The relationship between salivary Fibroblast Growth Factor-2 and cortisol reactivity and psychological outcomes prior to and during the COVID-19 pandemic

Abstract

Background

Fibroblast growth factor-2 (FGF2) is a biomarker that is associated with depression, anxiety and stress in rodents. In humans, we have previously demonstrated that salivary FGF2 increased following stress in a similar pattern to cortisol, and FGF2 (but not cortisol) reactivity predicted repetitive negative thinking, a transdiagnostic risk factor for mental illness. The current study assessed the relationship between FGF2, cortisol, and mental health before and during the COVID-19 pandemic.

Methods

We employed a longitudinal correlational design using a convenience sample. We assessed whether FGF2 and cortisol reactivity following the Trier Social Stress Task (TSST) were associated with DASS-21 depression, anxiety and stress, measured at the time of the TSST in 2019–20 (n = 87; time 1), and then again in May 2020 during the first wave of COVID-19 in Sydney (n = 34 of the original sample; time 2).

Results

FGF2 reactivity (but not absolute FGF2 levels) at time 1 predicted depression, anxiety, and stress across timepoints. Cortisol reactivity at time 1 was associated with stress over timepoints, and absolute cortisol levels were associated with depression across timepoints.

Limitations

The sample was comprised of mostly healthy participants from a student population, and there was high attrition between timepoints. The outcomes need to be replicated in larger, more diverse, samples.

Conclusions

FGF2 and cortisol may be uniquely predictive of mental health outcomes in healthy samples, potentially allowing for early identification of at-risk individuals.

1. Introduction

The global impact of the coronavirus disease (COVID-19) on social, economic and psychological factors has been monumental, with more than 114 million people impacted by unemployment and business closures (Deb et al., 2022; International Labour Organisation, 2021), as well as prolonged periods of social isolation and lockdowns. As COVID-19 developed, incidence of depression, anxiety, posttraumatic stress disorder, complicated grief and substance use increased (Bovero et al., 2022; Calina et al., 2021; Di Blasi et al., 2021; Dodge et al., 2021; Dragioti et al., 2022; Tang and Xiang, 2021). The increased prevalence of mental health conditions was particularly pronounced for people living in quarantine confinement, front-line workers and people with pre-existing mental health conditions (Bonati et al., 2022; Lewis et al., 2022a; 2022b). The mental health of student populations was also significantly impacted, e.g., more than 50% of university students in Australia and the United Kingdom reported low or very low psychological wellbeing during the COVID-19 pandemic, with a significant portion exceeding clinical cut-off levels for anxiety and depression (Chen and Lucock, 2022; Dodd et al., 2021).

Mental health experts have advocated for the prevention and early intervention of psychological distress before symptoms evolve into clinical disorders in order to improve prognostic outcomes (Jacka and Reavley, 2014). However, prevention of mental health disorders and early identification of vulnerable individuals continues to present a challenge for health professionals and government advisory bodies, as there are few reliable indicators of individual vulnerability (Galea et al., 2020). One potential way of identifying people at risk of developing mental illnesses is by using biomarkers (i.e., objective biological variants that predict clinical outcomes or hold diagnostic value, in that they reliably co-occur with symptoms or syndromes [Strimbu and Tavel, 2010]).

One potential biomarker for anxiety and mood disorders that is receiving increasing attention is fibroblast growth factor 2 (FGF2). FGF2 is a cell-signaling protein which is naturally produced in the brain and peripheral system (Benington et al., 2020). FGF2 plays a role in cell proliferation, differentiation, growth, and, of particular relevance to psychological outcomes, emotion modulation (Turner et al., 2012). In rodents, brain and peripheral (i.e., serum) levels of FGF2 are negatively associated with trait anxiety and expression of learned fear (Graham et al., 2017; Perez et al., 2009; Turner et al., 2016a). While scant research has investigated FGF2 in humans, one study has shown that salivary FGF2 is negatively associated with expression of learned fear in healthy humans (Graham et al., 2017), suggesting that there may be a relationship between FGF2 and psychological outcomes in humans. Furthermore, postmortem brain studies of individuals diagnosed with Major Depressive Disorder showed significantly reduced FGF2 expression in the prefrontal cortex compared to healthy controls (Kang et al., 2007). More recent research has demonstrated that children with anxiety and/or depression had lower serum FGF2 than healthy controls, and that levels of anxious/depressive symptoms were negatively correlated with serum FGF2 levels (Lebowitz et al., 2021).

