Use of human samples was approved by the Fred Hutchinson Cancer Research Center Institutional Review Board. For the discovery phase of this study, 50 subjects were randomly selected from women in the WHI trial who received and adhered to oral CEE 0.625 mg daily over the first year from randomization, and who did not experience a major clinical outcome during trial follow-up. This population is a substudy of the WHI CEE trial, which is composed of 10,739 women, 5,310 in the active CEE arm and 5,429 in the placebo arm. These women had each undergone hysterectomy, and most had never received hormone therapy prior to trial enrollment. Some were prior postmenopausal hormone therapy users who had stopped hormone therapy some months or years prior to trial enrollment. Rarely, subjects were current hormone therapy users at baseline screening and these subjects were required to undergo a 3 month 'wash-out' period of no hormone therapy use prior to randomization. Sera were collected before and after 1 year of CEE in 7 ml royal blue-stoppered serum tubes for trace elements, no additive, silicone coated (BD 367737), and frozen at -80°C until proteomic analysis. All subjects in this substudy were adherent to study medication (defined as taking >80% of study medication per protocol) throughout the first year from randomization. Sera from a second subgroup (n = 50) of women from the active CEE arm of the CEE trial who met the same selection criteria were included in an independent sample ELISA validation phase of this study.

Sera samples at baseline and 1 year after ET (50 women total) were divided in 5 experiments. For each experiment 30 μl aliquots of sera from 10 women at baseline, and 10 women 1 year after ET were pooled. Baseline and treated pools were then individually immunodepleted of the top six most abundant proteins (albumin, IgG, IgA, transferrin, haptoglobin and antitrypsin) using a Hu-6 column (4.6 × 250 mm; Agilent, Wilmington, DE, USA). Briefly, columns were equilibrated with buffer A at 0.5 ml/minutes for 13 minutes and aliquots of 75 μl of the pooled sera were injected after filtration through a 0.22 μm syringe filter. The flow-through fractions were collected for 10 minutes at a flow rate of buffer A of 0.5 ml/minute, combined and stored at -80°C until use. The column bound material was recovered by elution for 8 minutes with buffer B at 1 ml/minute. Subsequently, immunodepleted samples were concentrated using Centricon YM-3 devices (Millipore, Billerica, MA, USA) and re-diluted in 8 M urea, 30 mM Tris pH 8.5, 0.5% OG (octyl-beta-d-glucopyranoside; Roche Diagnostics, Indianapolis, IN, USA). Samples were reduced with DTT in 50 μl of 2 M Tris-HCl pH 8.5 (0.66 mg DTT/mg protein), and isotopic labeling of intact proteins in cysteine residues were performed with acrylamide. Baseline pools received the light acrylamide isotope (C12 acrylamide; >99.5% purity; Sigma-Aldrich (Fluka), St. Louis, MO, USA), and their corresponding 1 year ET pools received the heavy 1,2,3-C13-acrylamide isotope (C13 acrylamide; >98% purity; Cambridge Isotope Laboratories, Andover, MA, USA). Alkylation with acrylamide was performed for 1 h at room temperature by adding to the protein solution the appropriate quantity of C12-acrylamide or C13-acrylamide per milligram protein, diluted in a small volume of 2 M Tris-HCl pH 8.5 19. For each of the five experiments, the pool of baseline (C12) and estrogen-treated (C13) samples was then mixed together for further analysis.

The two-dimensional protein fractionation has been performed based on the previously described IPAS approach 202224. Briefly, after isotopic labeling and mixing of the two pools, the sample was diluted to 10 ml with 20 mM Tris in 6% isopropanol, 4 M urea pH 8.5 and immediately injected in a Mono-Q 10/100 column (Amersham Biosciences, Piscataway, NJ, USA) for the anion-exchange chromatography, the first dimension of the protein fractionation. The buffer system consisted of solvent A (20 mM Tris in 6% isopropanol, 4 M urea pH 8.5) and solvent B (20 mM Tris in 6% isopropanol, 4 M urea, 1 M NaCl pH 8.5). The separation was performed at 4.0 ml/minutes in a gradient of 0-35% solvent B in 44 minutes; 35-50% solvent B in 3 minutes; 50-100% solvent B in 5 minutes; and 100% solvent B for an additional 5 minutes. A total of 12 pools were collected from the anion exchange chromatography. The 12 pools were then subjected to a second dimension of separation by reversed-phase chromatography. The reversed-phase fractionation was carried out with a Poros R2 column (4.6 × 50 mm; Applied Biosystems, Foster City, CA, USA) using trifluoro-acetic acid/acetonitrile as buffer system (solvent A, 95% H₂O, 5% acetonitrile, 0.1% trifluoro-acetic acid; solvent B, 90% acetonitrile, 10% H₂O, 0.1% trifluoro-acetic acid) at 2.7 ml/minutes. The gradient used was 5% solvent A until absorbance reached baseline (desalting step) and then 5-50% solvent B in 18 minutes; 50-80% solvent B in 7 minutes and 80-95% solvent B in 2 minutes. Sixty fractions of 900 μl were collected during the run, corresponding to a total of 720 fractions for each experiment. Aliquots of 200 μl of each fraction, corresponding to approximately 20 μg of protein, were separated for mass-spectrometry shotgun analysis.

