Epigenetic GrimAge acceleration and cognitive impairment in bipolar disorder

Abstract

Bipolar disorder (BD) has been previously associated with clinical signs of premature aging, including accelerated epigenetic aging in blood and brain, and a steeper age-related decline in cognitive function. However, the clinical drivers and cognitive correlates of epigenetic aging in BD are still unknown. We aimed to investigate the relationship between multiple measures of epigenetic aging acceleration with clinical, functioning, and cognitive outcomes in patients with BD and controls. Blood genome-wide DNA methylation levels were measured in BD patients (n=153) and matched healthy controls (n=50) with the Infinium MethylationEPIC BeadChip (Illumina). Epigenetic age estimates were calculated using an online tool, including the recently developed lifespan predictor GrimAge, and analyzed with generalized linear models controlling for demographic variables and blood cell proportions. BD was significantly associated with greater GrimAge acceleration (AgeAccelGrim, β=0.197, p=0.009), and significant group-dependent interactions were found between AgeAccelGrim and blood cell proportions (CD4+ T-lymphocytes, monocytes, granulocytes, and B-cells). Within patients, higher AgeAccelGrim was associated with worse cognitive function in multiple domains (short-term affective memory (β=−0.078, p=0.030), short-term non-affective memory (β=−0.088, p=0.018), inhibition (β=0.064, p=0.046), and problem solving (β=−0.067, p=0.034)), age of first diagnosis with any mood disorder (β=−0.076, p=0.039) or BD (β=−0.102, p=0.016), as well as with current smoking status (β=−0.392, p<0.001). Overall, our findings support the contribution of epigenetic factors to the aging-related cognitive decline and premature mortality reported in BD patients, with an important driving effect of smoking in this population.

1. Introduction

Bipolar disorder (BD) is a chronic and often severe psychiatric disorder affecting around 1–3% of the population (Carvalho et al. 2020). Multiple studies have provided evidence for accelerated aging mechanisms in patients with BD (Rizzo et al. 2014; Fries et al. 2020), including a steeper age-related decline in executive function and cognitive control (Seelye et al. 2019) and a higher risk of dementia (Diniz et al. 2017). In addition, premature aging is thought to at least partly underlie the higher rates of age-related medical conditions seen in BD patients, including cardiovascular disease, hypertension, metabolic imbalances, cancer, reduced lifespan (Roshanaei-Moghaddam and Katon 2009), age-related changes in physiology (Mutz et al. 2022), and many age-related neuroanatomical alterations (Fries et al. 2020; Ballester et al. 2022).

Different biological clocks have been investigated to explore the molecular basis of accelerated aging in BD, most notably telomere length (Huang et al. 2018), mitochondrial DNA copy number (Spano et al. 2022), and oxidative stress alterations (Jiménez-Fernández et al. 2021). Recent studies have also reported accelerated epigenetic aging in blood and post-mortem brains of patients with BD (Fries et al. 2017; Fries et al. 2020), as well as a deceleration of such mechanisms associated with the use of mood stabilizers (Okazaki et al. 2020). Nevertheless, the clinical implications of these alterations have not been fully explored. In addition, while the first generation of epigenetic clocks were developed to specifically predict chronological age, more sophisticated, second-generation measures have been recently proposed to predict biological aging (Bergsma and Rogaeva 2020), none of which has been investigated in the context of BD. First generation clocks, which were trained on chronological age, include the Horvath’s DNA methylation (DNAm) Age (Horvath 2013) and the Hannum’s DNAm Age (Hannum et al. 2013). Of note, the ratio of the Hannum’s DNAm Age to the chronological age has been termed ‘apparent methylomic aging rate’ (AMAR) (Hannum et al. 2013). More recently, the so-called DNAm PhenoAge (Levine et al. 2018) and DNAm GrimAge (Lu et al. 2019) were developed from whole blood based on phenotypic age-related variables. Specifically, the DNAm GrimAge is based on the combination of DNAm surrogates of seven plasma proteins associated with various age-related conditions and tobacco smoking pack-years, in addition to sex and age (Lu et al. 2019).