FGF2 appears to be modulated by stress, with studies in rodents showing that restraint stress or administration of stress hormones leads to changes in FGF2 mRNA and protein in the brain and adrenal gland (Bland et al., 2006; 2007; Meisinger et al., 1996; Molteni et al., 2001; Salmaso et al., 2016). The direction and extent of these changes differs depending on the nature of the stressor, such as whether it was acute versus chronic, or controllable versus uncontrollable. For example, acute and controllable stress results in increases in hippocampal FGF2 mRNA expression (Bland et al., 2007), whereas chronic, uncontrollable stress reduces hippocampal FGF2 protein levels (Bland et al., 2006; Kirby et al., 2013) and mRNA expression (Bland et al., 2007; Kirby et al., 2013). Recent findings from our lab have established a relationship between FGF2 and stress in human samples (Bryant et al., 2022). We found that salivary FGF2 levels changed in response to psychosocial stress (i.e., the Trier Social Stress Test; TSST), in a similar pattern to salivary cortisol release (Bryant et al., 2022). Notably, cortisol and FGF2 reactivity (i.e., the magnitude of change in response to stress) were not associated and appeared to act somewhat independently as components of the human stress response, with respect to their association with the psychological impact of stress. For example, lower FGF2 reactivity (but not cortisol reactivity) was associated with fear of negative evaluation and higher levels of trait repetitive negative thinking (RNT), including rumination and post-event processing, as well as prospectively rated TSST-related RNT in the week following the experiment (Bryant et al., 2022). In summary, this study provided the first evidence that individual differences in the magnitude of stress-induced changes in FGF2 could be a novel biomarker for psychological processes (e.g., RNT) that are strongly aligned with mental health outcomes.

The onset of the COVID-19 pandemic presented a unique opportunity to test whether FGF2 reactivity (measured pre-pandemic) predicted psychological outcomes during a naturalistic stressor. In the current study, we re-contacted participants who took part in our earlier study in 2019–20 (Bryant et al., 2022; ‘time 1′, pre-pandemic) in May 2020 (‘time 2′, mid-pandemic), during Sydney's first wave of COVID-19 lockdown. We measured their psychological functioning via the Depression Anxiety Stress Scales Item 21 (DASS-21; Lovibond and Lovibond, 1995). We assessed whether FGF2 (absolute levels or reactivity following the TSST) measured at time 1 predicted DASS-21 outcomes measured at time 1 and time 2. The association between absolute and reactive cortisol levels with DASS-21 outcomes at both time 1 and time 2 were also analysed to see if cortisol related differentially to psychological outcomes, as seen in Bryant et al. (2022).

2. Materials and methods

2.1. Participants

Data collection for time 1 of this study began in September 2019 and concluded in March 2020. At the conclusion of data collection for time 1, early COVID-19 cases had been detected in Sydney, however, restrictive measures (i.e., lockdowns) had not yet been introduced. Participants were recruited through online advertising from the University of New South Wales (UNSW) undergraduate and public communities. Participants were predominantly first year university students. Exclusion criteria included current or past mental health issues, poor oral hygiene that could affect saliva analyses (e.g., gingivitis or tooth decay), current hormonal contraceptive use (which has been shown to blunt cortisol levels [Nielsen et al., 2013]) and antidepressant use (which impacts both cortisol and FGF2 basal levels [Manthey et al., 2011; Maragnoli et al., 2004]). Eighty-seven participants took part in the study at time 1, which involved providing five saliva samples (for the measurement of FGF2 and cortisol) before and after undergoing the TSST, responding to psychological questionnaires, and reporting on levels of RNT in the week following the TSST. Participants who took part in the study at time 1 were reimbursed with course credit or $20 per hour for their time. Study outcomes from time 1 reported in Bryant et al. (2022) include the change in FGF2 and cortisol over time in response to the TSST, the relationship between FGF2 and cortisol, and the relationship between FGF2, cortisol, and the following psychological outcomes: Repetitive Thinking Questionnaire- short form (McEvoy et al., 2014), Post-Event Processing Questionnaire (Rachman et al., 2000), State-Trait Anxiety Inventory- trait form (Spielberger et al., 1983) and the Brief Fear of Negative Evaluation Scale (Leary, 1983). Participants also completed the DASS-21 at time 1; these outcomes were not reported on in Bryant et al. (2022), but are reported as part of our longitudinal analysis in the present study.

Of the 87 participants who completed the study at time 1, 67 participants provided contact details for participation in future studies. These individuals were recontacted in May 2020 via email, with 34 participants responding and participating in data collection at time 2 (mid-COVID-19). The final sample at time 2 consisted of 10 males and 24 females, aged 18 to 28 years (mean = 20.8 years). Participants who took part in the study at time 2 were reimbursed with a $20 e-gift card per hour for their participation in the study.

2.2. Measures

2.2.1. Depression anxiety stress scales (DASS-21)

Participants completed the DASS-21, a self-report 21-item measure of state depression, anxiety and stress symptoms during the past week, at time 1 and time 2. Normative data in a non-clinical sample has shown that the DASS-21 has high reliability, and adequate convergent and discriminant validity (Crawford and Henry, 2003).

