Multiplexed quantification of estrogen receptor and HER2/Neu in tissue and cell lysates by peptide immunoaffinity enrichment mass spectrometry

Abstract

Access to a wider range of quantitative protein assays would significantly impact the number and use of tissue markers in guiding disease treatment. Quantitative mass spectrometry-based peptide and protein assays, such as immuno-SRM assays, have seen tremendous growth in recent years in application to protein quantification in biological fluids such as plasma or urine. Here, we extend the capability of the technique by demonstrating the application of a multiplexed immuno-SRM assay for quantification of estrogen receptor (ER) and human epidermal growth factor receptor 2 (HER2) levels in cell line lysates and human surgical specimens. The performance of the assay was characterized using peptide response curves, with linear ranges covering approximately 4 orders of magnitude and limits of detection in the low fmol/mg lysate range. Reproducibility was acceptable with median coefficients of variation of approximately 10%. We applied the assay to measurements of ER and HER2 in well-characterized cell line lysates with good discernment based on ER/HER2 status. Finally, the proteins were measured in surgically resected breast cancers, and the results showed good correlation with ER/HER2 status determined by clinical assays. This is the first implementation of the peptide-based immuno-SRM assay technology in cell lysates and human surgical specimens.

Diagnostic workup of newly diagnosed cancers frequently includes measurement of the expression of protein biomarkers in tumor cells, where biomarkers are widely used to prognosticate[1, 2], predict response to targeted therapies[3–9], and determine the tissue of origin in carcinomas of unknown primaries[10]. As more targeted therapies come online, the need for companion diagnostics is expected to increase dramatically[11]. For example, the molecular classification of breast cancer according to the expression of hormone receptors and growth factors, such as estrogen receptor (ER) and the human epidermal growth factor receptor 2 (HER2), can guide targeted treatment (e.g. hormone therapy for ER+ and Herceptin for HER2-overproducing tumors). The determination of ER and HER2 status in tissues is typically achieved using immunohistochemistry (IHC), yet, despite tremendous efforts in establishing guidelines and protocols[2, 12, 13], the assays are not quantitative and they have been difficult to standardize across laboratories, leading to development of alternative nucleic acid-based assays using fluorescence in situ hybridization (FISH) or reverse transcription polymerase chain reaction (RT-PCR)[13, 14].

In this report, we explore the use of a proteomic assay format based on targeted quantitative mass spectrometry using selected/multiple reaction monitoring (SRM/MRM) for classification of human breast cancer cell lines and surgical specimens. SRM/MRM has been used extensively for small molecule quantification and has recently been increasingly applied in the area of quantitative proteomics[15–20] wherein proteins are digested to component peptides using an enzyme such as trypsin, and one or more proteotypic peptides are measured as quantitative stoichiometric surrogates for protein concentration in the sample[21, 22]. Coupled to stable isotope dilution (SID) methods (i.e. using a spiked-in isotopic peptide standard), SRM can be used to measure concentrations of proteotypic peptides as surrogates for quantification of proteins in complex biological matrices[21, 22]. The technique is readily multiplexed, offers good performance, and can be transferred across laboratories[23], although sensitivities are limiting in complex biospecimens. To enhance sensitivity of SRM assays, anti-peptide antibodies can be used to enrich target peptides from complex proteomic samples (Stable Isotope Standards and Capture by Anti-Peptide Antibodies, SISCAPA)[24, 25], resulting in highly sensitive immuno-SRM assays that are readily multiplexed[26], amenable to automation, and have been implemented in a wide array of studies[17, 27–31].

