Selective TiO 2 Phosphopeptide Enrichment of Complex Samples in the Nanogram Range

: Phosphopeptide enrichment is a commonly used sample preparation step for investigating phosphorylation. TiO 2 -based enrichment has been demonstrated to have excellent performance both for large amounts of complex and for small amounts of simple samples. However, it has not yet been studied for complex samples in the nanogram range. Our objective was to develop a methodology applicable for complex samples in the low nanogram range, useful for mass spectrometry analysis of tissue microarrays. The selectivity and performance of two stationary phases (TiO 2 nanoparticle-coated monolithic column and spin tip ﬁlled with TiO 2 microspheres) and several loading solvents were studied. Based on this study, we developed an e ﬀ ective and robust method, based on a spin tip with a non-conventional 50 mM citric acid-based loading solvent. It gave excellent results for phosphopeptide enrichment from samples containing a few nanograms of a complex protein mixture.


Introduction
Phosphorylation is one of the most widely investigated post-translational modifications (PTMs) of proteins. This ubiquitous PTM is responsible for several biological functions-for example, signal transduction, switching on/off the functionality of a protein, or growth regulation during cancer [1]. The phosphate group is usually attached to serine, threonine, or tyrosine side chains with different abundance (pSer >> pThr >> pTyr) [2], but there has been evidence for histidine and arginine phosphorylation as well [3]. Despite its widespread presence, the overall ratio of phosphorylated proteins to non-phosphorylated proteins and the coverage of phosphosites are relatively small. The analysis is further complicated due to their low electrospray ionization efficiency compared to non-phosphorylated analogs [4]. Although mass spectrometry (MS) is an established and powerful analytical technique with its ever-growing sensitivity, low-abundance PTMs are still easily overlooked during the analysis due to sensitivity limitations. Taking all these into consideration, it is advisable to perform an enrichment/isolation step before MS measurements. The most common way is to two different stationary phases (spin tip filled with TiO 2 microspheres and a TiO 2 nanoparticle-coated monolithic column) and various sample loading solvent combinations (including trifluoroacetic acid, lactic acid, 2,5-dihydroxybenzoic acid, citric acid, heptafluorobutyric acid, and acetic acid).

Results and Discussion
We performed a detailed investigation of loading solvents providing different degrees of selectivity in TiO 2 -based MOAC systems for enriching complex samples in the nanogram range. A mixture of 500 ng HeLa tryptic digest (HeLa hereinafter) and 500 fmol synthetic phosphopeptide mix (PP mix hereinafter) was subjected to phosphopeptide enrichment. Half of the resulting sample was analyzed using nanoLC-MS/MS; therefore, theoretically, 250 ng HeLa and 250 fmol PP mix. Five parallel enrichments were performed with all the methods tested. The control sample was subjected only to C 18 clean-up. This type and amount of sample is a good model system for small amounts of complex samples such as tissue biopsies or tissue microarrays.
Two different stationary phases were investigated: spin tip filled with TiO 2 microspheres and TiO 2 nanoparticle-coated monolithic column. Five methods were investigated for the spin tip setup and two for the TiO 2 functionalized monolithic column, applying identical solvent compositions to two of the spin-tip methods. The tested methods are summarized in Table 1. One of the spin tip methods (ST-50CA) was further modified to decrease the non-specific binding of non-phosphorylated peptides (discussed separately). The enrichment performance of the tested methods is described by the number of identified PPs, selectivity, repeatability, and recovery.

Phosphopeptide Identification
Performance concerning phosphopeptide (PP) identification has been characterized using several factors: the total number of identified PPs in five parallel runs; the average number of identified PPs (average of five runs); and the enrichment factor (EF), which describes the change in the ratio of the number of PPs to the number of non-phosphorylated peptides (NPs).
Among preparation than from the control sample (no enrichment). This is likely due to insufficient binding of phosphopeptides to the stationary phase.
The best results were obtained using the ST-50CA and MON-AA methods, which both resulted in two times larger numbers of PPs identified. Among these, ST-50CA is preferable, as in this case, the enrichment factor is higher (i.e., over 90% of the NPs were removed; see Table 2). The results indicate that both the loading conditions and the TiO 2 stationary phases have a large influence on performance. Note also that while 50 mM citric acid-based loading is best with the spin tip, acetic acid-based loading performs better with the monolithic column. These results suggest that the stationary phase's morphology strongly determines the performance.
The next important parameter is the enrichment factor (EF), which describes the changes in the PP/NP ratio in comparison to that of the control sample (calculated based on average identification numbers; see Table 2). There is an extremely high EF for the ST-LA and ST-DHB methods due to the near-complete removal of the NPs; however, their overall performance is poor. The two methods applying citric acid (ST-50CA and ST-1000CA) showed outstanding results (EF = 35.1 and 77.8, respectively). This implies that citric acid is an effective displacing agent in this spin tip setup. The MON-AA method resulted in a moderate EF, implying that the NP removal of the setup should be further improved.

