Exploring the Association between Pro-Inflammation and the Early Diagnosis of Alzheimer’s Disease in Buccal Cells Using Immunocytochemistry and Machine Learning Techniques

Abstract

The progressive aging of the global population and the high impact of neurodegenerative diseases, such as Alzheimer’s disease (AD), underscore the urgent need for innovative diagnostic and therapeutic strategies. AD, the most prevalent neurodegenerative disorder among the elderly, is expected to affect 75 million people in developing countries by 2030. Despite extensive research, the precise etiology of AD remains elusive due to its heterogeneity and complexity. The key pathological features of AD, including amyloid-beta plaques and hyperphosphorylated tau protein, are established years before clinical symptoms appear. Recent studies highlight the pivotal role of neuroinflammation in AD pathogenesis, with the chronic activation of the brain’s immune system contributing to the disease’s progression. Pro-inflammatory cytokines, such as TNF-$\alpha$, IL-1$\beta$, and IL-6, are elevated in AD and mild cognitive impairment (MCI) patients, suggesting a strong link between peripheral inflammation and CNS degeneration. There is a pressing need for minimally invasive, cost-effective diagnostic methods. Buccal mucosa cells and saliva, which share an embryological origin with the CNS, show promise for AD diagnosis and prognosis. This study integrates cellular observations with advanced data processing and machine learning to identify significant biomarkers and patterns, aiming to enhance the early diagnosis and prevention strategies for AD.

1. Introduction

The progressive aging of the world’s population and the high impact of major human pathologies, including neurodegenerative diseases, have catalyzed the emergence of new tools that could lead to preventive, early diagnosis, or therapeutic strategies. Alzheimer’s disease (AD), the most common neurodegenerative disorder among the elderly, accounts for 60–70% of all dementia cases and carries a high morbidity rate, with serious social and economic consequences [1]. By 2030, it is estimated that approximately 75 million people in developing countries will be affected by AD. While each person experiences biological aging differently, aging is widely recognized as a gradual decline in molecular, cellular, tissue, and, ultimately, organ function, leading to a loss of physiological integrity, which includes the underlying pathology of AD [2]. Despite extensive research on the pathological and clinical aspects of AD, its precise etiology remains unknown due to the disease’s heterogeneity and complexity.

The hallmarks of AD pathology, namely amyloid-beta (A$\beta$) formation and extracellular deposits of plaques and hyperphosphorylated tau protein correlated with neurofibrillary tangles (NFTs), which are linked to synaptic and neuronal loss [3,4], are now recognized to develop several years before any clinical symptoms become apparent in patients. Despite extensive investigation into the mechanisms behind the pathology of amyloid-beta (A$\beta$) plaques, hyperphosphorylated tau aggregates, and neurofibrillary tangles (NFT), a clear understanding of AD’s pathogenesis remains elusive, hindering efforts to prevent the disease.

In recent years, numerous clinical and research studies have demonstrated the strong correlation between neuroinflammation and neurodegeneration, with a well-balanced innate and adaptive immune system participating in the inflammatory pathways of the central nervous system (CNS) [5,6]. Neuroinflammation, unlike other risk factors or genetic causes of AD, appears to have a pivotal role as a result rather than as a cause of the AD pathology background. Chronic and persistent activation of the innate immune system of the brain triggers the release and inflammatory cascade of pro-inflammatory and toxic products, such as cytokines and reactive oxygen species, which facilitates the A$\beta$ and NFT pathologies [7,8,9,10]. The increased presence of pro-inflammatory cytokines, TNF-α, IL-2, IL-1$\beta$, and IL-6 in the CNS and blood of AD and MCI patients [11] suggests a strong correlation between Alzheimer’s disease and MCI patients in a peripheral inflammation state, and its crucial role in the mechanisms associated with aging and neuroinflammation disorders [12]. It is well known that the upregulation of pro-inflammatory cytokines that can affect brain function is a consequence of a chronic low-grade state of inflammation [13,14,15]. Pro-inflammatory cytokines TNFa, IL-1$\beta$, and IL-6 have been described as having a fundamental role in the integrity of the blood–brain barrier (BBB) and its specific defensive functions. During AD progression, the upregulation of TNF-α and IL-6 is linked to the alteration of BBB function, facilitating the crossover of lymphocytes and macrophages from the periphery into the brain, supporting the strong relationship between the immune system and central nervous system (CNS) [16,17,18].

