Multiplexed detection of serological cancer markers with plasmon-enhanced Raman spectro-immunoassay

A plasmon-enhanced Raman spectroscopic assay has been developed for multiplexed detection of breast cancer markers—with high sensitivity and exquisite specificity, offering the potential of evaluating the breast cancer burden accurately.


Introduction
Despite recent advances in the understanding of breast cancer progression and in the development of therapeutic modalities, breast cancer remains a global problem with a signicant mortality rate and an equally substantial socio-economic burden. [1][2][3][4] Our rudimentary knowledge of local recurrence and distant metastatic breast cancer is primarily responsible for the continued loss of lives. While local breast cancer responds very well to therapy and has a 5 year survival near 98%, the 5 year survival rate for metastatic breast cancer that involves distant organs drops to a dismal 24%. 5 Extending life expectancies, therefore, requires sustained research in monitoring and managing recurrence and metastatic disease. Specically, sensitive measurement of changes in tumor burden will assist the development of optimal treatment strategies for metastatic breast cancer. Moreover, early detection of recurrence prior to diagnosis by conventional modalities such as radiographic imaging will allow surveillance of asymptomatic cancer survivors.
In this milieu, there has been a burgeoning interest in circulating biomarkers owing to their potential for diagnosis, prognostication and monitoring response to systemic therapies in the neoadjuvant, adjuvant and metastatic settings. 6 While promising data has recently been reported on circulating tumor cells and circulating tumor DNA, 7,8 serum-based glycosylated tumor markers, notably cancer antigen 15-3 (CA15-3), CA27-29 and carcinoembryonic antigen (CEA), represent the most mature panel for monitoring patients with metastatic disease. [9][10][11][12] These biomarkers are signicantly overexpressed in stage IV breast cancer patients, which contain much higher concentrations than normal levels of <30 U mL À1 , <38 U mL À1 and <10 ng mL À1 for CA15-3, CA27-29 and CEA, respectively. 9,13,14 Despite being endorsed by American Society of Clinical Oncology, however, their utility has been limited by the sensitivity and specicity of the individual markers. 15 To overcome this drawback, a shi in paradigm towards concomitant measurement of multiple markers has gained impetus. 16 Yet, current diagnostic techniques, including enzyme-linked immunosorbent assay (ELISA), radioimmunometric assay and Western blot, do not provide the necessary multiplexing functionality and additionally oen suffer from limited sensitivity and heavy interference from biological matrices. 17,18 Given these limitations, a single blood-based test for these tumor antigens is still to be incorporated into a clinical laboratory assay.
Here we present a multiplex surface-enhanced Raman spectroscopy (SERS)-based assay for sensitive and specic detection of the tumor antigen panel. Our approach combines spectroscopic imaging with tailored SERS probes, where the signal enhancement arises from the proximity of the Raman reporter molecule to the intense localized plasmonic elds created by the nanostructured metals. [19][20][21][22][23][24] The signal of this reporter transduces the presence (and concentration) of the tumor antigen at extremely low concentrations to a quantitative and reproducible spectral pattern. We designed a SERS chip that comprises pre-dened wells patterned in a quartz substrate. Each array is functionalized with monoclonal antibody (mAb) for different tumor antigens. Using a Raman microscope to scan the chip, the individual spectra are integrated into numerical algorithms for robust estimation of the expression levels. We show that this assay offers multiplexing capability in a single serum droplet ($2 mL) while achieving a high sensitivity and molecular specicity. We further developed a wide-area, compact Raman spectroscopic scanner that can sample the chip in a small fraction of the time necessary for standard chemical imaging. Collectively, these ndings underline the transformative potential of this assay for serum expression.

