Phospho-antibody Profiling Using Alkaline Phosphatase Treatment in Cell Lines
A particular strength of RPPA is phosphoantibody profiling within cellular systems. RPPA enable the investigation of the phosphorylation status of many proteins in large numbers of samples. An important issue that has previously prevented the widespread adoption of RPPA for this purpose is the dependence on high antibody specificity and the establishment of suitable controls that are amenable to high throughput investigations. Alkaline phosphatase (AP) is an enzyme that indiscriminately removes phosphate from phosphorylated Ser, Thr or Tyr residues in proteins. We have previously developed a lysis buffer (AGLyse) that is highly efficient at lysing cells while at the same time providing an environment where AP retains its enzymatic activity. Here we show that AGLyse can be used in combination with a panel of commercially available cancer cell lines to directly assess the suitability of the phosphorylation-specific antibodies for the investigation of the perturbation of signal transduction pathways.
A panel of 113 antibodies (106 anti-phospho-antibodies and seven protein-specific, but not phosphorylation-specific antibodies) was selected randomly from our antibody library and probed against eight cultured cell lysates that were treated or not treated with AP prior to printing on the RPPA. As expected, the removal of phosphates from the proteins in the cellular lysates resulted in a decrease of the binding of anti-phospho antibodies which was reflected in a clear separation of the observed relative fluorescence intensity (RFI) of untreated cellular lysates from that of AP-treated controls upon non supervised hierarchical clustering (Figure 1a). When the logFC was calculated for all 113 antibodies used for this experiment, followed by the sorting of antibodies by their response in 293T cells, the heatmap in Figure 1b was obtained. All of the non-phosphorylation-specific antibodies did not show a strong response to AP treatment (as would be expected) and were therefore situated at the right end of the heatmap in Figure 1b together with phosphorylation-specific antibodies that did not show the expected response to AP-mediated protein dephosphorylation (antibodies such as pIGF1R (Tyr1135/ Tyr1136), pFAK (Tyr576), pJAK (Tyr1022),and pSmad3 (Thr179)). This effect was more pronounced, when the average logFC response was employed to sort the tested antibodies, resulting in a further shift of the non-phosphorylation specific antibodies to the right in the heatmap in Figure 1c.
We then asked whether the AP treatment could serve as an independent predictive factor to assess the suitability of anti-phosphorylation antibodies without considering other spot quality measures. Previously, we determined the suitability of antibodies for RPPA purposes using an antibody score that was calculated by taking into account six different factors: 1. Spot quality score (percentage of the total sum of RFI excluding “poor” spots defined by ZeptoVIEW software); 2. Signal-to-noise ratio (the average fold difference between the RNFI of individual spots and the background obtained from individual spots); 3. Dilution linearity score (averaged linearity generated by 8-point dilution across all samples); 4. Fold reduction score (average fold reduction in response to AP across all samples); 5. Positive reference score (visual determination of the positive reference. This is a binary score: good 1 or bad 0); 6. Spot graininess/donut effect (visual determination of homogenous staining and other signal related effects. This is a binary score: good 1 or bad 0). Factors 1 to 4 are equally weighted and categorized into three classes respectively (scored 1,2,3) with 3 having the best performance. Factors 5 and 6 are Boolean value that determine whether a particular antibody can be used or not. The sum of factors 1-4 multiplied by the two binary values from factors 5 and 6 gave rise to a ranked antibody score ranging from 0-12 (Supplemental Data S-4).
Defining all antibodies with an antibody score of 8 or higher (67 antibodies) as “Good” antibodies and antibodies with an antibody score of less than 8 (46 antibodies) as “Bad” antibodies permitted asking the question, if the AP-treatment based logFC value on its own could serve as a predictor of an antibody being “Good” or “Bad”. The receiver operating characteristic curve (ROC curve) calculated for an antibody-score cutoff value of 8 and a logFC cutoff value of -0.792 resulted in an area under the curve of 0.825, indicating a reasonable ability of predicting the suitability of an anti-phosphorylation-specific antibody based on the AP-treatment dependent logFC value alone (Chi-square test p<0.001). To verify independently from the RPPA the veracity of results based on these antibodies, we selected 42 antibodies with logFC ≤-0.792 and performed Western Blots in all eight cell lines investigated here (Supplementary data S-2). 36 out of 42 antibodies (85%) showed meaningful single bands at the expected sizes, confirming the suitability of this screening method for high-throughput approaches.
