Increased academic stress is associated with decreased plasma BDNF in Chilean college students

Introduction

Stress is an organism’s multifaceted response to any demand (Selye, 1956). The initiating stressor prompts either an acute or chronic reaction (Selye, 1973; Tafet, 2022). Acute stress elevates heart rate, blood pressure, respiratory rate, glycemia, and coagulation factors (Tafet & Nemeroff, 2016; Tafet & Nemeroff, 2020). This leads to a coping phase, then relaxation, and finally, physiological homeostasis (Tafet, 2008; Tafet & Nemeroff, 2016; Tafet & Nemeroff, 2020). Chronic stress begins with an alert phase, then a resistance stage. An exhaustion phase follows, stemming from prolonged stressor exposure, disrupting homeostasis. This imbalance triggers pathophysiological events like increased proinflammatory cytokines, hippocampal structural changes, and hormonal shifts (Conrad, 2010; Hänsel et al., 2010; Nargund, 2015; Sribanditmongkol et al., 2015). Furthermore, chronic stress is linked to cardiovascular disease and cancer development or prognosis (DeSteno, Gross & Kubzansky, 2013; Dhabhar, 2014; Steptoe & Kivimäki, 2012; Tian et al., 2014).

Academic stress (AS) arises from educational experiences. It starts in primary students and intensifies through university (Dyson & Renk, 2006; Putwain, 2007). Higher education epitomizes this stress due to hefty workloads. Concurrently, students often face family separation, job entry, and new environments (Barraza-Macías, 2007; Castillo-Navarrete et al., 2023a). University students encounter notably stressful periods, requiring significant adaptation. This can lead to exhaustion, dwindling study interest, and environmental challenges (Guzmán-Castillo et al., 2022).

Brain-derived neurotrophic factor (BDNF) is a protein prevalent in the adult brain. It’s found in most cortical areas, several subcortical regions, and spinal cord sections. BDNF is vital for functions like dendritic branching, spine morphology, synaptic plasticity, and long-term potentiation (LTP). Consequently, it affects memory, learning, appetite, and sleep (Castillo-Navarrete et al., 2023a; Karege et al., 2002; Lommatzsch et al., 2005; Nagahara & Tuszynski, 2011).

The BDNF gene, on 11p13, features a nonconservative exonic SNP at nucleotide 196 (dbSNP rs6265, G/A). This SNP transforms valine to methionine in proBDNF 5′protein at codon 66 (Val66Met). This change impacts BDNF packaging, reducing its release-dependent activity. The Val/Met variant correlates with cognitive deficits, impaired memory, and reduced hippocampal activity (Nagahara & Tuszynski, 2011; Vyas & Basant, 2012).

BDNF in peripheral blood (PB) originates from platelets and the brain (Lommatzsch et al., 2005; Polacchini et al., 2018; Suliman, Hemmings & Seedat, 2013). Since it traverses the blood–brain barrier, peripheral BDNF indicates brain BDNF levels and cortical health (Karege et al., 2005; Polacchini et al., 2015; Suliman, Hemmings & Seedat, 2013). BDNF levels in PB are measured in serum, plasma, or platelets (Castillo-Navarrete et al., 2023a; Polacchini et al., 2015; Teche et al., 2013).

Stress profoundly impacts BDNF (Bath, Schilit & Lee, 2013; He et al., 2020). Chronic stress activates the hypothalamic-pituitary-adrenal (HPA) axis, elevating cortisol and neuroinflammatory cytokines while reducing BDNF levels (Brunoni, Lopes & Fregni, 2008; Castillo-Navarrete et al., 2023a). Assessing peripheral BDNF during acute stress can help gauge HPA axis activity (Castillo-Navarrete et al., 2023a; Linz et al., 2019; Schmitt, Holsboer-Trachsler & Eckert, 2016; Schulte-Herbrüggen et al., 2006).

