Improved phosphoproteomic analysis for phosphosignaling and active-kinome profiling in Matrigel-embedded spheroids and patient-derived organoids

Many attempts have been made to reproduce the three-dimensional (3D) cancer behavior. For that purpose, Matrigel, an extracellular matrix from Engelbreth-Holm-Swarm mouse sarcoma cell, is widely used in 3D cancer models such as scaffold-based spheroids and patient-derived organoids. However, severe ion suppression caused by contaminants from Matrigel hampers large-scale phosphoproteomics. In the present study, we successfully performed global phosphoproteomics from Matrigel-embedded spheroids and organoids. Using acetone precipitations of tryptic peptides, we identified more than 20,000 class 1 phosphosites from HCT116 spheroids. Bioinformatic analysis revealed that phosphoproteomic status are significantly affected by the method used for the recovery from the Matrigel, i.e., Dispase or Cell Recovery Solution. Furthermore, we observed the activation of several phosphosignalings only in spheroids and not in adherent cells which are coincident with previous study using 3D culture. Finally, we demonstrated that our protocol enabled us to identify more than 20,000 and nearly 3,000 class 1 phosphosites from 1.4 mg and 150 μg of patient-derived organoid, respectively. Additionally, we were able to quantify phosphosites with high reproducibility (r = 0.93 to 0.95). Our phosphoproteomics protocol is useful for analyzing the phosphosignalings of 3D cancer behavior and would be applied for precision medicine with patient-derived organoids.


Results
Removal of contaminants from Matrigel using acetone precipitation of digested peptides. In order to establish a protocol for phosphoproteomics compatible with Matrigel-embedded samples, we first sought to remove Matrigel-derived contaminants using centrifugation. As an experimental model, we used HCT116 cells cultured in Matrigel for eight days. The Matrigel-embedded HCT116 cells formed cystic spheroids (Fig. 1). We prepared phosphopeptides from 2.0 mg protein that was obtained from the HCT116 spheroids, which have been described in Fig. 1 and the Material and Methods section. Although 10,415 class 1 phosphosites are identified from the sample prepared with the centrifugation protocol (Table S1), the ion intensity and number of identified peptides per minute is significantly suppressed, as shown in the chromatographs (Fig. S1a). Thus, in order to remove contaminants such as phospholipids, we introduced acetone precipitation of tryptic peptides to the sample preparation workflow (Fig. 1). Using 2.0 mg proteins from HCT116 spheroids collected by one of two commonly used recovery methods, i.e., the Dispase method (hereafter termed D method) and Cell Recovery Solution method (hereafter termed CR method), the protocol involving acetone precipitation enabled us to identify 23,283 (D) and 19,671 (CR) class 1 phosphosites from triplicate experiments (Figs 2A,B, and Table S2). In contrast to the chromatographs of the samples prepared by the centrifugation protocol, the ion intensity in the chromatographs is significantly increased, as shown in Fig. S1b. Furthermore, we compared the number of MS/MS spectra that are identified as peptides for the centrifugation method, the D method, and the CR method. When using the centrifugation method, 9.9% of the MS/MS spectra was identified as peptides (Fig. 2C). Conversely, when using the two methods involving acetone precipitation, the average percentages are increased to 26.9% (D Method) and 22.8% (CR Method) (Fig. 2C). Together with the chromatography data, these results demonstrate that the protocol involving acetone precipitation efficiently removes contaminants that hamper ionization and the identification of phosphopeptides. Therefore, our protocol allows deeper phosphoproteomic analyses than those available through the centrifugation method.
Experimental bias in phosphoproteomic analysis due to the procedure used for sample collection from Matrigel. Cell Recovery Solution and Dispase are commonly used for the removal of Matrigel from embedded samples. Matrigel can be depolymerized in Cell Recovery Solution at 4 °C without the use of enzymes which cause degradation to the extracellular domains of cell surfaces 13 . Alternatively, Dispase, a neutral protease from Bacillus polymyxa 14 , digests Matrigel at physiological temperature (37 °C) without any stimulation of biochemical reactions in response to low temperature. Because the cellular conditions during the removal of Matrigel are quite different between the D and the CR methods, we applied both methods to our phosphoproteomic analysis to assess whether the cellular phosphorylation status is affected by the method employed.
