2] is the most widespread analytical technique for proteomics as it can be routinely applied for mapping and quantitative profiling of intact proteins from complex mixtures. Incremental improvements in 2-DE technology [1, 2], new detergents [3], new protein staining methods [4], and advances in sample pre-fractionation [5] have alleviated if not overcome most of former shortcomings of the 2-DE approach with respect to resolution, dynamic range, and display of very acidic and/or basic as well as highly hydrophobic proteins. One remaining limitation, the under-representation of membrane proteins in conventional 2-DE, can be addressed—at the expense of resolution—by alternative two-dimensional methods such as BAC/SDS [6, 7], CTAB/SDS [8] or SDS/SDS gel electrophoresis [9]. In contrast to one-dimensional gel electrophoresis, two-dimensional separations usually lead to pure proteins or at least protein mixtures of relatively low complexity. After in-gel digest with trypsin, these proteins can be identified by direct analysis of the proteolytic peptides with matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS). Here, the sequencing capability of modern TOF/TOF mass spectrometers [10–13] makes protein identifications by MALDI-TOF MS/MS more reliable than those obtained by peptide mass fingerprinting alone [14, 15] and extends the method to the identification of proteins in mixtures, post-translationally modified proteins, and small proteins. Thus, the potential to obtain peptide mass and sequence information together with the high sample throughput and the ease of automation qualifies MALDI-TOF MS as the method of choice for the identification of gel-separated proteins especially in high-throughput proteomics projects [16, 17].

It is obvious that gel-based large-scale analyses of proteomes require automation at all levels of the workflow [18]. In addition to the more general advantages of laboratory automation (saving of manpower, no mix-up of samples), this task is of particular importance in proteomics as it helps to reduce keratin contamination from human skin and hair which otherwise interferes with the identification of the target proteins. As a labor-intensive and time-consuming process, the in-gel digest of proteins is often the bottleneck in the proteomics workflow. Within automation of the in-gel digest process, especially liquid handling is technically challenging as often low volumes of liquids have to be dispensed to and aspirated from the gel plugs without touching them, which otherwise may lead to clogging of the pipette tip and/or loss of the gel plug. Thus, most of the systems available to date use multi-well plates with pierced bottoms or filter-bottoms to enable removal of liquids without pipetting. Removal of liquids is aided either by a nitrogen purge applied through a special sealing membrane via a dual needle design [19, 20] or by simple application of vacuum. However, although automated systems for in-gel digest have already been developed for about 10 years [16, 19–21] and different workstations have become commercially available (see overview in [22]), the common consent is that setup and validation of these systems is still far from being trivial (see for example the web-based discussion forum of the Association of Biomolecular Resource Facilities (http://www.abrf.org).

In our laboratory, automated in-gel digest is performed by using a commercially available robotic liquid-handling system equipped with a vacuum station. With the objective of reaching the same high performance of manual low-throughput in-gel digest procedures, we have modified this robotic system by implementing a novel stack-type digestion device composed of a base frame, a digestion plate holder, a 96-well plate with laser-cut holes, and a silicone-sealed removable lid. This device is fully compatible with robotic handling and allows for vacuum-aided removal of reagents as well as collection of the extracts containing the proteolytic peptides without manual intervention. Validation of the modified robotic system is presented on the basis of low picomole- to femtomole-level protein samples either processed in a fully automated manner or, for comparison, by using manual procedures.

The same robotic system is also used for subsequent sample preparation for MALDI-TOF MS according to the thin-layer affinity method originally described by Gobom et al. [23]. For coating of the pre-structured positions on the MALDI sample support by alpha-cyano-4-hydroxycinnamic acid (CHCA), we have developed a novel motor-operated matrix application device. The matrix preparation obtained with this device was tested for its applicability to the automated acquisition of both peptide mass fingerprints (PMF) and fragment ion spectra.

Apart from introducing the novel hardware components which facilitate automation of in-gel digest and sample preparation for MALDI-TOF MS, the rationale of the present study is also to provide the most detailed performance data so far of an automated protein identification workflow. The results presented here can be readily reproduced by other laboratories and used for benchmarking of automation with respect to in-gel digest and sample preparation. Based on our experience, this should be of great value especially during the early setup and validation phase of these automated procedures.