BDNF supports neuroplasticity and brain health, with its expression affected by stress types, including chronic, acute, and post-traumatic (Aksu et al., 2018; Martinotti et al., 2016; Zamani et al., 2019). Psychiatric disorders like depression and schizophrenia exhibit altered BDNF levels (Martinotti et al., 2016; Suliman, Hemmings & Seedat, 2013; Vinogradov et al., 2009). Cognitive training in schizophrenics elevates serum BDNF (Martinotti et al., 2012; Vinogradov et al., 2009). Therapies, including antidepressants and psychotherapy, also boost BDNF (Kishino et al., 2001; Martinotti et al., 2016; Vanicek et al., 2019). These findings highlight BDNF’s role in stress-related psychiatric disorders and therapeutic modulation potential.

Traumatic experiences may predispose individuals to stress disorders through epigenetic modulation (Sant & Goodrich, 2018). This can involve DNA methylation, histone changes, and non-coding RNA processes, influencing gene transcription (Hing, Sathyaputri & Potash, 2018). 5-methylcytosine (5-mC) is a methyl group attached to cytosine’s carbon 5. Global DNA methylation (G-DNA-M) measures 5-mC levels against total cytosine (Murgatroyd, 2014; Paul & Tollefsbol, 2014; Yamamoto et al., 2014). While many studies target specific genes, G-DNA-M’s role, present in coding and non-coding areas, is noteworthy (Makhathini et al., 2017).

University students experience distinct stress levels, possibly causing epigenetic shifts affecting BDNF gene expression. Despite its importance, few studies explore AS’s impact on peripheral BDNF and G-DNA-M percentages. This research seeks to discern if AS affects plasma BDNF and global DNA methylation percentages, offering insights into academic stress’s intricate regulatory mechanisms.

Methods

Statistical evaluation

Data source

Access the utilized data at https://doi.org/10.48665/udec/ORVXAA.

Descriptive and bivariate analysis

An initial descriptive examination of variables took place. This was followed by a bivariate analysis. Each independent variable was individually assessed to discern its relation to changes in plasma BDNF levels or G-DNA-M percentage (dependent variables). Graphical analysis was incorporated to ascertain the nature of the relationship, be it linear or otherwise.

Multivariate analysis

To address the proposed changes in plasma BDNF levels and/or G-DNA-M percentage during the final evaluation, we crafted empirical predictive models (refer to Fig. 1). Initially, for numerical variables, we graphically represented them to gauge the data curve’s attributes, both for the entire cohort and by gender. This visual exploration revealed that the relationships between the aforementioned predictive variables (changes in final plasma BDNF levels and global DNA methylation percentage) aren’t linear (see Fig. 1A). Hence, we employed restricted splines to capture the non-linear interplay between these variables (Gauthier, Wu & Gooley, 2020; Harrell, 2015). Restricted splines effectively represent a non-linear relationship through a linear one, essentially transforming an independent variable. The independent variable’s value range was segmented, with “knots” demarcating one segment’s conclusion and the next’s commencement. Consequently, splines ensure the resultant curve remains smooth and continuous, minimizing model misrepresentation and elucidating the predictor-outcome relationship (Gauthier, Wu & Gooley, 2020).

Figure 1C illustrates the quest for predictive strength (R²), focusing on the change in a variable without a definitive cut-off for a robust model. In the empirical development of a multivariate predictive model, it became evident that variable relationships weren’t linear, necessitating the use of restricted splines. Based on one model’s Anova, we proposed a subsequent model. This involved conceptually discarding non-pertinent variables and pinpointing those with insignificant linear and non-linear components. We then compared the new model to its predecessor using likelihood ratio tests for two nested models (see Fig. 1B). Relying on the computed adjusted R² (R²aj), we subjected the empirical model to validation via bootstrapping (over 10,000 iterations) as a model validation criterion (Efron, 1979; Harrell, 2015).

We began with a comprehensive model encompassing 16 predictors for changes in plasma BDNF and G-DNA-M percentage. Using the likelihood ratio test, variables were systematically eliminated until a final model emerged for each metric. This model underwent validation via Bootstrap (Efron, 1979; Ledesma, 2008). We derived the empirical predictive power (R²e), offering a tangible gauge of the model’s predictive prowess and future performance. Subsequently, for an average individual, predictions were made based on a specific independent variable, while controlling others. This determined changes in the final versus initial evaluations in plasma BDNF or G-DNA-M percentage. Database creation utilized Microsoft Excel 365, while R (R Core Team, 2023) facilitated statistical analysis, setting a significance threshold at α = 0.05.