We compared 6,747 phosphosites identified in both experiments using Matrigel-embedded HCT116 cells recovered with the D or the CR method (Fig. 3A). Two parameters, fold change (FC) and q value, were used as criteria for identifying phosphorylation sites showing significant differences by D and the CR methods. Based on these criteria, we observed that 288 phosphosites in the samples using the D method are increased relative to those using the CR method, whereas 166 phosphosites were decreased in comparison to those collected using the CR method (Fig. 3B, Table S3). We further searched for the class 1 phosphosites that were identified using either one of the samples collected using the D or CR method. We observed that 1,443 and 742 phosphosites are identified only from the samples collected using the D and CR methods, respectively (Table S3).
Then, we applied these phosphosites to pathway analysis by DAVID 15 and kinase enrichment analysis (KEA) 15,16 . In the case of the phosphosites increased or identified only from the sample in the D method, the pathway analysis revealed that several phosphosignaling pathways such as the MAPK and ErbB signaling pathways, are significantly enriched (Fig. 3C). KEA predicted significant activation of several kinases such as RPS6KA3, CDK2, and MEK1 (Fig. 3D). Activation of RPS6KA3 and CDK2 via phosphorylation itself (RPS6KA3 S78, CDK2 T160) using the D method were also confirmed by phosphoproteomic analysis (Table S3). Pathway analysis of the phosphosites enhanced by CR method did not identify activation of any phosphosignaling pathways (Fig. 3C). However, KEA analysis of the phosphosites increased or identified only from the CR method predicted activation of PRKACA and CDK12 (Fig. 3D). Coincident with the results of KEA, activation of CDK12 via phosphorylation itself (CDK12 S685) from the sample treated by the CR method was confirmed by phosphoproteomic analysis (Table S3). In summary, these data indicate that the method used for sample recovery greatly affects cellular phosphosignaling and kinome activity and can cause experimental bias in the phosphoproteomic status.
Comparison of the phosphoproteomic statuses of differential cellular phenotypes as revealed by 2D-and 3D-cultured samples. 3D cellular conformation is regulated through several phosphosignaling pathways 17 . Thus, we compared the phosphoproteomic statuses of 2D-cultured HCT116 cells and 3D-cultured HCT116 spheroids embedded in Matrigel to identify any modulation of phosphorylation signaling that is dependent on the 3D cellular environment. We first conducted phosphoproteomic analysis of 2D-cultured HCT116 cells with and without acetone precipitation. 15,419 class 1 phosphosites were identified from the 2D-cultured HCT116 cells with acetone precipitation (62.5% number of phosphosites being identified relative to those identified from the 2D-cultured sample without acetone precipitation) (Table S2). To investigate the difference between 2D-cultured and 3D-cultured HCT116 samples, we compared the number of phosphosites identified in all replicates using the 2D-cultured HCT116 cells with acetone precipitation to that identified using the 3D-cultured HCT116 spheroids (using either the D or CR method). As a result, of the phosphosites detected using the 2D-culture, 6,773 and 6,025 were commonly detected in all triplicates of 3D cultured samples using the D or CR methods, respectively (Fig. 4A,B). Of these 6,773 and 6,025 phosphosites, 1,613 and 815 phosphosites were significantly increased in the 3D-cultured samples recovered with the D and CR method, respectively, relative to those in the 2D-cultured samples (Fig. 4C). We also identified 2,395 and 1,397 phosphosites that were detected only in the 3D-cultured samples by the D and the CR method, respectively (Fig. 4C).
To reduce experimental bias dependent on the sample collection process as reported previously 18 , we excluded phosphosites with significant differences between the D and CR methods (listed in Table S3) from the 1,613 and the 815 phosphosites. Consequently, 1,439 and 808 phosphosites were increased in the 3D-cultured samples with the D and CR methods, respectively ( Fig. 4Cand Table S4). Similarly, phosphosites only detected using either the D or the CR methods were removed from the 2,395 and 1,397 phosphosites which were identified only in the 3D-cultured samples. As a result, 1,324 and 1,364 phosphosites were identified only in the 3D-cultured samples with the D and the CR methods, respectively ( Fig. 4C and Table S4).