Materials and methods

Gel electrophoresis

Dilutions of LMW marker (product code 17-0446-01, GE Healthcare) were separated by SDS PAGE on pre-cast NuPAGE 4–12% Bis-Tris gels (Invitrogen) using a MES buffer system according to the manufacturer instructions. Gels were run under non-reducing conditions. Proteins were visualized using a colloidal Coomassie staining with Coomassie Brilliant Blue G-250 according to Neuhoff et al. [24]. Gels were documented with a near-infrared fluorescence imager (Odyssey system; LI-COR) allowing for sensitive and quantitative protein detection on the basis of the Coomassie fluorescence that can be recorded in the 700 nm channel [25]. Gel plugs were excised manually with an OneTouch Plus spot picker of 1.5 mm inner diameter (The Gel Company, San Francisco, CA).

Robotic system

The robotic liquid handling system Genesis ProTeam 150 Advanced Digest (Tecan) was used for the automated in-gel digest and sample preparation for MALDI-TOF MS according to protocols provided by the supplier. Briefly, our system is equipped with an eight-channel low-volume pipetting arm for liquid dispensing and with a vacuum station for liquid removal through laser-cut holes of the digestion plates. Furthermore, the system contains a robotic manipulator arm for transport of movable components such as multi-well plates over the work table. The entire robotic system is kept under a wall-mounted cover built in-house to protect the samples from dust and thereby from keratin contamination. The novel stack-type digestion device introduced in the present study (see Results and discussion) was designed to be fully compatible with the system’s hardware and could be therefore easily implemented into existing procedures by adapting the flexible operating software of the instrument.

In-gel digest

For the automated in-gel digest, the gel plugs were placed into polypropylene 96-well plates with laser-cut holes (product code SP-0954, Abgene) allowing for vacuum-aided removal of liquids. The novel stack-type digestion device was assembled and inserted into the robotic liquid handling system. The manual in-gel digest was performed in V-shape polypropylene 96-well plates (product code 651201, Greiner) and removal of liquids was performed by manual pipetting. In the case of both automated and manual digest, the extracts containing the proteolytic peptides were collected in standard polypropylene 96-well plates (product code AB-0800, Abgene). Porcine sequencing grade modified trypsin (Promega) was used.

The individual steps of the in-gel digest protocols are summarized in Table 1. Briefly, after destaining (Table 1; step 1) and dehydration (step 2), the gel plugs were dried at 50 °C without a lid (step 3) to ensure efficient up-take of the reducing agent (5 mM DTT). After reduction (step 4), excess DTT was removed. For alkylation of the protein sulfhydryl groups, an iodoacetamide working solution (22.5 mM) was prepared fresh from a cooled stock solution (450 mM in water) prior to dispensing and added to the gel plugs. After incubation in the dark (step 5) and a short wash (step 6), gel plugs were dehydrated (step 7) and dried (step 8) as described above to ensure efficient up-take of trypsin. After cooling (step 9), the gel plugs were rehydrated for 15 min with 2 μl trypsin working solution (25 ng/μl) prepared fresh from a cooled stock solution (200 ng/μl in 1 mM hydrochloric acid) prior to dispensing (step 10). Due to the lack of a corresponding cooling position for the stack-type digestion device, cooling (step 9) and rehydration (step 10) within the automated protocol were performed at room temperature rather than on ice although the latter is recommended for minimizing trypsin autolysis. After addition of digest buffer (step 11), the cleavage was allowed to proceed for 2 h at 45 °C (step 12). The digest was stopped by acidification with 0.5% TFA/0.1% OGP (step 13) and the peptides were extracted for 30 min while shaking (manual protocol) or for 60 min by means of passive elution (automated protocol) due to the lack of plate shaker on the robotic system’s work table.

Table 1 Individual steps of the automated (A) and the manual (M) in-gel digest protocol
Full size table

Matrix and sample preparation for MALDI-TOF MS

Microcrystalline layers of CHCA were prepared on prestructured MALDI sample supports followed by application of the samples on the basis of the thin layer affinity method [16, 23, 26]. Briefly, 10 mg CHCA (Fluka) were dissolved in 90 μl tetrahydrofuran (THF) and mixed with 5 μl 1 M citric acid in 0.01% TFA and 5 μl 0.01% TFA according to a recently described protocol [27]. After sonication and centrifugation, 70 μl of the supernatant were applied to a prestructured MALDI sample support (AnchorChip 600/384, Bruker Daltonics) and coating of all sample positions was achieved with the novel motor-operated matrix application device within seconds (see Results and discussion).