Selectivity
Another important parameter besides enrichment efficiency is selectivity. This can be subdivided based on the degree and site of phosphorylation and also the length, the isoelectric point, and the hydrophobicity of the peptide backbone. These comparisons are based on the total number of identified PPs.
All methods, except MON-50CA, showed an increased selectivity towards doubly phosphorylated peptides ( Figure 1). In most cases, doubly phosphorylated peptides were enriched by a factor of 1.7-4.3 more than monophosphorylated peptides. In the case of samples enriched using the ST-AA method, the ratio of doubly phosphorylated peptides increased from 1.9% (control sample) to 13.0% (Figure 1a), a 7.7-fold increase. This is also the method which identified the largest number of doubly phosphorylated peptides (altogether 17). This is over 1.9 times more than that using the otherwise most favorable ST-50CA and MON-AA methods (6 and 9, respectively); therefore, ST-AA is the method of choice for analyzing doubly phosphorylated peptides. In contrast to a recent publication [25], we observed only a moderate increase in the selectivity of the ST-1000CA method towards doubly phosphorylated peptides compared to the ST-50CA method. The cause of the difference between the two findings is presumably the different form of TiO 2 used.
Regarding phosphorylation sites, remarkable differences were observed compared to the distribution of the control sample. All enrichment methods showed 2-4 times increased selectivity towards threonine phosphorylated (pT) peptides (Figure 1b). Most methods showed an approximately 1.9-2.5-fold increase towards enrichment and identification of tyrosine phosphorylation (pY, Figure 1b).
Separations 2020, 7, x FOR PEER REVIEW 5 of 16 1.9-2.5-fold increase towards enrichment and identification of tyrosine phosphorylation (pY, Figure  1b).  The peptide backbone bearing the phosphorylation also has a marked influence on phosphopeptide enrichment. The peptide length had a relatively small but well identifiable influence on the enrichment of phosphopeptides (Figure 2a). The peptide length distribution was similar for all methods, providing increased phosphopeptide identification (ST-AA, ST-1000CA, ST-50CA, MON-AA). In all of these cases, a preferred enrichment/identification of small-size peptides was shown. This might be expected, as the influence of the phosphate group on peptide behavior is more pronounced if the peptide is smaller. The other methods provided non-conventional distributions due to the improper binding of the phosphopeptides.
The acidity of the peptide backbone (as measured by the isoelectric point) has a large influence on the selectivity of phosphopeptide enrichment (Figure 2b). In the control sample, over 92% of the PP backbones identified were acidic, and the mean pI was 4.29. The MON-50CA method conserved this distribution. In the case of all other methods, the pI distribution of PPs identified shifted significantly towards the basic range. The mean pI in the case of the most favorable ST-50CA method was 5.47, while the largest shift (mean pI = 6.59) was observed in the case of the ST-DHB method. These findings indicate that the enrichment of neutral or basic phosphopeptides is more efficient than those having an acidic backbone ( Figure 2b).
The hydrophobicity of the enriched peptides as measured by the Grand Average of Hydropathy (GRAVY) index is also a good descriptor of selectivity. Enrichment slightly favors hydrophobic peptides, but the change was only very small ( Figure 2c).
We have compared the identity of PPs identified using the two most promising methods (ST-50CA and MON-AA) to that of the control sample. This showed a fairly large variability of PPs identified using different enrichment methods ( Figure 3). As noted above (Table 2)   We have compared the identity of PPs identified using the two most promising methods (ST-50CA and MON-AA) to that of the control sample. This showed a fairly large variability of PPs identified using different enrichment methods ( Figure 3). As noted above (Table 2)

