Resolving Water, Proteins, and Lipids from In Vivo Confocal Raman Spectra of Stratum Corneum through a Chemometric Approach

Summary

Here, we present a protocol for collection of confocal Raman spectra from human subjects in clinical studies combined with chemometric approaches for spectral outlier removal and the subsequent extraction of key features.

Abstract

Development of this in vivo confocal Raman spectroscopic method enables the direct measurement of water, proteins, and lipids with depth resolution in human subjects. This information is very important for skin-related diseases and characterizing skin care product performance. This protocol illustrates a method for confocal Raman spectra collection and the subsequent analysis of the spectral dataset leveraging chemometrics. The goal of this method is to establish a standard protocol for data collection and provide general guidance for data analysis. Preprocessing (e.g., removal of outlier spectra) is a critical step when processing large datasets from clinical studies. As an example, we provide guidance based on prior knowledge of a dataset to identify the types of outliers and develop specific strategies to remove them. A principal component analysis is performed, and the loading spectra are compared with spectra from reference materials to select the number of components used in the final multivariate curve resolution (MCR) analysis. This approach is successful for extracting meaningful information from a large spectral dataset.

Introduction

In clinical studies, in vivo confocal Raman spectroscopy has shown its unique ability for determining stratum corneum thickness and water content^1,2,3,4, and tracking the penetration of active materials topically applied to the skin^5,6. As a noninvasive approach, confocal Raman spectroscopy detects molecular signals based on vibrational modes. Thus, labeling is not needed⁷. In vivo confocal Raman spectroscopy provides chemical information with depth resolution based on the confocal nature of the technique. This depth-dependent information is very useful in studying the effects of skin care products^4,8, aging^9,10, seasonal changes³, as well as skin barrier function diseases, such as atopic dermatitis^11,12. There is a lot of information in the high frequency region of confocal Raman spectroscopy (2,500–4,000 cm^-1), where water produces distinct peaks in the region between 3,250–3,550 cm^-1. However, the Raman peaks of proteins and lipids, which are centered between approximately 2,800–3,000 cm^-1, overlap each other because the signals are mainly produced from methylene (-CH₂-) and methyl (-CH₃) groups¹³. This overlapped information presents a technical challenge when obtaining relative amounts of individual molecular species. Peak fitting^14,15 and selective peak position^12,16 approaches have been used to resolve this challenge. However, it is difficult for these single peak-based methods to extract pure component information because multiple Raman peaks from the same component change simultaneously¹⁷. In our recent publication¹⁸, an MCR approach was proposed to elucidate the pure component information. Using this approach, three components (water, proteins, and lipids) were extracted from a large in vivo confocal Raman spectroscopic dataset.

The execution of large clinical studies can be demanding on individuals collecting in vivo spectroscopic data. In some cases, spectral acquisition can require operating equipment for many hours in a day and the study can extend up to weeks or months. Under these conditions, spectroscopic data may be generated by equipment operators that lack the technical expertise to identify, exclude, and correct for all sources of spectroscopic artifacts. The resulting data set may contain a small fraction of spectroscopic outliers that need to be identified and excluded from the data prior to analysis. This paper illustrates in detail a chemometric analysis process to "clean up" a clinical Raman dataset before analyzing the data with MCR. To successfully remove the outliers, the types of outliers and the potential cause for the generation of the outlier spectra need to be identified. Then, a specific approach can be developed to remove the targeted outliers. This requires prior knowledge of the dataset, including a detailed understanding about the data generation process and the study design. In this dataset, the majority of outliers are low signal-to-noise spectra and originate primarily from 1) spectra collected above the skin surface (6,208 out of 30,862), and 2) strong contribution to the spectrum from fluorescent room light (67 out of 30,862). Spectra collected above the skin surface produce a weak Raman response, as the laser focal point approaches the skin surface and is mostly in the instrument window below the skin. Spectra with a strong contribution from fluorescent room light are generated due to either instrument operator error or subject movement, which produces a condition where the confocal Raman collection window is not fully covered by the subject’s body site. Although these types of spectral artifacts could be identified and remediated during spectral acquisition by a spectroscopic expert at the time of data acquisition, the trained instrument operators used in this study were instructed to collect all data unless a catastrophic failure was observed. The task of identifying and excluding outliers is incorporated into the data analysis protocol. The protocol presented is developed to resolve this challenge. To address the low signal-to-noise spectra above the skin surface, the location of the skin surface needs to be determined first to allow removal of spectra collected above the skin surface. The location of the skin surface is defined as the depth where the Raman laser focal point is half in the skin and half out of the skin as illustrated in Supplemental Figure 1. After removing low signal-to-noise spectra, a principal component analysis (PCA) is implemented to extract the factor dominated by fluorescent room light peaks. These outliers are removed based on the score value of the corresponding factor.

