Discrimination of molecular thin films by surface-sensitive time-resolved optical spectroscopy

An optical discrimination technique, tailored to nanometric-sized, low optical absorbance molecular complexes adhering to thin metal ﬁlms, is proposed and demonstrated. It is based on a time-resolved evanescent-wave detection scheme in conjunction with hierarchical cluster analysis and principal value decomposition. The present approach aims to differentiate among molecular ﬁlms based on statistical methods, without using previous detailed knowledge of the physical mechanisms responsible for the detected signal. The technique is open to integration in lab-on-a-chip architectures and nanoscopy platforms for applications ranging from medical screening to material diagnostics. V C 2015 Author(s). All article content, except where otherwise noted, is licensed Creative Commons

Discrimination of molecular thin films by surface-sensitive time-resolved optical spectroscopy Simone Peli, 1,2 Nicola Nembrini, 1,2,3 Francesco Damin, 4 Marcella Chiari, 4 Claudio Giannetti, 1,2 Francesco Banfi, 1 An optical discrimination technique, tailored to nanometric-sized, low optical absorbance molecular complexes adhering to thin metal films, is proposed and demonstrated. It is based on a timeresolved evanescent-wave detection scheme in conjunction with hierarchical cluster analysis and principal value decomposition. The present approach aims to differentiate among molecular films based on statistical methods, without using previous detailed knowledge of the physical mechanisms responsible for the detected signal. The technique is open to integration in lab-on-a-chip architectures and nanoscopy platforms for applications ranging from medical screening to material diagnostics. Optical detection of biomolecules plays a pivotal role in modern biophysics 1,2 and biophotonics, 3-5 especially for molecular complexes immobilized on surfaces. 6 Optical spectroscopy and fluorescence emission are routinely used to investigate structural conformation changes of molecular beacons on decorated surfaces 7,8 and detect surface-bound engineered polypeptides. 9 Since the molecular fingerprint is encoded in either the spectral reflectance/transmittance or the spectral fluorescence emission, optical spectroscopies based on multi-wavelength detection have been widely employed. Despite their success, conventional optical spectroscopies suffer some major drawbacks, such as the lack of specific surface sensitivity and time-resolved capabilities, i.e., the potential to follow transient molecular excitations initiated by an external trigger. In order to overcome some of the above-mentioned shortcomings, we developed a molecular discrimination technique, based on time-resolved evanescent wave 10 optical spectroscopy and data mining techniques.
The experiment is based on a pump and probe optical scheme to investigate biomolecular layers with typical vertical dimensions of the order of less than 10 nm, 2 orders of magnitude smaller than the optical wavelength. The molecular layer is bound to an Au thin film deposited on the hypotenuse of a glass prism. A pump laser pulse excites the Au film. A time-delayed evanescent wave probe pulse, created at the glass-air interface of a prism in the Kretschmann configuration, 11 probes a volume comprising the Au film and the molecular layers (see the inset of Fig. 1). The relative reflectivity variation of the probe pulse is recorded as a function of the time delay from the pump pulse arrival.
While time-resolved spectroscopy techniques are not new to biochemistry, 12 they are usually employed to study the relaxation behavior of known molecular species. Here, we use the information contained in time-resolved reflectivity variation to sort out the differences among different molecular aggregates on thin metallic films. To highlight the specificity of the present approach, it is appropriate to make comparisons with already developed optical surface sensitive techniques: second harmonic generation spectroscopy (SHGS), 13 sum-frequency generation vibrational spectroscopy (SFG-VS), 14 surface cavity ring-down spectroscopy (s-CRDS), 15,16 and attenuated total reflection (ATR) infrared spectroscopy. 17  Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 159.149.197.139 On: Wed, 13 Apr these techniques allows one to investigate the orientation of surface molecules and gather important insights into their chemistry.
SHGS and SFG-VS are nonlinear techniques requiring relatively high intensity laser pulses. The signal may originate from bulk multipolar contributions near the surface of a centrosymmetric or isotropic medium. In addition, contributions may arise from thin metal films supporting the investigated molecular layers. For these reasons, caution must be taken, and the use of theoretical models is mandatory in interpreting data. 14 ATR and s-CRDS are based on evanescent waves and linear absorption. The latest refinements in CRDS with stabilized frequency combs reached exquisite sensitivity, 16 but both are spectroscopic techniques that do not encompass time resolved capabilities. The approach proposed in this work combines the time resolved capability and surface sensitivity of non-linear techniques with linear probe absorption as in s-CRDS or ATR at a single wavelength. Instead of relying on spectral absorption features, we use the complex interplay of physical mechanisms that lead to the variation of absorption after pump excitation as a distinctive feature of the molecular layers under investigation. 18 Moreover, as in s-CRDS, the polarization information could be used to determine the molecular orientation. 15 The time-resolved spectra are analyzed using two approaches: (1) a hierarchical binary cluster tree dendrogram built using a Euclidean distance (Ward method) 19 and (2) a singular value decomposition (SVD) analysis. 20 These approaches are compared against the standard analysis of the time-dynamics of molecular complexes based on the sum of exponentially decaying functions and are shown to provide the best performance.
