Automated Measurement of Multiple Cancer Biomarkers Using

Mar 31, 2015 - Phone: +82-31-920-2011., *E-mail: [email protected]. ... For the automated measurement of biomarkers, a cytokeratin-based biomarker ...
0 downloads 0 Views 2MB Size
Article pubs.acs.org/ac

Automated Measurement of Multiple Cancer Biomarkers Using Quantum-Dot-Based Microfluidic Immunohistochemistry Seyong Kwon,† Chang Hyun Cho,† Eun Sook Lee,*,‡ and Je-Kyun Park*,†,§ †

Department of Bio and Brain Engineering, Korea Advanced Institute of Science and Technology (KAIST), 291 Daehak-ro, Yuseong-gu, Daejeon 305-701, Republic of Korea ‡ Research Institute and Hospital, National Cancer Center, 323 Ilsan-ro, Ilsandong-gu, Goyang-si, Gyeonggi-do 410-769, Republic of Korea § KAIST Institute for the NanoCentury, 291 Daehak-ro, Yuseong-gu, Daejeon 305-701, Republic of Korea ABSTRACT: We report an automated multiple biomarker measurement method for tissue from cancer patients using quantum dot (QD)-based protein detection combined with reference-based protein quantification and autofluorescence (AF) removal. For multiplexed detection of biomarkers in tissue samples, visualization of QDs on cytokeratin was performed to create a multichannel microfluidic device on sites with dense populations of tumor cells. Three major breast cancer biomarkers (i.e., estrogen receptor, progesterone receptor, and human epidermal growth factor receptor 2) were labeled using QDs successively on cancer cells in tissue sections. For the automated measurement of biomarkers, a cytokeratin-based biomarker normalization method was used to measure the averaged expression of proteins. A novel AF-removal algorithm was developed, which normalizes the reference AF spectra reconstructed from unknown AF spectra based on random sampling. For accurate quantification of QDs, we automatically and accurately removed the AF signal from 344 spots of QD-labeled tissue samples using 240 reference AF spectra. Using analytical data with 10 tissue samples from breast cancer patients, the measured biomarker intensities were in good agreement with the results of conventional analyses.

C

protein detection.6 Although these analytical systems have resulted in improvements in biomarker detection in cancer tissue samples, biomarker analysis remains semiautomated and semiquantitative, because important parameters, such as the threshold intensity of the background signal must be determined manually by pathologists.7−9 There is a demand for multiple biomarker analysis, as well as the requirements of quantitative biomarker analysis, both in the study of cancer and in diagnosis. New biomarkers have been discovered because of improvements in the understanding of the causes of cancers and the development of new drugs,10−13 and there is a higher demand for new technologies or assays that can be used to analyze a large number of biomarkers simultaneously.2 Although multicolor fluorescence labeling methods can enable multiplexed protein detection, the number of proteins that can be detected simultaneously is limited, and such methods also require considerable time.14 Therefore, a lack of suitable detection technologies has limited the advancement of biomarker discovery. We have previously exploited microfluidic technology for the multiplexed detection of proteins.15,16 Nevertheless, the use of

onventional cancer biomarker measurement systems that are based on immunohistochemistry (IHC) are limited in the operational efficiency and analytical accuracy. For example, immunoperoxidase staining is commonly used to visualize target proteins in tissues, using a color-producing reaction of peroxidase.1 This method is adequate for qualitative analysis of proteins to determine the tissue morphology and protein expression as either negative or positive; however, because of the heterogeneous nature of cancers, quantitative analyses of cancer biomarkers are important to better understand and diagnose individual cancers.2 Quantitative biomarker analysis is also required in hospitals. Consider, for example, of the determination of equivocal (2+) scores of epidermal growth factor receptor 2 (HER2), which requires further genomic tests.3 The conventional scoring method is dependent on the judgment of an individual pathologist, and so the credibility of the scoring results based on a semiquantitative approach may be problematic due to its subjective nature.4,5 To address these quantification issues with characterizing protein expression, graphical image analysis software products have been developed to quantitate protein expression in cancer tissue samples. These software packages can help to remove background signals and measure the intensity of stained proteins numerically. Moreover, various types of fluorescencelabeling molecule have been used to increase the sensitivity of © 2015 American Chemical Society

Received: October 3, 2014 Accepted: March 31, 2015 Published: March 31, 2015 4177