Protein Extraction Optimization in FFPE and FF Tissue Specimen
Following the establishment of the AP treatment-derived logFC value as a critical assay acceptance parameter in cell culture lysates, it was important to check the suitability of AGLyse based extraction for the analysis of clinically more relevant lysates of FFPE or FF tissue samples. For the analysis of breast-derived FFPE samples, five different lysis methods were compared (Table 1a): 1. A commercial Qproteome FFPE sample extraction buffer; 2. A Tris/SDS based method that had previously been adopted for RPPA analyses (15, 18, 21, 26); 3. The Tris/SDS based method (2) with an additional sonication step; 4. AGLyse extraction buffer procedure (including heating to 80 oC, sonication and AP inhibitors); 5. AGLyse extraction buffer procedure (including heating to 80 oC, sonication without AP inhibitors)
A panel of 14 pre-validated antibodies (11 anti-phosphorylation antibodies and 3 total protein antibodies) was used to compare extraction methods 1-5. None of the extraction buffers produced a signal when it was applied to the array without any tissue sample. The correlation between the RPPA RFI signals generated from samples prepared using the 5 different extraction methods fluctuated but was acceptable for all antibodies (Figure 2A). The inter-method correlations for Methods 1-4 all showed correlation coefficients R2>0.9 and a p-value of p<0.01) (Figure 2b). Interestingly, method 4 showed a higher correlation with the other three methods than with method 5, despite the presence/absence of the AP inhibitors being the only difference between method 4 and method 5. It is possible that either the AP inhibitor present in method 4 contributes to the solubilization of proteins from FFPE samples or that it inhibits intrinsic phosphatases in the FFPE lysates, a feature that is not necessary in the more denaturing methods 1-3.
For FF samples, the well characterized CLB1 lysis buffer was compared with Method 5 (Table 1b): 1. CLB1 buffer extraction; 2. CLB1 buffer extraction with sonication and heating to 80 oC; 3. CLB1 buffer extraction with heating to 80 oC. The same panel of 14 antibodies was used to compare extraction methods 5-8 on the FF tissue samples. Generally, using method 5, lower RFI signals were observed than for the extraction with CLB1 extraction buffer (Figure 3a). However, the correlation of the readout of the lysates prepared from FF tissue samples was very high for all 14 antibodies (R2≥0.93, p<0.01, Figure 3b). Therefore, we continued to apply alkaline phosphatase treatment with the established Method 5 (AGLyse without phosphatase inhibitor) in 3 parallel breast tissue selections and each with 3 independent replicates. Upon removal of the phosphatase inhibition, a signal reduction was observed for all 14 antibodies used in above experiments. Of the anti-phospho-antibodies, pP38 (Thr180/Tyr182), pHER2 (Tyr1221/Tyr122) and pGSK3β (Ser9) and pERK1/2 (Tyr204) showed a pronounced reduction of the signal intensity whereas negative controls (total antibodies against Akt, GSK-3β and α-tubulin) had a lesser reduction of the signal intensity in response to AP treatment (Figure 4).
To assess the reproducibility of the method, we analyzed additional FF samples from two independent breast tissue sites each with two parallel extraction replicates using 36 anti-phosphorylation-antibodies validated in the previous screening experiment with AP treatment. A general reduction obtained from all phospho-proteins in response to AP treatment in FF tissues was observed (Figure 5a) (paired t-test p<0.05). Site 2 displayed a lower RFI than site 1 (Figure 5a and 5b), but the overall phosphorylation site occupation between site 1 and site 2 was very similar. Meanwhile, the overall expression exhibited no difference between two sites (Figure 5b). (non-paired t-test p=0.11) and high expression concordance was maintained significantly between experimental replicates (Figure 5c) (R2>0.9 and p<0.001).
These data together provided initial evidence for the suitability of the AGLyse-based protein extraction method for FFPE and FF tissue specimen prior to RPPA profiling. It also demonstrates the feasibility of an AP treatment workflow for FF tissues indicating its potential for the validation of anti-phosphorylation-antibodies in clinical tissue samples.