Figure 1: Illustrates the multivariate analysis process, the empirical development of a predictive model, and its theoretical foundation. Figure 1A displays the relationship between final physical and psychological reactions and the logarithm of BDNF change. The non-linear nature of these variables prompted the use of restricted splines in our predictive model. These splines effectively transform a non-linear relationship into a linear one. Figure 1B depicts the relationship between final physical and psychological reactions and the logarithm of BDNF change, with the variable controlled based on the model. Comparing the variable curve to the no-change line reveals a confidence interval gap (24 to 31 points). Figure 1C delves into the quest for predictive power (R²). It emphasizes the challenge of pinpointing a definitive cut-off value for a robust model. Figure 1D outlines the empirical development of a multivariate predictive model. Given the non-linear relationships among variables, the analysis incorporated non-linear effects using restricted splines. In the context of the population, when scores for physical and psychological reactions in the final assessment range between 24 and 31 points, a decline in end-of-study plasma BDNF levels compared to initial levels is anticipated. Using an iterative approach and based on one model Anova, a subsequent model was proposed. This model conceptually removed non-essential variables and pinpointed those with insignificant linear and non-linear components. The new model was then compared to its predecessor using likelihood ratio tests for nested models. With the adjusted R² (R²aj) as a guide, the empirical model underwent validation through bootstrapping, exceeding 10,000 iterations (Harrell, 2015). This process yielded the empirical predictive power (R²e).

Results

Discussion

In our quest to understand the relationship between AS and its potential impact on plasma BDNF and G-DNA-M percentage, we embarked on the creation of empirical models. Although there are overarching guidelines for such modelling, the specificity of our predictive objectives called for a distinct data set. The construction and interpretation of these models were driven by a blend of statistical rigour and theoretical understanding.

To investigate the link between AS and plasma BDNF shifts, we employed a 16-variable model. This model showed clear significance (R²a = 0.187; p < 0.001). Later, seven non-impactful variables were discarded. One anticipated exclusion was the variable representing excessive alcohol consumption (AUDIT) from the final evaluation. The average participant score (2.88 ± 2.86 points) was beneath the 6–8 point risk consumption threshold (Alvarado et al., 2009; Donoso, 2015). Previous studies have underscored alcohol’s influence on peripheral BDNF levels (Umene-Nakano et al., 2009). It’s noteworthy that while serum BDNF levels experience an uptick 14 days after halting alcohol consumption, they don’t parallel the levels seen in non-drinking counterparts even after half a year (Girard et al., 2020; Heberlein et al., 2016). Notably, BDNF levels in adolescents demonstrated a decline among those who consumed alcohol (Miguez et al., 2020).

The variables from the Beck-II depression inventory, both final and baseline, were excluded. Despite a significant rise in the Beck-II score at the final assessment (p < 0.001), the average was 17.65 (11.23) points, under the 18-point threshold. This evaluation occurred amidst academic stress, potentially highlighting undetected depressive symptoms from the baseline.

Peripheral BDNF decrease is noted in major depressive disorder patients. Various therapies, including pharmacological and psychotherapeutic, only achieve partial BDNF increase (Aldoghachi et al., 2019; Bocchio-Chiavetto et al., 2010). They don’t match levels in healthy individuals (Brunoni, Lopes & Fregni, 2008; Kishi et al., 2018). Studies using Beck-II find a BDNF-depressive symptomatology link (Cheon et al., 2018; Just, Fraszczak & Araszkiewicz, 2016). Male university students with the disorder showed reduced plasma BDNF levels. Their levels were measured using Beck-II, without linking to academic overload (Tavakoli et al., 2017).

In the SISCO-II-AS framework, stressors were excluded from both initial and concluding assessments. Stressors encapsulate scenarios students perceive as academic pressures (Castillo-Navarrete et al., 2020). A comparison between the assessments revealed discernible differences (Table 1). This, combined with other SISCO-II-AS elements, indicates the manifestation of AS in the study’s latter phase.

