Assessment and Refinement of Sample Preparation Methods for Deep and Quantitative Plant Proteome Profiling

Abstract

A major challenge in the field of proteomics is obtaining high-quality peptides for comprehensive proteome profiling by LC–MS. Here, evaluation and modification of a range of sample preparation methods using photosynthetically active Arabidopsis leaf tissue are done. It was found that inclusion of filter-aided sample preparation (FASP) based on filter digestion improves all protein extraction methods tested. Ultimately, a detergent-free urea-FASP approach that enables deep and robust quantification of leaf and root proteomes is shown. For example, from 4-day-old leaf tissue, up to 11 690 proteins were profiled from a single sample replicate. This method should be broadly applicable to researchers working with difficult to process plant samples.

Plant tissues present a significant challenge for isolating proteins and generating clean peptides suitable for MS. Plants produce relatively large amounts of interfering compounds such as phenolics, terpenes, pigments, organic acids, lipids, and polysaccharides. Furthermore, while interfering compounds are challenging in all plant tissues, they are particularly abundant in green (i.e., photosynthetically active) tissues.^[1,2] As a result, numerous protein extraction methods using various combinations of acetone, methanol, chloroform, phenol, detergents, and molecular weight cutoff filters have been evaluated on plant tissues^.[1–18] However, nearly all the evaluations of extraction methods have examined compatibility with 2-DE rather than reverse phase LC, which is necessary for deep and quantitative proteome profiling.

We recently completed a large-scale protein atlas comprising 33 different tissues generated from both vegetative and reproductive stages of development.^[19] Notably, profiling leaf tissue samples resulted in ≈31% fewer quantified proteins compared to nonphotosynthetically active tissues, which is typical in analyses of plant-derived samples. To address this discrepancy, we examined if an improved sample processing protocol could enhance the depth of proteome coverage for photosynthetically active tissues.

We initially tested three-protein extraction/digestion methods using 2D-LC–MS/MS. Specifically, our preliminary test examined 1) a dilute SDS in-solution digestion method we have used extensively;^[19–23] 2) SDS–FASP (where FASP is filter-aided sample preparation);^[24] and 3) a urea extraction (UA) and in-solution digestion method. We found that the UA method yielded slightly more detected proteins (Table S1, Supporting Information). Because other groups have reported incomplete removal of SDS by the SDS–FASP method,^[25] coupled with the observation that SDS–FASP did not perform markedly better than UA, we next explored several SDS-free sample processing approaches.

We next tested the UA method in further detail and also compared it to methanol/chloroform (Chloro) and phenol-based methods. First, we examined the amount of protein that was recovered from these extraction approaches. We performed extractions from 100 mg of green leaf tissue from 3-week-old Arabidopsis plants and then quantified the amount of protein recovered, prior to in-solution digestion, using Bradford assays. The UA method yielded, on average, 1004.7 μg of protein, which was the largest amount of protein recovered (Table S2, Supporting Information). Additionally, both the phenol and UA performed better than the Chloro method, in our hands, in terms of amount of protein recovered. Therefore, in situations where there is limited starting material, the phenol and UA methods have a distinct advantage.

Next, we evaluated these methods using MS. Because molecular weight cutoff filters should offer additional removal of nonprotein contaminates, we also examined whether the performance of UA, Chloro, or phenol methods was improved with inclusion of FASP-based on-filter digestion (Figure S1, Supporting Information). For this experiment, we extracted protein from 250 mg leaf tissue using the UA, Chloro, or phenol method. The extracted protein, from each method, was then split in half and processed by either in-solution digestion or FASP (see Supporting Information Methods for details). Furthermore, because the different methods yielded differing amounts of protein, we extracted protein from more tissue than necessary (250 mg) to generate an excess of peptides for LC–MS/MS (>300-fold) and thus avoid biases between methods stemming from limited material. We analyzed triplicate technical replicate extractions for each method using 1 μg of peptides by 1D-LC–MS/MS (Figure 1 and Table S3, Supporting Information). The resulting data exhibited high quantitative reproducibility based on Pearson correlation values of MaxQuant label free quantification (LFQ) intensity among the replicates of each extraction method (Figures S2–S7, Supporting Information). All extraction methods were improved by addition of FASP-based on filter digestion (Figure 1). Specifically, FASP resulted in increases in the number of proteins identified by 10.2%, 7.1%, and 12.8% for the Chloro–FASP, phenol–FASP, and UA–FASP methods, respectively. In particular, the increase due to FASP was evident at the MS/MS level where it resulted in a 20.7%, 15.3%, and 44.6% increase for Chloro–FASP, phenol–FASP, and UA–FASP, respectively. Finally, in terms of proteins detected, unique peptides detected, and MS/MS identified the phenol–FASP and UA–FASP methods perform similarly to one another and both outperform the Chloro–FASP method (Figure 1).

Figure 1.