Evaluation of RPPA Performance in Melanoma FFPE Samples
As most clinically available samples are FFPE derived, we expanded our validation using a set of melanoma FFPE specimen. Three serially sectioned samples from a total of 63 patients were lysed and protein was extracted with AGLyse buffer. Extracted protein lysates were either analyzed using RPPA or using western blotting with a pre-validated Melan-A antibody, a specific melanocyte marker that is routinely used in histopathological diagnosis. Although extraction replicate No.2 had relatively higher yields in comparison to other two replicates, the overall expression correlations between RPPA and western blotting were all retained at similar levels (R2=0.44-0.53, p<0.05) (Figure 6a and Supplemental Figure S2).
Due to the semi-quantitative nature of western blotting as well as its sub-optimal reproducibility for inter-experimental quantification, we also equally divided western blotting data into three categories based on their signal intensities (0, low and high) and plotted against RPPA data. Two out of three extraction replicates showed significantly different expression at least between low and high groups (ANOVA p<0.05) indicating largely the consistency between these two methods (Figure 6b). We then interrogated further with 4 additional antibodies validated for RPPA (GAPDH, LAG3, PD-L1 and S100) and performed pair-wise comparison between replicates and for all targets tested, strong correlations were observed (R2 ranging from 0.54 to 0.97, p<0.05) (Figure 7). Taken together, by assessing a cohort of 63 melanoma patient samples, we further warranted the robustness of the established extraction methodology with FFPE samples and proved their compatibility for downstream RPPA analysis strengthening its application potential in clinical settings.
RPPA Characterization of Lung Cancer FFPE Tissues
Performing RPPA based analysis of FFPE-tissue protein extracts generated using the AGLyse buffer was further evaluated through the analysis of 3 subtypes of lung cancers versus normal lung tissue samples. This analysis was performed using a panel of 13 histopathological protein expression biomarkers some of which are routinely used for lung cancer subtyping. 20 Non-Small Cell Lung Cancer (NSCLC) patients, of which 10 were adenocarcinoma (ADC) and 10 were squamous cell carcinoma (SCC) and 10 small cell lung cancer (SCLC) patients were analyzed. 10 adjacent non-cancerous lung epithelium samples were included as control. Of these patients a total of 120 FFPE slides were acquired comprising 3 serial sections per patient for protein extraction and subsequential RPPA profiling. Protein quantification showed similar extraction efficiency between parallel sectioned samples despite one set (rep2) having significantly lower protein yields than others (p<0.05) (Supplemental Figure S3). For RPPA profiling, adenocarcinoma markers Napsin A, cytokeratin 7, TTF1, squamous cell carcinoma markers p40, p63 and small cell lung cancer markers TTF1 were used. Additionally, EGFR, VEGFR3, VEGF, PD-1, topoisomerase, tubulin-βIII and ROS1 were also used to obtain in-depth protein expression profiles on the same sample sets. Unsupervised hierarchical clustering revealed distinct expression patterns of proteins in the three lung cancer subtypes separating them from non-cancerous epithelium. This was supported by displaying the score plot of a PCA analysis that indicated a partial separation not only of non cancerous tissue extracts from cancer tissue extracts, but also a partial separation of the three lung cancer subtypes from each other (Figure 8a/b). Interestingly, the results confirmed what is known about the biology of the different lung cancer subtypes. This can be more clearly appreciated when the tissue sample sets are grouped according to their clinical subtypes. ADC tissue samples showed high expression levels of Ck7 and Napsin A and reduced protein expression of p63 and p40. SCC tissue samples in contrast showed the opposite pattern with minimal protein expression levels of CK7 and Napsin A and comparatively higher levels of p40 and p63. Both subtypes, ADC and SCC had EGFR expression levels that were significantly higher than those observed in SCLC extracts and in control tissue. SCLS extract displayed higher protein expression levels than the three other tissue extracts for TTF1, which was also reflected at IHC level (Figure 10 This protein had the lowest expression level in the SCC tissue samples, as expected). Of interest, as for other markers, all of them developed a similar increasing pattern with baseline expression in normal lung tissues, slight up-regulation in ADC and further elevated level in SCC and highest expression in SCLC suggesting a potential expression signature associated with the aggressiveness of cancer and probable for SCLC molecular stratification (Figure 9). For EGFR and p40, as two antibody strains were available for RPPA respectively, both were tested against each other and our RPPA data showed optimal consistency for both targets (R2=0.94, p<0.001) again reconfirming the technical robustness and compatibility for FFPE derived tissue specimen (Supplemental Figure S4).