While the immuno-SRM technology has been prototyped for measuring biomarkers in plasma and other fluids [24, 29, 30], it has not yet been refined and adapted for quantification of proteins in tissues. To our knowledge, this is the first evaluation of peptide-based immuno-SRM in cell lines and tissues, building upon our work in plasma and other reports showing peptide immunoaffinity enrichment from cell line and tissue samples. In one early report, Gobom et al. identified and quantified endogenous neurotensin from human hypothalamus lysates using immunoaffinity purification with antibodies coupled to Sepharose G beads. The enriched samples were analyzed by MALDI-TOF-MS and nano-ESI-MS/MS, and the measurements were compared to radioimmunoassay[32]. In another report, Ukena et al. characterized an RFamide-related peptide in rat hypothalamus by immunoaffinity enrichment and tandem mass spectrometry on a Q-TOF-MS[33]. Neither report required proteolytic digestion, since the peptides were bioactive peptides and were not part of a larger protein, and neither report combined the enrichment with SRM quantification. Other reports have included proteolytic digestion followed by immunoaffinity enrichment of peptides, yet again not using SRM for subsequent analyses. For example, Rush et al. profiled tyrosine-phosphorylated peptides in cancer cell lines by enriching the peptides using a phosphotyrosine-specific antibody and LC-MS/MS[34]. Jiang et al. measured a peptide specific to epidermal growth factor receptor (EGFR) in human breast cancer cell lines and tissues, with quantification achieved using an isotopically heavy-labeled peptide and MALDI-MS[35]. In addition, Hoeppe et al. used antibodies raised against epitopes based on only three to four amino acids to reduce the complexity of human HEK293 cell lysate digests and characterized the simplified samples by MALDI-TOF-MS/MS[36]. Last, immunoaffinity capture of proteins (that is, classic immunoprecipitation) followed by proteolytic digestion and SRM has been reported[37]; our work here focuses on immunoaffinity capture of peptides, not proteins.

To evaluate the utility of the immuno-SRM assay platform for quantification of biomarkers in human cells, we developed assays for the well-established HER2 and ER biomarkers, and tested their performance in a set of highly characterized breast cancer cell lines[38] and surgical specimens. Experimental details are given in the Supplemental Materials. We targeted one peptide from each protein (LLFAPNLLLDR for the ER protein and AVTSANIQEFAGCK for HER2) and developed polyclonal and monoclonal antibodies to each, respectively, as described previously[39–41]. The open-source Skyline software[42, 43] was used to generate a list of transitions and their associated collision energy and declustering potential values for the 2+ and 3+ charge states of the peptides, the peptides were analyzed in pure form by SRM-MS, and subsequently the most sensitive parent peptide charge states and transitions were selected for all further analyses (Table 1). Nine transitions for LLFAPNLLLDR and seven for AVTSANIQEFAGCK were included in the SRM method, without incurring appreciable sensitivity losses. The y7 transition ion for LLFAPNLLLDR and the y9 transition ion for AVTSANIQEFAGCK were used for quantification, while the additional transitions were used to verify the identity of the peptide ion being monitored.

Table 1. Protein and peptide information for ER and HER2.

For LLFAPNLLLDR, the C-terminal arginine (bold and underlined) was 13C and 15N isotopically labeled. For AVTSANIQEFAGCK, the label consisted of only the 13C isotope. The cysteine in AVTSANIQEFAGCK was carbamidomethylated. Collision energies (CE) and declustering potentials (DP) were obtained using the regression equations in Skyline for an AB Sciex triple quadrupole.