Onto these pre-coated positions of the MALDI sample support, 2 μl of the extracts containing the proteolytic peptides were deposited using the same robotic liquid handling system as for the in-gel digest. After 3 min incubation, the liquid was aspirated. For washing, 4 μl 0.1% TFA were deposited on the sample positions and immediately aspirated. Similarly, 1 μl of Peptide Calibration Standard II (Bruker Daltonics, Germany) diluted in 0.1% TFA/0.1% OGP was deposited onto the calibrant positions of the prestructured MALDI sample support and processed as described above. The manual sample application procedure was identical with the only difference that the peptide solutions were not aspirated prior to the TFA wash.

Mass spectrometry

Mass spectra were acquired on an Ultraflex I MALDI-TOF/TOF mass spectrometer [12] equipped with panoramic mass range focussing and high-precision calibration technology (Bruker Daltonics, Germany). For peptide mass fingerprinting, positively charged ions in the mass-to-charge (m/z) range of 500–4,000 were analyzed automatically in the reflector mode. Sums of 25 single-shot spectra (600 spectra in total) were recorded from at least six different sample spot positions using the fuzzy logic control of the FlexControl 2.4 operation software (Bruker Daltonics, Germany). Automatic annotation of monoisotopic peptide signals in the m/z range of 800–4,000 was performed using the SNAP algorithm implemented into the FlexAnalysis 2.4 post-processing software (Bruker Daltonics, Germany). Here, background signals corresponding for example to trypsin autolysis peptides were removed from the peak list on the basis of a reference mass list and thereby excluded from being searched in the database and from being selected as precursor ion for MALDI-TOF/TOF analysis (see below). PMF spectra were calibrated externally on the basis of near-neighbor calibrant spectra automatically recorded and processed as described above.

As for the automated recording of PMF and calibrant spectra, software-controlled acquisition was used for MALDI-TOF/TOF analysis on the Ultraflex mass spectrometer operated in the MS/MS mode. Precursor ion selection for the recording of fragment ion spectra was performed by the BioTools 3.0 software (Bruker Daltonics, Germany). Fragment ion spectra (600 spectra in total) of different precursor ions selected from each PMF spectrum were automatically recorded and processed using the SNAP algorithm.

Database search

PMF and MS/MS data sets were batch-processed using the BioTools 3.0 software (Bruker Daltonics, Germany) as interface to the Mascot 2.0 software [28] licensed in-house. Database searches were performed in the Swiss-Prot database without any limitations by means of taxonomies. Carboxamidomethylation of cysteine was set as fixed and oxidation of methionine as variable modification. The monoisotopic mass tolerance was set to 100 ppm and one missed cleavage was allowed. Database searches of MS/MS data sets were performed as described above with the fragment mass tolerance set to 0.7 Da. MALDI-TOF/TOF was selected as instrument type.

Results and discussion

Test samples

For validation of the modified robotic system, low picomole- to femtomole-level protein samples were subjected to either fully automated or manual in-gel digest. A standard protein mixture (Table 2) containing six proteins of different size (molecular weight ranging from 14.4 to 97 kDa) and hydrophobicity (GRAVY index −0.555 to −0.006) rather than a single standard protein was used as test system to diminish protein-dependent effects on the performance of the protein identification workflow. Three different dilutions of the protein mixture were separated by SDS PAGE (Fig. 1a) to obtain samples spanning the concentration range of 8 ng to 1.5 μg protein per band and 29 fmol to 34 pmol protein per gel plug, respectively (Table 2). From each band, two gel plugs were excised (Fig. 1b). For best possible comparability, gel plugs originating from the same band were divided into the corresponding 96-well plates for automated or manual processing. Two gels were run with each dilution applied in duplicate resulting in 144 samples in total. Thus, two 96-well plates containing 36 samples each (duplicates from three different concentrations of six different proteins) were subjected to two independent automated digests. Similarly, the corresponding pools of samples were processed manually by using the same set of reagents as for the automated protocol to exclude reagent-dependent differences between the two procedures.