For protein identification we performed in-solution trypsin digestion with the lyophilized aliquots of the 720 individual fractions. Individual digested fractions 4 to 60 from each reversed-phase run were pooled in 11 pools, corresponding to a total of 132 fractions for analysis from each experiment. Tryptic peptides were analyzed by a LTQ-FT mass spectrometer (Thermo-Electron, Waltham, MA USA) coupled to a nano-Aquity nanoflow chromatography system (Waters, Milford, MA, USA). The liquid chromatography separation was performed in a 25 cm column (Picofrit 75 μm ID; New Objective, Woburn, MA, USA), in-house-packed with MagicC18 (Michrom Bioresources, Auburn, CA, USA) resin using a 90 minutes linear gradient from 5-40% of acetonitrile in 0.1% formic acid at 250 nl/minute. The spectra were acquired in a data-dependent mode in a m/z range of 400 to 1,800, and selection of the 5 most abundant +2 or +3 ions of each mass spectrometry (MS) spectrum for MS/MS analysis. Mass spectrometer parameters were: capillary voltage of 2.1 KV; capillary temperature of 200°C; resolution of 100,000; and FT target value of 1,000,000.

The acquired LC-MS/MS data were automatically processed by the Computational Proteomics Analysis System (CPAS) 25. For the identification of proteins with a false discovery rate (FDR) <5%, database searches were performed using X!Tandem against the human IPI (International Protein Index) database v.3.13 using tryptic search 25. Cysteine alkylation with the light form of acrylamide was set as a fixed modification and with the heavy form of acrylamide (+3.01884) as a variable modification. The database search results were then analyzed by PeptideProphet 26 and ProteinProphet 27 programs. Our high confidence list of identifications retained proteins with ProteinProphet scores ≥0.95 (5% error rate) and two or more peptides per protein.

Quantitative ratios of proteins comparing 1-year to baseline samples were obtained by differential labeling of peptides containing cysteine with acrylamide isotopes (heavy or light). Quantitative information was extracted using a script designated 'Q3ProteinRatioParser' that was developed in-house to obtain the relative quantification for each pair of peptides identified by MS/MS that contains cysteine residues 19. Only peptides with a minimum PeptideProphet score of 0.75, and mass deviation <20 ppm were considered for quantification. Ratios of heavy-to-light acrylamide-labeled peptides were plotted on a histogram (log2 scale) and the median of the distribution was centered at zero. This normalization approach was chosen since the great majority of proteins were not expected to be deregulated in 1-year ET compared to baseline samples. All normalized peptide ratios for a specific protein were averaged to compute an overall protein ratio. Proteins for which only peptides labeled with the heavy form of acrylamide were detected were included in the final list of proteins with quantitative information presented as '1-year ET only'. All peptide and protein ratios were calculated on a logarithmic scale. Statistical significance of the protein quantitative information was obtained via two procedures: for those proteins with multiple peptides quantified, a P-value for the mean log-ratio, which has mean zero under the null hypothesis, was calculated using one-sample t-test; and for proteins with a single paired MS event, the probability for the ratio was extrapolated from the distribution of ratios in a baseline-baseline experiment whereby the same sample was labeled with heavy and light acrylamide. The raw data and summary list of identified and quantified proteins are available through the Computational Proteomics Analysis System upon request.