In the SISCO-II-AS model, Pprx’s importance is clear. In the final assessment, it rose significantly (p < 0.001). Items in it suggest self-regulation loss. Stress affects neurotrophin expression, especially BDNF (Bath, Schilit & Lee, 2013). Stress outcomes are shaped by form, duration, timing, and the individual. The HPA axis is activated by psychological stress, amplifying cortisol levels while diminishing BDNF (Brunoni, Lopes & Fregni, 2008). Sexual dimorphism in stress responses, notably in BDNF levels, showed pronounced variations: men at 17.14% and women at 3.56%. BDNF’s interplay with cortisol during acute stress suggests assessing peripheral BDNF can shed light on HPA axis operations (Linz et al., 2019). Persistent stress impacts the HPA axis, lowering BDNF (Miller, Chen & Zhou, 2007; Schmitt, Holsboer-Trachsler & Eckert, 2016; Schulte-Herbrüggen et al., 2006). This axis plays a role in initiating and concluding sleep (Han, Kim & Shim, 2012). The mention of sleep disorders in Pprx emphasizes sleep’s importance in cognitive processes. Predictably, individuals with insomnia have reduced BDNF levels, with a direct correlation to the severity of their insomnia (Elliott et al., 2014; Grønli, Soulé & Bramham, 2014; Schmitt, Holsboer-Trachsler & Eckert, 2016).

Based on the SISCO-EA’s AS construct, responses emphasize stressors’ effects, not causes. Responses, whether conscious or unconscious, are physiological and psychological. This amplifies the total reaction (Pprx and Sbrx), indicating regulatory capacity loss and disrupted homeostasis (Castillo-Navarrete et al., 2020). Furthermore, coping strategies rise in reaction to stressor consequences. This is evident as Pprx and coping variables increased significantly in the final assessment.

The constant intellectual challenge posed by university studies provides ongoing cognitive stimulation. Such cognitive training has the potential to counteract the BDNF-reducing effects of academic stress, thereby potentially maintaining or even increasing peripheral BDNF levels (Al-Thaqib et al., 2018; Angelucci et al., 2016; Damirchi, Hosseini & Babaei, 2018; Jahangiri, Gholamnezhad & Hosseini, 2019; Kallies et al., 2019; Ledreux et al., 2019; Miyamoto et al., 2018). Alongside this, the practice of hobbies, particularly those that involve physical activity, can enhance BDNF levels, offering another potential buffer against academic stress (Al-Thaqib et al., 2018; Damirchi, Hosseini & Babaei, 2018; Jahangiri, Gholamnezhad & Hosseini, 2019; Kallies et al., 2019; Ledreux et al., 2019; Miyamoto et al., 2018).

The percentage of G-DNA-M showed no significant differences between baseline and final assessments. Thus, linking these variables to plasma BDNF change remains uncertain. In African-Americans, increased G-DNA-M was observed in post-traumatic stress individuals (Smith et al., 2011). This suggests psychosocial stress may alter G-DNA-M patterns, affecting perception and physiological responses to stressors. Since G-DNA-M state reduction inhibits transcription, a potential BDNF level decline could result from heightened DNA methylation. To clarify this, specific BDNF gene methylation and related genes should be examined.

By removing non-linear components and simplifying non-linearity, we derived the most straightforward predictive model. This validated model (R²aj =0.368; R²e =0.172) applies to populations under evaluated conditions and procedures. It predicts future BDNF plasma levels under these conditions. This model aids in predicting plasma BDNF changes at the final assessment, partially addressing the study’s aim. Predictions indicated a pronounced sex difference in plasma BDNF changes: women’s plasma BDNF decreased by 3.56% from baseline at the final assessment, while men’s showed a more considerable decrease of 17.14% (Table 3). This disparity underscores the need to further explore the role of sex in modulating BDNF responses to academic stress.

We empirically crafted a multivariate model predicting G-DNA-M level changes at final assessments. The initial model, comprising 16 variables, showed notable predictive capacity (R²a = 0.226; p < 0.001). Variables like baseline stressors, baseline coping, and final Pprx were then discarded. Intriguingly, the final Pprx, a primary AS indicator and a strong self-regulation loss signal, wasn’t significant. This might suggest an inadequate sample size or that chronic stress doesn’t impact G-DNA-M levels. However, these propositions require validation through dedicated studies.