Table 2 Characteristics of the standard proteins subjected to the automated identification procedure
Full size table
Fig. 1
figure 1

Gel electrophoretic separation of the standard protein mixture. A Coomassie-stained gel is shown before (a) and after (b) the excision of gel plugs. Three different dilutions of LMW marker were applied, resulting in 7,200, 720, and 72 ng total protein load as indicated above the lanes (see Table 2 for the corresponding load of the individual standard proteins). The minor band slightly below the major ovalbumin band at 45.0 kDa was also identified as ovalbumin in a separate experiment. On the basis of the fluorescence signal (see section Material and Methods for details), the upper band at 45.0 kDa was found to contain 80% of the total ovalbumin

Full size image

Design of the stack-type digestion device

For automated in-gel digestion, a novel stack-type digestion device has been developed which is assembled by inserting the 96-well plate containing the gel plugs into the digestion plate holder (Fig. 2a). The holder fits the plate tightly enough to allow for vacuum-aided removal of liquids through the laser-cut holes. This construct is mounted on a silicone-sealed base frame (Fig. 2a) to facilitate robotic handling of the entire stack and to prevent the wells from contacting the ground which would lead to contamination and leakage. Finally, the stack is completed with a metallic lid containing a pre-sliced silicone inlay (Fig. 2a) to protect the samples from contamination and evaporation. The lid is also compatible with robotic handling and can be automatically removed to facilitate drying of the gel plugs. Due to the high self-weight of the lid, the stack is not lifted when the pipetting tips are retracted through the silicone inlay after liquid dispensing. Thus, all pipetting steps can be performed while the stack is placed on the vacuum station without the need for any additional mechanic down-holding.

Fig. 2
figure 2

Design of the stack-type digestion device. The exploded view of the device (a) shows its four components (from bottom to top): base frame (BF), digestion plate holder (DPH), 96-well digestion plate with laser-cut holes (DP), and silicone-sealed removable lid (L). The different assemblies of the stack on the vacuum station during vacuum-aided reagent removal into waste (b) and during vacuum-aided recovery of the extracts containing the proteolytic peptides into the collection plate (c) are shown in the schematic drawings. For liquid dispensing, the pipetting needles pierce through the pre-sliced silicone inlay of the metallic lid as indicated. Note that clogging of the laser-cut holes in the bottom of the digestion plate is prevented by the eccentric configuration of the two holes

Full size image

Whereas removal of reagents into the waste is readily possible with the stack placed on the vacuum station (Fig. 2b), vacuum-aided collection of the extracts containing the proteolytic peptides, the last step of the in-gel digest protocol, is usually not straightforward. The drawback is that the wells of the upper digestion plate have to dip into the wells of the lower collection plate to ensure quantitative transfer of the peptide extracts without any spill-over. Here, this requirement is met by the stack-type design of the digestion device. After the collection plate has been automatically inserted into the vacuum station, robotic handling can be also used to detach the upper part of the digestion stack from the base frame followed by placing it on the vacuum station. In this configuration, the pierced wells enter the wells of the collection plate by 7 mm which ensures safe transfer even of low volumes (Fig. 2c). Thus, the entire in-gel digest protocol (Table 1), from destaining of the gel plugs to quantitative recovery of the peptide extracts, can be performed without any manual intervention.

Validation of the automated in-gel digest

For direct comparison, gel plugs from the same protein band (see above) were processed either in an automated manner using the novel stack-type digestion device implemented into the robotic liquid handling system, or manually using a corresponding protocol (Table 1). Accordingly, the robotic liquid handling system was used to apply the tryptic peptides obtained from the automated procedure to the MALDI sample support, whereas the peptides obtained from the manual procedure were deposited and washed by manual pipetting. Both sets of samples were prepared on one MALDI sample support pre-coated with CHCA and the spectra were acquired within the same automated data acquisition. The sequence coverage as given in the Mascot output of the PMF search was used as criteria to compare the automated in-gel digest and sample preparation procedure with the manual one (Fig. 3). Thereby, it was found that the performance of the automated and the manual procedure were almost identical, possibly with some advantages of the automation for low-level samples in the concentration range around 300 fmol per gel plug and below (Fig. 3). With regard to the differences between automated and manual protocol (see Table 1 and related text in Materials and Methods), it should be noted that neither the rehydration with trypsin at room temperature nor the passive extraction of the proteolytic peptides significantly impaired the quality of automated in-gel digest by means of increased trypsin autolysis or inefficient peptide extraction (data not shown). However, we are currently implementing a cooling device into our robotic system to allow for rehydration with trypsin at lower temperature and to render this process independent of the ambient temperature.

