Comparative analysis of serum and saliva samples using Raman spectroscopy: a high-throughput investigation in patients with polycystic ovary syndrome and periodontitis

This case–control study was undertaken in line with the Strengthening the Reporting of Observational Studies in Epidemiology guidelines [32]. A total of 107 Chinese patients diagnosed with PCOS (aged 20–34 years) following the 2003 Rotterdam definition criteria [3] and 22 systemically healthy female volunteers were recruited from both Departments of Gynecology and Stomatology at the Shenzhen Maternity & Child Healthcare Hospital (SZMCH) in Guangdong, China from October 2021 to August 2022. The diagnosis of PCOS was made on the basis of any two of the following criteria: i) clinical and/or biochemical hyperandrogenemia, such as signs of hirsutism, seborrheic alopecia, acne, and/or serum total testosterone (T) levels exceeding 7.5 μg/L; ii) menstruation < 8 times yearly or menstrual cycle > 35 days; and iii) transabdominal or transvaginal ultrasound showing a unilateral ovarian volume of ≥ 10 ml and/or ≥ 12 follicles of 2–9 mm diameter on the same surface of one ovary. Meanwhile, other conditions accounting for abnormal ovulation or hyperandrogenemia (e.g., congenital adrenocortical hyperplasia, androgen-secreting tumours, Cushing's syndrome, and thyroid abnormalities) were excluded. Systemically healthy female volunteers (aged 20–34 years) had menstrual cycles of 26–30 days, with normal results of medical examinations performed within one year other than metabolic disease (including insulin resistance, hypertension and hyperlipoidemia). Additionally, all regular sex hormone indicators were within the reference range 3 months prior to the enrolment. The following subjects were excluded, including i) taking antibiotics within 3 months and/or receiving periodontal treatments within 12 months; ii) systemically unhealthy prior to the diagnosis of PCOS or having systemic comorbidities other than metabolic complications following the diagnosis of PCOS; iii) receiving assisted reproductive technology; iv) continuously taking immunosuppressive agents, bisphosphonate osteoporosis drugs, steroids and immune-related drugs; and v) unwilling to participate in this study. In the sampling procedure, the non-PCOS subjects were selected to match the characteristics of PCOS patients, including age, body mass index (BMI) and waist-to-hip ratio (WHR), and then three PCOS patients per non-PCOS subjects were enrolled. Oral and informed written consent was received from all subjects prior to the study. This research work was approved by the Ethics Committee of SZMCH (No. SFYLS [2021]050), and the study was performed following the 2013 Declaration of Helsinki. The personal information of all the subjects and the obtained datasets were only used for scientific research and not disclosed to the public.

At the time of enrolment, the following information and datasets were obtained from all participants through interviews and questionnaires, including demographic characteristics (age and education attainment), lifestyles (smoking, drinking and sleep problems), gestation records, history of PCOS (diagnosis, duration and treatments), psychological status (depression, anxiety and stress), oral hygiene practice (regular dental visit, symptom of gingival bleeding during brushing, awareness of periodontal diseases), other oral/periodontal conditions (periodontal abscess and endodontic periodontal lesions, mucogingival deformities, occlusal trauma and dental prosthesis) and general health status. Psychological health status was evaluated by the Depression Anxiety and Stress Scale-21 [33]. The modified Ferriman-Gallwey (mFGS) scoring system [34] was used to evaluate the severity of hirsutism, one of the hyperandrogenic symptoms in patients with PCOS. Physical examination was performed to record BMI, WHR as well as systolic and diastolic blood pressure (SBP and DBP). The results of various serum biochemical parameters, including triglycerides (TG), total cholesterol (TC), high-and low-density lipoprotein cholesterol (HDL and LDL), fasting plasma glucose (FPG) and levels of LH, FSH, T, prolactin (PRL), progesterone (P) and E2 were retrieved from the dataset of SZMCH.