Repeatability
The repeatability of the methods can be assessed from two points of view: the variations in the PPs identified and variations in the relative amount of the various PPs identified (qualitative and quantitative aspects). Qualitative analysis in LC-MS is usually based on data-dependent analysis (DDA), which is very sensitive but has a significant chance factor when identifying compounds of low abundance. For this reason, it is typical that in parallel runs, various compounds are sometimes

Repeatability
The repeatability of the methods can be assessed from two points of view: the variations in the PPs identified and variations in the relative amount of the various PPs identified (qualitative and quantitative aspects). Qualitative analysis in LC-MS is usually based on data-dependent analysis (DDA), which is very sensitive but has a significant chance factor when identifying compounds of low abundance. For this reason, it is typical that in parallel runs, various compounds are sometimes identified and some other times not found. We have performed five parallel experiments, both in the case of the control sample and in the case of samples enriched by various methods. A qualitative measure of the repeatability is the comparison of the number of PPs which were identified in all five runs, with the number of PPs which were identified in only one run, but not in the other four. In the control (unenriched) sample, there were nearly three-times more PPs identified in one run than those PPs which were identified in all five runs (Figure 4). The enriched samples, in general, showed much better repeatability. Results of the two most promising methods (ST-50CA and MON-AA) are shown in Figure 4. Repeatability is best in the case of the ST-50CA method, where more PPs are identified in all five parallel runs than in those observed only once. This is due to two favorable circumstances: first, the enriched samples contain a larger portion of PPs. Second, and probably more importantly, the amount of non-phosphorylated peptides is significantly reduced, simplifying the sample complexity. Both effects help to diminish the negative influence of DDA analysis. Note that despite this negative feature, DDA analysis is essential for the sensitive analysis of complex samples. Further details on repeatability are given in the Supplementary Information.
The assessment of quantitative repeatability was evaluated based on the relative standard deviation (RSD) of the peak areas of four PPs mixed in low amounts (250 fmol) into the HeLa cell lysate (Table 3). In the case of the control sample (no enrichment), the average RSD was on average 24.2%, which is a typical value when analyzing complex mixtures (without any normalization). The enriched samples showed similar RSD values, indicating that sample treatment did not increase the RSD values significantly. In fact, RSD values in the case of the ST-50CA and MON-AA methods were somewhat better, on average 21.5% and 16.0%, respectively.
Separations 2020, 7, 74 9 of 15 in Figure 4. Repeatability is best in the case of the ST-50CA method, where more PPs are identified in all five parallel runs than in those observed only once. This is due to two favorable circumstances: first, the enriched samples contain a larger portion of PPs. Second, and probably more importantly, the amount of non-phosphorylated peptides is significantly reduced, simplifying the sample complexity. Both effects help to diminish the negative influence of DDA analysis. Note that despite this negative feature, DDA analysis is essential for the sensitive analysis of complex samples. Further details on repeatability are given in the Supplementary Information. The assessment of quantitative repeatability was evaluated based on the relative standard deviation (RSD) of the peak areas of four PPs mixed in low amounts (250 fmol) into the HeLa cell lysate (Table 3). In the case of the control sample (no enrichment), the average RSD was on average 24.2%, which is a typical value when analyzing complex mixtures (without any normalization). The enriched samples showed similar RSD values, indicating that sample treatment did not increase the RSD values significantly. In fact, RSD values in the case of the ST-50CA and MON-AA methods were somewhat better, on average 21.5% and 16.0%, respectively.

Improving the Selectivity of the Spin Tip Method
Based on the results and practical considerations presented above, the ST-50CA gave the best overall performance. The results discussed above suggest that reducing non-specific binding of NPs to the stationary phase may result in further improvement. With this in mind, we tested two variants:

Improving the Selectivity of the Spin Tip Method
Based on the results and practical considerations presented above, the ST-50CA gave the best overall performance. The results discussed above suggest that reducing non-specific binding of NPs to the stationary phase may result in further improvement. With this in mind, we tested two variants: in one case, we increased the TFA content of the loading solvent from 0.1% to 1.5% (ST-50CA-B method), while in the other case, we used a more complex loading solvent (0.1% TFA + 50mM CA + 0.2% HFBA + 1.5% AA, indicated as the ST-50CA-C method). Table 4 summarizes the changes in their performance compared to the original ST-50CA method. There was very little change in the number of PPs identified, but the number of NPs identified decreased significantly, especially in the case of the ST-50CA-B method. This simplified the chromatograms and improved the enrichment factor by nearly tenfold. There was no significant change in selectivity, concerning peptide length, GRAVY score, pI, phosphorylation site, and phosphorylation extent, compared to the ST-50CA method. Only a minor difference was observed among the PPs identified: 60% of PPs were identified using all the three methods, and over 70% of the PPs, were commonly identified in each pair of methods ( Figure S3 in Supplementary Information). This overlap is significantly higher than that observed with any other method pairs discussed in Section 2.2. There was only a marginal change in repeatability compared to the ST-50CA method ( Figure S5 in the Supplementary Information). In summary, the ST-50CA-B method gave the best overall performance; therefore, this is the method of choice for the phosphopeptide enrichment of complex protein mixtures in the nanogram range.
For testing the enrichment performance of the different setups, we used the mixture of 500 ng Thermo Fisher Pierce HeLa cell line tryptic digest (Unicam Plc, Budapest, Hungary) and 500 fmol Enolase MassPrep Phosphopeptide mix (Waters Hungary, Budapest, Hungary). The Enolase phosphopeptide mix contains four synthetically phosphorylated peptides (one serine, threonine, tyrosine, and double serine phosphorylated, respectively) besides the tryptic digest of the protein. After the enrichment and C 18 clean-up, the resulting samples were reconstituted in 10 µL 98% H 2 O, 2% ACN, 0.1% FA injection solvent, of which 5 µL was injected (250 ng, theoretical, HeLa peptide and 250 fmol, theoretical, synthetic phosphopeptide content). In the case of control samples, only C 18 clean-up was performed.

Loading Solvent Dependence of TiO 2 Spin Tip Setup
Pierce™ TiO 2 Spin Tips (ST) were used for all the below-listed methods. The following centrifugation speeds were applied: 1000× g, 10 min for loading; 1000× g, 5 min for elution; 2000× g, 2 min for all the other steps.
The impact of lactic acid on phosphopeptide enrichment was assessed using the slightly modified version of the manufacturer's protocol (ST-LA). The spin tips were first conditioned with 20 µL of Loading Solvent A (LBA; 80% ACN, 20% H 2 O, 2% TFA), followed by equilibration with 20 µL of Loading Solvent B (LBB; 26% lactic acid in LBA). The samples were loaded and re-loaded onto the column in 150 µL LBB, then washed with 20 µL LBB and 3 × 20 µL LBA. The first elution step was performed using 50 µL of 1.5% ammonia (in 100% H 2 O), followed by 50 µL 5% pyrrolidine (in 100% H 2 O). After the elution, the samples were lyophilized and prepared for the C 18 clean-up.
The impact of 2,5-dihydroxybenzoic acid and citric acid (ST-DHB, ST-50CA, and ST-1000CA) on the enrichment performance was tested using the following protocol. The spin tips were first conditioned with 2 × 50 µL washing solvent (WB; 40% ACN, 60% H 2 O, 0.1% TFA), then they were equilibrated with 2 × 50 µL sample loading solvent. The samples were loaded and re-loaded onto the column in 150 µL loading solvent consisting of 80% ACN, 20% H 2 O, 0.1% TFA, and the displacing agent. The displacing agents were as follows: 250 mg mL −1 DHB, 50 mM citric acid, and 1000 mM citric acid. The columns were then washed with 2 × 50 µL loading solvent and 2 × 50 µL washing solvent. The first elution step was performed with 50 µL of 4% ammonia (in 80% ACN, 16% H 2 O), followed by 2 × 50 µL of 4% ammonia (in 100% H 2 O). After the elution, the samples were lyophilized and prepared for the C 18 clean-up.

Loading Solvent Dependence of CIM-OH-TiO 2 Setup
The monolithic column was operated off-line; solvents and samples were loaded on the column with a syringe pump at 100 µL min −1 for loading and elution and 200 µL min −1 for all the other steps.
The loading solvent dependence of the TiO 2 functionalized monolithic column (MON) was tested with two different solvent setups, both comparable with the spin tip methods: one of the methods was a slightly modified version of a method previously optimized specifically for this column (same solvents as the acetic acid method for the spin tips), while the other was the most promising method (50 mM CA, 0.1% TFA) in the spin tip setup.
The All the other solvents and the loading, washing, and elution procedures were the same as described in Section 3.2.1.