Aging or peripheral inflammatory conditions can also lead to alterations in the blood–brain barrier, allowing immune cells to enter the brain tissue and potentially supporting neuropathological processes [19,20] depending on the duration and intensity of stimulation. The in-depth interpretation and translation of these pathological pathways that could form a potential connection between normal aging and risk factors that lead to neuroinflammation and AD is of great importance for prevention and potential lifestyle interventions [21,22]. Since prevention is crucial in managing AD, the development of new trial designs, diagnostic approaches, and technological solutions using data mining techniques are anticipated to enhance its effectiveness and have a greater impact in the field of prevention.

To date, there has been no minimally invasive, straightforward, and cost-effective procedure available for the early detection or prevention of AD. Apart from the cerebrospinal fluid (CSF) approach, which is unpleasant and unsuitable for routine use, a range of candidate biomarkers from blood samples have been proposed. Although these biomarkers have not yet been validated, they represent a potentially more cost-effective and relatively less invasive approach that could offer valuable information for the early diagnosis or prognosis of the disease or serve as a screening tool [5,6,23,24,25,26]. In recent years, buccal mucosa cells and saliva have emerged as promising tool for the potential diagnosis, prognosis, or screening of AD. This interest stems from their shared embryological origin with the central nervous system (CNS) from differentiated ectodermal tissue. These materials are expected to provide insights into changes related to brain neurodegeneration and the AD pathology [11,27,28,29]. Many studies have shown significant structural and buccal epithelium differences during normal aging or in AD and MCI patients compared with healthy individuals [7,8,9,10].

Buccal cells, associated with the immunocytochemical expression of peripheral pro-inflammatory cytokines, including TNF-α, IL-1$\beta$, and IL-6, have demonstrated the association between peripheral inflammation and CNS degeneration in individuals with MCI and AD or the elderly. This includes the activation and release of various peripheral pro-inflammatory cytokines and antigens, such as TNF-$\alpha$, IL-1$\beta$, and IL-6, that can penetrate the BBB and provoke specific responses within the brain. Additionally, factors such as insulin levels, insulin receptor resistance, and adipose tissue are implicated in increasing the risk of AD [30,31,32,33,34,35].

This study aims to bridge the gap between clinical and computational methodologies to enhance the understanding and detection of Alzheimer’s disease. By integrating detailed cellular observations with advanced data processing and machine learning techniques, this research seeks to identify significant biomarkers and patterns associated with AD. The goal is to develop minimally invasive, cost-effective, and reliable methods for the early diagnosis and prevention of AD, thereby contributing to better management and potentially mitigating the impact of this neurodegenerative disorder on the aging population.

2. Materials and Methods

2.1. Sample Collection

A cohort of 162 individuals was included in this study, comprising 140 healthy individuals with no symptoms of dementia or cognitive deficits and 22 patients diagnosed with mild cognitive impairment (MCI) or Alzheimer’s disease (AD). The participants, ranging in age from 18 to 80 years, were randomly selected, without gender, literacy, or socioeconomic restrictions. Importantly, none of the participants presented signs of oral injury at the time of buccal cell collection. Buccal cell samples were obtained from the inner cheeks of each participant using soft cyto-brushes. The collected material was partially prepared as smears, immediately fixed in 95% alcohol, and stained using the Papanicolaou (Pap) method for cytomorphological analysis.

2.2. Immunological Analysis

For immunological analysis, additional slides were prepared from the buccal cell samples. These slides were transferred to 4% formaldehyde in PBS for a minimum of 30 min, air-dried for 1 h, and stored in sealed boxes at −4 °C until immunostaining procedures were performed. The antibodies used for immunocytochemistry included TNF$\alpha$ (52B83), IL-1$\beta$ (E7-2-hIL1$\beta$), and IL-6R$\alpha$ (H-7), all mouse monoclonal antibodies from Santa Cruz Biotechnology Inc., each at a dilution of 1:50. The antibodies showed brown cytoplasmic expression when stained with DAB Quanto Chromogen.