All subjects received full-mouth periodontal examination at 6 sites of each tooth by a single calibrated examiner (HJL), using a UNC-15 probe (Hu Friedy, USA). All examinations were employed in line with the concurrent time points of physical and biochemical check-ups. The intra-examiner reliability was evaluated by repeated measurements at the site level in four subjects, with the intraclass correlation coefficient of 0.887 (95% CI: 0.860–0.909) for the full agreement on probing depth (PD) and 0.873 (95% CI: 0.829–0.905) on clinical attachment loss (CAL). All examinations were undertaken in line with the concurrent time points of medical check-ups. The documented periodontal parameters consisted of full-mouth plaque score (FMPS, %), bleeding on probing (BOP, %), PD, tooth mobility, number of remaining teeth and number of tooth losses due to periodontitis (excluding the third molars). CAL was calculated with reference to PD and gingival recession. On the same day, the subjects also underwent panoramic X-rays for assessing alveolar bone levels. To avoid the possible effects on embryo in case of pregnancy, all the subjects were suggested to use contraception for the menstrual cycle after radiographic examination. Subsequently, the subjects were generally identified as non-periodontitis (periodontal health and gingivitis) and periodontitis groups, following the current classification of periodontal diseases [35‐38].

All samples were collected in the morning following a 12-h fast within the early follicular phase (Days 2 to 5) of the spontaneous or progesterone-induced menstrual cycle (Fig. 1). Prior to oral examination, the participants were instructed not to drink (except water) or take chewing gum. Approximately 5 ml of unstimulated whole saliva samples were collected, and then centrifuged at 1,000 rpm for 10 min at 4 °C. The supernatant was transferred to a 500 μL storage tube and stored at -80 °C for further analysis. Venous blood (5 mL) was obtained from the antecubital vein following the standard venipuncture. The serum samples were obtained by centrifugation (3,500 rpm for 10 min at 4 °C), transferred to a 500 μL storage tube and stored at -80 °C until the biochemical analysis was performed.

Enzyme-linked immunosorbent assay (ELISA) kits (Finetest, Guangzhou Chenxue Biotech Co., Ltd., China) were used to quantitate the levels of IL-6, IL-17A and MMP-8 in both saliva and serum samples.

After thawing at room temperature and appropriately mixing, 0.8 μL of the saliva samples were taken and deposited as a drop on the calcium fluoride substrate at room temperature. In addition, 2 μL of serum samples were mixed with 14 μL of ultrapure water in an Eppendorf tube. Then, 0.8 μL of the mixture was taken and deposited as a droplet on a CaF2 substrate at room temperature. Subsequently, Raman spectra were acquired from the Clinical Antimicrobial Susceptibility Test Ramanometry system (CAST-R, Qingdao Single-cell Biotechnology, Qingdao, Shandong, China) equipped with a 532 nm Nd: YAG laser, a 100 × air objective (Olympus, Japan) with a numerical aperture of 0.9 nm and a spectral resolution of ∼2 cm⁻¹ for the observation of body fluids. In the pre-experiments, the Raman spectroscopy parameters were first provided based on the previous experience of the specialist technicians in the detection of serum and saliva samples, and then adjusted according to the signal-to-noise ratio and stability of the obtained Raman spectroscopy signals. The parameters of the equipment currently displayed for the detection of saliva and serum samples were finally determined. The laser power was 100 mW. The spectra ranged from 420 to 3050 cm⁻¹, and the accumulative times of a single point were one time using a 5/1 s acquisition time (saliva/serum samples). Ten scattered points of each sample from the ring region were selected to minimize the bias in testing based on the preliminary experimental results.

The flow chart is presented in Fig. 1 and Figure S1.