DOI: 10.1021/acs.analchem.5b00199 Anal. Chem. 2015, 87, 4177−4183

Article

Analytical Chemistry

Figure 1. Schematic diagram showing the biomarker multiplexing method for tissue samples. Multiple biomarkers are labeled using the corresponding antibodies (Abs) using a multichannel microfluidic device on the tumor-specific antigenic site in a tissue section. (A) A 4-μm section of cancer stained using anticytokeratin Abs and QD525-IgGs. (B) Following excitation of QD525, the cancer target location is selected. (C) The multichannel microfluidic device is aligned on the cancer-cell area of the tissue section. Abs for each biomarker are incubated in the respective microchannels. (D) QD605-IgGs are then incubated to label the Abs tagged on biomarkers. (E) Following excitation of QD525 and QD605, the intensities of these QDs are quantified spectroscopically.

procedure was completed, tumor specimens were cut into 4-μm sections, dried for 1 h at room temperature, and then incubated for 1 h at 60 °C. ER, PR, and HER2 proteins in the tissue samples were stained using the conventional diaminobenzidine staining method, which is a colorimetric assay, and biomarker expressions were scored by pathologists at the National Cancer Center. Materials for Immunostaining. Rabbit polyclonal Abs were used for the detection of breast cancer biomarkers, and were supplied by Novus Biologicals, USA (ER) Novus Biologicals (PR) and Dako, Denmark (HER2). Mouse monoclonal Abs for the targeting of cytokeratin protein (Dako) were used to determine the tumor-specific antigenic sites. QD-tagged secondary Abs were used to label the primary Abs. QD525 goat antimouse IgGs (Invitrogen, USA) and QD605 goat antirabbit IgGs (Invitrogen) were used to label cytokeratin with QD525 and to label the biomarkers with QD605. TRIS-buffered saline and Tween 20 were used as a washing buffer, and a mixture of 2% BSA, 5% goat serum, and phosphate-buffered saline (PBS) was used as a blocking solution. Target retrieval solution (Dako) was used to implement a microwave antigen-retrieval technique. Design and Fabrication of the Microfluidic IHC Device. A microfluidic device consisting of 40-μm-high, 400μm-wide, and 5 mm-long rectangular microchannels arranged in parallel was used to incubate the proteins in each microchannel. The gap between microchannels was 100 μm to partition the channels. Fluid flow was driven by the withdrawal mode of a syringe pump connected to the outlet. The diameter of the biomarker reservoirs was 1.5 mm for compatibility with micropipette tips. The resistances of each of the microchannels were equal to provide equal flow rates. Multiplexed Immunostaining of Tumor-Specific Site in Cell Blocks or Tissue Sections. Tissue sections and cell blocks were dewaxed in xylene and rehydrated using a graded series of ethanol solutions. Following hydration, a microwavebased antigen-retrieval technique was used, and the samples

a multichannel microfluidic devices for multiplexed antibody (Ab) incubation requires improvement, because the microchannel-based staining procedure lacked the ability to locate densely populated tumor regions in tissues. Here, we describe a novel concept for simultaneous detection and automated quantification of the expression of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2) in breast cancer tissue using quantum dots (QDs). To achieve parallel multiplexed biomarker detection at the tumor-specific sites in tissue sections, we exploit a visualization method using a cytokeratin-based tumor-specific antigenic site selection strategy. To automate the measurements, QD-based quantitative spectroscopy is introduced, which can obtain averaged signals for the biomarkers using QD double staining and cytokeratin. Measurements of ER, PR, and HER2 using our system are compared with pathologically scored values using a conventional IHC method. Furthermore, a novel autofluorescence (AF) removal process is introduced by reconstructing unknown AF spectra from reference AF spectra; this ensures accurate removal of the background signal from the original protein signals, which leads to more accurate quantification of QDlabeled biomarkers.



EXPERIMENTAL SECTION Preparation of Tissue Specimens. Human tissue samples were obtained from the National Cancer Center (Goyang, Korea), with the corresponding written consent provided by the patients or their relatives. This study was approved by the Institutional Review Boards at the National Cancer Center and the Korea Advanced Institute of Science and Technology. Tissue samples were fixed for 24 h in 4% neutral-buffered formalin, Bouin’s fixative, acetic formalin alcohol, or 4% or 10% unbuffered formalin, 4 h in PreFer (Anatech; Battle Creek, MI) or Pen-Fix (Richard Allen Scientific; Kalamazoo, MI), or 48 h in 4% neutral-buffered formalin. When the paraffin embedding 4178