Results and discussion
We employed gold nanostars (GNS) as the basis for designing SERS probes with substantive signal enhancement and exceptional multiplexing capability (Fig. 1A). 22,23 By modulating the protrusion length and density as well as the core size, we optimally tuned the localized surface plasmon resonance (LSPR) of the GNS to 734 nm and observed that the thin silica coating caused a slight red-shi to 748 nm (Fig. 1B). The interplay between plasmonic enhancement and optical extinction causes the GNS with LSPR blue-shied (off-resonant) from the 785 nm excitation wavelength to provide the maximum net amplication in the colloidal suspension. 22,25 We embedded Raman reporter, 4-nitrothiophenol (4-NTP), on the GNS surface, which was then coated with a thin silica layer ($5 nm thickness, Fig. 1C and S3 †). The silica coating enables exible surface functionalization rendering the desired molecular specicity and prevents the leaching of 4-NTP during the processing and assay operations. Next, we used standard amine coupling chemistry to gra antibodies (CA15-3 monoclonal antibody (mAb), CA27-29 mAb and CEA mAb) to carboxyl group-modied SERS tags (Fig. S1 †). 23 Using the Raman microscope, we acquired spectra from 4-NTP, SERS tags and the mAb-modied SERS tags (SERS probes) for probe characterization (Fig. 1D). The acquisition conrmed that the signatures of the SERS tags and the CA15-3 targeted probes were identical to that of 4-NTP ( Fig. 1D and Table S1 †). Similar results were also observed for CA27-29 and CEA targeted probes (Fig. S2 †).
Additionally, each well in the SERS chip was functionalized with carboxyl group and activated with the standard amine coupling chemistry, followed by conjugation with the respective antibodies ( Fig. 1E and S1 †). Here the mAb molecules immobilized on the quartz slide act as the capture probe and the mAb molecules on the SERS tag surface serve as the recognition moiety on the detection probes for the biomarkers. Bovine serum albumin was used as the surface blocking reagent to avoid nonspecic adsorption of extraneous species on the chip surface ( Fig. S1 †). 26 The chip bound with biomarkers was then incubated in a solution containing SERS probes forming the sandwich assay conguration. Aer removal of the free SERS probes, the chip was subjected to spectral acquisition. We performed spectroscopic imaging, as opposed to single point measurements, to improve signal robustness through spatial averaging and to minimize sampling errors.
To determine the feasibility of the SERS chip for biomarker detection, we rst performed proof-of-concept experiments in PBS buffer media (Fig. 2). The sandwich conguration, in the presence of 100 U mL À1 CA15-3, faithfully reproduces the signal of the Raman reporter. In contrast, the control experiments conrm that there was no observable signal in the "blank" as also when only the SERS probe was added. The latter can be attributed to the fact that the SERS probes are easily washed away when the sandwich conguration via the antibody-antigen binding is not formed. In order to display the SERS chip response, we constructed spectral images based on the integral area of the 1570 cm À1 Raman peak ( Fig. 2A). We observed substantially brighter SERS images in the presence of CA15-3 antigen-the imaging equivalent of the single point spectral acquisition shown in Fig. 2B.
To investigate its applicability as a quantitative assay, we next examined the SERS response upon varying the biomarker concentrations in the ranges encountered in clinical practice ( Fig. 3A and B). 9,13,14 Concentrations both lower and higher than the clinically relevant levels were also included to obtain a comprehensive assessment of the dynamic range. Specically, six concentrations of CA15-3, CA27-29 and CEA were spotted on the SERS chip. The SERS response shows a consistent increase in intensity (brightness) due to more captured SERS probes per well with rising concentration for all three biomarkers. We also correlated the relative SERS response, in relation to the control, with the various biomarker concentrations (Fig. 3B). Substantive linearity was observed in the log-log calibration curve over the examined concentration ranges, 0.1 U mL À1 to 500 U mL À1 for CA15-3 (coefficient of determination R 2 ¼ 0.94) and CA27-29 (R 2 ¼ 0.95), and 0.1 ng mL À1 to 500 ng mL À1 for CEA (R 2 ¼ 0.97).