DNAm Age acceleration, the residual obtained from fitting the predicted DNAm Age to chronological age, is hypothesized to reflect the cellular aging of a person’s body relative to their chronological age (Horvath 2013; Hannum et al. 2013; Marioni et al. 2015). Such acceleration indices include the intrinsic epigenetic age acceleration (IEAA) derived from the original Horvath DNAm Age (Chen et al. 2016), the intrinsic epigenetic age acceleration (IEAA) and the extrinsic epigenetic age acceleration (EEAA) derived from Hannum’s DNam Age (Chen et al. 2016), the PhenoAge acceleration (PhenoAgeAccel) (Levine et al. 2018), and the GrimAge acceleration (AgeAccelGrim) (Lu et al. 2019). Of these, AgeAccelGrim differs from prior clocks in having demonstrated superior predictive ability for lifespan and all-cause mortality, time-to-death, time-to-coronary heart disease, and time-to-cancer, as well as exhibiting a strong relationship with visceral adiposity/fatty liver, a general medical comorbidity index, and general physical functioning levels (Lu et al. 2019).

Based on evidence of accelerated aging in BD using the first-generation Horvath clock (Fries et al. 2017; Fries et al. 2020), we hypothesize that the newer, more robust epigenetic aging markers focused on biological features of aging would significantly outperform previous markers in the study of aging in BD. To investigate this, we assessed multiple epigenetic clocks in a large sample of BD patients and controls and assessed their relevance in association with multiple clinical and cognitive variables. A better understanding of aging mechanisms and their clinical implications in BD will provide key targets for the development of robust anti-aging approaches that may ultimately prevent premature aging-related clinical conditions in patients.

2. Experimental procedures

2.1. Subjects

One hundred and fifty-three patients with BD and 50 healthy controls were recruited at the UTHealth Center of Excellence on Mood Disorders, Houston, TX, with group-matching for age, sex, and race/ethnicity. BD diagnosis was confirmed through the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I), and a standardized protocol was used for collection of socio-demographic data. Current manic and depressive symptoms were assessed with the Young Mania Rating Scale (YMRS) (Young et al. 1978) and the Montgomery-Asberg Depression Rating Scale (MADRS) (Montgomery and Asberg 1979), respectively. Participants presented with no other medical conditions at the time of enrollment, including neurological disorders and traumatic brain injury, schizophrenia, developmental disorders, eating disorders, and intellectual disability. Controls were excluded if they presented a history of any Axis I disorder in first-degree relatives or if they had taken a prescribed psychotropic medication at any point in their lives. All interviews were administered to participants by trained evaluators and later reviewed by a board-certified psychiatrist. Female participants of reproductive age underwent a urine pregnancy test, and all participants underwent a urine drug screen to exclude illegal drug use. Informed consent was obtained from all participants upon enrollment and prior to any procedure, and the protocol for this study was approved by a local institutional review board.

2.1.1. Functioning and cognitive status

The functioning status of all subjects was assessed by the Global Assessment of Functioning Scale (GAF) (Aas 2011) and the Functioning Assessment Short Test (FAST) (Rosa et al. 2007). Cognitive function from N = 136 patients and all healthy controls was assessed with the Brief Assessment of Cognition in Affective Disorders (BAC-A) (Bauer et al. 2015), a comprehensive cognitive battery specifically developed for BD. Eight cognitive function measures were obtained: short-term affective memory, short-term non-affective memory, verbal fluency, delayed affective memory, delayed non-affective memory, inhibition, problem solving, and token motor speed. Preliminary results for analyses comparing cognitive function between patients and controls have been previously published (Bauer et al. 2015) and were available for integration with the epigenetic clocks in the present study.