Protein Description	Prot.	Gene Symbol	Accession Number	Peptide sequence	Parent ion m/z	Parent ion charge state	Fragment ion m/z	Ion type	Fragment ion charge state	CE (V)	DP (V)
Estrogen receptor alpha	ER	ESR1	P03372	LLFAPNLLLDR	642.8874	2+	1171.683	y10	1+	32.4	78
							1058.599	y9	1+	32.4	78
							911.5309	y8	1+	32.4	78
							840.4938	y7	1+	32.4	78
							743.441	y6	1+	32.4	78
							629.3981	y5	1+	32.4	78
							516.314	y4	1+	32.4	78
							420.7505	y7	2+	32.4	78
							445.2809	b4	1+	32.4	78
				LLFAPNLLLDR	647.8915	2+	1181.692	y10	1+	32.4	78
							1068.608	y9	1+	32.4	78
							921.5392	y8	1+	32.4	78
							850.502	y7	1+	32.4	78
							753.4493	y6	1+	32.4	78
							639.4064	y5	1+	32.4	78
							526.3223	y4	1+	32.4	78
							425.7547	y7	2+	32.4	78
							445.2809	b4	1+	32.4	78
Receptor tyrosine- protein kinase erbB-2	HER2	ERBB2	P04626	AVTSANIQEFAGCK	748.3641	2+	1325.615	y12	1+	38.4	85.7
							1066.499	y9	1+	38.4	85.7
							952.4557	y8	1+	38.4	85.7
							839.3716	y7	1+	38.4	85.7
							582.2704	y5	1+	38.4	85.7
							435.202	y4	1+	38.4	85.7
							663.3114	y12	2+	38.4	85.7
				AVTSANIQEFAGCK	751.3742	2+	1331.636	y12	1+	38.4	85.7
							1072.519	y9	1+	38.4	85.7
							958.4758	y8	1+	38.4	85.7
							845.3917	y7	1+	38.4	85.7
							588.2906	y5	1+	38.4	85.7
							441.2222	y4	1+	38.4	85.7
							666.3214	y12	2+	38.4	85.7

Performance metrics for the assays were determined by generating response curves using synthetic standard peptides. These curves were not used as calibration curves; rather, the response curves were used to determine the limits of detection, limits of quantification, and linear ranges of the assays. Curves were generated in a trypsin-digested cellular lysate matrix derived from the MCF10A breast epithelial cell line.

For the peptide AVTSANIQEFAGCK (HER2), there were detectable amounts present in the available cell lines, precluding the use of a lysate free of endogenous signal for the curve matrix. Therefore, we prepared a reverse response curve to estimate the limit of detection (LOD), limit of quantification (LLOQ), and linear range[44]. (Of note, these curve parameters were not used to back-calculate concentrations.) The heavy:light peptide ratio was measured and plotted versus the concentration ratio of heavy and light peptide in the sample (Figure 1A and 1B). The assay showed a linear response for all peptide amounts above 0.3 fmol. This linear range covers over 4 orders of magnitude, spanning protein concentrations of 3 to 50,000 fmol/mg lysate. The LOD was 6 fmol/mg lysate determined by calculating three times the signal-to-noise in the blank. The lowest point above the limit of detection with a CV <20% was 12 fmol/mg lysate; this value was used as the LLOQ. The median CV for all points on the response curve was 2.5%, and the CV at the LLOQ was 10.8% (the range of CVs was 1.5% to 10.8%).

Figure 1. Response curves for HER2 and ER peptides in cell lysate.

Peptides were serially diluted in digested lysate of MCF10A cells. 100 ug lysate was used for each antibody capture. For the HER2 peptide AVTSANIQEFAGCK, the response was determined by varying the heavy stable isotope-labeled peptide relative to the constant light peptide plotted in linear space (A) and as log10 transformed values (B). The ER peptide LLFAPNLLLDR response was determined by varying the light synthetic peptide relative to a constant heavy peptide amount and plotting in linear (C) and log10 space (D). Weighted (1/y) linear regression was performed using R. Insets magnify the low concentration portion of the curves. Error bars indicate the range of measurements.

The peptide LLFAPNLLLDR from the ER protein was not detected in the MCF10A lysate, allowing for generation of a response curve without endogenous interference. However, there was a signal due to a peptide contamination associated with the antibody. Residual peptide was observed even though the peptide was covalently linked to the resin used for affinity purification. Such residual peptides have been reported previously[27, 28, 39]. Although the exact source of the contaminating peptides has not been determined, Hoofnagle et al.[27] propose that the peptides hydrolyze slowly from the resin. Because this signal affects the detection of analyte peptide at low concentrations, it must be characterized as part of the assay and thus effectively raises the LOD of the assay. To determine the performance characteristics of the assay, light peptide for LLFAPNLLLDR was serially diluted into digested lysate from the MCF10A cell line, keeping the heavy labeled peptide constant. The peptide response curves are plotted in Figure 1C and 1D. The practical LOD (7 fmol/mg lysate) was calculated as the signal producing three times the standard deviation of the background, including the contaminant peptide signal. The coefficient of variation at the limit of detection was 15.9%, so the limit of quantification (defined as lowest point measured with CV<20%) was also 7 fmol/mg. Omitting the lowest points on the response curve, below the practical LOD, resulted in a linear range of nearly 4 orders of magnitude, spanning protein concentrations of 7 fmol/mg to 50,000 fmol/mg. The median CV for all points on the response curve was 4.2%, and the CV at the LLOQ was 15.9% (the range of CVs was 0.8% to 15.9%).