Those phosphosites were subjected to pathway analysis and KEA to compare differences in phosphosignaling between the 3D-cultured HCT116 spheroids and 2D-cultured cells. Results from pathway analysis revealed that several phosphosignaling pathways, such as the insulin and mTOR signaling pathways, are commonly enriched in 3D-cultured cells collected by either D or CR method (Table 1). KEA predicted significant activation of MAPK9, RPS6KA3, and SGK1 under 3D conditions using both the D and the CR methods ( Table 2). SGK1 is known to be associated with both the insulin and mTOR signaling pathways 19 . In addition, two phosphosites (BAG3 S289 and WNK1 S2032) assigned as mTOR substrates in PhosphositePlus 20 were increased in both 3D samples (Table S4). Results of these phosphosites are consistent with those of the pathway analysis, further emphasizing the activation of the mTOR signaling pathway in 3D-cultured samples. Activation of RPS6KA3 via phosphorylation itself (S227) in 3D-cultured HCT116 cells was also revealed by the phosphoproteomic data from samples recovered by both methods, indicating agreement with the results of KEA analysis (Table S4). Thus, our phosphoproteomic method enables us to identify the modulation of phosphosignaling and kinome activity under 3D conditions. This information would help to rationalize the cellular conformation of 3D architectures in terms of phosphorylation-mediated regulatory mechanisms.
Global phosphoproteomic analysis using patient-derived organoids by fractionated or one-shot proteomic methods. The application of organoid systems has been rapidly extending to various types of cancers, including prostate cancer, pancreatic cancer, and hepatocellular carcinoma [6][7][8]21 . Although personal omics profiling with patient-derived organoids has steadily progressed 7 , research into phosphoproteomic profiling from patient-derived organoids is still lacking owing to the technical challenges associated with sample preparation.
Thus, we sought to determine whether our protocol is applicable to phosphoproteomic analysis using patient-derived organoids at the large scale (1.4 mg) and small scale (150 μg). Patient-derived organoids were collected with the D method because a higher number of phosphosites were identified in HCT116 spheroids by the D method relative to that by the CR method (Fig. 5A). Patient-derived organoids show crypt-like structures that resemble the 3D architecture of colorectal cancer tissue (Fig. 5B) 7,22 . First, we collected 1.4 mg (samples from 32 wells in 48 well-plate) of protein lysate from patient-derived organoids. In total, 21,516 phosphopeptides were identified from triplicate experiments ( Fig. 5C and Table S5), indicating that our protocol enables phosphoproteomic analysis of patient-derived organoids at equal sensitivity to that achieved for cultured cell lines.
To apply proteomics to precision medicine, it is essential to acquire comprehensive data from a small amount of patient-derived sample in an expeditious manner. Therefore, for the purpose of high-throughput phosphoproteomics from patient-derived organoids, we attempted a small-scale phosphoproteomic analysis without peptide fractionation before LC-MS/MS. We collected 150 μg of protein lysate from patient-derived organoids (4 wells in a 48 well-plate). This one-shot phosphoproteomic analysis resulted in identification of 2,979 phosphopeptides from triplicate experiments ( Fig. 5D and Table S5). Subsequently, we evaluated the quantitative reproducibility of the phosphosites data from the large-scale (fractionated) and small-scale (one-shot) analyses. Pearson's correlation coefficients for the one-shot phosphoproteomic analysis (0.93 to 0.95) were virtually equivalent to those of the fractionated phosphoproteomic analysis (0.94 to 0.95) (Fig. 5E). This result demonstrates the excellent reproducibility available with our phosphoproteomic protocol, even for small-scale samples.
Thus, these data indicate that our protocol could be applied for obtaining global phosphoproteomics data from patients-derived organoids.

Discussion
In this study, we examined acetone precipitation for phosphoproteomic analysis of Matrigel-embedded samples in order to reduce contaminants from Matrigel. We obtained global phosphoproteomic data from Matrigel-embedded spheroids and patient-derived organoids.
In previous studies, precipitation of proteins with acetone, acetone/trichloroacetic acid, and methanol/chloroform has been reported to be utilized for the elimination of contaminants 12 . However, in this study, we have applied the precipitation of digested peptides, and not proteins, by adding acetone and have achieved a significant decrease in contaminants and identification of global phosphoproteomics (Fig. 2). These results indicate that our protocol may be potentially applied in case of peptidome using samples that contain large amount of contaminants such as plasma 23 and neoantigen in the cancer tissue 24 . Additionally, a previous report demonstrated the selective precipitation of phosphopeptides using Ba 2+ /acetone from whole digested peptides 25 . Therefore, our protocol may be further applied to not only eliminate contaminants but also to selectively enrich the digested peptides without the usage of immunoprecipitation.