Fig. 3
figure 3

Validation of the automated in-gel digest. Gel plugs from the same protein band were subjected to automated in-gel digest followed by automated sample preparation, and to manual in-gel digest followed by manual sample preparation. The sequence coverage obtained by the automated procedure (black bars) is displayed in comparison to those obtained by the manual procedure (white bars). Results are shown for all six standard proteins applied in three different concentration levels spanning two orders of magnitude. The bars represent the average sequence coverage obtained from two independent in-gel digests/sample preparations of duplicate samples (n=4). The error bars indicate the standard deviation

Full size image

Taken together, it was concluded that the in-gel digest and the sample preparation procedure was successfully automated. In our view, reaching the high performance of the manual procedure with an automated system is already one of the best possible scenarios, especially if one considers that visual inspection of the pipetting process and location of the gel plugs is not possible in automation. With regard to sample throughput, the time needed for 96 samples to be processed in automated in-gel digest and sample preparation was 6.5 h and 2.5 h, respectively, which results in a throughput capability of 192 samples per day.

It is worth mentioning that relatively high sequence coverage of 35–50% was obtained even from low-level samples (Fig. 3). This was even more obvious when the coverage was corrected for parts of the protein sequence which are present in the sequence entry in the database but could not be covered by the analysis. These included signal and propeptide sequences, N-glycosylated peptides (Table 2), and large tryptic peptides with m/z>4,000 which were outside of the acquisition range set for the mass spectrometer. On the basis of these considerations, the following correction factors have to be applied to the sequence coverage obtained in order to correlate it to the theoretically possible coverage: for albumin 1.04, phosphorylase b 1.06, ovalbumin 1.09, carbonic anhydrase 1.19, trypsin inhibitor 1.19, and α-lactalbumin 1.75, with the latter protein being an extreme case as only four tryptic peptides were recordable under the conditions applied. The high sequence coverage observed here is explained by certain features of the in-gel digest protocol. By comparing the Mascot output of PMF searches with carboxamidomethylation of cysteine set to either fixed or variable modification, it was found that the sequence coverage was identical (data not shown) indicating that the in-gel reduction/alkylation of the proteins was quantitative. Even for 2-DE samples usually subjected to reduction/alkylation prior to the second dimension, this is often not the case if reduction/alkylation during the in-gel digest has not been performed in a complete manner. Quantitative reduction/alkylation, however, improves the substrate accessibility for trypsin and prevents the release of disulfide-bonded peptides, both contributing to high sequence coverage. For example, the sequence coverage of albumin dropped by approximately 50% when reduction/alkylation was omitted from the in-gel digest protocol (data not shown). Similarly, identification of small proteins containing disulfide bonds such as trypsin inhibitor and α-lactalbumin by peptide-mass fingerprinting alone was nearly impossible without prior reduction/alkylation (data not shown). Finally, quantitative reduction/alkylation should be ensured when carboxamidomethylation of cysteine is set to fixed rather than variable modification in the Mascot search engine which is generally recommended to improve the score for the identification of cysteine-containing proteins.

Another and probably even more important feature of the in-gel digest protocol with respect to high sequence coverage was the addition of the nonionic detergent OGP at both the digestion and extraction step. During digestion, OGP probably serves to improve the substrate accessibility for trypsin and to solubilize the nascent tryptic peptides [29, 30]. Similarly, addition of OGP during the extraction is thought to assist the recovery of large and/or hydrophobic tryptic peptides from the gel plugs [29, 30]. This is of particular importance for the workflow presented here as organic solvents cannot be alternatively used when the samples are subsequently prepared for MALDI-MS according to the thin layer affinity method, a preparation limited to aqueous samples [23]. More generally, OGP may also help to decrease the loss of peptides due to adsorption to plastic surfaces. In addition to its beneficial effects during the in-gel digest procedure, the presence of OGP has also a significant impact on sample preparation for MALDI-MS and data acquisition. First, with regard to sample preparation according to the thin layer affinity method, OGP almost completely prevents methionine- and tryptophan-containing peptides from being oxidized during the adsorption process, an artefact otherwise frequently observed with this preparation method [23]. Thereby, a decrease in detection sensitivity for these peptides is prevented. Second, OGP is not only MALDI-MS compatible but is even able to enhance the MALDI-MS response especially of larger peptides [31]. Thus, the addition of OGP at virtually all levels of the workflow—digestion, extraction, sample preparation—is probably the key to the high sequence coverage observed in the present study. As a proof of principle, 40% sequence coverage was even obtained from sub-picomole amounts of ovalbumin, a hydrophobic protein known to be problematic with regard to peptide recovery after tryptic digestion [32, 33].