2.3. Salivary collection and analysis

Levels of FGF2 were obtained from saliva (as opposed to blood serum), as saliva collection is a non-invasive method of obtaining multiple samples in a short period of time. Research has demonstrated that levels of FGF2 are present and detectable in human saliva using enzyme-linked immunosorbent assay (ELISA) analysis (Graham et al., 2017; van Setten, 1995). Salivary levels of FGF2 have also been shown to be comparable to serum levels of FGF2 (Huang et al., 2012), hence sensitivity was not compromised by this methodological choice. Saliva collection was conducted at time 1. Participants were instructed not to eat in the hour before saliva collection, refrain from consuming alcohol, nicotine or caffeine in the 12 h prior to participation, and to refrain from intense exercise in the 24 h prior to participation. Saliva was collected between 12:00 and 19:00 to control for the diurnal effects of cortisol (Adam and Kumari, 2009) and the potential diurnal effects of FGF2 (Turner et al., 2016b). Saliva samples of 1.5 mL were collected using the passive drool method. This method is recommended to reduce bacterial contamination during sampling (Bhattarai et al., 2018). The saliva samples were stored in a −30° freezer and were analysed within six months of collection, in line with best-practise guidelines for salivary research (Bhattarai et al., 2018; Garde and Hansen, 2005). Samples were centrifuged prior to analysis and salivary levels of cortisol and FGF2 were analysed using commercially available ELISAs, following manufacturer instructions for cortisol and FGF2 (S1–3002, Salimetrics; RDSDFB50, R&D Systems, respectively).

FGF2 and cortisol values were square root transformed prior to analysis. After square root transformation, the average values for FGF2 levels obtained across the five timepoints were 2.09 pg/mL, 2.05 pg/mL, 2.18 pg/mL, 2.13 pg/mL, and 2.09 pg/mL. The average cortisol values after square root transformation across the five timepoints were 0.46 μg/dL, 0.45 μg/dL, 0.48 μg/dL, 0.47 μg/dL, and 0.40 μg/dL (see Bryant et al., 2022, for the original report of this data). FGF2 and cortisol reactivity values were obtained by applying area under the curve calculations (Pruessner et al., 2003) with respect to increase (AUCi) and absolute levels were obtained by calculating area under the curve with respect to ground (AUCg). AUC calculations produce a single value for each participant's change in FGF2 and cortisol levels from pre- to post-TSST, relative to their individual baseline. AUCi represents the magnitude of change over time, whereas AUCg captures absolute levels over time in relation to the distance from ground, or zero.  For further details on how AUC values were calculated and the relationship between FGF2 and cortisol, refer to Bryant et al. (2022). Here, the association between FGF2 AUCi, FGF2 AUCg, cortisol AUCi, cortisol AUCg and DASS-21 scores at time 1 and time 2 will be reported.

2.4. Procedures

All procedures were carried out in accordance with the Declaration of Helsinki and approved by the UNSW Human Research Ethics Committee (approval numbers: HC190368; HC200228), with written consent from each participant. At time 1 in 2019–2020, participants (n = 87) attended a laboratory visit at UNSW and provided two baseline saliva samples before completing the TSST to elicit a psychosocial stress response. The TSST is a well-validated paradigm for eliciting physiological and psychological stress responses which requires participants to deliver a video-recorded five-minute speech with minimal preparation and complete verbal mathematical calculations in front of an evaluator (Birkett, 2011). Participants provided additional saliva samples post-TSST at 0-, 15- and 45-minute intervals. Participants completed the DASS-21 before the stress exposure, between the two baseline saliva samples, which provided baseline (or pre-stress) levels of depression, anxiety and stress prior to time 1. At time 2 in 2020, participants (n = 34) completed the DASS-21 again to assess psychological outcomes during the COVID-19 outbreak in May 2020. These measures were completed by participants remotely via Qualtrics. Refer to Fig. 1 for a simplified timeline of the procedure (see Bryant et al., 2022 for information on additional questionnaires that were administered and not depicted in this figure).

Fig. 1.

Visual depiction of the timeline of the procedure.

2.5. Statistical analysis

Statistical analyses were performed using SPSS Statistics Version 26 with values of p < .05 considered statistically significant. Absolute levels of FGF2 and cortisol (AUCg), as well as FGF2 and cortisol reactivity (AUCi) following stress obtained at time 1 (i.e., during the TSST experiment), were used for analysis in this study.

To address the question of whether FGF2 or cortisol AUCi or AUCg at time 1 was associated with DASS-21 scores across the two timepoints, linear mixed models (LMM) were conducted as a means of analysing repeated measures data with missing data points. LMMs were conducted with time 1 and time 2 DASS-21 subscale scores (i.e., anxiety, depression, and stress) as the three time-varying dependent variables, and with the linear effect of time, age, sex, and FGF2 or cortisol AUCi or AUCg as covariates. The linear mixed models were run using a stepped approach (i.e., each parameter/variable was added sequentially in order to identify factors driving main effects and/or interaction terms). Each step, including an intercept only model, is presented in models 1–4 in Table 3, Table 4, Table 5, with the final step for each LMM presented as models 5a-d after the inclusion of FGF2 AUCi, FGF2AUCg, CORT AUCi, or CORT AUCg, respectively. All models contained a scaled identity covariance matrix. A random effect was included in each model to assume variations in baseline response values for each factor. The reported beta values estimate the increase in the dependent variable as a function of a one unit increase in the independent variable. Follow up correlational analyses were conducted to assess the relationship between FGF2/cortisol AUCi/AUCg and depression, anxiety, and stress at the two separate time points.