DOI: 10.1021/acs.analchem.5b00199 Anal. Chem. 2015, 87, 4177−4183

Article

Analytical Chemistry

signal from the QDs was measured using a spectroscopic system (Figure 1E). Prior to demonstrating our system with tissue samples, cell blocks were fabricated to simulate tissue blocks, which simplified the analysis because of the homogeneous composition. To visualize the Ab staining, anti-ER and anticytokeratin Abs were incubated in MCF7 cell-block sections with QD605IgGs and QD525-IgGs, respectively. Anti-HER2 and anticytokeratin Abs were also labeled on the SKBR3 cell-block sections using QD605 and QD525, respectively. Interestingly, proteins were stained with QDs more intensely at the edges than the center of the cell blocks (data not shown). This appears to be due to nonuniform delivery of reagents to the cell block during the fixation process, and is the reason that there was more staining at the edges of cell block than at the center, because reagent delivery is strongly affected by diffusion. We infer that the tissue samples will exhibit the same problem. Consequently, inhomogeneous staining may result in misidentification, even if the same cell block or tissue sample is used. For example, the maximum signal for HER2 in a SKBR3 cell-block section was 133.37 and the minimum was 81.45; it follows that if the value of 133.37 is correct, improper choice of sample location will result in a drop in the signal by 39%. This is a serious problem in cancer diagnosis because it can lead to inaccurate biomarker scoring, and the accuracy of biomarker scoring is critical in determining the treatment strategy. To overcome this problem, we have previously reported a loading-control based protein quantification method;19 however, β-actin is not an optimal choice as a loading-control for tissue samples due to a lack of specificity toward tumor cells. To validate the use of cytokeratin as a loading control, QD525cytokeratin was colabeled using QD605-biomarkers, and the intensities of the QD605 signals were normalized to those of QD525. Fluorescence spectra were obtained from multiple locations in each cell block, and the coefficient of variation (CV) was compared, not only between locations within a single sample but also between samples. As shown in Figure 2, the intensities of the QD605 signal prior to normalization varied significantly among locations, even though the cell numbers were similar. Following the cytokeratin-based biomarker normalization, the CVs decreased from 24.82−49.58% to 9.69−13.03% (Figure 2C and D). In addition, the CVs of the averaged sample intensity (average intensity values between samples) decreased markedly following cytokeratin-based biomarker normalization (from 9.01−17.14% to 2.35−6.21%). This suggests that cytokeratin is suitable for use as a loading control for cancer cells, and can compensate for variations in the staining not only between locations within a sample but also among samples. Although we used cytokeratin as a cancercell-specific loading-control here, other cancer-specific markers can be used with this approach. AF Removal in Tissue Samples. Conventional AF removal methods depend on the color separation of fluorescence images, and the acquisition of signals with a bandwidth that corresponds to the particular fluorescence molecules used. Some of fluorescence signals can be lost due to the overlap between the original signal and the AF. Accordingly, each fluorescence image should be postprocessed using image analysis software, which adds complexity and leads to problems with standardization. Because different image analysis software packages implement different analysis algorithms, some of parameters (such as threshold values for the rejection of background signals) should also be determined

were treated for 20 min in target retrieval solution (Dako) at pH 9 and 750 W. After the sample slide was cooled, it was treated in blocking solution to prevent nonspecific binding of secondary Abs. To determine the tumor-specific antigenic sites, a mouse monoclonal anticytokeratin Ab was added to the on the section slide and incubated for 30 min. After they were washed with TRIS buffered saline and Tween 20, QD525 goat antimouse IgGs (QD525-IgG) were added, followed by incubation for 1 h, slides were then placed under a fluorescence microscope (Olympus IX72; Japan) to visualize the region with the deposited QD525. A multichannel microfluidic device was then aligned on the area where the cytokeratins were highly and widely expressed. After mounting of the microfluidic device on the section slide, the inlets were filled with different rabbit primary Abs, and a syringe pump in withdrawal mode was used to induce flow of primary Abs over the surface of the tissue section; thereby primary Abs were effectively bound to their target proteins. This process took 30 min, and required only 5 μL of each primary Ab solution, due to the small volume of the microchannels. Following incubation for 1 h, images and spectroscopy data of the QD605 goat antirabbit IgGs (QD605IgG) were obtained from the mounted slides. Biomarker Signal Processing. Quantitative spectroscopy was carried out to quantify the QDs. A spectrometer (QE65000; Ocean Optics) equipped with a fluorescence microscope (IX72; Olympus) was used, and the QD signal was obtained using a 40× objective lens. Each protein corresponds to a QD605 signal at each stripe, where a multichannel microfluidic device was located. Ultraviolet excitation and acquisition of the emission signal from the QDs were carried out using the same objective lens; thereby the selection of measurement locations is flexible. To acquire AF spectra from tissue sections, they were deparaffinized in xylene and were mounted with a cover slide using mounting media. Fluorescence spectra were obtained from 240 randomly selected locations in the tissue sections, which were obtained from eight breast cancer patients. Protein quantification based on QD labeling was processed in the following order: (1) spectroscopic acquisition of the selected location in the tissue specimen; (2) AF removal; (3) obtain peak values from the QD525 and QD605 intensities; (4) normalize the QD605 signal to the QD525 signal (the QD605 signal corresponds to QDs on the target protein, and the QD525 signal to QDs on the cytokeratin).