Next, we used the SERS chip for biomarker detection in serum ( Fig. 4A and S4 †). The logarithm of the SERS responses increases linearly with the logarithm of biomarker concentrations investigated with R 2 values equal to 0.98, 0.90 and 0.99 for CA15-3, CA27-29 and CEA, respectively. We further analyzed the binding characteristics of the biomarkers by tting the experimental data to Langmuir isotherms, which yielded dissociation constants of 95.9 U mL À1 , 83.1 U mL À1 and 113.2 ng mL À1 for CA15-3, CA27-29 and CEA, respectively (Fig. S5 †). Although the spectral intensity values are lower in sera than those obtained in buffer, the acquired proles and the response curves highlight the molecular specicity via the lack of interference from the myriad endogenous constituents of the sera. Additionally, we used multivariate regression analysis for concentration prediction as it exploits the entire spectral information (rather than focusing on a single peak) and has the associated advantage of Blank indicates that the SERS measurement is directly performed on CA15-3 mAb-modified quartz substrate without addition of either CA15-3 SERS probe or CA15-3 antigen; no biomarker indicates SERS acquired from the SERS platform in the presence of CA15-3 SERS probe but in the absence of CA15-3 antigen. SERS responses represent acquisition intensities when 100 U mL À1 concentration of CA15-3 antigen is incorporated to complete the sandwich assay. All experiments are triply performed in parallel, and the relative SERS response with respect to the blank is used to generate the SERS image. Scale bar in (A) is 20 mm. Highlighted area (1500 cm À1 to 1630 cm À1 ) in (B) indicates the area surrounding the characteristic Raman peak (1570 cm À1 ) that is used for construction of SERS images in (A) and the ensuing analysis.
noise averaging across the spectrum. Leave-one-out cross-validation was performed using partial least squares (PLS) regression ( Fig. 4B and S6 †). 27 Evidently, there is close agreement between the predicted and reference concentrations with R 2 values of 0.98, 0.99 and 0.99 for CA15-3, CA27-29 and CEA, respectively. Importantly, the limits of detection (LOD) were computed to be 0.99 U mL À1 , 0.13 U mL À1 and 0.05 ng mL À1 for CA15-3, CA27-29 and CEA, respectively. These values are signicantly smaller than the corresponding LOD values reported from the conventional methods, such as commercial ELISA kits (widely treated as the gold standard for proteomics assays): 5.0 U mL À1 for CA15-3, 3.8 U mL À1 for CA27-29 and 1.0 ng mL À1 for CEA. 28 We note that the SERS responses shown in Fig. 3 and 4A are slightly different for each antigen, which may be attributed to the different antibody-antigen binding affinities.
A key advantage of our platform is its multiplexing ability. To test this feature, we architected a 3 Â 3 array of sensing units with each row dedicated to measurement of a specic antigen and the three columns enabling triplicate measurements. A single drop of serum ($2 mL) spiked with differing quantities of the three cancer antigens was pipetted to cover the whole chip, followed by sequential addition of the mAb-SERS probes (Fig. 5A). During the incubation period, the serological markers and mAb-SERS probes together form the sandwich assay conguration with the capture probes on the corresponding wells. Without any other pretreatment, we employed spectroscopic imaging on the chip to render direct and simultaneous readout of the tumor antigen concentrations. We examined two serum samples spiked with different concentrations of the antigens. The antigen concentrations in the rst sample resembled the levels of a healthy individual whereas the concentrations in the second sample were consistent with observations in metastatic breast cancer patients (Fig. 5B). We observe that the rst sample generates a weak, yet observable, SERS signal. In contrast, the SERS intensity from the second specimen exhibits a signicantly larger response in each case. We also quantied the antigen concentrations on the basis of the acquired spectra and the previously formulated PLS calibration models. The predicted values show excellent agreement with the reference concentrations with relative errors of prediction of 10.4%, 3.0% and 6.0% for CA15-3, CA27-29 and CEA, respectively (Fig. S7 †). Relative standard deviations were calculated to be 13.5% (CA15-3), 4.0% (CA27-29) and 8.4% (CEA), which are deemed to be clinically acceptable. Furthermore, our result demonstrates the low interference from other biomarkers, i.e. robustness to cross-reactivity (stemming from the antibody-antigen affinity), despite the high biomarker concentrations in the serum specimen representative of the patient sample.
Finally, we assessed the feasibility of higher throughput SERS measurements using a simpler, portable imaging system. We developed a wide-eld compact scanning setup to address the limited sampling area and the substantive costs of a Raman microscope. Consisting of a laser diode and an air-cooled CCD imager, the atbed scanner offered a large eld of view (100 cm 2 ). Despite the system's relatively lower detection sensitivity, we observed that the acquired images still allow clear differentiation between the two spiked serum samples (Fig. 5C) with the sample mimicking breast cancer patient antigen levels exhibiting markedly higher SERS response. The wide eld of view enables direct visualization of the entire 3 Â 3 panel with a 5-fold reduction in acquisition time. Coupled with the facile readout of the atbed scanner, the SERS chip promises a highly sensitive and specic tool that can be further rened to create an inexpensive, point-of-care platform.