Table 3.

Linear mixed model with depression as the dependent variable.

Model	Parameter	Estimate (β)	Std. Error	t	P value	95% Confidence Interval
						Lower Bound	Upper Bound
1	Intercept	4.099	.374	10.956	< 0.001	3.355	4.843
2	Intercept Age	−1.115 .254	3.071 .149	−0.363 1.710	.717 .091	−7.222 −0.042	4.991 .550
3	Intercept Age Sex	−0.517 .213 .645	3.165 .158 .807	−0.163 1.354 .799	.871 .179 .426	−6.809 −0.100 −0.958	5.775 .527 2.249
4	Intercept Age Sex Time	−2.745 .144 .992 2.837	3.259 .161 .818 .682	−0.842 .896 1.212 4.159	.402 .373 .229 < 0.001	−9.221 −0.176 −0.634 1.471	3.731 .463 2.617 4.203
5a	Intercept Age Sex Time FGF2 AUCi	−2.456 .131 1.259 2.762 −0.030	3.165 .156 .804 .692 .014	−0.776 .839 1.565 3.994 −2.128	.440 .404 .121 <0.001 .037	−8.750 −0.179 −0.340 1.376 −0.058	3.838 .441 2.859 4.148 −0.002
5b	Intercept Age Sex Time FGF2 AUCg	−4.909 .136 1.019 2.773 .013	3.460 .159 .811 .678 .007	−1.419 .855 1.256 4.091 1.742	.160 .395 .212 <0.001 .085	−11.785 −0.181 −0.592 1.415 −0.002	1.968 .453 2.630 4.130 .027
5c	Intercept Age Sex Time CORT AUCi	−2.792 .152 .728 2.861 .060	3.238 .160 .829 .676 .037	−0.862 .954 .879 4.232 1.649	.391 .343 .382 <0.001 .103	−9.228 −0.165 −0.918 1.508 −0.012	3.643 .470 2.374 4.215 .133
5d	Intercept Age Sex Time CORT AUCg	−3.970 .073 .784 2.815 .068	3.256 .162 .810 .679 .034	−1.219 .452 .968 4.142 2.038	.226 .652 .336 <0.001 .045	−10.443 −0.248 −0.825 1.454 .002	2.504 .394 2.393 4.175 .135

Table 4.

Linear mixed model with anxiety as the dependent variable.

Model	Parameter	Estimate (β)	Std. Error	t	P value	95% Confidence Interval
						Lower Bound	Upper Bound
1	Intercept	3.565	.364	9.789	< 0.001	2.840	4.289
2	Intercept Age	4.738 −0.057	3.047 .148	1.555 −0.388	.124 .699	−1.326 −0.352	10.803 .237
3	Intercept Age Sex	3.204 .051 −1.769	3.051 .152 .773	1.050 .334 −2.287	0.297 .739 .025	−2.865 −0.252 −3.307	9.273 .353 −0.232
4	Intercept Age Sex Time	2.810 .040 −1.708 .477	3.117 .154 .783 .656	.902 .260 −2.182 .727	0.370 .796 .032 .470	−3.387 −0.266 −3.263 −0.839	9.007 .346 −0.152 1.794
5a	Intercept Age Sex Time FGF2 AUCi	3.179 .022 −1.390 .406 −0.036	2.962 .146 .753 .669 .013	1.073 .152 −1.846 .607 −2.746	.286 .879 .069 .547 .008	−2.718 −0.268 −2.889 −0.936 −0.062	9.076 .313 .109 1.749 −0.010
5b	Intercept Age Sex Time FGF2 AUCg	1.352 .034 −1.689 .445 .008	3.344 .154 .783 .652 .007	.404 .224 −2.156 .682 1.120	.687 .824 .034 .498 .230	−5.296 −0.272 −3.247 −0.863 −0.005	7.999 .341 −0.132 1.753 .022
5c	Intercept Age Sex Time CORT AUCi	2.787 .044 −1.826 .489 .027	3.125 .154 .800 .656 .035	.892 .284 −2.283 .745 .766	.375 .777 .025 .460 .446	−3.427 −0.263 −3.416 −0.828 −0.043	9.002 .350 −0.236 1.805 .097
5	Intercept Age Sex Time CORT AUCg	2.099 −0.001 −1.827 .483 .039	3.171 .157 .789 .653 .033	.662 −0.007 −2.317 .740 1.197	.510 .994 .023 .463 .235	−4.208 −0.314 −3.394 −0.827 −0.026	8.407 .312 −0.259 1.793 .104

Table 5.

Linear mixed model with stress as the dependent variable.