The sample size was estimated using software (G*Power 3.1.9.7). Based on the results of the study by Dursun et al. [11] on PCOS patients, a considerable effect value (d > 0.80) was expected. Therefore, the difference in PD values of 0.50 ± 0.3 mm between the PCOS and non-PCOS groups was used as a reference. Setting a significance level of 5% and detection power of 85%, the sample size was considered to be at least 62 (1:1). To account for the unequal distribution between the groups (1:3), a sample size of at least 82 participants was determined.

The non-Raman spectroscopy data mainly include demographic information, anthropometric parameters, hormonal and metabolic indicators, periodontal parameters and the levels of inflammatory mediators (IL-6, IL-17A and MMP-8). The results were presented appropriately (mean ± SD or median with IQR for continuous variables, and frequency for categorical variables). Intergroup differences in continuous variables were evaluated by t-tests or Mann–Whitney U tests, while categorical variables were examined with the χ² tests. Binary logistic regression was employed to analyze the odds ratio (OR) for assessing the possible influence of clinical indicators on periodontitis. Intergroup analysis of variability and logistic regression models was undertaken to examine the relationship among PCOS and periodontal conditions and hormonal/metabolic indicators. The statistical analyses were undertaken with a software tool (SPSS version 26, IBM Corporation, NY, USA). A two-sided P < 0.05 was deemed to be statistically significant. Considering the potential risks of false positive results, the statistical significances was adjusted as 0.05/6 = 0.0083 for IL-6, IL-17A and MMP8 as well as 0.05/7 = 0.0071 for T, P, E2, PRL, LH, LH/FSH and FSH, respectively.

The raw Raman spectrum was presented as a signal map with a high individual variability requiring certain processing prior to subsequent analysis. Labspec 5 software (HORIBA Jobin Yvon Ltd., U.K.) was then used to preprocess the Raman spectroscopy datasets following wavelet baseline correction and vector normalization steps. After importing the raw Raman data into the software, the ‘baseline correction’ and ‘normalization’ buttons were clicked on sequentially, and the preprocessed Raman spectra were obtained. Ten spectra were recorded, preprocessed and averaged for each sample to generate a representative, stable and reliable spectrum for further analysis. Average spectra were statistically assessed and graphed using Origin2021 (Version 9.80). After obtaining stable and reliable salivary Raman spectra, literature searching for characteristic peaks with high repetition rates in the same group of Raman spectra for attribution was helpful for analyzing the components of different groups. The differences among multiple groups for selected peaks of Raman spectra were appropriately assessed using analyses of variance (ANOVA) or the Kruskal‒Wallis test. Partial least squares discriminant analysis (PLS-DA) was conducted to evaluate the differences between the spectra of the same sample. The spectra were then classified and distinguished accordingly. PCA was used to analyze the differences in Raman spectra between samples and grouping factors. Permutational multivariate analysis of variance was performed to analyze the extent to which different grouping factors could account for the difference in samples, and statistics analysis was performed using permutation tests (two-sided P < 0.05 was considered statistically significant). The receiver operating characteristic (ROC) curve was employed to demonstrate the predictive ability of the model for different disease states. Spearman correlation analysis was used to explore the correlations among the clinical indicators, inflammatory mediators, sex hormones and metabolic indicators, periodontal parameters and Raman spectroscopic results. Internal scripts were used for further analysis and visualization under the R environment (Version 4.2.2). The ‘mixOmics’ package in R (Version 4.2.2) was employed for above analysis. The relevant script code is available at http://​mixomics.​org/​ (Table S1).

After assessing and matching the 129 initially recruited participants, a total of 66 PCOS patients (26.0 ± 3.3 years) and 22 healthy subjects (25.7 ± 3.5 years) were finally included in the data analysis (Fig. 2). Just over half (56.8%) of the subjects showed a monthly household income of over 9,000 CNY, 75.0% received college education or above, and 30.7% of them had regular dental visits (Table S2). It is worth noting that 31.8% exhibited mild to moderate periodontitis (15.9% at Stages I and II, respectively). The rest were non-periodontitis subjects including gingivitis (36.4%) and periodontal health (31.8%). More than four-fifths (80.7%) presented over 50% of sites with plaque, and 62.5% with 10–50% of BOP sites.