Design of the motor-operated matrix application device

The prerequisite for the thin-layer affinity method is the preparation of microcrystalline layers of CHCA on prestructured MALDI sample supports. This is easily achieved by spreading out a CHCA solution over the prestructured sample support as the hydrophilic anchor positions adsorb a small volume of the solution which results in a matrix layer after evaporation of the solvent [23]. Ideally, on the one hand, these matrix layers should be thin with a highly homogenous surface structure to improve resolution and mass accuracy in MALDI-TOF-MS, but, on the other hand, should be thick enough to prevent matrix ablation during the recording of fragment ion spectra often requiring significantly higher laser energies. Furthermore, especially when high throughput is desired, the entire matrix application process should be reproducible to ensure a reliable automated data acquisition. Apart from the CHCA concentration and aqueous content of the matrix solution, the layer thickness is dependent on the speed of dragging the matrix solution across the sample support. Therefore, a motor-operated matrix application device was designed (Fig. 4a) to move the sample support beneath a Teflon blade contacting the support only via lateral spacers. The matrix solution was applied at the Teflon blade, along which it spreads out immediately forming a liquid film. This liquid film could then be dragged across the sample support with defined and adjustable speed. Thereby, microcrystalline layers of CHCA were reproducibly obtained (Fig. 4b) which exhibit a highly homogenous surface structure (Fig. 4c) while being relatively thick and compact (Fig. 4d). At least in our hands, this was not the case when coating was achieved by manually dragging the matrix solution across the prestructured sample support using Teflon rods or blades. By microscopic inspection, the appearance of the matrix coating was found to be operator-dependent and of poor reproducibility, which often resulted in PMF spectra with impaired resolution and mass accuracy or even prevented the recording of a sufficient number of fragment ion spectra (data not shown). On the basis of these preliminary observations, the matrix preparation obtained with the motor-operated matrix application device was exclusively used in all further experiments.

Fig. 4
figure 4

Design of the motor-operated matrix application device. The key components of the device a are the movable adaptor (A) for the prestructured MALDI sample support (“target”, T), the Teflon blade (TB) with lateral spacers to drag the liquid film of CHCA solution across the sample support, and the adjustable motor (M) to move the adaptor via an electric cylinder. As shown in the inset, after evaporation of the solvent, microcrystalline layers of CHCA are formed in a highly reproducible manner (b). These matrix layers exhibit a highly homogenous surface structure (top view, c) while being relatively thick and compact (side view, d)

Full size image

Automated identification of proteins

After successful automation of in-gel digest and sample preparation for MALDI-MS, the matrix preparation obtained with the motor-operated matrix application device was integrated into the workflow and its performance in automated identification of proteins was investigated. As the availability of both peptide mass and sequence information is a prerequisite for high confidence protein identification, the matrix preparation was tested for its applicability to the automated acquisition of PMF and fragment ion spectra from the same sample position. Therefore, the two sets of peptide extracts obtained from the automated in-gel digest (36 samples each, see above) were again independently prepared on sample supports pre-coated with the motor-operated device and subjected to automated data acquisition. With external near-neighbor calibration, the average relative root-mean-square (RMS) error for all PMF spectra matched to the respective standard proteins was 27 ppm (Fig. 5). Furthermore, the low sample consumption during recording of the PMF spectra together with compactness of the matrix allowed for the recording of several fragment ion spectra from each sample spot. Under the present conditions, it was found that at least four fragment ion spectra could be reproducibly acquired which is in agreement with the limitations resulting from the use of a conventional nitrogen laser for desorption from matrix thin layers. Whereas each of the four fragment ion spectra obtained from the high- and medium-level samples matched the respective standard proteins in almost all cases, less than four spectra could be positively matched for most of the low-level samples. However, even sequence information from only a single peptide can significantly boost the confidence of protein information in comparison to peptide mass fingerprinting alone. With regard to the fragment mass accuracy, the average absolute RMS error for all fragment ion spectra matched to the respective standard proteins was 0.30 Da (Fig. 5) meeting the instrument specifications [12]. Thus, as already expected on the basis of the microscopic inspection, the CHCA layers prepared with the motor-operated matrix application device allowed for the recording of both accurate masses in the MS mode as well as several accurate fragment ion spectra in the MS/MS mode. With the objective to increase the number of fragment ion spectra to be recorded from one sample spot, the samples were prepared as before but with additional on-spot recrystallization. However, with respect to their amorphous surface structure, these samples resembled dried-droplet preparations with consequential impairment in resolution and mass accuracy of PMF spectra and disadvantages in automated data acquisition (data not shown). Furthermore, recrystallization led to a significant increase in methionine and tryptophan oxidation and thereby to a decrease in detection sensitivity for peptides containing these amino acids (data not shown). Because of these drawbacks, recrystallization was omitted from the sample preparation protocol although recording of more than four fragment ion spectra might have been possible.