Model	Parameter	Estimate (β)	Std. Error	t	P value	95% Confidence Interval
						Lower Bound	Upper Bound
1	Intercept	5.338	.395	13.520	< 0.001	4.553	6.123
2	Intercept Age	7.581 −0.110	3.297 .160	2.300 −0.685	0.024 .495	1.025 −0.428	14.137 .209
3	Intercept Age Sex	6.838 −0.057 −0.845	3.391 .169 .861	2.016 −0.340 −0.982	0.047 .735 .329	.096 −0.393 −2.555	13.579 .278 .865
4	Intercept Age Sex Time	5.563 −0.091 −0.678 1.547	3.493 .172 .877 .714	1.593 −0.530 −0.773 2.165	0.115 .598 .441 .035	−1.379 −0.434 −2.420 .115	12.505 .251 1.064 2.979
5a	Intercept Age Sex Time FGF2 AUCi	5.941 −0.109 −0.345 1.454 −0.035	3.365 .166 .855 .728 .015	1.765 −0.656 −0.403 1.998 −2.353	.081 .514 .688 .051 .022	−0.753 −0.439 −2.046 −0.006 −0.065	12.635 .221 1.356 2.914 −0.005
5b	Intercept Age Sex Time FGF2 AUCg	4.693  −0.095  −0.668  1.534 .005	3.760 .173 .881 .714 .008	1.248 −0.546 −0.759 2.147 .638	.215 .586 .450 .036 .525	−2.781 −0.439 −2.419 .102 −0.011	12.168 .250 1.082 2.966 .021
5c	Intercept Age Sex Time CORT AUCi	5.516 −0.077 −1.087 1.544 .094	3.398 .168  .869  .708 .038	1.624 −0.462 −1.250 2.180 2.460	.108 .645 .215 .033 .016	−1.237 −0.411 −2.814 .125 .018	12.269 .256 .641 2.963 .171
5d	Intercept Age Sex Time CORT AUCg	5.130 −0.116 −0.751 1.547 .024	3.564 .177 .886 .716 .037	1.439 −0.656 −0.847 2.159 .649	.154 .513 .399 .035 .518	−1.957 −0.468  −2.511 .110 −0.049	12.217 .236 1.010 2.983 .097

3. Results

3.1. Descriptive statistics

Participant demographics, means and standard deviations for the measures used in the current study are presented in Table 1 . Descriptive statistics for participants who did (i.e., responders) and did not (i.e., non-responders) respond to the participation invitation at time 2 are presented in Table 2 . All outcomes presented in Table 1-2 were collected at Time 1.

Table 1.

Participant demographics, mean scores and standard deviations (in parentheses) for DASS-21 subscales.

	Time 1 (pre-COVID-19); n = 87	Time 2 (mid-COVID-19); n = 34
Ethnicity	Asian: n = 61 Caucasian: n = 16 Other: n = 10	Asian: n = 27 Caucasian: n = 5 Other: n = 2
Level of Education	High school: n = 67 Bachelor's degree: n = 10 Master's degree: n = 4 Diploma: n = 4 Other: n = 2	High school: n = 24 Bachelor's degree: n = 9 Master's degree: n = 1 Diploma: n = 0 Other: n = 0
Relationship Status	Single: n = 64 In a relationship: n = 20 De facto: n = 2 Married: n = 1	Single: n = 26 In a relationship: n = 7 De facto: n = 1 Married: n = 0
Occupation	Student: n = 55 Paid (part time): n = 24 Paid (full time): n = 2 Looking for work: n = 4 Self-employed: n = 1 Did not wish to say: n = 1	Student: n = 20 Paid (part time): n = 10 Paid (full time): n = 1 Looking for work: n = 1 Self-employed: n = 1 Did not wish to say: n = 1
Depression	3.43 (3.25)	5.79 (5.14)
Anxiety	3.41 (3.63)	3.85 (3.95)
Stress	4.98 (3.79)	6.09 (4.74)

Table 2.

Descriptive statistics for time 1 DASS-21 subscale scores and FGF2 and cortisol AUCi/AUCg for responders and non-responders.

	Mean	Standard Deviation	Standard Error	Range
Responders
Depression	2.12 (Normal range)	2.33	.40	0–10
Anxiety	2.88 (Normal range)	3.36	.58	0–16
Stress	3.74 (Normal range)	3.43	.59	0–15
FGF2 AUCi	.88	29.61	5.08	−113.70–61.65
FGF2 AUCg	197.93	50.67	8.69	120.53–291.00
CORT AUCi	−1.74	9.84	1.69	−20.55–27.00
CORT AUCg	40.61	11.82	2.03	23.10–72.90
Non-Responders
Depression	4.26 (Normal range)	3.49	.48	0–14
Anxiety	3.75 (Mild range)	3.78	.52	0–14
Stress	5.77 (Normal range)	3.83	.53	0–16
FGF2 AUCi	2.42	23.50	3.23	−46.58–99.60
FGF2 AUCg	184.84	53.66	7.37	100.73–371.40
CORT AUCi	−0.07	11.14	1.53	−27.60–33.15
CORT AUCg	40.50	11.43	1.57	17.40–76.35