This protocol provides detailed information for how six principal components are determined in the MCR process. This is done through a PCA analysis followed by spectral shape comparison between the loadings for models generated with a different number of principal components. The experimental process for data collection of reference materials as well as the human subjects is also explained in detail. 

Protocol

This study was approved by the institutional review committee of Beijing Children’s Hospital in compliance with the ethical guidelines of the 1975 Declaration of Helsinki. It was conducted according to ICH guidelines for Good Clinical Practice. The study took place from May to July 2015.

1. Collection of in vivo confocal Raman spectra from human subjects with atopic dermatitis

1. Include subjects in compliance with the following criteria.
   1. Include subjects between the ages of 4–18.
   2. Include subjects with mild to moderate atopic dermatitis (score of 2 or 3 according to the Physician’s global assessment) with active disease symptoms on 5%–30% of the body surface, with at least two lesions on the arms.
   3. Include subjects that are in good health, excluding symptoms directly related to the AD disease.
   4. Include subjects that provide written informed consent.
   5. Include subjects that have an individual topology angle (ITA) value higher than 40 at the testing location.
2. Exclude subjects that meet any of the following criteria.
   1. Exclude subjects that are either currently participating or have previously participated in a clinical study at any test facility within the previous 4 weeks.
   2. Exclude subjects with cancer or that have been diagnosed or treated for cancer within 5 years prior to the study.
   3. Exclude diabetic subjects.
   4. Exclude subjects who have an immunologic or infectious disease that could place the subject at risk or interfere with the accuracy of the study results (i.e., hepatitis, tuberculosis, HIV, AIDS, lupus, or rheumatoid arthritis).
   5. Exclude subjects that have skin conditions that might interfere with instrumental measurements or will prevent the clear assessment of the skin only to atopic dermatitis. Examples include extremely dry skin, damaged skin, cuts, scratches, sunburn, birthmarks, tattoos, extensive scarring, rashes, excessive hair growth, or acne.
   6. Exclude subjects that use oral immunosuppressive drugs, antibiotics, or other systemic therapies within the past month, except for minor tranquilizers.
   7. Exclude subjects with any other medical conditions that, in the opinion of the investigator, precludes them from study participation.
   8. Exclude subjects with higher pigmentation in the testing area.
3. Label the lesion area of the atopic dermatitis study participant and mark with a 3 cm x 4 cm area on or near the lesion site as shown in Figure 1A.
4. Label the non-lesion area with the same marker in the counterpart body site (e.g., left forearm vs. right forearm) as shown in Figure 1B.
5. Place the marked body site in close contact with the window of the in vivo confocal Raman instrument as shown in Figure 2A and Figure 2B. Cover the whole window to avoid the impact of room light on the body site.
6. Perform the Raman data collection.
   NOTE: The instrument has a spectral resolution of 2 cm^-1 and 50x microscopy objective (NA = 0.9 oil immersion), using a 671 nm laser with a power of 17 mW. Wavelength is calibrated using the spectrum of a built-in neon-argon lamp. The intensity calibration is done by measuring the spectrum of a NIST (National Institute of Standards) glass calibration standard.
   1. Move the focus until a spectrum as illustrated in Figure 2C is observed, then move the focus away from the skin surface 10 μm.
   2. Start the data collection for 26 steps with a 2 μm step size in the 2,510 cm^-1–4,000 cm^-1 frequency region. Use an exposure time of 1 s and measure eight replicates for each area lasting ~10–15 min total.

2. Collection of confocal Raman spectra from reference materials

1. Place the reference materials, the major components in human skin stratum corneum¹⁹, on the window of the confocal Raman instrument (see Table of Materials: Bovine Serum Albumin (BSA), deionized water (DI water), ceramide, cholesterol, free fatty acid, and squalene).
2. Collect the reference materials' Raman spectra consecutively from the outside of the material to the material center using the same collection parameters as described above.
3. Integrate the area under each Raman spectra between the range of 2,510–4,000 cm^-1 and identify the top three maximum value points in these 26 measurements. Average the Raman spectra from those three points to obtain the final reference material spectra.