It has been reported that alteration of phosphoproteomic data can be caused by processes during sample collection, such as ischemia in surgery and washing before lysis 18,26 . Similarly, in this study, we have observed that phosphoproteomic data is biased dependent on the method used for the removal of the Matrigel. Pathway analysis indicted that several phosphosignals are activated in the samples isolated by the D method compared to those by the CR method. Although the number of phosphosites identified is higher using the D method than that using the CR method, the CR method is a better way to harvest Matrigel-embedded samples in terms of minimizing the fluctuation of phosphosignaling during sample collection. Recently, the stiffness of the scaffold has been reported to be important for efficient expansion of organoids 27 . Accordingly, the results of this study indicate that our protocol allows phosphoproteomic analysis of samples embedded in such hard conditions. Further improvements of the current protocol are expected to minimize experimental bias during sample collection, and to enable phosphoproteomic analysis of organoids cultured in much stiffer scaffolds. 3D cancer models have been utilized to investigate the mechanisms of cancer growth and invasion in conditions closer to in vivo environments. For example, previous studies with 3D cancer spheroids have demonstrated the differential relationship between extracellular conditions and intracellular signaling pathways in malignant cancer and normal tissues 17 . In the current study, phosphoproteomic analysis of 3D-cultured HCT116 spheroids revealed the activation of several phosphosignaling pathways, such as the AMPK, mTOR, and ErbB pathways (Table 1). Because the different results obtained from the 2D and 3D culture condition were not caused by methodological bias (Fig. 4), we concluded that the signaling pathways were activated because of the difference between the culture conditions that were observed in the 2D and 3D circumstances. Collectively, these results support the conclusions of previous studies, i.e., that those pathways are associated with the 3D growth and metastasis of malignant cancer [28][29][30] . This fact indicates that our approach has the potential to reveal the precise modulation of phosphosignaling in cancers under 3D conditions. Therefore, our protocol for Matrigel-embedded samples would contribute comprehensive understanding of phosphorylation-mediated regulation in 3D behavior of cancer.
In recent years, methods for high-throughput phosphoproteomics using relatively small amounts (<1 mg) of protein lysate have been developed, such as the EasyPhos technique, which can identify more than 10,000 phosphosites from 1 mg of protein lysate 31 . In combination with pharmacological response data, quick and sensitive phosphoproteomic analysis could facilitate the prediction of drug sensitivity and provide important details of phosphosignaling under drug treatment. In the current study, our method enabled us to identify almost 3,000 phosphosites from 150 μg lysate of patient-derived organoids (Fig. 5C). The combination of our protocol with multiplex labeling techniques may provide further improvement to deep and high-throughput phosphoproteomics.
In summary, we have achieved the accurate identification of cellular phosphorylation characteristics not only from large-scale samples, but also from small scale samples. Therefore, the phosphoproteomic method developed in this study may contribute significantly to the profiling of individual phosphorylation networks in cancer patients from limited amount of patient derived samples such as organoids. Furthermore, our results indicate that our phosphoproteomic procedure represents a significant breakthrough for providing novel insights into the 3D architectures of cancers, thus facilitating the identification of novel druggable targets.

Cell cultures and sample collection. HCT116 cells were maintained in DMEM supplemented with 10%
FBS and 1% Penicillin-streptomycin at 37 °C under 5% CO 2 . The Matrigel-embedded cultures used in the current study were prepared by mixing 4,000 HCT116 cells with 25 μL of ice-cold Matrigel followed by mounting in a 48-well plate. After incubation for 10 min at 37 °C, DMEM supplemented with FBS and penicillin-streptomycin was overlaid. Patient-derived organoids were established from surgically resected tumor samples at the JFCR Cancer Institute (Tokyo, Japan). Informed consent was obtained from all donors. The experimental protocol was approved by the ethics committees of the National Institute of Biomedical Innovation Health and Nutrition, and the JFCR Cancer Institute. All methods were carried out according to relevant guidelines and regulations. Organoids were cultured as previously described 21 . Based on the previous study, organoids were derived only from the malignant part of the surgical cancer tissue. The culture medium was exchanged with fresh medium every two days. HCT116 cells and patient-derived organoids were collected after washing with ice-cold PBS buffer containing PhosSTOP and cOmplete and lysed in phase transfer surfactant (PTS) buffer (12 mM sodium deoxycholate, 12 mM sodium lauroyl sarcosinate, 50 mM ammonium bicarbonate) 32 . Lysates were boiled at 95 °C  Digestion with trypsin, removal of surfactant, and acetone precipitation of tryptic peptides.