RESULTS AND DISCUSSION Biomarker Quantification and Multiplexing Strategy for Breast Cancer Tissue Samples. Figure 1 shows the scheme for biomarker multiplexing based on visualization of the cytokeratin-stained area. Cytokeratin is a protein with keratincontaining intermediate filaments, is found in epithelial cells and has been used in IHC as a masking marker to distinguish epithelia from stromal elements of tissue.17,18 The tumor area was revealed by incubating the anticytokeratin Abs and labeling QD525-IgGs to locate a multichannel microfluidic device on the site of tumor-specific antigens (Figure 1A−C). After the multichannel microfluidic device was aligned on the tissue section, anti-ER Abs, anti-PR Abs and anti-HER2 Abs were injected through each microchannel and incubated. The QD605-IgGs were then labeled using the anti-ER, anti-PR and anti-HER2 Abs, which were tagged on the tissue section (Figure 1D). To quantify the expression of biomarkers, the 4179

DOI: 10.1021/acs.analchem.5b00199 Anal. Chem. 2015, 87, 4177−4183

Article

Analytical Chemistry

is because it measure only the intensities at every wavelength; therefore, the method is not affected by differences among spectrometric equipment. Furthermore, highly accurate quantification of biomarkers is possible with this method, in case AF of tissues was precisely removed from the spectra. Nevertheless, this approach has not been widely used, due to the absence of a robust protocol and difficulties of tissue AF removal. Tissue AF results from light that is absorbed and scattered by the microand nanostructures of endogenous proteins such as fibronectin, elastin and collagen, and from nucleic acids and tissue preparation processing.20,21 Accurate AF removal during QD signal acquisition from tissue samples is critical for quantification of the cellular components. AF removal in tissue is not straightforward, however, and has been a challenging issue in QD-based biomarker tissue analysis. This is because of the geometrical heterogeneity of the tissue sections. For quantitative analysis of AF in tissue, fluorescence spectroscopy was carried out using 240 observation locations which were randomly selected from samples from 10 breast cancer patients. The intensity of the AF signal from tissue samples is large at all wavelengths, and the spectral shapes are complex. Ghazani et al. subtracted AF background from tissue samples using preacquired AF data from a fixed area of the tissue microarrays.22 However, this is not effective for tissue samples because of the difficulties in aligning the same area where the AF data were acquired prior to QD-staining. Furthermore, the tissue AF vary with time; indeed, such a timedependence of the AF signal was observed (data not shown). For these reasons, we developed a new AF-removal method to eliminate AF from QD-labeled tissue samples. The technique is based on AF reconstruction using AF spectral libraries. We assume that some of the AF spectra share common features, but with different ratios. With this hypothesis, we obtained 240 reference tissue AF spectra from unstained tissue samples from 10 patients as reference data, and created an AF removal

Figure 2. Site-dependent intensity of the QDs labeled on proteins in each cell section, and protein quantification using a cytokeratin-based normalization method. Quantitative analysis of the QD intensities for (A) MCF7 and (B) SKBR3 cell block sections. The QD605 intensities were spectroscopically collected and normalized to the QD525 intensity of cytokeratin from multiple spots using ER in MCF7 and HER2 in SKBR3. For each cell-block sample, the CV values were calculated with multiple locations in (C) the MCF7 and (D) the SKBR3 cell block sections. Two parameters were compared, that is, cytokeratin-based normalized intensities and the QD605 labeled biomarker intensities. CV values obtained from the average intensity value (AVG) of the cell-block samples are compared in the spectra.

by the user. Consequently, biomarker measurements may differ depending on the type of image analysis package and userspecific settings. The QD-based quantitative spectroscopy has numerous significant advantages in the quantification of cancer biomarkers in tissue compared with conventional graphical image analysis approaches, in terms of the accuracy of analysis, the ease of data processing, as well as the ease of quantification of proteins. This