Microscope slides containing buccal cell smears were evaluated by visual scoring. Buccal cells were classified as basal, intermediate, differentiated, or karyolitic based on their cytoplasmic and nuclear features and ratios (Pap stain). This stain is valuable in staining a variety of bodily secretions and cell smears (nuclei: blue, superficial cells: pink, intermediate cells: blue). Hematoxylin Gill was selected as it is the lightest basophilic stain and does not overstain the specimen, but it complements the other aspects of the specimen, such as the cytoplasm. The frequency and proportion of these different cell populations were analyzed. Additionally, the cytoplasmic expression of buccal cells stained by the immunocytochemical method was evaluated and scored as mild–negative (<10%), medium (10–50%), and high (>50%) antibody expression according to two variable factors, i.e., counting the number of positively stained cells and scoring the intensity of the staining.

2.3. Data Analysis

Data preprocessing was carried out using Python’s pandas library [36]. The dataset comprised 162 instances across 13 variables, with 9 variables retained for analysis. Categorical variables were converted to numerical values; for example, sex was encoded as 1 for male and 0 for female. Age was categorized into six groups (0–19, 20–25, 26–35, 36–50, 51–65, 65+) and encoded numerically from 0 to 5. The tag for analysis was transformed into three categorical values: 0 for health, 1 for neuro, and 2 for other disease profiles. Health represents a cohort with no symptoms of dementia or cognitive deficits, neuro represent patients diagnosed with MCI and AD, and other denotes a cohort with the positive expression of TNF-$\alpha$, IL-1$\beta$, and IL-6Rα; smoking; and a neurodegenerative history.

Data visualization was performed using the matplotlib [37] and seaborn [38] libraries. A correlation heatmap using Pearson’s correlation coefficient was created to examine the inter-variable relationships and their associations with the dataset’s tag (profile). Dimensionality reduction was conducted using principal component analysis (PCA) [39], Uniform Manifold Approximation and Projection (UMAP) [40], and t-Distributed Stochastic Neighbor Embedding (t-SNE) [41], with visual representations generated in both 2D and 3D formats.

Feature importance was assessed using a hybrid approach [42] combining three gradient boosting classifiers: XgBoost [43], CatBoost [44], and LightGBM [45]. These algorithms ranked the features based on their importance, and the rankings were aggregated using a Borda rank-based count method. A line/dot plot was constructed to display the feature importance values.

Multi-class classification was conducted using the random forest algorithm [46], with a 10-fold cross-validation procedure to evaluate the performance. This particular classifier was selected due to its well-established strong performance on complex biomedical datasets, combined with its interpretability; it offers a valuable feature importance metric to assess the significance of individual features. The results were synthesized into a confusion matrix. Additionally, a decision tree classifier [47] was configured with a maximum depth of 10, a minimum of 15 samples required to split an internal node, and a minimum of 5 samples per leaf node to maintain model interpretability.

These combined biological and computational methodologies ensured a comprehensive analysis, integrating detailed cellular observations with advanced data processing and machine learning techniques to provide robust insights into the dataset. This study aims to bridge the gap between clinical and computational methodologies to enhance the understanding and detection of Alzheimer’s disease. By integrating detailed cellular observations with advanced data processing and machine learning techniques, the research seeks to identify significant biomarkers and patterns associated with AD. The goal is to develop minimally invasive, cost-effective, and reliable methods for the early diagnosis and prevention of AD, thereby contributing to better management and potentially mitigating the impact of this neurodegenerative disorder on the aging population.

3. Results

The results of this study revealed significant findings from both the biological and computational analyses. Microscope slides containing buccal cell smears were evaluated to classify buccal cells based on their cytoplasmic and nuclear features as can be seen in Figure 1. The frequencies and proportions of basal, intermediate, differentiated, and karyolitic cells were analyzed. Additionally, the cytoplasmic expression of buccal cells stained by the immunocytochemical method was evaluated and scored. The analysis showed varied expression levels of pro-inflammatory cytokines TNF$\alpha$, IL-1$\beta$, and IL-6, with distinct patterns observed in individuals with AD and MCI compared to healthy controls.