Sexual dimorphism in stress response has been well-documented, and our findings support these established notions. Besides the observed differences in G-DNA-M between genders, it is essential to highlight the disparity in BDNF changes between men and women. The pronounced decline in men’s BDNF levels compared to women suggests an inherent difference in how chronic academic stress impacts the neurotrophic support system between genders (Castillo-Navarrete et al., 2023b). This divergence, especially within the context of the HPA axis activity, provides a nuanced layer to the understanding of the relationship between stress, G-DNA-M, BDNF, and gender. Such findings mandate a deeper exploration of sex-specific mechanisms and interventions in managing academic stress.

Significant variables in this intricate model are final Beck-II and baseline Beck-II. Global DNA hypomethylation is documented in major depressive disorder (Bakusic et al., 2017; Nantharat et al., 2015). Additionally, some studies link major depressive disorder to stressful events (Bakusic et al., 2017; Duman & Canli, 2015). Early life stresses and psychosocial factors are central to these analyses (Bakusic et al., 2017; Duman & Canli, 2015; Nantharat et al., 2015; Tseng et al., 2014; Unternaehrer & Meinlschmidt, 2016).

Psychosocial stress might impact specific genes’ G-DNA-M patterns. These genes concern individual physiological responses and conditioning factors (Lisoway et al., 2019; Smith et al., 2011; Wolf et al., 2018). A stressor’s presence activates the HPA axis. This activation encompasses both the stimulatory pathways and the feedback loops which are involved in regulating the production of stress hormones, including cortisol (Miller, Chen & Zhou, 2007; Schmitt, Holsboer-Trachsler & Eckert, 2016; Schulte-Herbrüggen et al., 2006). About post-traumatic stress, an increase in G-DNA-M has been documented (Archer et al., 2019; Mehta et al., 2020; Notaras & Van den Buuse, 2020; Roth et al., 2011; Smith et al., 2011). For a more precise understanding, studies focusing on the specific methylation of particular genes associated with this system would be beneficial.

While our study focused on G-DNA-M and BDNF as indicators of stress response, we acknowledge the absence of direct cortisol measurements. Cortisol, a primary stress hormone, would offer immediate insights into HPA axis activity during stress events. Future studies should incorporate cortisol levels to provide a comprehensive assessment of the stress response.

The model, after validation and refinement, showed a negative R²e (−0.448), raising questions about its prediction accuracy. In our sample, women’s G-DNA-M percentage rose by 15.06% in the final check. For men, it was an 18.96% rise (Table 3). The SISCO-II-AS model predicts a G-DNA-M percentage increase in the final assessment. While these variables suggest the AS phenomenon, the influence of unidentified variables remains uncertain. They might activate the HPA axis and other stress-regulating neuroendocrine systems. These could also cause specific and global methylation changes (Archer et al., 2019; Bakusic et al., 2017; Duman & Canli, 2015; Lisoway et al., 2019).

In this study, a rigorous approach was used. We employed psychometric variables with a validated history and an explicit application method, which is a strength of this study. An adjusted R² was obtained, suggesting the possibility of extrapolating the results to a broader population. However, the robustness of the study is potentially limited by the small sample size. Furthermore, it should be noted that our sample consisted only of students with strong biological and scientific backgrounds. This could significantly influence how the phenomenon of AS is manifested and managed, possibly biasing the overall applicability of the results. This specificity in the sample represents both a strength, in terms of its well-defined scope, and a limitation, given the potential challenges in broader applicability.

Regarding plasma BDNF levels, the study recognizes certain limitations. Several pre-analytical variables were not accounted for. These exclusions extend not only to the individual participants’ characteristics but also to the technical aspects like the plasma separation process after obtaining the peripheral blood sample.

The decision to use the percentage determination of G-DNA-M has its limitations, underscoring the study’s cautious approach. Drawing expansive conclusions from this measure is restrained. A more comprehensive understanding would necessitate specific methylation studies targeting the promoter level of different BDNF gene exons and genes associated with the HPA axis function. This represents an area for potential improvement and further inquiry.