Figure 3. Autofluorescence (AF) removal based on AF reconstruction and biomarker quantification in cancer tissue. (A) The AF-removal process of a QD-labeled tissue sample using the AF-reconstruction method. Spectra for QD-labeled tissue sample (i.e., QD-tissue signal) is inserted to the library of reference tissue AF spectra (panel i). The profile of the QD-tissue signal remains fixed, and each reference AF spectrum in the library is normalized to the QD-tissue signal at the wavelength of interest (red point in panel j). This wavelength range is not affected by QD signal (yellow region in panel j). One of the normalized reference AF signal is selected (i.e., selected AF in panel k), which has the smallest intensity difference between the QD-tissue signal and the normalized reference AF signal. AF-removed signal is obtained by subtraction of selected AF from the QDtissue signal (panel l). (B) Spectra of randomly sampled AF signals (i.e., the input AF) and reconstructed AF spectra (i.e., reconstructed AF) using the AF reconstruction method. (C) The similarity of the reconstructed AF spectra for selected numbers of AFs in library. 4180

DOI: 10.1021/acs.analchem.5b00199 Anal. Chem. 2015, 87, 4177−4183

Article

Analytical Chemistry

Comparison of Our AF Removal Method with a Conventional Method. To verify our AF removal method, we carried out multiplexed biomarker labeling, biomarker quantification, and AF removal using breast cancer tissue using our method and the conventional software-based image analysis method. We first identified the cytokeratin-stained area using anticytokeratin Abs and QD525-IgGs, then Abs for each biomarker (i.e., ER, PR, and HER2) were labeled using QD605 in the region of the tumor cells. Figure 4A shows fluorescence

algorithm which uses reference tissue AF library for the accurate removal of AF from the QD-labeled tissue samples. Figure 3 shows our AF removal method used to process with QD-stained tissue samples based on the AF-reconstruction method. Spectra for QD-labeled tissue sample (i.e., QD-tissue signal) is inserted to the library of reference tissue AF spectra (reference AFs in Figure 3A-i). The profile of the QD-tissue signal remains fixed, and reference AF spectra in the library are normalized to the QD-tissue signal at the particular wavelengths. This wavelength (shown by the red point in the spectra in Figure 3A-j) should be in the wavelength range that is not affected by QD signal (in this case, yellow region of wavelengths in Figure 3A-j). Although one wavelength point is selected in this case, more selection of wavelength point improves accuracy of AF removal. Because QD525 and QD605 were used in our case, multiple points in the wavelength ranges 430−470 and 685−700 nm were used. The intensity differences between the QD-tissue signal and the normalized reference AF were calculated in these wavelength ranges. Consequently, one of the normalized reference AF signal is selected (i.e., selected AF in Figure 3A-k), which has the smallest intensity difference between QD-tissue signal and normalized reference AF signal. Finally, an AF-removed signal is obtained by subtracting the selected signal from the QDtissue signal (Figure 3A-l). To verify this hypothesis quantitatively, we reconstructed randomly sampled AF spectrum (i.e., input AF) using the reference AF spectra in the library (Figure 3B). To verify the accuracy of AF reconstruction, we defined the similarity rate of reconstructed AF spectra, A, as a new parameter, which represents the similarity of reconstructed AF spectra as A=

∑w |Iwd − R wd| d

∑w Iwd

Figure 4. Comparison of biomarker quantification between our spectroscopic method and the conventional software-based image analysis method. (A) Fluorescence images of a tissue slide from a breast cancer patient were labeled with QD605 (ER, PR, and HER2) and QD525 (cytokeratin). (B) Fluorescence spectra of proteins obtained from the same locations as the images shown in panel A. (C) Biomarker (i.e., ER, PR, and HER2) quantification results for our spectroscopic method and the conventional software-based image analysis method.

× 100(%)

d

where wd is the desired wavelength range, Iwd is the input AF spectrum, and Rwd is the reconstructed AF spectrum. Using a preselected number of reference AF spectra, we reconstructed 30 randomly sampled AF spectra from 10 tissue slides, and calculated the similarity rate of the reconstructed AF spectra (i.e., reconstructed AF in Figure 3B). For each of the selected numbers of AF libraries, a random selection of AF spectra from the entire AF library of 240 was taken at least more than three times. Interestingly, even with a small number of reference AF spectra (i.e., 20), the similarity rate of reconstructed AF spectra was 89−90% (Figure 3C). Our algorithm was able to reconstruct AF spectra with more than 90% similarity of unknown AF spectra with over 100 samples. Using this algorithm, we successfully removed AF signals from QD-labeled tissue spectra obtained from 344 locations on QD-stained tissues using 240 reference tissue AF spectra, which were collected from various subtypes of breast cancer tissue, including basal-like, luminal, and HER2 overexpressive. This was verified from observations of Gaussian peaks from the AFremoved tissue spectra. To use this algorithm, sampling of reference tissue AF spectra should be completed for each laboratory or hospital, considering the interlaboratory or the interhospital quality variation and differences in spectroscopy apparatus. Clearly, the more reference tissue AF spectra in the library, the more accurate the reconstruction. Note that each AF removal process was carried out automatically, because the algorithm is relatively straightforward.