Fig. 5
figure 5

Automated identification of proteins. The two sets of peptide extracts which were obtained from the automated in-gel digest and already used to generate the data in Fig. 3, were again independently prepared on CHCA layers generated with the motor-operated matrix application device. PMF and fragment ion spectra were automatically acquired under software control, postprocessed, and searched in the database using the Mascot search engine. Probability based Mowse scores as given in the Mascot output were used to generate the graphs. Protein scores from the protein summary report of the PMF searches alone (black bars) are displayed in comparison to those from the protein summary report of the combined MS/MS ion searches (white bars). The latter scores demonstrate how the confidence of protein identification is increasing when the peptide sequence information derived from up to four fragment ion spectra is considered in addition to the PMF data. The significance threshold of 66 (p<0.05) is depicted as a horizontal line. Results are shown for all six standard proteins applied in three different concentration levels spanning two orders of magnitude. The bars represent the average protein scores obtained from two independent in-gel digests/sample preparations of duplicate samples (n=4). The error bars indicate the standard deviation. The numbers within the black bars are the average relative RMS errors (in ppm) of the PMF spectra (n=4), whereas the numbers within the white bars are the average absolute RMS errors (in Da) of the fragment ion spectra (n≤16)

Full size image

Obviously, the potential of obtaining peptide mass and sequence information makes protein identifications by MALDI-TOF MS/MS more reliable than those obtained by peptide mass fingerprinting alone. Accordingly, significantly higher protein scores were obtained when the fragment ion spectra were included into the database search (Fig. 5). Thereby, even small proteins such as α-lactalbumin were reliably identified (Fig. 5) which is often not possible by peptide-mass fingerprinting alone due to the low number of peptide masses available for database search. High confidence protein identification was obtained even for low amounts of the hydrophobic protein ovalbumin (Fig. 5) usually hard to identify by in-gel digestion and subsequent peptide-mass fingerprinting. Again, mainly due to the contribution of high quality fragment ion rather than PMF spectra alone (Fig. 6), protein scores far above the significance threshold were obtained for ovalbumin in the database search.

Fig. 6
figure 6

Typical mass spectra of high-, medium-, and low-level samples of ovalbumin (45.0 kDa). a PMF spectrum corresponding to 11 pmol protein per gel plug. 86% of the total intensity in the spectrum was matched to ovalbumin peptides resulting in a sequence coverage of 57% (relative RMS error=12 ppm). Mascot PMF score was 146 (significance threshold=66). b PMF spectrum corresponding to 1.1 pmol protein per gel plug. 92% of the total intensity in the spectrum was matched to ovalbumin peptides resulting in a sequence coverage of 50% (relative RMS error=8 ppm). Mascot PMF score was 139. c PMF spectrum corresponding to 110 fmol protein per gel plug. 54% of the total intensity in the spectrum was matched to ovalbumin peptides resulting in a sequence coverage of 33% (relative RMS error=11 ppm). Mascot PMF score was 84. As the intensities of signals representing ovalbumin peptides decrease approximately by a factor of two with each dilution step (note the absolute intensity axes), the intensities of signals representing trypsin autolysis peptides such as m/z=2211.10 (labeled with T) become more dominant. d–f) Fragment ion spectra of the ovalbumin peptide 127–142 labeled in red in the PMF spectra (\({\left[ {M + H^{ + } } \right]}_{{calc}}\)=1,687.84; amino acid sequence: GGLEPINFQTAADQAR). For the sake of clarity, only the y- and b-ions are annotated and marked in blue and red, respectively. P indicates the precursor ion. d Fragment ion spectrum derived from the precursor at m/z=1,687.84 in PMF spectrum (a). 37 fragment ion signals were matched to the peptide sequence (absolute RMS error=0.18 Da) resulting in almost complete C- and N-terminal ion series. The MS/MS ion score as given in the peptide summary report of the Mascot search was 148 (significance threshold=36). e Fragment ion spectrum derived from the precursor at m/z=1,687.84 in PMF spectrum (b). 29 fragment ion signals were matched to the peptide sequence (absolute RMS error=0.28 Da) resulting in an almost complete C- and a partial N-terminal ion series. The MS/MS ion score as given in the peptide summary report of the Mascot search was 112. f Fragment ion spectrum derived from the precursor at m/z=1,687.84 in PMF spectrum (c). 19 fragment ion signals were matched to the peptide sequence (absolute RMS error=0.43 Da) resulting in partial C- and N-terminal ion series. The MS/MS ion score as given in the peptide summary report of the Mascot search was 61. Note that the most dominant fragment ion signals y3 and y12 are due to the known facile amide-bond cleavages C-terminal of Asp and N-terminal of Pro, respectively [39]