3.2. Analytical statistics

3.2.1. Depression (DASS-21)

See Table 3 for full details of the LMMs with depression as the dependent variable. Participants’ symptoms of depression increased from time 1 (pre-COVID-19) to time 2 (mid-COVID-19) (β= 2.762, p < .001). Age and sex were not significant covariates for depression, however, lower FGF2 AUCi, but not FGF2 AUCg, significantly predicted higher depressive symptoms across time (β= −0.030, p= .037). Higher cortisol AUCg, but not cortisol AUCi, predicted higher depressive symptoms across time (β= 0.068, p= .045). As there were significant effects of time, FGF2 AUCi and cortisol AUCg, we tested models with interaction terms time*FGF2 AUCi and time*cortisol AUCg, respectively. Neither interaction term was significant (β= −0.037, p= .133 and β= 0.106, p= .064, respectively), suggesting that neither FGF2 AUCi nor cortisol AUCg predicted the amount of change in depression over time. Follow up partial correlational analysis (with sex and age as covariates) suggests the main effect of FGF2 on depression levels over time was driven by a large and significant association between FGF2 AUCi and depression at time 1 (r= −0.547, p= .001), and a medium and significant association between FGF2 AUCi and depression at time 2 (r= −0.371, p= .037). The main effect of cortisol AUCg on depression levels over time was driven by small, non-significant associations between cortisol levels and depression at both time 1 (r = 0.155, p= .398) and time 2 (r = 0.244, p= .179). Independent samples t-tests were conducted to assess for potential differences in DASS-21 scores between responders and non-responders. The mean depression score for non-responders was significantly higher than responders at time 1 (t = 3.157, p= .002; M = 4.26 and M = 2.12, respectively).

3.2.2. Anxiety (DASS-21)

See Table 4 for full details of the LMMs with anxiety as the dependent variable. Symptoms of anxiety did not significantly change over time (β= 0.406, p= .547). Age and sex were not significant covariates for anxiety. Lower FGF2 AUCi, but not FGF2 AUCg, significantly predicted higher anxiety symptoms across time (β= −0.036, p= .008). Neither cortisol AUCi nor AUCg were significant predictors of anxiety symptoms across time.  Follow up partial correlations suggest that the main effect of FGF2 on anxiety levels over time was driven by a large and significant association between FGF2 AUCi and anxiety at time 1 (r= −0.667, p= < 0.001), with a small and nonsignificant association between FGF2 AUCi and anxiety at time 2 (r= −0.264, p= .145). There was no difference in levels of anxiety at time 1 between responders and non-responders (t = 1.095, p= .277).

3.2.3. Stress (DASS-21)

See Table 5 for full details of the LMMs with stress as the dependent variable. The change in stress from time 1 to time 2 just missed the threshold for significance (β= 1.454, p= .051). Age and sex were not significant covariates for stress. Lower FGF2 AUCi, but not FGF2 AUCg, significantly predicted higher stress levels over time (β= −0.035, p= .022). Higher cortisol AUCi, but not cortisol AUCg, significantly predicted higher stress levels over time (β= 0.094, p= .016). We also tested models with interaction terms time*FGF2 AUCi and time*cortisol AUCi, respectively. Neither interaction term was significant (β= 0.016, p= .526 and β= 0.093, p= .192, respectively), suggesting that neither FGF2 AUCi nor cortisol AUCi predicted the amount of change in stress over time. Follow up partial correlations suggest the main effect of FGF2 on stress levels over time was driven by a large and significant association between FGF2 AUCi and stress at time 1 (r= −0.757, p= < 0.001), and a small and nonsignificant association between FGF2 AUCi and stress at time 2 (r= −0.232, p= .202). The main effect of cortisol AUCi on stress levels over time was predominantly driven by a medium, non-significant association between cortisol AUCi and stress at time 2 (r = 0.321, p= .073) and a small, nonsignificant association between cortisol AUCi and stress at time 1(r = 0.089, p= .626). There was a significant difference in stress levels at time 1 between responders and non-responders (t = 2.522, p= .014), with non-responders having significantly higher stress levels (M = 5.77) than responders (M = 3.74) at time 1.

4. Discussion

The current study provides evidence that human salivary FGF2 may be a stable and reliable predictor of depression over time. Specifically, individuals with higher FGF2 AUCi in response to a psychosocial stressor had lower depression symptoms, both at the time of FGF2 measurement and 2–8 months later during a naturalistic stressor. Higher FGF2 AUCi was also associated with lower levels of anxiety and stress at the time of saliva collection. Therefore, FGF2 AUCi may be an adaptive biological response that is involved in protecting against some of the negative psychological effects of stress. Notably, the current findings showed that FGF2 reactivity (i.e., FGF2 AUCi; the magnitude of change in response to stress), rather than absolute levels of FGF2 (i.e., FGF2 AUCg), uniquely predicted these psychological outcomes. This suggests that the magnitude of FGF2 release in response to stress (relative to an individual's baseline level) may be particularly important in the psychological stress response, and this is not merely a consequence of generally higher levels of FGF2.