After measuring the protein concentration of the cell lysate with a DC protein assay, the lysate samples were reduced, alkylated, and subsequently trypsinized as described previously 33 . The surfactant was then removed as also described previously 32 . Briefly, the proteins that were dissolved in the PTS buffer were digested with trypsin that was followed by vigorous mixing with an equal volume of ethyl acetate containing 1% trifluoroacetic acid. Further, centrifugation at 14,000 g was conducted for 3 min at 4 °C. The surfactant in the PTS buffer was transferred to an organic phase, and the aqueous phase of the mixture was subjected to the following centrifugation or acetone precipitation step. For the centrifugation protocol, the aqueous phase containing tryptic peptides was centrifuged at 12,000 g for 5 min. For the acetone precipitation protocol, the pelleted peptides were mixed with 1 mL of ice-cold acetone, sonicated for 10 min, and stored at −20 °C for 2 h. Then, the tryptic peptides were centrifuged at 12,000 g for 5 min. After discarding the supernatant, the samples were lyophilized for 10 min. The pellets were resuspended in 2 M urea and 1% TFA solution and subjected to desalination using an OASIS HLB column.
Enrichment of phosphopeptides, TMT labeling, and fractionation with C18/SCX stage-tips. Enrichment of phosphopeptides with Fe 3+ IMAC resin was performed as described previously 34 .
Phosphopeptides were divided into seven fractions with C18-SCX stage-tips as described previously 35 . For the one-shot phosphoproteomics, desalting with C18 stage-tips was conducted after phosphopeptide enrichment.
Mass spectrometry analysis. Equipment for LC-MS/MS analysis was corresponding to a previous study 36 .
The nano-LC gradient was performed at 280 nL/min and consisted of a linear gradient of buffer B from 5 to 30% B over 135 min. Parameters in Q Exactive instrument was coincident with the condition in the previous study 36 .
Data processing for identification and quantification of peptides. Phosphopeptide identification was conducted with MaxQuant 1.5.1.2 supported by the Andromeda search engine 37 . The UniProt human database (release 2011_11) combined with 262 common contaminants was used for the analysis of MS/MS spectra. Enzyme specificity was set to a C-terminal of Arg or Lys with the allowed cleavage at the proline bond. Up to two miss-cleavages were tolerated. Fixed modification was performed by carbamidomethylation of cysteine residues. Variable modifications were performed by methionine oxidation and/or serine, threonine, and tyrosine phosphorylation. Protein group, peptide, and PTM site levels with <0.01 FDR were accepted. Peptides annotated as "Reverse" or "Potential Contaminant" were omitted. The cut-off criteria for phosphopeptides were those used in a previous study 38 . Details for the identification of phosphopeptides are described below. Phosphorylated and nonphosphorylated peptides are distinguished by the mass shift corresponding to the modification of phosphorylation (molecular weight = 79.966) at the MS1 level. Additionally, the MS2 scans (fragmented ionic products of phosphopeptides) are utilized to determine the localization of phosphorylation in the sequence of phosphopeptides 39 . Further, phosphopeptides with highly reliable identification were extracted and subjected to further analysis. The criteria of phosphopeptides included an Andromeda delta score >8 and class 1 phosphosites (localization probability >0.75). "Class 1 phosphosite" was a qualitative standard for phosphoproteomics that have been defined in a previous study 40 .
Statistical analysis of identified class 1 phosphosites. Statistical analysis was carried out with Perseus 1.5.5.3 (www.perseus-framework.org) 41 . The data in label-free quantification were log 2 transformed and normalized using the sum of all intensities and median centering of the values in each sample. p values were calculated with the two-tailed welch t test, then adjusted to a q value with the permutation test. Fold change and q value were used for eliminating phosphosites that showed significant differences. The cutoff criteria for fold change were more than twice or less than half that of a control. Additionally, the phosphosites with a q value lower than 0.05 were used for subsequent analysis. Proportional Venn diagrams were prepared with eulerAPE 42 . The correlation matrix was constructed with the use of the R package "Correlation Matrix. " Pathway analysis of the extracted phosphosites was performed with KEGG in DAVID 15 . KEA ver. 2 was employed according to the reported protocol 16 .
Data availability. Accession code of MS data in this study is PXD009032 in jPOST (http://jpostdb.org/) 43 .