Next, we evaluated the performance of the immuno-SRM method using 17 cell lines derived from human breast cancers; the ER/HER2 status of this extensively studied cell line panel has been well-characterized[38]. Samples were analyzed in process triplicate (each sample’s triplicate processed on the same day) using the optimized immuno-SRM assays, and the data are reported in Table 2 and plotted in Figures 2A and 2B. Results for the HER2 assay showed good separation between cell lines that do vs. those that do not overproduce HER2 (p-value of 0.0012); the lowest HER2-overproducing cell line (MDAMB361) had 3.62E+03 fmol/mg, whereas the highest HER2- cell line (MDAMB453) had 1.65E+03 fmol/mg. Results for the ER assay also showed good separation between cell lines that do vs. those that do not express ER (p-value of 0.0019), although there was one outlier ER- sample (HCC1419) that had a relatively high concentration of 258 fmol/mg.

Table 2. Quantification of HER2 and ER in cell lines and tissues.

The multiplexed immuno-SRM assay was applied to a set of highly characterized cell lines and tissue samples. For tissues, HER2 and ER status were determined by immunohistochemistry and confirmed by FISH for HER2 when intermediate. Concentration values were based on a single point calibration of the immuno-SRM assay relative to the amount of heavy stable isotope-labeled peptide. 100 ug of lysate was measured for each sample in process triplicate. Tissue samples with an asterisk were analyzed in duplicate, for which CV was reported as the percent difference. For entries marked <LLOQ, peak areas were below the limit of quantification.

	HER2		ER
Sample	fmol/mg lysate	CV	fmol/mg lysate	CV	HER2 status	ER status
HCC1419	3.47E+04	16%	2.58E+02	6%	+	−
ZR7530	3.45 E+04	5%	3.04E+02	8%	+	+
HCC1569	2.71E+04	12%	<LLOQ		+	−
BT474	1.98E+04	18%	9.68E+02	17%	+	+
HCC1954	1.70E+04	4%	1.04E+02	7%	+	−
UACC812	1.44E+04	7%	2.94E+02	16%	+	+
HCC202	1.42E+04	4%	9.7E+01	10%	+	−
MDAMB361	3.62E+03	1%	2.27E+02	6%	+	+
MDAMB453	1.65E+03	28%	<LLOQ		−	−
ZR751	1.17E+03	11%	7.86E+02	9%	−	+
T47D	7.72E+02	4%	6.45E+02	2%	−	+
CAMA1	6.66E+02	24%	1.74E+02	12%	−	+
HCC1428	6.10E+02	11%	3.67E+02	18%	−	+
MCF7	2.45E+02	7%	6.66E+02	8%	−	+
HCC70	2.32E+02	6%	7.9E+01	16%	−	−
HS578T	1.30E+02	5%	7.2E+01	10%	−	−
HCC1395	1.02E+02	11%	<LLOQ		−	−
tissue 080127A1*	1.19E+04	16%	1.40E+02	15%	+a	−a (2 of 8)c
tissue 090163B2	4.37E+03	3%	1.25E+02	11%	+a	+a (4 of 8)c
tissue 080189A1	1.01E+03	10%	<LLOQ		−a	−a (0 of 8)c
tissue 080077A1	8.72E+02	2%	6.71E+02	5%	−a	+a (3+ of 3+)a
tissue 080028A1*	3.84E+02	4%	2.71E+02	61%	−a (not amplified)b	+a (8 of 8)c
tissue 090252A1	2.69E+02	16%	<LLOQ		−a (not amplified)b	−a (0 of 8)c
tissue 090023E1	2.30E+02	11%	<LLOQ		−a	−a (0 of 8)c
tissue 080031A2	1.67E+02	5%	<LLOQ		−a	−a

^adetermined by IHC

^bintermediate by IHC, confirmed by FISH

^cAllred scores, which refer to discrete scores (scaled from 0 to 8) based on IHC measurements that correspond to a metric encompassing staining intensities and percentage of positive cells; a higher score corresponds to higher ER levels.