images collected from the ER, PR and HER2 stained regions. Figure 4B shows spectra of respective fluorescence images from Figure 4A. The AF removal process was performed, and the results of this also are shown in Figure 4B, where the peak wavelength of 525 nm corresponds to the intensity of QDs, that is, cytokeratin. The peak wavelength corresponding to ER, PR, and HER2 was 605 nm. Although the absolute peak at 605 nm for PR was stronger than that for ER, cytokeratin-based normalized value for ER was similar to that for PR (Figure 4B). This is because the quantified values using our method for ER and PR were normalized to cytokeratin (with a peak at 525 nm) in each of the spectra, and the intensities of cytokeratin (QD525) in each of the spectra shown in Figure 4B reflect the cell density and staining quality in respective spots (Figure 4A). We compared our spectroscopic biomarker quantification results (i.e., cytokeratin-based biomarker normalization and AF removal based on reconstruction) with quantification results obtained using the conventional software-based image analysis method (Figure 4C). Conventional software-based image analysis for biomarker quantification was carried out using color-specific channel separation of fluorescence images and threshold selection for rejection of the background signal. The dimension of the tumor cell area were measured to obtain an 4181

DOI: 10.1021/acs.analchem.5b00199 Anal. Chem. 2015, 87, 4177−4183

Article

Analytical Chemistry average of the biomarker signal from the cancer cells. The QDtissue signals represent biomarker intensities at 605 nm normalized to the intensities at 525 nm without AF removal. As shown in Figure 4C, the biomarker quantification results using the conventional software-based image analysis method were similar to the spectroscopic results without AF removal. We may therefore infer that the conventional software-based biomarker quantification results contain AF signals. Using our analytical methods, the ER and PR expressions were measured and found to be lower than when using the conventional method, whereas the HER2 expression was found to be higher than with the conventional method. Based on the spectroscopic measurements, the QD605 intensities of ER and PR were 20.31 and 55.07, respectively; however, without AF removal, the intensities 124.41 and 174.85, respectively. Without AF removal, the signals for ER and PR were stronger by factors of 6.13 and 3.18, respectively, than the intensities measured using our method. Consider that pathological scores of this tissue sample was ER-/PR-/HER2 3+ (as determined by pathologists); the results from our measurement system distinguish negative and 3+ scores more than those obtained from the conventional software-based image analysis method. If the AF signal for the tissue sample shown in Figure 4 were large, and if there were insufficient AF removal using the software-based image biomarker analysis method, then the measured signals from the three biomarker (i.e., ER, PR, and HER2) would overestimate the biomarker expressions. These will result in a scoring result of ER+/PR+, which may in turn result in poor treatment decisions for the expression of ER and PR. If the biomarker expressions were originally positive for all biomarkers, and if the fluorescence intensities were sufficiently strong to allow us to neglect the AF signal, accurate removal of AF would not be necessary. However, if the expression levels of biomarkers were relatively low, such as with the sample shown in Figure 4, inaccurate removal of the background signal may lead to poor treatment decisions. In this case, if ER and PR were falsely determined as “positive”, this may lead to prescription of an inappropriate drug, such as tamoxifen or another hormonal drug that targets ER. This shows that inaccurate or insufficient AF removal may have serious consequences, especially when quantifying weak or negative biomarker expressions. Simultaneous Detection and Quantification of ER, PR, and HER2 in Cancer Tissue. Breast cancer tissue samples were obtained from 10 patients, and tested using our biomarker multiplexing system, and ER, PR and HER2 were quantified using the cytokeratin-based biomarker normalization analysis and the AF-removal. The three biomarkers were stained in parallel, depending on the shape of microchannel (Figure 5A). To investigate heterogeneity of parallel tissue biomarker analyses, we verified our results using pathologically scored tissues for ER, PR, and HER2. Tissue samples from 10 breast cancer patients were stained with QDs and compared with the score obtained by pathologists using a conventional IHC method (Table 1). When the cutoff value for biomarker expression was (arbitrarily) set to 0.5 to determine positive or negative scores for ER and PR, the results of the conventional scoring were in good agreement with the results from our system. Furthermore, scores of 0, 1+, and 3+ for HER2 were distinguished precisely. Although this demonstration involved only 10 patients, the results show that this system has considerable potential for analysis of major breast cancer biomarkers. Furthermore, a sample with dimensions of