Full size image

With regard to the sensitivity of the entire process consisting of automated in-gel digest, sample preparation, and data acquisition, high confidence protein identifications on the basis of both PMF and fragment ion spectra were obtained for samples in the concentration range of as low as 100 fmol protein per gel plug. As can be concluded from the low contribution of the fragment ion spectra to the identification of proteins in the concentration range below this level (Fig. 5), one limiting factor was the lower sensitivity of the mass spectrometer in the MS/MS mode than in the MS mode, a limitation that may be overcome by new generation MALDI-TOF/TOF instruments. Furthermore, it has to be mentioned that only 10% of the extract containing the proteolytic peptides (2 μl out of approximately 20 μl total extract volume) was used to generate the data presented here, providing potential to further increase the sensitivity. On the other hand, peptide extract volumes of 10–20 μl open the possibility not only to repeat sample preparation and data acquisition, but also to use complementary mass spectrometric techniques for protein identification.

At least to our knowledge there is no other study in the literature which can be used for direct comparison with the automated workflow presented here, a task that is anyway hard to accomplish as the overall performance of automated protein identification depends on multiple factors such as protein input, sample preparation, instrumentation, etc. However, on the basis of the data from previous reports which also deal with automation in proteomics, but are only marginally related to ours [34, 35], it was concluded that the sensitivity of the entire process as presented here is superior to those of the automated workflows described so far. Finally, it is worth mentioning that our automated in-gel digest and sample preparation procedure is cost-effective with respect to consumables. Standard plastic ware is used as digestion plate and time-consuming and expensive desalting of the samples by using reversed-phase chromatography media prior to mass spectrometry is replaced by sample preparation according to the thin layer affinity method.

Conclusion

By implementing a novel stack-type digestion device into a commercially available robotic liquid handling system, we have established a cost-effective automated platform capable of performing in-gel digest, extraction of proteolytic peptides, and subsequent sample preparation for MALDI-MS. As manual intervention was not needed during this entire process, keratin contamination was minimized. The automated platform enabled high confidence protein identifications even for samples in the concentration range of as low as 100 fmol protein per gel plug. Due to the flexible design of the stack-type digestion device, this concept may be applicable with slight modifications to other robotic systems than the one used here.

With the objective of reproducibly obtaining a homogenous matrix preparation of high quality, we have also developed a motor-operated matrix application device. Using this matrix preparation, PMF spectra as well as several fragment ion spectra could be automatically acquired from the same sample spot without any further intervention such as recrystallization. Thereby, peptide mass and sequence information was available as a prerequisite for high confidence protein identifications.

Both technical innovations, the stack-type digestion device and the motor-operated matrix application device, are now essential parts of our standard application for the automated identification of proteins. Currently, we combine different protein prefractionation strategies with 2-DE to conduct various large-scale proteomics analyses of biological starting material as diverse as mouse brain tissue, frog oocytes, and plant leaves [36]. In the framework of these projects, 100–400 gel plugs including those from only faintly stained spots are manually excised from 20×20 cm two-dimensional gels after visualization by colloidal Coomassie staining. Due to sample preparation by the automated platform introduced here combined with MALDI-TOF MS/MS analysis, high confidence protein identification is now routinely achieved for 80–90% of the samples.