Our previous study found dissociations in the relationship between FGF2 and cortisol with a range of psychological outcomes, including fear of social evaluation, rumination and post-event processing (Bryant et al., 2022). The current findings further indicate that FGF2 may act somewhat independently of cortisol as a component of the stress response, as there were differential patterns of associations between cortisol, FGF2 and DASS-21 outcomes, both with respect to the presence and direction of these relationships. Higher cortisol AUCg predicted more depressive symptoms across time, however, was not associated with anxiety or stress across time. Furthermore, cortisol AUCi was positively associated with stress across time, but not depression or anxiety, in contrast to FGF2 AUCi which was negatively related to all DASS-21 outcomes across time. Our findings with respect to cortisol are consistent with research finding that people with clinical levels of depression show higher levels of cortisol compared to healthy controls in the recovery period following stress (Burke et al., 2005). The fact that the relationship between FGF2 AUCi and psychological outcomes is in an opposing direction to that of cortisol suggests that FGF2 potentially plays a unique role that is distinct from cortisol in impacting psychological outcomes following stress.

The relationship between FGF2 AUCi and depression, anxiety, and stress may emerge as a function of FGF2’s role in recovery and regeneration as part of the stress response. Research suggests that FGF2 release is modulated by the hypothalamic-pituitary-adrenal (HPA) axis (Callaghan et al., 2013), which primarily modulates the biological and physiological stress responses. Although little is known about the specific mechanisms underpinning the HPA axis modulation of FGF2, FGF2 knockout mice (i.e., mice that are genetically modified to inhibit FGF2 expression) have increased HPA axis activity and reduced glucocorticoid receptor activity in the hippocampus, effects that are reversed with FGF2 administration (Salmaso et al., 2016). As FGF2 is involved in recovery and regeneration following physical stress (Ornitz and Itoh, 2015), it is possible that FGF2’s response to psychosocial or emotional stress operates in a similar manner. Salivary expression of FGF2 is increased in individuals diagnosed with oral cell carcinoma (Gorugantula et al., 2012) and the topical application of FGF2 to mouth ulcers in rabbits accelerated healing of the lesions (Fujisawa et al., 2003). Similar applications of FGF2 have been adopted for human skin wound healing after research evidenced that the topical administration of FGF2 accelerated wound closure, to the extent that this has been accepted into Japanese medical practice since 2001 (Koike et al., 2020). FGF2 therefore may not only facilitate growth and repair for damaged cells and tissue, but also act as a protective buffer during situational stress which poses similar potential negative impacts on the body. The present study suggests that FGF2 may also be part of a response system to psychological stress, supporting the idea that the brain and body process physical and psychological stress similarly. For example, although emotional or psychological stress appear to activate unique neural systems, there is considerable overlap in the physiological and neurobiological responses to various types of stress (Kemeny, 2003; Weiner, 1992). Changes in the autonomic nervous system (e.g., increased heart rate) and HPA axis responses (e.g., cortisol release) are observed in both physical and psychological stress responses (Koenig et al., 2015; Ulrich-Lai and Herman, 2009). Therefore, FGF2 release may be a biological response to stress which is both physically and psychologically protective. Furthermore, FGF2’s protective role following physical and psychological stress may be genetically predisposed, with recent evidence demonstrating that human serum FGF2 levels were positively correlated between mothers and their offspring (Lebowitz et al., 2023). This research also demonstrated that children with higher serum FGF2 levels had lower symptoms of anxiety and depression, suggesting that there may be a cross-generational link between parental FGF2 levels and psychiatric vulnerability in children, although further research is required to substantiate this possibility (Lebowitz et al., 2023).

5. Limitations

Several limitations in the current experiment should be considered when evaluating the capacity of FGF2 as a predictive biomarker. While this study provided evidence that FGF2 AUCi is linked to depression, anxiety and stress at time 1, and depressive symptoms in a future stressful context (i.e., time 2), FGF2 AUCi was not significantly associated with anxiety or stress at time 2. Previous research has demonstrated that FGF2 levels and the relationship with affective outcomes are relatively stable over time. For example, fear expression (i.e., freezing) in rodents was related to central FGF2 protein levels three months after behavioural testing (Walters et al., 2016) and serum FGF2 levels in children have been shown to be stable after six months (Lebowitz et al., 2021). Given these findings, we would expect the relationship between FGF2, stress, and anxiety at time 1 to be similar at time 2. The failure to replicate this relationship at time 2 may be due to an underpowered sample, as more than half of the original participant pool did not respond to participate at time 2. Indeed, as the correlations between FGF2 AUCi and stress and anxiety at time 2 were negative (consistent with the correlations at time 1), it is possible that the same relationship existed at time 2 yet was undetected in a smaller sample. Furthermore, the small sample size in this study precluded the investigation of whether depression, anxiety and stress symptoms are differentially associated with FGF2 reactivity. Future research may recruit samples with distinct clinical presentations (i.e., individuals diagnosed solely with an anxiety disorder, stressor-related disorder, or depression) in order to investigate potential differential profiles of FGF2 reactivity between these diagnoses. However, as the comorbidity between these presentations is estimated to be as high as 70% (Kalin, 2020), it may prove challenging to isolate these presentations in respect to their relationship with FGF2.