Figure 5. Biomarker quantification in tissue samples using the cytokeratin-based normalization method and AF-removal. (A) Parallel biomarker multiplexing with tissue samples. Cytokeratin was stained with anticytokeratin Abs and QD525. The bright blue corresponds to AF spectra of the tissue samples, and the red corresponds to QD605labeled biomarkers. These biomarkers are expressed at tumor-antigenic sites, as revealed by cytokeratin staining. (B) The results of cancer biomarker quantification for the three major breast cancer biomarkers ER, PR, and HER2. Results for ten patients were quantified using spectroscopy with the cytokeratin-based normalization method.

approximately 400 × 5000 μm could be utilized to characterize the breast cancer biomarkers ER, PR, and HER2. Our system has some limitations, however, and may not be adequate for, for example, rare cell discovery using IHC. Therefore, to use our system for clinical evaluation and treatment decision-making, further validations are required using a larger number of cancer patients to determine parameters for quantitative clinical evaluations, such as threshold values that can distinguish HER2 1+ and 3+. Although there was good agreement between the scores for HER2 3+ between our system and the scores obtained by pathologists using conventional IHC, we observed that there were significant variations in the intensities. As shown in Figure 5B, the HER2 intensity from patient 5 was significantly higher than those of the others. In addition, using our system, we find a score of 0.45 for ER for patient 2, which was scored as negatively by the pathologists using conventional IHC (Table 1). The scores for ER using our system for other patients who were conventionally scored positive were as follows: 0.59 (patient 3), 0.68 (patient 8), and 0.58 (patient 10). The conventional ER score for patient 2 may be misleading, as the ER intensity for patient 2 was determined to be “weak” using conventional method. These results imply that our system may enable more accurate quantitative analysis than conventional techniques, and therefore has considerable potential for quantitative cancer research applications.



CONCLUSION We have demonstrated a cytokeratin-based biomarker quantification system for automated measurements of major breast cancer biomarkers in tissue samples. Cytokeratin was used as a loading control to compensate for variations in the staining quality, not only between different locations within a sample but also between tissue samples. A novel AF-removal algorithm was developed to compensate for unknown AF signals using normalization to reference AF spectra to achieve accurate quantification of the QD signals that correspond to biomarkers. Using this simple algorithm, we succeeded in removing the AF signals from 344 spectra of QD-labeled tissue samples. Visualization of the tumor-cell area was carried out to realize microfluidic parallel biomarker multiplexing using cancer tissue samples. Using 10 samples from breast cancer patients, 4182

DOI: 10.1021/acs.analchem.5b00199 Anal. Chem. 2015, 87, 4177−4183

Article

Analytical Chemistry

Table 1. Comparison of Patients’ Biomarker Evaluation Results between Conventional Score and Quantified Value by Our Spectroscopic Biomarker Quantification Method Patient code biomarker evaluation ER PR HER2

conventional scorea quantified valueb conventional score quantified value conventional score quantified value