Figure 2. Quantification of HER2 and ER in cell lines and tissues.

The concentration determined by immuno-SRM was plotted for the cell lines (A and B) and tissues (C and D) reported in Table 2. HER2 values were plotted on a log10 scale due to the wide range in values. For ER, samples measured below the LLOQ were plotted at the LLOQ.

Following demonstration of the assays in breast cancer cell lines, we analyzed eight surgically resected human breast cancers whose ER and HER2 statuses had been determined as part of routine diagnostic workup of the patients using standard clinical assays (e.g. IHC for ER and IHC±FISH for HER2). Samples were analyzed in process duplicate or triplicate (depending on sample availability) using the optimized immuno-SRM assays, and the data are reported in Table 2 and plotted in Figures 2C and 2D (each sample’s replicates were processed on the same day). For the HER2 assay, the CVs ranged from 1% to 28% across all cell line and tissue samples (including percent differences for samples measured in duplicate), and the median CV was 7%. The ER assay CVs ranged from 2% to 61% (also including percent differences), with a median CV of 10%. Median CVs for analyses of the cell lines and surgical specimens were slightly higher for the process duplicate or triplicate samples compared to the response curve samples; this is because trypsin digestion was part of the cell line and surgical sample analyses (but not the response curve generation, which used a single common digest as a matrix), indicating a portion of the overall CV is due to trypsin digestion, as described previously[23]. Remarkably, although breast cancer tissues are highly heterogeneous with respect to their composition and cellularity, results for the HER2 assay showed good separation between tumors that do vs. those that do not overproduce HER2 (Figure 2C). The ER assay also showed good separation (Figure 2D), although as with the cell lines (Figure 2B) there was one ER- tumor (080127A1) that showed a relatively high concentration of ER (140 fmol/mg). Significantly, a comparison of the tissue immuno-SRM results with Allred scores obtained for ER showed good agreement, see Table 2 (Allred scores refer to discrete scores (scaled from 0 to 8) based on IHC measurements that correspond to a metric encompassing staining intensities and percentages of positive cells; a higher score corresponds to higher ER levels[45]). The IHC classification for the HER2 tissue samples was also in good agreement with the immuno-SRM results, Table 2. In addition, FISH assays indicating HER2 status were performed on two of the tissue samples, and again those results corresponded well with the immuno-SRM data (Table 2).

These results demonstrate the feasibility of developing immuno-SRM assays of sufficient sensitivity and analytical performance to quantify protein biomarkers in human cell lines and surgical specimens. Additionally, the amount of material used in this demonstration (100 ug) is readily obtainable from core biopsies. However, for immuno-SRM to become a viable platform for tissue-based clinical diagnostics, technical hurdles will need to be overcome. For example, the composition of human tumors is highly heterogeneous, consisting of tumor cells, blood and lymphatic vessels, infiltrating immune cells, stromal elements, etc. Additionally, the proportion of the tumor that corresponds to each of these elements is also highly variable, not only amongst different patients but also amongst different anatomical locations within the same tumor. Because tumor biomarkers generally are produced by only one component of the tumor (commonly the cancer cell component), simply normalizing the total amount of input material (i.e. tumor protein lysate) in the assay will not account for heterogeneous composition and thus may yield unreliable results.