From a computational perspective, the correlation heatmap indicated significant positive correlations between the target variable (profile) and the constructed age range, and, to a lesser extent, with the TNF, IL-6, and IL-1β variables as shown in Figure 2. This suggests that these factors are strongly associated with the disease profiles being studied.

Dimensionality reduction visualizations revealed that the data points clustered together distinctly, with the most defined groupings observed. Four distinct clusters were identifiable, with one cluster comprising samples characterized by both ‘neuro’ and ‘other’ profiles, while the remaining clusters encompassed samples with ‘neuro’ and/or ‘other’ profiles as shown in Figure 3. This indicates the clear separation between the different disease profiles based on the analyzed variables. In the 2D PCA plot, 74% of the total variance is captured, while the 3D PCA plot accounts for 81.4% of the variance. Dimensionality reduction was also performed using the top four features identified through the feature importance consensus method. Under this approach, the first two principal components accounted for 87% of the total variance, and the first three principal components explained 95%. However, it is important to highlight that the final PCA visualizations, both in 2D and 3D, did not reveal any clear separation patterns among the samples. As a result, the findings related to the top four features are not included in this manuscript.

The feature importance analysis highlighted the age range as the predominant factor in differentiating among the three categories of the output variables (profiles) as shown in Figure 4. Other significant features included IL-6, IL-1β, and TNF, underscoring their relevance in the context of the disease profiles.

The performance of the random forest algorithm showed the highest accuracy in correctly identifying samples from the ’other’ category, with moderate effectiveness for the ’healthy’ and ’neuro’ categories as can be seen in Figure 5. Specifically, of the 55 samples labeled as healthy, the algorithm accurately classified 37. For the neuro category, it correctly identified 16 out of 21 samples. In the other category, the algorithm achieved the accurate classification of 60 out of a total of 85 samples, underscoring its particular strength in recognizing this class.

The decision tree classifier that was also applied has been configured to balance complexity and interpretability. The tree facilitated the clear visualization of the feature space’s segmentation, capturing essential data patterns and avoiding spurious correlations as can be seen in Figure 6. This approach allowed for transparent and interpretable model results, contributing to a comprehensive understanding of the data.

The integration of detailed cellular observations with advanced data processing and machine learning techniques provided robust insights into the dataset. The findings underscore the potential of combining biological and computational methodologies to enhance the understanding and detection of Alzheimer’s disease, paving the way for the development of effective early diagnosis and prevention strategies.

4. Discussion

This study demonstrates the potential of integrating biological observations with advanced computational methodologies to enhance the understanding and detection of Alzheimer’s disease (AD). The analysis revealed significant patterns and correlations, emphasizing the role of neuroinflammation and its biomarkers in the progression of AD. The findings suggest that peripheral inflammation is critically involved in the disease’s pathogenesis, supporting the hypothesis that immune system dysfunction contributes to neurodegenerative processes.

The application of artificial intelligence (AI) and machine learning (ML) techniques in this research has proven to be instrumental in managing and interpreting complex datasets. These advanced computational methods allowed for the identification of significant biomarkers and the differentiation of disease profiles with high accuracy. As the volume and complexity of biomedical data continue to grow, AI and ML will become increasingly essential in uncovering subtle patterns and relationships that are not readily apparent through traditional analysis.

The use of minimally invasive sampling methods, such as buccal cells, combined with sophisticated data processing, offers a promising approach for the early diagnosis and monitoring of AD. This study highlights the potential of such integrated methodologies to develop reliable, cost-effective diagnostic tools that can be widely applied in clinical settings.

The continuous evolution of AI and ML technologies holds great promise for the future of biomedical research. By enabling the analysis of large and complex datasets, these tools can help to identify novel biomarkers and therapeutic targets, paving the way for more effective prevention and treatment strategies. Future research should focus on leveraging these advancements to explore the intricate interactions between genetic, environmental, and biological factors in AD, ultimately improving patient outcomes and advancing our understanding of this debilitating disease.