Another limitation is that this study lacked the sensitivity to assess whether FGF2 predicted a change in psychological outcomes over time, as anxiety and stress did not significantly change from time 1 to time 2. This may reflect our restricted sample of predominantly healthy participants, who remained healthy at time 2, evidenced by the majority of DASS-21 scores being in the normal to mild range, even during the COVID-19 pandemic. Due to ethical considerations, individuals with mental health issues were excluded from recruitment in the initial study as it involved a psychosocial stressor. Another consideration is that the current sample were majority university students and aged between 18–35 years old. While research has clearly demonstrated that the mental health of young adults was significantly affected by COVID-19, (Sojli et al., 2021), the current sample may not reflect the populations most affected by COVID-19, such as full-time workers at risk of unemployment, parents home schooling their children and front-line workers. Furthermore, there were significant group differences in depression and stress scores at time 1 between participants who responded to recruitment compared to those who did not respond, as participants who did not respond had higher levels of depression and stress. It is possible that FGF2 AUCi predicts the amount of change in psychological distress at a future timepoint for populations that showed a more pronounced change in psychological symptoms over the course of the pandemic, or had higher symptom severity overall. Future prospective studies should include clinical populations to allow for more variability in symptom levels and change in psychological outcomes over time.

6. Implications and conclusions

Notwithstanding these limitations, the present study raises the possibility that FGF2 AUCi could be causally related to psychological outcomes. If so, then it is possible that procedures designed to augment FGF2 levels during periods of stress could be clinically beneficial. Some of the behaviours that are already recommended for remediating depressive symptoms and anxiety are also known to elevate FGF2 levels. For example, exercise is commonly incorporated into the treatment of mental health disorders (Carek et al., 2011) and physical exercise has been shown to induce hippocampal mRNA expression in rodents (Gómez-Pinilla et al., 1997). Adequate sleep is another factor that is encouraged in the treatment of mental health difficulties, and sleep restoration following a period of deprived sleep leads to increases in FGF2 in rats (Hairston et al., 2004). It is therefore possible that FGF2 plays some mechanistic role in remediating the symptoms of depression and anxiety via these common therapeutic interventions. However, given the current findings indicate that FGF2 reactivity (rather than absolute levels) are predictive of psychological outcomes, future research should investigate the impact of these therapeutic manipulations on FGF2 reactivity, which is yet to be assessed. Furthermore, these findings question the practicality of measuring stress-related changes in FGF2 as a biomarker, as opposed to obtaining a single basal measure that does not require a stress manipulation. As existing research has demonstrated that higher FGF2 levels obtained from blood serum are associated with favourable psychiatric outcomes (He et al., 2014; Lebowitz et al., 2021), basal serum levels of FGF2 may hold more utility as a psychiatric biomarker. However, measuring FGF2 reactivity through saliva samples (as this is the least invasive method to obtain multiple measurements) may be beneficial for screening specific stress-related coping, as opposed to general mental health outcomes. This may be useful for the early identification of vulnerable individuals in populations who are exposed to higher levels of stress and trauma (e.g., first-responders and military personnel).

The present study provides further evidence for the potential role of FGF2 in mental health outcomes and coping in the context of stress. While an abundance of research has demonstrated that depression, anxiety and stress have peaked during the COVID-19 pandemic, less is understood about individual differences that are associated with vulnerability or greater resilience. The current findings suggest that biological factors such as FGF2 and cortisol may be involved in physiological and cognitive processes that are elicited in the context of stress. This study builds upon previous work from our lab demonstrating that not only is FGF2 a reactive component of the human stress system, but it is negatively associated with cognitive processes typically seen in affective disorders (Bryant et al., 2022). We have demonstrated in this study that FGF2 is associated with depression, stress and anxiety at the time of an acute laboratory social stressor, and is also predictive of depressive symptoms at a later timepoint during a naturalistic stressor. Nonetheless, the potential of FGF2 as a biomarker of clinical utility for psychological conditions remains unclear. Future studies with larger cohorts, as well as the examination of people with current psychological conditions, are required to further elucidate the role of FGF2 in mental health disorders.

CRediT authorship contribution statement

Emma M. Bryant: Conceptualization, Data curation, Methodology, Project administration, Writing – original draft, Writing – review & editing. Rick Richardson: Funding acquisition, Writing – review & editing. Bronwyn M. Graham: Conceptualization, Formal analysis, Funding acquisition, Methodology, Supervision, Writing – review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.