2.2. Epigenetic age estimates

Peripheral blood was collected from fasting subjects by venipuncture into EDTA-containing vacutainers, which were immediately processed for the isolation of buffy coat and later stored at −80°C freezers until further analyses. DNA was isolated from buffy coat using the DNeasy Blood & Tissue Mini Kit (Qiagen, Hilden, Germany), according to the manufacturer’s instructions, and quantified with NanoDrop (Thermo, Waltham, MA, USA). Five hundred nanograms of DNA were bisulfite-converted with the EZ DNA Methylation^™ Kit (Zymo Research, Irvine, CA, USA), followed by interrogation of genome-wide DNA methylation levels using the Infinium EPICMethylation BeadChip (Illumina). Poor quality probes were excluded based on detection p-values < .01 using the minfi R package (Aryee et al. 2014). Epigenetic age estimates were calculated using the New DNA Methylation Age Calculator available online (https://dnamage.genetics.ucla.edu/). Specifically, we obtained the following measures for further statistical analyses: Horvath DNAm age (Horvath 2013), Hannum DNAm age (Hannum et al. 2013), PhenoAge (Levine et al. 2018), AMAR (Hannum et al. 2013), and GrimAge (Lu et al. 2019). Measures of aging acceleration were estimated by regressing the predicted epigenetic ages on chronological ages and using the residuals as ‘acceleration indices’ (Horvath IEAA, Hannum IEAA, Hannum EEAA, PhenoAgeAccel, and AgeAccelGrim). We also obtained estimates of blood cell count (CD8+ T-lymphocytes (CD8T), CD4+ T-lymphocytes (CD4T), natural killer (NK), B-lymphocytes (Bcell), granulocytes (Gran), and monocytes (Mono)) based on DNA methylation data using the Houseman method (Houseman et al. 2012).

2.3. Statistical Analyses

Analyses relied on generalized linear modeling (GLM) to model the cross-sectional relationship between multiple epigenetic aging acceleration measures and diagnosis (BD vs. healthy controls). Continuous predictors were z-scored prior to all analyses to provide a common scale for interpreting model effects. The primary set of models in the current workflow evaluated the unadjusted effect of BD on epigenetic aging and epigenetic aging acceleration. Eight measures were evaluated in separate models: (1) Horvath DNAm age, (2) Horvath IEAA, (3) Hannum DNAm age, (4) Hannum EEAA, (5) Hannum IEAA, (6) AMAR, (7) PhenoAgeAccel, and (8) AgeAccelGrim. All but one of these measures were kept in the original metric: AgeAccelGrim was log-transformed to ameliorate potentially biased inferences that would have resulted from violating the normality of residuals. Consequently, model coefficients for the AgeAccelGrim outcome were exponentiated to provide interpretation of predictor effects in terms of the percentage change in the outcome (e.g., a one unit increase in the predictor would be related to a given percentage change in AgeAccelGrim). The log-transformation was determined to be not necessary for the other measures.

Preliminary analyses evaluated the influence of potentially confounding variables including sex, ethnicity, and race; however, none of these met criteria for confounding (i.e., none demonstrated a relationship with both diagnosis and any given epigenetic aging measure). Data collection also gathered information regarding years of education, blood cell counts (CD8T, CD4T, NK, Bcell, Mono, Gran), and mood (total scores of MADRAS and YMRS, and current mood episode); however, given the cross-sectional nature of the data, it is not possible to clearly delineate the direction of influence these variables may have on the relationship between diagnosis and epigenetic aging acceleration. Conceivably, any of these could be influenced by either BD and aging acceleration (or both); as such, these may be better conceptualized as mediators (i.e., implicated in a chain between diagnosis and epigenetic aging acceleration) or colliders (i.e., influenced by both epigenetic aging acceleration and diagnosis). However, in line with evidence of differential DNA methylation in independent blood cell types (Reinius et al. 2012) and recent literature on epigenetic aging (Marioni et al. 2015; Katrinli et al. 2020), analyses also investigated the relationship between epigenetic aging acceleration and diagnosis with adjustment for blood cell counts.

All subsequent analyses examined relationships between AgeAccelGrim specifically and other variables of interest. First, exploratory models examined the potential for interactions between each blood cell count type and diagnosis, adjusted for constituent main effects. Unadjusted p-values were derived for all models in this set (and each following set) of exploratory models. Although these models were considered exploratory and hypothesis-generating, as due diligence, adjusted p-values were also derived via false discovery rate (FDR) correction for Type I error.