Protein ratios were analyzed to identify proteins whose average ratio (1 year of CEE/baseline), averaged over the five proteomic experiments, differed from zero on a log2 scale. All analyses were performed using the statistical package R 28. Protein log-ratios were normalized across experiments by a median location shift to ensure the distributions of proteins for each IPAS experiment were centered at zero. Protein log-ratios were standardized by forming a sample variance from the (up to five) log-ratios for each protein, and adding a corresponding sample variance from a corresponding set of (up to five) log-ratios from a completely analogous set of five proteomic experiments from the WHI estrogen plus progestin trial. Statistical testing was performed by using a weighted moderated t-statistic 29 implemented in the R package LIMMA 30. A weighted average ratio was calculated for each protein by weighting the (up to five) log-ratios by the number of quantified peptides for each protein and a matrix of weights was included in the linear model. Benjamini and Hochberg's method for controlling the FDR was used to compute adjusted P-values 31.

To improve our estimate of the posterior standard deviation used in the moderated t-statistics, protein ratios from an additional five IPAS experiments that compare estrogen plus progestin and whose quantification followed exactly the same protocol were also included in the linear model. Specifically, average ratios were calculated by fitting a linear model where the design matrix consisted of two dummy variables indicating estrogen or estrogen plus progestin use. All results in this manuscript are based on inferences for the dummy variable of estrogen use (that is, the average ratio for ET use). Including the estrogen plus progestin data does not affect the estimated values of the ET ratios, but does increase the degrees of freedom and consequently increases power.

For network analysis, the unfiltered list of gene names of proteins, and their ratios and P-values from all five IPAS experiments were uploaded into the MetaCore analytical suite version 4.7 (GeneGO, Inc., St. Joseph, MI, USA), and analysis was performed as described previously 32.

Measurements were performed on the same sera from the 50 women utilized for proteomic analysis using ELISAs according to the manufacturer's protocols: human IGFBP1, IGFBP2, IGFBP4, and IGFBP6 (R&D Systems, Minneapolis, MN, USA); IGF1 (Diagnostic Systems Laboratories, Webster, TX, USA); factor IX (F9), factor X (F10), and PROZ (protein Z, vitamin K-dependent plasma glycoprotein) (Hyphen Biomed, Neuville-Sur-Oise, France); ceruloplasmin (US Biological, Swampscott, MA, USA); vitamin D binding protein (Alpco Diagnostics, Salem, NH, USA); fetuin-A (AHSG) (Biovendor, Candler, NC, USA); vitronectin (Innovative Research, Novi, MI, USA); KNG1 (Affinity Biologicals, Ancaster, ON, Canada); MMP2 (Calbiochem, Gibbstown, NJ, USA). Individual serum samples and standards were run in duplicate and absorbance measured using a SpectraMax Plus 384 and results calculated with SoftMax Pro v4.7.1 (Molecular Devices, Sunnyvale, CA, USA). P-values and testing whether there was a significant change from baseline to year 1 for individual proteins were computed using the non-parametric t-test on the log2 scale. For a particular protein, validity of IPAS results was gauged by comparing means (95% confidence intervals) of protein ratios to results from standard ELISA kits. The t-statistic and moderated t-statistic were used to calculate 95% confidence intervals for ELISA and IPAS data. For comparison of discovery and validation findings we also report Pearson's correlation coefficients for log-ratios.

Some characteristics at baseline of the 50 subjects included in the discovery phase are summarized in Table 1. There were no statistically significant differences in any baseline characteristics noted between pools. The average age of the subjects was 61.4 ± 7.9 years (mean ± standard deviation).

There were 2,576,869 tandem mass spectra with >0.05 PeptideProphet score acquired in these experiments (Table 2); 1,760,094 spectra yielded proteins identified with a <5% error rate. To our knowledge, this serum protein dataset is the largest obtained from a human observational study or clinical trial to date. This remarkable size of the serum protein dataset is a result of the extensive fractionation and large number of mass spectra collected in these experiments. The number of proteins identified and quantified showed some variation between experiments (16% coefficient of variation for number of quantified proteins), which may be related to sample processing and MS sampling. However, this variation is not expected to affect quantitative ratios, as each experiment consisted of combined baseline and post-therapy sera that were differentially isotopically labeled prior to mixing. Labeling efficiency was evaluated and the results are shown in Figure 1. The log-ratio histograms were all approximately Gaussian shaped.