The pump and probe laser pulses are generated by two synchronized femtosecond fiber lasers with a repetition rate of 100 MHz and a pulse time width of 120 fs. The fundamental wavelength of 1560 nm of the first laser is used as a pump, while the second harmonic at 780 nm of the second laser is used as a probe. The delay between the pump and probe pulses is obtained by the Asynchronous Optical Sampling (ASOPS) technique. 21 This technique allows one to measure relative reflectivity variation as small as 10 À6 over a delay window extending from 100 fs to 10 ns.
The experiments are performed on a BK7 prism hypotenuse, where a nominal thickness of 4 nm of gold is evaporated. The gold-coated prism is functionalized with a multifunctional copolymer of dimethylacrylamide (DMA), N-acryloyloxysuccinimide (NAS), and 3-(trimethoxysilyl) propyl methacrylate (MAPS)-copoly(DMA-NAS-MAPS)providing reactive groups suitable for immobilizing molecular structures. 9 The gold coated prism constitutes the reference system. The gold film coated with the copoly (DMA-NAS-MAPS) constitutes our first model. A biotinylated IgG antibody covalently immobilized by the copolymer on the surface of the prism constitutes the second model of our study. 18 In Fig. 1, we show the signal measured on the reference sample in the evanescent probe configuration, which is 20 times bigger than the signal obtained in the conventional reflection geometry. This result demonstrates the surfacesensitivity of the ASOPS-based time-resolved technique in the evanescent-wave configuration. We point out that both the pump (k pump ¼ 1560 nm) and probe (k probe ¼ 780 nm) wavelengths are far from the Au surface plasmon resonances, which play no role in the present experiment. This enlarges the range of exploitable coating materials and probe wavelengths, a fact of relevance in view of applications where disparate bio-systems may call for a specific coating material due to technical or cost-related issues. It is important to note that the estimated temperature increase due to pump absorption in our experimental conditions is less than 10 K, and no damage to the biomolecular film is expected. 18 The variation of the evanescent probe reflection induced by the optical pump has the ubiquitous decaying exponential shape (Fig. 2). The dynamics measured on the different samples looks very similar, the signature of the specific timeresponse of individual molecular aggregates residing in tiny differences. We use signal analysis methods to extract these differences and exploit them to discriminate among different molecular aggregates.
To illustrate the method, we consider the time resolved measurements relative to three samples: the bare gold film (black lines, from 1 to 5 in Fig. 2); the gold film coated with the copoly(DMA-NAS-MAPS) (blue lines, from 6 to 9 in Fig. 2); the gold film coated with the copoly(DMA-NAS-MAPS) that immobilize the IgG antibody (red lines, from 10 to 13 in Fig. 2).
Since the measurements on the same sample may vary when repeated after some days due to different ambient conditions (temperature and humidity), or because the experiment is carried out at a different spot on the sample, we made multiple sets of measurements for each of the three samples. The total 13 measurements have been organized in a 2000 Â 13 data matrix X. Each column represents a complete experimental trace. Rows map time delays. Note that the traces are affected by evident variations of the noise superposed on the exponential decay dynamic. We included in the analysis even traces that are visually not appealing, such as trace 6 in Fig. 2, and traces where noise is visually greater than optimum as in traces 1, 8, 9, and 13. The aim is to demonstrate the robustness of the analysis against the presence of external perturbations.
To measure the similarity among spectra, we first adopt cluster analysis to perform an exploratory inspection. Specifically, we aggregate the data using a hierarchical cluster analysis known as Ward's minimum variance method. 19 This is an agglomerative method that clusters data with the criterion to minimize the total within-cluster variance once an appropriate metric has been defined to "measure" the distance among experimental traces. The aim is to reveal the internal similarities among traces and to structure these similarities hierarchically. To calculate the variance, a squared Euclidean distance between traces is used. The results of hierarchical clustering are presented in the dendrogram shown in Fig. 3(a). The data group in three distinct clusters, each associated with one of the three investigated samples, according to the principle that similar samples cluster together because they have similar traces. 18 SVD 20 is a complementary data analysis approach. SVD is used here to reduce the dimensionality of the data, projecting the data on a reduced eigenvector basis of particular relevance. In this case, clustering emerges when different objects have similar projections.