Next, a set of 10 exploratory models evaluated potential moderators of the relationship between AgeAccelGrim and BD. For these analyses, AgeAccelGrim was modeled as a function of the interaction between BD and one moderator, controlling for constituent main effects and each of the blood cell type counts. Each model evaluated one variable as a potential moderator, including 2 clinical measures (GAF and FAST total scores) and the 8 cognitive function measures from the BAC-A battery (short-term affective memory; short-term non-affective memory; verbal fluency; delayed affective memory; delayed non-affective memory; inhibition; problem solving; token motor speed).

Finally, subgroup analyses modeled AgeAccelGrim as a function of each of the blood cell type counts within individuals with BD. Then, within those same individuals, a set of 23 exploratory models fit AgeAccelGrim as a function of one cognitive or clinical measure, controlling for each of the blood cell type counts as covariates. Models evaluated 3 mood measures (MADRS total score; YMRS total score; current mood episode), the 8 cognitive measures, and 12 clinical measures (current medication status (any); current lithium use (yes/no), number of total hospitalizations, BD subtype, smoking status, family history of any mental disorder (Köhler-Forsberg et al. 2020), total number of psychiatric comorbidities, lifetime psychotic symptoms (yes/no) (Özyıldırım et al. 2010), current psychotic symptoms (yes/no), length of illness (in years), age of first diagnosis with any mood disorder, and age of diagnosis with BD (Joslyn et al. 2016)).

3. Results

3.1. Descriptive statistics

The present sample (N = 203; BD = 153, controls = 50) was predominantly female (70.9%), with race either African American (34.5%) or Non-Hispanic White (36.9%). The sample had average age M = 36.6 (SD = 11.0) years, with M = 14.3 (SD = 2.4) years of education. Of the individuals with BD, the majority were diagnosed with Type I (N = 132; 86.3%), relative to Type II (N = 21; 13.7%). Patients and controls did not differ for age, sex, race/ethnicity, body mass index (BMI), CD8T, CD4T, Bcell, Mono, or Gran cells; however, smoking status, years of education, and NK cells were significantly different between groups. Complete sample details are provided in Table 1 and blood cell type composition are shown in Table 2. Mood states in patients included euthymia (35.9%), depression (37.9%), mania (15.7%), hypomania (9.1%) and mixed state (1.3%), as determined by MADRS and YMRS scores. Almost all patients were on medication (89.5%) at enrollment (lithium – 21.5%, anticonvulsants – 43.1%, antidepressants – 43.1%, atypical antipsychotics – 52.3%, typical antipsychotics – 3.9%, benzodiazepines – 26.1%, stimulants – 3.3%). Psychiatric comorbidities among patients included generalized anxiety disorder (17.0%), post-traumatic stress disorder (22.8%), social phobia (13.7%), panic disorder (25.5%), agoraphobia (18.3%), bulimia (3.3%), anorexia (1.3%), and binge eating disorder (7.8%). In addition, most of the patients (58.8%) self-reported substance abuse (alcohol – 14.3%, cannabis – 9.1%, cocaine – 7.2%, opiates – 1.9%, other substance – 9.1%) or dependence (alcohol – 34.6%; cannabis – 25.5%; cocaine – 9.1%; opiates – 3.9%, other substance – 15.0%).

Table 1.

Sample demographics

	Bipolar disorder (n=153)	Controls (n=50)	p-value
Age (years), mean (SD)	37.0 (11.2)	35.5 (10.4)	0.404†
Sex (%)
Female	71.9	68.0	0.360‡
Male	28.1	32.0
Race/ethnicity (%)
Non-Hispanic White or Caucasian	40.5	26.0	0.310‡
Hispanic or Latino	15.7	20.0
Black or African American	31.4	44.0
Others	11.7	10.0
Missing	0.65	0
Smoking status (%)
Yes	30.7	4.0	< 0.001 ‡
No	66.0	96.0
Missing	3.2
Education categorical (%)
Elementary school grade (1 to 12)	9.1	2.0	0.010 ‡
High school	19.6	6.0
Part college	37.9	30.0
Graduated college	26.8	54.0
Graduated professional	6.53	8.0
Body Mass Index (%)
Underweight	1.9	0	0.287‡
Normal	22.2	32.0
Overweight	24.2	18.0
Obese	47.7	38.0
Missing	3.92	12.0
Mood state (%)
Euthymia	35.9	NA
Mania	15.7	NA
Hypomania	9.1	NA
Depression	37.9	NA
Mixed	1.3	NA

^†Mann–Whitney test,

^‡Chi-square test. NA - not applicable.