3. Removal of the outlier spectra through chemometrics analysis

1. Determine the skin surface and remove the out-of-skin spectra.
   1. Change the file extension from '.ric' to '.mat' and load the .mat file to the MATLAB software platform.
   2. Correct the baseline using Baseline(Automatic Weighted Least Squares) by right-clicking the imported dataset under Analyze | Other Tools | Preprocessing in the PLS_Toolbox software with default setting.
   3. Sum up the values between 2,910–2,965 cm^-1 to obtain the intensity values under each Raman spectrum from the 26 consecutive steps measurement as shown in Figure 3A via the sum function in MATLAB.
   4. Interpolate the instrument offset value (value in the X-axis in Figure 3B) from 26–260 using the linspace function in MATLAB.
   5. Interpolate the intensity value from 26 to 260 using the spline method in MATLAB, leveraging the newly-generated 260 position values.
   6. Use MATLAB's polyfit and polyval functions to obtain the new set of intensity values with 260 points. First, use the 260 position and intensity values as X and Y inputs for the polyfit function, respectively. Set the degree value to 20. Then, use the output coefficients and the 260 extended position values as the input for polyval to obtain the final 260 intensity values.
   7. Use MATLAB's max and min functions to identify the maximum and minimum points from the newly interpolated 260 intensity values.
   8. Calculate the mean intensity value by dividing the sum of the maximum and minimum intensity value by two.
   9. Identify the intensity value from the 260 intensity values (calculated from step 3.1.6) that is closest to the mean intensity value and set its corresponding position value as skin surface. Set this position value as the zero point in the X-axis as illustrated in Figure 3B.
   10. Change all the other position values according to the zero point and the known 2 μm step size.
   11. Remove all the spectra collected above the skin surface according to their position value.
   12. Import the rest of the data to PLS_Toolbox to create a dataset and rename it "RamanData.mat".
2. Remove the outlier spectra with the room light effect.
   1. Load the Raman spectra dataset (RamanData.mat) after removal of the out-of-skin spectra and implement the PCA analysis.
   2. Load the dataset into the PLS_Toolbox software under the MATLAB platform and right-click the dataset to choose Analyze | PCA.
   3. Select Normalize as the preprocessing approach and choose None for the cross validation.
   4. Use the three components for the PCA decomposition analysis as shown in Supplementary Figure 2.
   5. Remove the cover on the in vivo Raman instrument's collection window and collect the room light spectra in the high frequency region using the same parameters used for the reference materials data collection.
   6. Identify the room light effect factor through comparison with the room light background spectra as shown in Supplementary Figure 3.
   7. Remove the spectra with a significantly higher corresponding score value than normal (More than 99.8% of the score values of the whole dataset, which is 0.16 in this study).

4. Selection of the number of the components in MCR decomposition analysis

1. Correct the Raman spectra baseline using the same approach described above (section 3.1.2).
2. Perform the PCA analysis on the preprocessed dataset as described above (section 3.2) except for selecting “PQN” – “Mean Center” rather than “Normalize”, and plot the eigenvalues in logarithmic scale along with the number of components (20) as the default number in the decomposition analysis by clicking the Choose Components button and select log(eigenvalues) as the Y value. Choose three to eight as the number of the components used for the MCR analysis.
3. Perform MCR analysis.
   1. Load the dataset into the MCR_main software²⁰ through the Data Selection button.
   2. Choose the number of components (three to eight) by clicking the Determination of the number of components button.
   3. Click the Pure button under the Initial Estimation tab, select Concentration under the Direction of the variable selection tab, and click the Do button.
   4. Click the OK button and then the Continue button to the next page.
   5. Click Continue in the next page and then select fnnls and 6 under the Implementation and Nr. of species with non-negativity profiles tabs, respectively. Click the Continue button.
   6. Choose the same parameters as section 4.3.5 for this page and click Continue.

Results

In this clinical study, in vivo confocal Raman spectra were collected from 28 subjects from 4–18 years old. A total of 30,862 Raman spectra were collected with the data collection protocol mentioned above. This large spectral dataset contains 20% spectral outliers as shown in Figure 4A. The low signal-to-noise outlier spectra were removed after determining the skin surface, followed by the PCA to identify the spectra with room light features. The third factor in this PCA model is identif...

Discussion

During the data collection, as described in section 2 and 3 of the protocol, each depth profile was collected in an area with contact between the instrument window and the skin by finding the darker areas from the microscopic images highlighted in the red circles in Figure 2C. Once these areas were located, it was critical to start the depth profile above the skin surface to accurately determine the location of the skin surface for the data analysis procedure. The location o...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Bovine Serum Albumin	Sigma-Aldrich
Cholesterol	Sigma-Aldrich
Cholesterol 3-sulfate sodium	Sigma-Aldrich
D-Erythro-Dihydrosphingosine	Sigma-Aldrich
DI water			Purified with Milipore(18.2MΩ)
Gen2-SCA skin analyzer	River Diagnostics, Rotterdam, The Netherlands	Gen2
Matlab 2018b	Mathwork	2018b
N-behenoyl-D-erythro-sphingosine	Avanti Polar Lipids, Inc.
N-Lignoceroyl-D-erythro-sphinganine(ceramide)	Avanti Polar Lipids, Inc.
Oleic Acid	Sigma-Aldrich
Palmitic Acid	Sigma-Aldrich
Palmitoleic Acid	Sigma-Aldrich
PLS_Toolbox version 8.2	Eigenvector Research Inc.	8.2
RiverICon	River Diagnostics, Rotterdam, The Netherlands	version 3.2
Squalene	Sigma-Aldrich
Stearic Acid	Sigma-Aldrich