The datasets of demographic, anthropometric, periodontal and inflammatory indicators from the subjects with or without PCOS are presented in Table 1. Overall, there were similar demographic profiles and lifestyles in these two groups. As anticipated, notable inter-group difference existed in the levels of LH (P < 0.001), T (P < 0.001), E2 (P < 0.05) and LH/FSH (P < 0.001). While there were no significant differences in the levels of FSH, PRL and P between the two groups. Notably, mean PD was significantly higher in PCOS patients than that in non-PCOS participants (P < 0.05). Whereas, there was no notable difference in the presence of periodontitis between PCOS and non-PCOS subjects. Moreover, MMP-8 level in serum samples was significantly higher in PCOS patients than that in the controls (P < 0.01). In addition, as shown in Table S3, salivary MMP-8 level was significantly higher in periodontitis group than that in non-periodontitis group (P < 0.001).

Binary logistics regression models were used to analyze the potential links of sex hormones, metabolism, inflammatory indicators and periodontal parameters with PCOS conditions (Table 2). Notably, the levels of TG, PD and mFGS were positively correlated with PCOS (OR: 4.108, P < 0.05; 7.027, P < 0.05; and 1.806, P < 0.01; respectively).

Overall, the Raman spectra ranged from 423 cm⁻¹ to 3050 cm⁻¹, and 88 averaged spectra in each type of sample were obtained after preprocessing. Figure 3A shows the average Raman spectra of serum samples from subjects with non-periodontitis (periodontal health and gingivitis) and periodontitis. The major peaks in the serum samples were observed at approximately 996 cm⁻¹, 1146 cm⁻¹, 1255 cm⁻¹, 1324–1328 cm⁻¹, 1437 cm⁻¹, 1507 cm⁻¹, 1595 cm⁻¹, 1646 cm⁻¹ and 2916 cm⁻¹ Raman shifts in both statuses (Table S4). The peak at 996 cm⁻¹ was attributed to aromatic ring breathing, especially phenylalanine [39]. The peak at 1146 cm⁻¹ was mainly assigned to the carbon‒carbon bonding mode of lipids, while the peaks at 1255 cm⁻¹, 1324 cm⁻¹, 1328 cm⁻¹ and 1646 cm⁻¹ were attributed to protein and amino acid vibrations [40‐43]. Moreover, the peak at 1507 cm⁻¹ was related to carotenoids [44] with various clinical implications such as cancer treatment, cardiovascular disease prevention, cataract, antioxidant and anti-ageing effects [45]. The peak at 2916 cm⁻¹ had the highest frequency value and corresponded to lipid structures and methyl stretching vibrations [44].

The average Raman spectra of saliva samples from subjects with different periodontal statuses are presented in Fig. 3B. The major peaks in saliva samples were observed at approximately 996 cm⁻¹, 1118 cm⁻¹(lipid), 1439 cm⁻¹(collagen), 1597 cm⁻¹ and 1652 cm⁻¹ (protein and amide I band) Raman shifts in both statuses [40‐44]. There were some peaks in different statuses, such as 1300 cm⁻¹(fatty acids), 1573 cm⁻¹(protein) and 2920 cm⁻¹(protein), which were only present under the periodontitis condition.