Table 2.

Blood cell type proportions in patients with bipolar disorder (BD) and controls

	Controls	BD	p-value☨	Adjusted p-value*
CD8+ T-lymphocytes	0.991 (0.10)	0.098 (0.09)	0.255	0.306
CD4+ T-lymphocytes	0.148 (0.12)	0.121 (0.10)	0.024	0.072
B-lymphocytes	0.032 (0.05	0.022 (0.04)	0.178	0.267
Natural killer cells	0.028 (0.06)	0.003 (0.02)	0.006	0.036
Monocytes	0.043 (0.05)	0.052 (0.04)	0.570	0.570
Granulocytes	0.562 (0.22)	0.595 (0.18)	0.052	0.104

Values are presented as median (interquartile range).

^☨Mann-Whitney test.

^*Benjamini-Hochberg adjusted p-value.

3.2. Diagnosis as predictor of epigenetic aging estimates

All predicted epigenetic clocks were significantly correlated with chronological age (Figures 1A and S1–3). The primary set of models examined the unique effect of diagnosis on epigenetic aging acceleration. Eight different measures were evaluated in separate models: (1) Horvath DNAm age, (2) Horvath IEAA, (3) Hannum DNAm age, (4) Hannum EEAA, (5) Hannum IEAA, (6) AMAR, (7) PhenoAgeAccel, and (8) AgeAccelGrim. Only AgeAccelGrim demonstrated a relationship with BD in the unadjusted models, finding that individuals with BD demonstrated 21.8% higher AgeAccelGrim (b = 0.197, p = .009), relative to healthy controls (Figure 1B). The same models with adjustment for each of the blood cell counts did not support a relationship between diagnosis and any of the measures, including AgeAccelGrim (b = 0.069, p = .299).

Figure 1. DNAm GrimAge acceleration in bipolar disorder.

A) Scatterplot illustrating the significant and positive correlation between GrimAge (epigenetic age, in years, predicted based on surrogate biomarkers for blood plasma proteins related to morbidity and mortality, cigarette smoking, sex, and age) and chronological age (years). Analysis was performed by Pearson correlation coefficient. B) Higher GrimAge acceleration (AgeAccelGrim) in patients with bipolar disorder (BD). Bars represent mean ± standard error. GrimAge acceleration was calculated by regressing the predicted GrimAge to the chronological age of the subjects and using the residuals as an estimate of the difference between them. Negative and positive values represent younger and older GrimAges compared to their chronological ages, respectively.

3.3. BD, AgeAccelGrim, and blood cell counts

A set of follow-up exploratory analyses modeled AgeAccelGrim as a function of the interaction between diagnosis and one of the blood cell counts, controlling for main effects. These analyses supported potential interactions between diagnosis and CD4T (p = .007), Mono (p = .009), Bcell (p = .010), and Gran (p < .001). Each of these interactions remained significant when adjusting for the other blood cell type counts and after FDR correction for Type I error. Interaction plots for each of these models are provided in Figure 2.

Figure 2. Relationships between blood cell proportions and GrimAge acceleration (AgeAccelGrim) in patients with bipolar disorder (BD) and controls.

Significant negative relationships have been found for CD4+ T-lymphocytes (CD4T), B cells (Bcell), and monocytes (Mono, only in controls), while a positive association was found between AgeAccelGrim and granulocytes (Gran) in both groups.