C 18 Sample Clean-Up
For desalting and clean-up, Thermo Fisher Scientific Pierce TM C 18 spin columns were used. The column was conditioned with 2 × 200 µL of 50% MeOH, 50% H 2 O, then washed with 2 × 200 µL 0.5% TFA, 95% H 2 O, 5% ACN and equilibrated with 2 × 200 µL loading solvent (0.1% TFA in H 2 O). The samples were loaded and re-loaded onto the column in 50 µL loading solvent and washed twice with 100 µL loading solvent. Elution was performed with 2 × 50 µL 0.1% TFA in 70% ACN, 30% H 2 O. After the elution, the samples were lyophilized and stored at −20 • C until reconstitution for injection.

Mass Spectrometry and Chromatography Analysis
Samples were dissolved in 10 µL injection solvent (98% H 2 O, 2% ACN and 0.1% FA) out of which 5 µL was subjected to nanoLC-MS/MS analysis using a Dionex Ultimate 3000 RSLC nanoLC (Dionex, Sunnyvale, CA, USA) coupled to a Bruker Maxis II Q-TOF (Bruker Daltonik GmbH, Bremen, Germany) via CaptiveSpray nanoBooster ionization source (0.1% FA in ACN as booster liquid). Trapping was performed on an Acclaim PepMap100 C 18 (5 µm, 100 µm × 20 mm, Thermo Fisher Scientific, Waltham, MA, USA) trap column with 0.1% TFA (H 2 O) as the transport liquid. Peptides were separated on an Acquity M-Class BEH130 C 18 analytical column (1.7 µm, 75 µm × 250 mm Waters, Milford, MA, USA) using 1-step linear gradient elution (4-50% eluent B in 90 minutes); Solvent A was 0.1% FA in H 2 O, Solvent B was 0.1% FA in ACN, flow rate 300 nL min −1 . Spectra were collected using a fixed cycle time of 2.5 sec and the following scan speeds: MS spectra at 3 Hz, while CID was performed on multiply charged precursors at 16 Hz for abundant ones and at 4 Hz for low-abundance ones. Internal calibration was performed by infusing sodium formate and data were automatically recalibrated using the Compass Data Analysis (v4.3; Bruker Daltonik GmbH, Bremen, Germany) software.

Isoelectric Point and GRAVY Score Calculation
The isoelectric points were calculated using the IPC-Isoelectric Point Calculator-by Kozlowsky [29], and GRAVY (Grand Average of Hydropathy) scores [30] were calculated with Microsoft Excel.

Data Visualization
Data visualization was done using Microsoft Excel, VIB-BEG Venn-diagram maker [31], and the R package eulerr [32].

Data Availability
The mass spectrometry proteomics data have been deposited in the ProteomeXchange Consortium via the PRIDE partner repository [33] with the dataset identifier PXD019439.

Conclusions
We have thoroughly analyzed the performance of two stationary phases and multiple loading solvents for phosphopeptide enrichment originating from very small amounts of complex samples. A large difference in terms of enrichment effectiveness and repeatability regarding both the stationary phase and the mobile phase chemistries was observed. We have presented methods with different selectivity for both the TiO 2 nanoparticle-coated monolithic column and the TiO 2 spin tip setup. Among these, the method based on the TiO 2 spin tip with 50 mM citric acid loading solvent provided the best results regarding all aspects of enrichment performance. Increasing the TFA content of the loading solvent to 1.5% further improved the figures of merit (ST-50CA-B method). This successfully improved selectivity towards phosphopeptides. Note that all of these methods were tested using complex samples in the low nanogram range, so these should be useful for analyzing tissue microbiopsies or tissue microarrays (TMAs). As compared to previous literature data, we can conclude that our methods are the first TiO 2 -based PP enrichment methods applicable for nanogram-range complex protein samples [34]. The final spin tip method provides identification performance similar to other methods working with milligram protein samples. However, there have recently been reported IMAC methods that are more complicated to use but can provide similar enrichment performance and selectivity to our optimized TiO 2 -based method [17][18][19].