Conclusions
Rapid, multiplexed tumor antigen analysis could improve early disease diagnosis and therapy response monitoring. We have developed a new liquid biopsy tool for multiplex detection of a panel of circulating tumor antigens based on plasmon-enhanced spectroscopic imaging. The structured nanoprobes realize substantive signal amplication while the attached Raman reporter independently tailors the spectral response. Moreover, the nanoprobes have the excitation and emission spectral signatures in the clear near infrared window and are designed to suppress both intimate contact (Raman) and through-space (uorescence) enhancement of endogenous markers. We demonstrate that the proposed SERS platform shows high detection sensitivity. The strong proof-of-concept data generated here provides the needed momentum to pursue clinical feasibility studies for metastatic cancer diagnosis and longitudinal monitoring of chemotherapy response in breast cancer patients. This approach can also, in principle, be extended to detect circulating genetic and epigenetic markers such as microRNA and hypermethylated tumor DNA by substituting only the recognition moiety in the capture and detection probes. Finally, while we have employed breast cancer as the paradigm, this approach is generally applicable to other diseases including prostate, lung and colorectal adenocarcinomas, where pathologic conditions are complexly manifest in patterns of multiple biomarker expression levels.

SERS tag synthesis
SERS tags were synthesized according to our previously reported method with a slight modication. [22][23][24] Briey, gold nanostar (GNS) nanoparticles with the LSPR band maximum of 734 nm in aqueous solution were synthesized by employing the gold seed-mediated method. [22][23][24] The GNS nanoparticles were dispersed into deionized water with a concentration of 1.7 pM for further use. To prepare the SERS tag, a freshly prepared solution of Raman reporter (4-NTP, 10 mM) was added dropwise to 15 mL GNS colloid while subject to rapid magnetic stirring. Stirring was continued for another 30 min before adding 10 mL of freshly prepared APTMS ethanolic solution (50 mM). Aer After the incubation, a mixture of SERS probes functionalized with various mAb molecules are dropped. After vigorously washed with PBS buffer solution, the SERS assay panel is subject to the SERS assay. (B) Relative SERS response for healthy (red) and patient (blue) serum samples (for healthy sample, CA15-3:10 U mL À1 , CA27-29: 30 U mL À1 and CEA: 1 ng mL À1 ; for patient sample, CA15-3: 150 U mL À1 , CA27-29: 180 U mL À1 and CEA: 200 ng mL À1 ). The mean integral area over the examined region is divided by that from the blank to give the relative SERS response. Three independent experiments are performed in parallel for each type of serum sample. (C) SERS images acquired by Raman spectroscopic scanner. Images from spiked serum samples mimicking the concentrations observed in (i) healthy serum sample and (ii) patient serum sample. For both i and ii, the first row is for CA15-3, the second row for CA27-29 and the third row for CEA. Each experiment is triply performed in parallel. Scale bar in (C) is 1.5 mm.
stirring for another 30 min, the pH value of reaction solution was adjusted to 9-10 by addition of NaOH aqueous solution. Following this, 200 mL of freshly prepared 0.54 wt% sodium silicate solution was added slowly, and then stirred for one day. 5 mL anhydrous ethanol was subsequently added to generate a condensed silica layer. The reaction solution was kept standing for one more day, centrifuged and washed with anhydrous ethanol and deionized water, respectively. Finally, the solid was dispersed into 0.5 mL 1Â PBS buffer solution for further use.