Evaluation of the performance of sample processing methods for LC–MS/MS. A) Technical replicate extractions were performed in triplicate for each method. Data are means of 3 ± SEM. p Values were determined using t tests and are listed in Table S4, Supporting Information. B) Overlap in the number detected proteins between the sample processing methods. The six-way Venn diagram was generated using InteractiVenn in order to visualize the unions of proteins detected from one or more of the extraction method.^[28] Data are the total identified proteins from triplicate runs.

We examined the subcellular localization of the proteins detected by each method using gene ontology (GO) cellular component annotations. This analysis revealed no major biases in the localization profile of the proteins extracted from each method (Figure 2). Furthermore, we performed GO term enrichment analysis on the proteins specifically detected by each method. This analysis did not return enrichment for any subcellular “component” GO term for any of the sample preparation methods. Based on these data, it does not appear that any of the methods we tested exhibit a subcellular compartment extraction bias relative to the other tested methods.

Figure 2.

Comparison of GO cellular component. Total number of proteins per subcellular compartment identified for each sample processing method.

Encouraged by these results, we sought to test the performance of UA–FASP processed peptides for deep proteome profiling by employing online SCX fractionation coupled with MS. We focused on UA–FASP because it enabled detection of a similar total number of peptides and proteins as the phenol–FASP method (Figure 1), but does not require the use of hazardous phenol or the inclusion of phase separation steps. We performed UA–FASP in duplicate by extracting protein from independent biological replicates of 3-week-old Arabidopsis leaf tissue. Significantly, this analysis revealed that the UA–FASP method enables deep profiling of an average of 10 242 proteins and 76 241 unique peptides (Figure 3 A–C and Table S5, Supporting Information). Furthermore, we examined reproducibility of the biological replicate analyses and found high Pearson correlation values for quantification using either MaxQuant LFQ intensity (r = 0.966) or spectral counting (r = 0.938) (Figure 3 D and E).

Figure 3.

UA–FASP enables deep profiling of leaf tissue. Proteins were extracted from 3-week-old Arabidopsis leaf tissue and processed into peptides using the UA–FASP method. A–C) Data are means of two independent biological replicates ± SEM. Scatterplots of the two biological replicates where the proteins are quantified as D) LFQ intensity or E) spectral counts.

The cellular composition of leaves changes during development, which may impact sample preparation performance. Therefore, we next tested the UA–FASP method using 4-day-old leaf tissue. We identified up to 11 690 proteins from a single sample and an average of 11 055 proteins and 89 298 unique peptides from the two independent biological replicates (Tables 6 and 7, Supporting Information). The run-to-run reproducibility between biological replicates was also high with Pearson correlation values of 0.95 (LFQ) and 0.98 (spectral count) (Table S6, Supporting Information).

Our original goal was to establish an extraction method that enhanced the depth of proteome coverage for photosynthetically active tissues, which typically suffer from lower proteome coverage relative to nonphotosynthetically active tissues. To determine if the UA–FASP method enabled similar coverage of photosynthetically active leaf tissue as nonphotosynthetically tissue, we performed 2D-LC–MS/MS profiling on two independent biological replicates of 4-day-old root tissue. We identified an average of 10 864 proteins and 79 076 unique peptides from the two biological replicates, which exhibited high reproducibility with Pearson correlations of 0.99 (LFQ) and 0.98 (spectral count) (Tables 8 and 9, Supporting Information). Taken together, these data reveal that the UA–FASP method enables deep and reproducible proteome quantification of both nonphotosynthetically and photosynthetically active tissues.

Conclusion

Obtaining clean peptides for comprehensive proteome profiling remains challenging. This is especially the case for difficult samples such as photosynthetically active green leaf tissues. While there are many sample extraction and processing methods that have been developed, most evaluations of their efficacy on plant tissues has been carried out using 2-DE rather than reversed-phase LC prior to MS, which enables deeper proteome coverage by providing greater separation power.^[26] To address this, we evaluated several approaches for sample processing that are based on urea, Chloro, or phenol and tested whether these methods were improved using FASP-based on-filter digestion. We find that inclusion of FASP based on filter digestion improves all protein extraction methods tested. We found that phenol–FASP and UA–FASP methods enabled the best coverage of protein, peptides, and MS/MS based on 1D-LC–MS/MS analyses (Figure 1). We chose the UA–FASP method for further testing because it enabled detection of a similar total number of peptides and proteins as the phenol–FASP method, but does not require the use of hazardous phenol or additional phase separation steps. Ultimately, using the UA–FASP method, we were able to robustly and reproducibly quantify over ten thousand proteins from each analyzed leaf or root sample. In the future, other methods to improve the number of proteins detected per sample can be evaluated such as the incorporation of hybrid search engines,^[27] basic pH fractionation, and/or long gradient reversed-phase separation.^[26] We believe this method will enable comprehensive profiling of proteomes from difficult to process plant samples.