Analyses then evaluated simple slopes for each significant interaction. The relationship between CD4T and AgeAccelGrim was negative for both groups; however, this was more pronounced for healthy controls (β = −0.323 [−0.418,−0.229]) than BD (β = −0.271 [−0.230,−0.088]). The relationship between Mono and AgeAccelGrim was distinctly negative for healthy controls (β = −0.170 [−0.268,−0.073]) but only marginally negative for BD (β = −0.003 [−0.083, 0.079]). Bcell demonstrated a similar pattern, with a negative relationship for healthy controls (β = −0.230 [−0.335, −0.124]) and no relationship for BD (β = −0.060 [−0.135, 0.015]). Finally, the relationship between Gran and AgeAccelGrim was positive for both groups, but stronger for healthy controls (β = 0.289 [0.192, 0.388]) than BD (β = 0.077 [0.001, 0.152]).

3.4. Interaction models

Group differences for functioning and cognitive variables are presented in Table 3. Patients presented with significantly lower GAF and higher FAST scores than controls (indicative of functioning impairment) and lower scores for short-term affective memory, short-term non-affective memory, and problem solving (after controlling for age). These variables (including each of the cognitive measures, GAF, and FAST total scores) were explored as potential moderators of the relationship between diagnosis and AgeAccelGrim. Each model fit AgeAccelGrim as a function of the interaction between diagnosis and potential moderator, controlling for the set of six blood cell type counts. This set of 10 models did not demonstrate any significant interactions (p > .05) between diagnosis and functioning/cognitive measures.

Table 3.

Functioning and cognitive measures in patients with bipolar disorder (BD) and controls

	Controls	BD	p-valuec	Adjusted p-value*
GAF total scorea	90.0 (7)	55.0 (20)	< 0.001	< 0.001
FAST total scorea	2.0 (7)	31.00 (27)	< 0.001	< 0.001
STAMb	0.116 ± 0.92	−0.274 ± 0.99	0.026	0.052
STNMb	0.241 ± 1.02	−0.237 ± 0.96	0.005	0.016
Fluencyb	0.063 ± 0.91	−0.089 ± 1.04	0.407	0.407
DAMa	0.400 (0.0)	0.400 (0.9)	0.076	0.126
DNMa	0.487 (0.8)	0.487 (0.9)	0.265	0.294
INHIBa	−0.120 (0.6)	0.144 (0.8)	0.213	0.266
PSa	0.541 (0.4)	0.402 (0.6)	0.009	0.022
Token motor speeda	0.012 (0.1)	−0.007 (0.2)	0.085	0.121

Legend: DAM - delayed affective memory; DNM - delayed non affective memory; FAST - Functioning Assessment Short Test; Fluency - verbal fluency; GAF - Global Assessment of Functioning; INHIB - inhibition; PS - problem solving; STAM - short-term affective memory; STNM - short-term non-affective memory. Cognitive scores from the Brief Assessment of Cognition in Affective Disorders (BAC-A) were z-scored for these analyses.

^aMedian (interquartile range).

^bMean ± standard deviation.

^cLinear regression controlling for age.

^*Benjamini-Hochberg adjusted p-value

3.5. Subgroup models

Within individuals with BD, analyses first modeled AgeAccelGrim as a function of all six blood cell types. This model supported relationships between AgeAccelGrim and three of blood cell types: CD4T (β = −0.280, p < .001; −14.4% AgeAccelGrim per sd of CD4T), NK (β = −0.097, p = .045; −9.3%), and Gran (β = −0.246, p = .008; −21.9%). Given previous reports of differential blood levels according to mood states in BD (Fusar-Poli et al. 2021; Kalelioglu et al. 2015), we also explored the association between blood cell type proportions with mood symptoms (MADRS and YMRS total scores) and current mood episodes (euthymia, mania, hypomania, depression, mixed) within patients. As seen in Supplementary Tables S1 and S2, we found no significant associations between mood and blood cell types.