Antibody conjugation to quartz chip and SERS tags
Assay panel functionalization. Our underlying principle here is that by associating a set of antibodies with a particular row, a combinatorial utilization of the same nanoparticle-surface species with multiple antibodies can be implemented. To make the SERS assay panel, quartz slides were used and cleaned by subsequent sonication in ethanol and water. To pre-dene the functional assay regions for different biomarkers, the Paralm was bonded onto the cleaned quartz chip with punched wells (3 rows Â 7 columns). In our assay, each row was dened for one type of biomarker. All of the following operations were carried out in a home-built humid chamber. Antibodies were immobilized onto the panel with pre-dened patterns through standard amine coupling chemistry. First, all wells were incubated over-night in an ethanol solution containing 100 mM TEPSA and then sequentially washed with ethanol and 1Â PBS buffer solution to achieve a carboxyl group-modied surface. The resulting carboxyl-terminated quartz array panel was activated by immersion in a PBS buffer solution containing 50 mM NHS and 200 mM EDC. Aer washing with the PBS buffer solution to remove excess NHS and EDC, the panel was incubated overnight in the buffer solution of 100 mg mL À1 CA15-3 mAb (CA27-29 mAb or CEA mAb). The non-specically bound antibodies were washed away with the 1Â PBS buffer.
To achieve high assay specicity, it is crucial to minimize the non-specic biomarker adsorption. In this work, we used BSA as the surface blocking reagent because of its excellent stability and biocompatibility. 25 The antibody-immobilized patterned panel was spotted and then incubated for 2 hours in a 1Â PBS buffer solution of 1 mg mL À1 BSA, followed by rinsing with 1Â PBS buffer solution. Next, the resultant assay was kept in the humid chamber for further SERS assembly.

Antibody-conjugated SERS tags
The antibody-SERS tag conjugates (SERS probes) were synthesized as detailed elsewhere in the literature. 23 First, the SERS tags with carboxyl groups were prepared by incubating 200 mL SERS tags overnight in a 0.12 M TEPSA buffer solution. The carboxyl group-modied SERS tags were washed twice with a 1Â PBS buffer solution, and then dispersed into a 1Â PBS solution contained 50 mM NHS and 200 mM EDC to activate the carboxyl terminal group. Aer the 2 hour incubation period, 100 mg mL À1 CA15-3 mAb (CA27-29 mAb or CEA mAb) was added onto the activated SERS tags in PBS buffer solution, and then incubated overnight. Unbound mAb residues were removed by centrifugation at 4000 rpm and subsequent washing with 1Â PBS buffer solution at least three times. The resultant SERS probes were re-dispersed into 0.5 mL 1Â PBS buffer solution for further use.

SERS assay of biomarkers
We performed SERS assay experiments in two different matrices, namely, in 1Â PBS buffer solution and in sera.
SERS assay in buffer. Various amounts of the biomarkers (CA15-3, CA27-29 or CEA) were spiked into 1Â PBS buffer solution to achieve a range of biomarker concentrations (0.1, 1.0, 10, 50, 100 and 500 U mL À1 for CA15-3 and CA27-29 antigens, 0.1, 1.0, 10, 50, 100 and 500 ng mL À1 for CEA antigen). These concentrations are selected as it spans the clinically relevant range from that typically encountered in healthy individuals to patients with advanced breast cancer. 9,13,14 2 mL of each biomarker solutions with various concentrations was spotted onto the corresponding pre-dened pattern (i.e., matrix arrangement of wells) on the SERS assay panel, and incubated for one hour. Next, the panel was carefully washed with 1Â PBS buffer solution to remove any traces of unbound biomarkers. Subsequently, 2 mL of SERS probes were spotted on the corresponding patterns and incubated for one more hour, followed by vigorous rinsing with 1Â PBS buffer solution. Finally, the sandwich assay was subjected to the SERS measurements.
SERS assay in serum. Similarly, the biomarker serum solutions at the aforementioned range of CA15-3, CA27-29 or CEA concentrations were prepared by spiking various amounts of asreceived biomarkers into FBS serum. The lack of pre-existing biomarkers in FBS serum precluded any potential interference. SERS assays of biomarkers in sera were prepared following a similar procedure to that outlined above for the buffer solution.