One possible approach to dealing with tumor composition heterogeneity is to use laser capture microscopy (LCM) to isolate the specific tumor component producing the biomarker. LCM has been used to harvest cells for LC-MS/MS analyses, using anywhere from 30,000–200,000 cells to obtain microgram quantities of protein[46, 47]. The advantage of this technique is that specific populations of cells (e.g. those expressing the biomarker of interest) can be targeted for harvest and immuno-SRM analyses. Disadvantages are that it is labor-intensive, requires specialized equipment, and samples a small amount of tissue that may not be representative of the entire tumor[48]. A second possible approach is to develop a panel of protein markers that are specific to each of the tumor components, and to use the relative abundances of these “component markers” to normalize the biomarker measurements across samples, thus accounting for heterogeneity in tumor composition.

Of note, there are precedents for measuring tissue biomarkers from grossly dissected tumors, by ensuring that the input material is representative of the tumor en masse. For example, the American Society of Clinical Oncology recommends measurement of urokinase plasminogen activator and plasminogen activator inhibitor 1 (uPA/PAI-1) by ELISAs on a minimum of 300 mg of fresh or frozen breast cancer tissue for the determination of prognosis in patients with newly diagnosed, node negative breast cancer[2]. In a second example, the RNA-based Oncotype DX diagnostic test (a clinically validated, high-complexity, multi-analyte reverse transcription–PCR genomic test that predicts the likelihood of breast cancer recurrence in early-stage, node-negative, estrogen receptor–positive breast cancer) selects tissue blocks containing the greatest amount of invasive breast carcinoma that is morphologically consistent with the submitting diagnosis and has the least amount of noninvasive mammary epithelium; samples with metabolically active nontumor elements constituting >50% of the tissue have those elements dissected out before sample extraction[49].

There are limitations of immuno-SRM assays compared with IHC. First, IHC can be used to determine the subcellular localization of proteins (e.g. nuclear versus cytoplasmic). While it may be possible to develop extraction methods to differentially recover nuclear versus cytoplasmic protein from tissues for immuno-SRM analysis, this has not yet been demonstrated. Hence, for rare biomarkers whose diagnostic utility is based not on total protein abundance but rather on mis-localization, IHC may be required. A second potential disadvantage of immuno-SRM assays for tissue diagnosis is that these assays will not be able to distinguish between tumors containing a small subpopulation of biomarker-positive cells versus tumors showing low level biomarker expression in 100% of the cells. Although there may be some instances where this distinction is important, in many cases pathologists have generated composite scores combining the percent of positive cells with a subjective measure of expression level (e.g. the Allred score for ER status in breast cancer[45]). In these cases, immuno-SRM assays may actually be advantageous because the scoring of expression level would no longer be subjective. Third, immuno-SRM measurements depend upon reproducible trypsin digestion. While many proteins are readily digested with trypsin, some are notoriously difficult. Hence, for each proteotypic peptide, the reproducibility of digestion will need to be assessed.

We note that quantification of membrane proteins by immuno-SRM in adherent cell lines presents a unique potential preanalytical variable. This is because adherent cells typically are recovered from the growth plate by trypsinization, which could cleave extracellular protein domains and thus affect the recovery and reproducibility of quantification of membrane proteins. In such cases, where possible, selecting proteotypic peptides from the intracellular domain may help to mitigate this effect.

Finally, ER and HER2 proteins have been measured in multiplex from breast cancer biopsies using bead-based sandwich ELISA assays[50, 51]. An advantage of these assays is that they can be run in higher throughput compared to currently configured immuno-SRM assays. In addition, up to nine different proteins were assayed as part of the multiplex samples. The plex number was limited, however, by non-specific interference between the individual assays. Conversely, we have previously shown that peptide-based immuno-SRM assays can be readily multiplexed using as many as 31 antibodies[17]. Immuno-SRM relies on mass spectrometry for detection, which provides absolute specificity, a decided advantage over ELISA assays. The need for only one antibody also makes immuno-SRM assays relatively easier to construct, less costly, and more rapid to configure than ELISA assays.

Abbreviations

ER

estrogen receptor

FISH

fluorescence in situ hybridization

HER2

receptor tyrosine-protein kinase erbB-2

IHC

immunohistochemistry

SID

stable isotope dilution

SRM

selected reaction monitoring