Then, a set of 23 exploratory models evaluated the relationships between AgeAccelGrim and one measure of mood (MADRS; YMRS; current mood episode), cognition (the 8 BAC-A measures), or clinical status (current medication status; current lithium status; SCID total hospitalizations; BD subgroup; family history of BD; total number of comorbidities; lifetime psychotic symptoms; current psychotic symptoms, length of illness; age at onset of any mood disorder; age at onset of bipolar disorder; smoking status), controlling for blood cell types. Of these, 7 predictors demonstrated a significant relationship with AgeAccelGrim: short-term affective memory (β = −0.078, p = .030; −7.6%), short-term non-affective memory (β = −0.088, p = .018; −8.5%), inhibition (β = 0.064, p = .046; +6.6%), problem solving (β = −0.067, p = .034; −6.5%), not currently smoking (b = −0.392, p < .001; −32.5% lower than smokers), age at onset of any mood disorder (β = −0.076, p = .039; −7.3%), and age at onset of bipolar disorder (β = −0.102, p = .016; −9.8%). Associations from selected models are highlighted in Figures 3 and 4. No other significant associations were found, including with acute mood symptoms or with current mood episodes (p > .05 for all). FDR adjustment for Type I error across the entire subset of 23 models found that only the effect of not currently smoking retained statistical significance (p < .001).

Figure 3. Association between AgeAccelGrim and cognitive status within patients with bipolar disorder.

A) Short-term affective memory (STAM; β = −0.078, p = 0.030); B) Short-term non-affective memory (STNM; β = −0.088, p = 0.018); C) Inhibition (INHIB; β = 0.064, p = 0.046); D) Problem solving (PS; β = −0.067, p = 0.034).

Figure 4. GrimAge acceleration (AgeAccelGrim) in patients with bipolar disorder according to their current smoking status.

The analysis included N = 101 non-smokers and N = 47 smokers (unknown, N = 5). Columns represent mean + standard error. β = −0.392, p < 0.001 (controlled for blood cell proportions).

A similar set of analyses was then performed within controls. The first model fitting AgeAccelGrim as a function of all six blood cell types found support for the effects of CD4T (β = −0.627, p < .001, −46.6%), NK (β = −0.235, p = .002, −29.1%), Mono (β = −.0205, p = .011, −18.5%), Gran (β = −0.808, p = .022, −55.4%), and Bcell (β = −0.201, p = .023, −18.2%). A set of 11 exploratory follow-up models within controls (the same model set for the BD subgroup, controlling for blood cell types, without models for the clinical measures) only supported a significant effect for problem solving (β = −0.183, p = .037, −16.8%). This effect was not significant after FDR adjustment for Type I error across the set of 11 models.

4. Discussion

The goal of this study was to investigate multiple measures of epigenetic aging in BD and explore their clinical and cognitive correlates in a large sample of patients. Our main findings indicate that: (i) BD was significantly associated with higher (21.8%) AgeAccelGrim compared to controls, but not with other epigenetic aging markers investigated; (ii) differential levels of blood leukocytes (CD4+ T-lymphocytes, monocytes, B-cells, and granulocytes) significantly impact AgeAccelGrim, with distinct patterns between patients and controls; and (iii) higher AgeAccelGrim was significantly associated with cognitive dysfunction and current smoking status within BD patients.

To our knowledge, this is the first study reporting AgeAccelGrim in BD and its association with clinical and cognitive outcomes. Of note, AgeAccelGrim has been shown to outperform the first generation of epigenetic age estimators in predicting all-cause mortality and time to onset of several serious illnesses (Lu et al. 2019; McCrory et al. 2021). In addition, epigenetic aging has been previously associated with many other conditions, including exposure to violence in childhood (Jovanovic et al. 2017), summative lifetime stress (Zannas et al. 2015), completed suicide (Okazaki et al. 2020), and all-cause mortality (Lu et al. 2019). Our results of higher AgeAccelGrim are in line with previous studies that strongly suggest BD as a disease of accelerated aging (Rizzo et al. 2014; Fries et al. 2020) with excess medical morbidity, including evidence of premature mortality (Hayes et al. 2015; Crump et al. 2013; Hoang et al. 2013). More specifically, we found that smoking status is a major contributor to the accelerated GrimAge in patients, which is supported by previous studies showing that cigarette smoking significantly accelerates epigenetic aging (including GrimAge) in blood (Cardenas et al. 2022) and respiratory organs (Wu et al. 2019). Accordingly, multiple studies have reported cigarette smoking-induced alterations in DNA methylation levels (Joehanes et al. 2016) resulting from increased DNA methyltransferase 1 (DNMT1) expression (Kwon et al. 2007). Smokers are known to have a shorter life expectancy than non-smokers in the general population (Mamun et al. 2004; Sakata et al. 2012) and in patients with BD (Chesney et al. 2021), and our results further support these previous findings with a DNA methylation-based biomarker of lifespan.