While we’ve examined global DNA methylation, we acknowledge the absence of specific BDNF promoter methylation analyses. Indeed, examining DNA methylation at BDNF gene promoters would provide more nuanced insights into BDNF regulation (reference to BDNF promoter studies). We concede that correlating global methylation with BDNF levels is a preliminary approach. It is crucial to remember that BDNF regulation is multifaceted, encompassing DNA methylation, histone modification, and non-coding RNA interactions. Our current findings only touch on a portion of these intricate regulatory mechanisms. In future projects, we aim to delve deeper into BDNF’s epigenetic regulation. Understanding how academic stress impacts BDNF through epigenetics, especially DNA methylation, is paramount.

From the developed predictive models, a proposal emerges. AS may relate to plasma BDNF level changes in the wider population. Also, it could correlate with G-DNA-M percentage shifts in our analyzed sample, considering the noted limitations. Given these findings, future endeavours should validate the model’s efficacy for predicting final BDNF levels in an expanded cohort of university students, specifically its link with AS.

There is a pressing and undeniable need to delve deeper into the intricacies of AS and its potential impacts on genetic architecture. Specifically, meticulous studies focusing on the methylation patterns at the exon level of the BDNF gene are paramount. Recognizing the critical importance of direct measurements of cortisol, we have initiated a new project assessing peripheral blood stress conditions and salivary cortisol levels in university students with varying degrees of academic stress. Preliminary protocols and methodologies for this study have been outlined (Castillo-Navarrete et al., 2023a). Equally vital is the investigation of genes intricately linked with the hypothalamic-pituitary-adrenal (HPA) axis. Such granularity in analysis can uncover nuanced mechanisms of stress response and regulation. Comprehensive examinations encompassing both genetic markers and cortisol determinations will be crucial in furthering our understanding of the AS phenomenon.

Study Limitations: In this investigation, certain limitations must be acknowledged. Primarily, our study utilized a convenience sample, which might not be fully representative of all university students. This sampling method could introduce biases that might influence our findings’ generalizability. Additionally, our reliance on self-report scales can present challenges. While these scales have been validated and widely used, there is always a possibility of participant bias, recall errors, or misinterpretation of questions, potentially affecting the accuracy of the data.

Our findings provide preliminary insights into the neurobiological markers associated with academic stress among university students. Recognizing the potential relationship between academic stress, plasma BDNF levels, and G-DNA-M percentages could offer new avenues for early intervention and support. Educational institutions, in collaboration with healthcare providers, might consider incorporating strategies such as regular mental health screenings, stress management workshops, and promoting activities known to elevate BDNF levels, like exercise and cognitive training. Such measures could proactively address academic stress, fostering a holistic well-being environment for students.

While our focus has been on academic stress and its influence on BDNF levels, we acknowledge the multifaceted nature of BDNF regulation. Other protective factors such as diet, social interactions, and mental well-being practices undertaken by some students might also play a role in BDNF dynamics. Genetic predispositions, where certain individuals might possess inherent resilience to stress-induced BDNF changes, should also be considered. Further studies investigating these additional factors in conjunction with academic stress would provide a more comprehensive understanding of BDNF regulation in this context.

Conclusions

Our study revealed notable differences between the initial and final assessments concerning SISCO-II-AS and its components. This underscores the presence of academic stress in the concluding evaluation. Yet, there were no discernible differences in terms of plasma BDNF and G-DNA-M percentages. We constructed and validated multivariate models, considering the SISCO-II-AS-derived variables, and acknowledging the inherent limitations of our approach. The predictive model focused on final plasma BDNF levels. It indicated a decline in the average final plasma BDNF percentage for both genders. This observation spanned the broader population. Conversely, another model projected G-DNA-M percentage fluctuations by the study’s end. It foresaw an increase in average G-DNA-M percentage change for our sample’s men and women. Taking study constraints into account, our data offers initial insights. There is a possible connection between academic stress and plasma BDNF level changes. Similarly, There is a link with G-DNA-M percentage changes among university students.

Supplemental Information