1

2

3

4

5

6

7

8

9

10

P 1.20 P 3.48 1+ 0.70

N 0.45 P 2.80 0 0.14

P 0.59 P 2.31 1+ 0.71

P 0.96 P 0.84 1+ 0.99

N 0.16 N 0.18 3+ 9.04

P 0.94 N 0.24 1+ 1.16

N 0.31 N 0.17 3+ 1.69

P 0.68 N 0.36 3+ 2.41

N 0.07 N 0.08 3+ 1.92

P 0.58 N 0.15 0 0.07

a

Conventional score was obtained by conventional IHC and pathologists (P = positive, N = negative). bQuantified value was acquired by our spectroscopic biomarker quantification method. (9) Rizzardi, A. E.; Johnson, A. T.; Vogel, R. I.; Pambuccian, S. E.; Henriksen, J.; Skubitz, A. P. N.; Metzger, G. J.; Schmechel, S. C. Diagn. Pathol. 2012, 7, 42. (10) Ferrara, N.; Hillan, K. J.; Gerber, H.-P.; Novotny, W. Nat. Rev. Drug Discovery 2004, 3, 391−400. (11) Rakha, E. A.; El-Sayed, M. E.; Green, A. R.; Lee, A. H. S.; Robertson, J. F.; Ellis, I. O. Cancer Am. Cancer Soc. 2007, 109, 25−32. (12) Dowsett, M.; Dunbier, A. K. Clin. Cancer Res. 2008, 14, 8019− 8026. (13) Yarden, Y.; Pines, G. Nat. Rev. Cancer 2012, 12, 553−563. (14) Liu, J.; Lau, S. K.; Varma, V. A.; Kairdolf, B. A.; Nie, S. Anal. Chem. 2010, 82, 6237−6243. (15) Kim, M. S.; Kim, T.; Kong, S.-Y.; Kwon, S.; Bae, C. Y.; Choi, J.; Kim, C. H.; Lee, E. S.; Park, J.-K. PLoS One 2010, 5, No. e10441. (16) Kim, M. S.; Kwon, S.; Kim, T.; Lee, E. S.; Park, J.-K. Biomaterials 2011, 32, 1396−1403. (17) Lane, E. B.; Alexander, C. M. Semin. Cancer Biol. 1990, 1, 165− 179. (18) Barak, V.; Goike, H.; Panaretakis, K. W.; Einarsson, R. Clin. Biochem. 2004, 37, 529−540. (19) Kwon, S.; Kim, M. S.; Lee, E. S.; Sohn, J. S.; Park, J.-K. Integr. Biol. 2014, 6, 430−437. (20) Baschong, W.; Suetterlin, R.; Laeng, R. H. J. Histochem, Cytochem. 2001, 49, 1565−1571. (21) DaCosta, R. S.; Wilson, B. C.; Marcon, N. E. Dig. Endoscopy 2003, 15, 153−173. (22) Ghazani, A. A.; Lee, J. A.; Klostranec, J.; Xiang, Q.; Dacosta, R. S.; Wilson, B. C.; Tsao, M. S.; Chan, W. C. W. Nano Lett. 2006, 6, 2881−2886.

the measured biomarker intensity was well correlated with the scoring results from pathologists using a conventional IHC method. This shows that our method is compatible with conventional scoring techniques, and that it can provide highly accurate quantification of QD-labeled proteins.



AUTHOR INFORMATION

Corresponding Authors

*E-mail: [email protected]. Phone: +82-31-920-2011. *E-mail: [email protected]. Phone: +82-42-350-4315. Author Contributions

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS This research was supported by a National Leading Research Laboratory Program (Grant NRF-2013R1A2A1A05006378) and a Converging Research Center Program (Grant 2011K000864) through the National Research Foundation of Korea funded by the Ministry of Science, ICT and Future Planning. The authors also acknowledge a Research Program (Grant 1211200-1) supported by the National Cancer Center, Korea.



REFERENCES

(1) Ramos-Vara, J. A. Vet. Pathol. 2005, 42, 405−426. (2) de Castro, D. G.; Clarke, P. A.; Al-Lazikani, B.; Workman, P. Clin. Pharmacol. Ther. 2013, 93, 252−259. (3) Dowsett, M.; Hanna, W. M.; Kockx, M.; Penault-Llorca, F.; Rüschoff, J.; Gutjahr, T.; Habben, K.; van de Vijver, M. J. Mod. Pathol. 2007, 20, 584−591. (4) Taylor, C. R.; Levenson, R. M. Histopathology 2006, 49, 411− 424. (5) Wolff, A. C.; Hammond, M. E. H.; Schwartz, J. N.; Hagerty, K. L.; Allred, D. C.; Cote, R. J.; Dowsett, M.; Fitzgibbons, P. L.; Hanna, W. M.; Langer, A.; McShane, L. M.; Paik, S.; Pegram, M. D.; Perez, E. A.; Press, M. F.; Rhodes, A.; Sturgeon, C.; Taube, S. E.; Tubbs, R.; Vance, G. H.; van de Vijver, M.; Wheeler, T. M.; Hayes, D. F. J. Clin. Oncol. 2006, 25, 118−145. (6) Xing, Y.; Chaudry, Q.; Shen, C.; Kong, K. Y.; Zhau, H. E.; Chung, L. W.; Petros, J. A.; O’Regan, R. M.; Yezhelyev, M. V.; Simons, J. W.; Wang, M. D.; Nie, S. Nat. Protoc. 2007, 2, 1152−1165. (7) Welsh, A. W.; Moeder, C. B.; Kumar, S.; Gershkovich, P.; Alarid, E. T.; Harigopal, M.; Haffty, B. G.; Rimm, D. L. J. Clin. Oncol. 2011, 29, 2978−2984. (8) Tuominen, V. J.; Ruotoistenmäki, S.; Viitanen, A.; Jumppanen, M.; Isola, J. Breast Cancer Res. 2010, 12, R56. 4183

DOI: 10.1021/acs.analchem.5b00199 Anal. Chem. 2015, 87, 4177−4183