A list of weighted, quantified protein products of 611 distinct genes resulted from the serum proteomic analysis (Additional data file 1), after filtering protein identifications to remove proteins without associated gene name (hypothetical proteins) and false identifications based on manual verification of mass spectra. The log2 ratios of protein levels (1 year CEE/baseline), derived from the isotopic labeling of cysteine residues, and their P-values is provided as volcano plots (Figure 2a). We found that 116 of the 611 proteins quantified in the serum met a nominal 0.05 significance level criterion for change after 1 year of CEE, compared to about 31 expected by chance. A similar view was obtained when adjusted P-values (FDR <0.05) were considered (Figure 2b). We found that 64 of the 611 proteins quantified (10.5%) in the serum had estimated FDRs of P < 0.05 for change from baseline to 1 year from randomization (Additional data file 2), while a strongly overlapping set of 64 proteins had nominal P < 0.05 and also had estimated log-ratios >1.20 or <1/1.20 (Additional data file 3). A network analysis of the 64 proteins with statistically significant changes relative to all quantified proteins and with an FDR <0.05 (MetaCore version 4.7) 32333435 yielded a significant enrichment in five networks: blood coagulation, kallikrein-kinin system, cell adhesion-platelet-endothelium-leukocyte interactions, complement system, and ossification (Table 3). We further classified these 64 proteins in relation to the known biological processes they are involved in through a search of the Gene Ontology (GO) database (Table 4). A search of the literature yielded prior associations with ET for 13 of the 64 proteins (ceruloplasmin (CP), plasminogen (PLG), tissue factor pathway inhibitor (TFPI), sex hormone binding globulin (SHBG), IGFBP1, IBFBP4, apolipoprotein A-II (APOA2), vitamin D binding protein (GC), apolipoprotein D (APOD), IGF1, AHSG, lactotransferrin (LTF), angiotensinogen (AGT); Table 4). Thus, novel associations were observed for 41 proteins. These proteins are associated primarily with blood coagulation, metabolism regulation, complement/inflammation/innate immunity, ossification, cellular growth, cell-cell/cell-matrix interactions, vessel morphogenesis/angiogenesis and blood pressure maintenance processes.

A critical step in estrogen effect on gene expression is recognition of the estrogen response elements (EREs) via estrogen receptors. For the differentially expressed proteins, we checked for the presence of conserved (between mouse and human) EREs in their corresponding genes. The sequence match was performed against a publicly available ERE database 36. Four proteins - AGT, galectin-1 (LGALS1), LTF, and trefoil factor 3 (TFF3) - found to be significantly elevated with CEE in our study, had conserved EREs upstream of the coding region. None of the down-regulated proteins had conserved EREs upstream of the coding regions of their genes. However, one down-regulated protein (matrix metalloproteinase 2 (MMP2)) had an ERE in the downstream region of its corresponding gene.

We sought to validate proteomic data by ELISA analysis of individual non-pooled sera from the same subjects in the study. Proteins were selected for assay among the set of 64 proteins meeting statistical criteria for change following CEE, based on availability of a pair of antibodies with the requisite specificity for ELISA-based validation. Thus, assays were available for IGF1, IGFBP4, IGFBP1, IGFBP6, F9, F10, AHSG, vitronectin (VTN), GC, CP, MMP2, kininogen (KNG1), and PROZ. In addition, IGFBP2 was tested as a negative control. SHBG was separately analyzed in a set of 50 women in the trial, who had similar characteristics to those in the training set. High-density lipoprotein and low-density lipoprotein were previously tested and, therefore, were not subjected to additional validation in our study 6. Figure 3 presents the data at baseline and 1 year for each protein. The correlation between IPAS proteomic log-ratios and ELISA log-ratios was strong (correlation = 0.83 without SHBG and 0.86 with SHGB; Figure 4). We obtained a correlation of 0.85 between spectral counts (number of tandem mass spectrometry (MS2) events/protein) and the known serum concentrations of more than 80 proteins (Figure 5a). The measured abundance range of the proteins subjected to ELISA (Figure 5b) is indicative of the depth of proteomic analysis in this study, which was achieved through extensive fractionation of intact proteins and reliance on high-resolution MS, spanning seven logs of protein abundance. However, low abundance proteins are somewhat under-sampled, given that proteins quantified in more than two proteomic experiments only spanned some four logs of protein abundance.

We further analyzed an additional, independent validation set of 50 non-overlapping randomly selected women, who adhered to CEE over the first year of randomization in the CEE trial, for IGF1, IGFBP4, IGFBP1, F10, AHSG, GC, CP, MMP2, and PROZ and for IGFBP2 as a negative control (Figure 2). The correlation between ELISA results for the training set and the independent test set was 95%, and between the independent set tested by ELISA and the training set tested by IPAS it was 87%. Elevated concentrations at 1 year from randomization compared to baseline were observed in these independent samples for all ten proteins studied.