SERS measurements
SERS measurements were performed using a home-built, confocal, inverted Raman microscope. A Ti:Sapphire laser of 785 nm wavelength (3900S, Spectra-Physics) was used as the excitation source and a 1.2 NA, 60Â water immersion objective lens (Olympus UPLASPO60XWIR) was used to focus the laser light to and collect the Raman-scattered light from the assay, as detailed in our previous work. 29 The backscattered light was collected by a 50 mm multimode ber (Thorlabs M14L01), delivered to a spectrograph (Holospec f/1.8i, Kaiser Optical Systems) and the dispersed light was nally detected by a TEcooled, back-illuminated, deep depletion CCD (PIXIS:100BR eXcelon, Princeton Instruments). The SERS microscopic images were obtained using dual-axis galvo mirrors (CT-6210, Cambridge Technology). The SERS response (RSERS) at 1570 cm À1 , characteristic of the Raman reporter (4-NTP), was computed by considering the integral of the area under the curve in the range of 1500 cm À1 to 1630 cm À1 and was used to construct the SERS images. The spatial average ( R SERS ) over the scanning region was used to calculate the SERS response in order to improve prediction robustness, given by: IðuÞdu (1) where N is the number (20 Â 20) of spectra obtained over the scanning region, u is the Raman shi in the integral range (1500 cm À1 to 1630 cm À1 ), and I(u) is the Raman peak intensity at Raman shi, u. All spectral measurements were obtained with an exposure time of 0.5 s at 4 mW laser power on the sample, unless otherwise noted. Furthermore, in order to investigate the feasibility of high throughput, low-cost measurements, a atbed Raman spectroscopic scanning system was constructed. Updated in design from our diffuse reectance and autouorescence scanner reported previously, 30 wide area spectroscopic imaging capability is achieved by mechanically scanning the sample on top of inverted Raman imaging system with a quartz substrate. Spectral recording time was 100 ms per pixel. Here, a 785 nm compact solid-state laser is used as the excitation source and the collected light is recorded on a portable spectrometer.

Characterization
Extinction spectra for the GNS and SERS tags were recorded on a Shimadzu UV-2401 spectrometer. Transmission electron micrographs (TEM) were acquired using the FEI Tecnai G2 Spirit TWIN transmission electron microscope at an accelerating voltage of 120 kV. The sample was dropped onto ultrathin Formvar-coated 200-mesh copper grids (Ted Pella, Inc.) and le to dry in air.

Data analysis
To evaluate the efficacy of the present assay for quantitative concentration measurements, we performed partial least squares (PLS) regression analysis. Specically, PLS calibration models were tested using the leave-one-sample-out cross-validation procedure for each biomarker. In this routine, one concentration is le out when developing the calibration model and the resultant model is used to predict concentrations of the le out concentration spectra. 30 This procedure is repeated until all concentrations are le out and each of the concentrations has been predicted. In particular, the calibration models are developed using 50 spectra (5 concentrations with 10 spectra per concentration for each biomarker), and the predictions are performed on the remaining 10 spectra (1 concentration). Furthermore, the limit of detection (LOD) of the developed SERS assay is calculated from the best t line obtained between the predicted concentrations and reference concentrations according to the IUPAC denition: 31 where s y/x is the standard deviation of the residuals and is a measure of the average deviation of the prediction values from the regression line, N is the number of spectra in the dataset, c i indicate the reference concentrations andĉ i the predicted concentrations.
We performed the similar PLS analysis for the spiked sera samples mimicking the concentrations of a healthy individual and a patient with advanced metastatic breast cancer to examine the accuracy and precision of quantitative measurements. Relative error of prediction (REP) and relative standard deviation (RSD) were calculated, which correlate directly with the accuracy and precision of SERS assay respectively. REP is calculated by the following equation: 30 The RSD of predicted concentrations is given by: 30 where N conc is the number of distinct concentrations in the dataset, p is the number of spectra per concentration and s c k is the standard deviation obtained at concentration c k .