To further investigate the effect of PCOS and periodontitis on the different Raman spectra, all the subjects were categorized as non-PCOS and non-periodontitis group (H–H, n = 17), PCOS and non-periodontitis group (PC-H, n = 43), non-PCOS and periodontitis group (H-Perio, n = 5), and PCOS and periodontitis group (PC-Perio, n = 23). Figure 4 reveals the peaks of serum and saliva samples presented in Fig. 3. The mean Raman intensities of the major peaks in the serum samples were not significantly different among the four groups. However, there were significant intergroup differences with salivary Raman peaks at 747 cm⁻¹, 1120 cm⁻¹, 1300 cm⁻¹, 1328 cm⁻¹ and 1574 cm⁻¹ (P < 0.05). The mean Raman intensities of the significant peaks at 747 cm⁻¹ (hemoglobin), 1120 cm⁻¹ (lipid), 1300 cm⁻¹ (fatty acids), 1328 cm⁻¹ (adenine) and 1574 cm⁻¹ (protein) were all higher in the PC-Perio group than those in the H-Perio group (P < 0.05). The mean Raman intensity at 1120 cm⁻¹ (lipid) was higher as well in the PC-Perio group than that in the H–H group (P < 0.05). The peaks at 1574 cm⁻¹ (protein) had greater mean Raman intensities in the PC-H group than that in the H-Perio group (P < 0.05).

To verify whether the differences detected may contribute to the establishment of a classification model for discriminating the signals collected from serum and saliva samples, PCA and PLS-DA were then performed on the collected spectra. After PCA analysis, the four groups were clustered to a certain extent, and each sample was distributed fairly evenly within its own 95%CI range, with no obvious discrete points (Fig. 5A and B). A PLS-DA classification model analysis of the Raman spectra of the serum samples revealed no aggregation among the four groups and a more concentrated distribution of samples within each group (Fig. 5C), similar to the Raman spectra of the saliva samples (Fig. 5D). There was no marked inter-group difference in the Raman spectra of the serum samples (R² = 0.03, P = 0.453) (Fig. 5E). However, there was a significant difference among the four groups in the saliva samples (R² = 0.08, P < 0.01) (Fig. 5F). Figure 5G and H showed that there could also have influences on the components between samples of the same type. However, the same groups of Raman spectra from the same sample type tended to cluster into one group and showed more similarity. The aggregation effect of grouping was more evident for saliva samples than that for serum samples. ROC analysis further indicated that the area under the curve (AUC) of saliva samples was more than 0.069 (P < 0.001).

Most of the major peaks related to periodontal status and PCOS belonged to lipids and proteins (Table S4). The differences in periodontal parameters, sex hormone levels and cytokine levels between the PCOS and periodontitis groups are presented in Table 1 and Table S3. It was concerned that these two diseases could be interlinked with changes in certain chemical compounds. Therefore, Spearman correlation tests were performed for related hormones and cytokines as well as the average Raman intensities corresponding to proteins and lipids (Tables S5 and S6).

The most significantly correlated components existed in the H-Perio group for both serum and saliva samples, followed with the PC-Perio group and the least in the H–H group. These associations, most positively, occurred mainly between hormones and cytokines. In the PC-H group, the level of IL-6 in serum was positively correlated with BOP, PD and Raman average spectral intensity of proteins (value = 0.9, P < 0.05). The levels of T in PC-H group were positively correlated with the levels of MMP-8 (value = 0.9, P < 0.05). The levels of MMP-8 in serum were also positively correlated with BOP (value = 0.462, P < 0.05) and PD (value = 0.538, P < 0.01) in PC-Perio group. The mean salivary Raman spectral intensity attributed to proteins in the H-Perio group was positively correlated with LH/FSH (value = 0.314, P < 0.05) and negatively correlated with the level of IL-17A in saliva (value = -0.373, P < 0.05). The mean salivary Raman spectral intensities of the attributed lipids were positively correlated with LH/FSH (value = 0.357, P < 0.05) in H-Perio group, and the mean salivary Raman intensity of the attributed proteins (value = 0.450, P < 0.01) in PC-Perio groups, and negatively correlated with IL-17A (value = -0.374, P < 0.05) and BOP (value = -0.363, P < 0.05) in H-Perio group. However, in the PC-H group, the mean salivary Raman spectral intensities of the attributed lipids were negatively correlated with the levels of MMP-8 (value = -0.900, P < 0.05).