Entering in more details, the data matrix is expanded in the form X ¼ USV T , where U is a matrix with the same dimensions as X whose columns are the orthonormal basis vectors for the time-resolved measurements and the rows of V contain the coordinates of each measurement projected on the eigenvectors space. S is a diagonal matrix that gives the eigenvalues spectrum, scaling the basis vectors according to their importance in reconstructing the data. The matrix M ¼ SV T gives the coordinates of the data, so that X ¼ UM.
As usual in this context, we reduce the data dimensions by projecting the original data into the subspace spanned by the two eigenvectors with the highest eigenvalues. Visualizing the coordinates as points in a scatter plot (Fig.  3(b)), three clusters are evidenced, each composed of points corresponding to the same sample. 18 In summary, cluster analysis evidences similarity derived from minimizing distances whereas SVD evidences similarity upon inspection of corresponding projections.
These findings demonstrate that time-resolved measurements alone contain sufficient information to highlight the differences between surfaces covered with different molecular aggregates.
We now compare the previous results to the standard multi-exponential decay analysis. When the pump pulse excites the system, the relaxation mechanisms manifest themselves invariably in a reflectivity decay that is well mimicked by a sum of exponentially decaying functions. The amplitude and decay time of each function are the fitting parameters. By analyzing the level of residues and the scaling of fitting parameters, we qualitatively find out that four exponentials are the optimal number to describe the data. Thus, eight parameters, four amplitudes (a i ; i ¼ 1…4) and four time decay constants (s i ; i ¼ 1…4), are needed to fit each experimental trace as f ðtÞ ¼ a 1 expðÀt=s 1 Þ þ a 2 exp ðÀt=s 2 Þ þ a 3 expðÀt=s 3 Þ þ a 4 expðÀt=s 4 Þ. In Fig. 4, the coordinates of each point represent the time decay constants or the amplitudes retrieved from the fitting procedure of a single trace. The color code is the same as the one adopted in Fig. 2.
Regardless of the specific plot one considers in Fig. 4, the formation of clusters identifying different samples is incomplete. Blue triangles (Copoly) never cluster in a visually distinct group, while black circles (Au) and red squares (IgG) cluster to a certain extent in Figs. 4(c) and 4(a) only. Similar considerations can be made for the patterns obtained visualizing data using different formats. The fitting parameters alone are hence not sufficient to group the samples in distinct sets. This points to the power of multidimensional analysis, like the Ward method and the SVD analysis, to capture more similarities than those uncovered by simply comparing fitting parameters.
Our strategy constitutes a change of paradigm with respect to traditional molecular screening methods. We exploit the information encoded in time domain, which are disclosed by multi-dimensional analysis, as opposed to traditional approaches relying on information retrieved from frequency domain techniques. Our results suggest a possible taxonomy of time-resolved spectra aimed at the discrimination of thin molecular layers based on data analysis methods. Our approach entails several advantages. It provides enhanced surface sensitivity and is label-free. The molecular recognition does not rely on the detailed modeling of the mechanism underlying the optical response. The technique has a high data throughput and can be integrated with existing microscopy platforms and lab-on-a chip architectures.
In perspective, a microfluidic lay-out, carved in a polydimethylsiloxane (PDMS) matrix, could be conformally bonded on the prism's Au-coated surface, the latter being functionalized with specific antibodies. The flow of antigens to be sensed will be driven across the chip to several sensing areas-one patch for each antibody-where the pump and probe measurement will be performed. Comparison-via hierarchical cluster analysis and SVD-of the acquired spectra against benchmark time-resolved traces for antigenbound antibodies will allow identifying the presence of specific antigens and, possibly, their concentration. Within this scheme, the present technique could serve as a high throughput screening method for some pathogens.
Combined with scanning microscopy, 22 the approach outlined in this work would serve as a platform to detect spatial variations in surface chemical properties arising from compositional inhomogeneity of molecular films. Acquisition of a pump and probe trace at each sample's coordinates, followed by association of a false color to each pixel on the basis of clustering analysis, would render a spatial map of the compositional contrast. The typical ASOPS acquisition time per pixel is of the order of few seconds, leading to the acquisition of 32 Â 32 pixel images in less than 1 h. 23