Our study also found interesting associations between AgeAccelGrim with blood cell proportions, specifically negative associations with CD4T, monocytes, and B cells (the latter only in controls), and a positive association with granulocyte levels (in both groups). Similar GrimAge associations with CD4T, B cells, and granulocytes have been previously reported in a sample of trauma-exposed subjects (Katrinli et al. 2020), although monocytes were positively associated in that population (in contrast to our findings in BD and controls). This is also in accordance with evidence of differential cell aging and lifespan for individual types of leukocytes (Spyridopoulos et al. 2008). Granulocytes, monocytes, and lymphocytes are all important players in the inflammatory response, with evidence of an important increase in low-grade inflammation with the normal aging process (Ferrucci and Fabbri 2018). Accordingly, differences in the levels of specific leukocytes and inflammatory mediators have been repeatedly reported in BD (Giynas Ayhan et al. 2017; Barbosa et al. 2014; Melo et al. 2019) and with acute mood episodes (Fusar-Poli et al. 2021). Although we found no significant associations between mood symptoms or states with blood cell type proportions in our sample, our results indicate that the relationship between AgeAccelGrim and blood cell proportions (CD4T, B cells, monocytes, and granulocytes) is stronger in healthy controls than in patients, suggesting a BD-related disruption in the normal association between aging and inflammation.

Within patients, AgeAccelGrim was significantly associated with worse short-term affective memory, short-term non-affective memory, problem solving abilities, higher inhibition, and age at diagnosis. Accordingly, a steeper age-related decline in cognitive functioning has been previously found in BD (Seelye et al. 2019; Chen et al. 2021), although this has not been replicated across multiple cohorts (Schouws et al. 2016; Bora and Özerdem 2017). In fact, a recent study found that the number of previous episodes with psychotic features is a significant risk factor for cognitive decline in BD patients, suggesting an important heterogeneity among patients for longitudinal age-related cognitive decline (Chen et al. 2021). A previous study from our group also found a greater accelerated epigenetic aging in BD patients’ hippocampus compared to healthy controls (Fries et al. 2020), which may suggest that BD patients are more likely to experience the cognitive effects of accelerated aging than healthy individuals. Finally, the identified negative association between AgeAccelGrim and age of first diagnosis with any mood disorder or BD suggests that earlier onset of illness, which has been previously suggested to predict higher severity and worse prognosis (Verma et al. 2021; Joslyn et al. 2016), may do so along with alterations in aging mechanisms.

Importantly, the notion of BD as a condition of accelerated aging is also supported by evidence suggesting that some medications used to treat BD may be protective against accelerated aging effects. For instance, long-term treatment with lithium, a first-line mood stabilizer in BD, has been suggested to exert a protective influence against telomere length shortening in leukocytes from BD patients (Pisanu et al. 2020; Coutts et al. 2019). Lithium’s ability to elongate telomere length has also been reported in vitro (Fries et al. 2020) and has been suggested to involve an upregulation of telomerase reverse transcriptase (Lundberg et al. 2020; Squassina et al. 2016). Additionally, mood stabilizers have been previously associated with a deceleration of epigenetic aging in BD patients (Okazaki et al. 2020). Of note, the current study did not find any significant association between AgeAccelGrim and medication status (or lithium use, specifically) within patients. However, almost all patients enrolled in this study were on medication, which likely precluded us from having statistical power for this specific comparison.

Although we were able to explore the association between AgeAccelGrim with multiple clinical variables indirectly linked to clinical severity and progression, limitations of our study include a lack of data on the number of previous episodes for all patients. In addition, participants in this study were younger than those in the Framingham Heart Study cohort (upon which the AgeAccelGrim was trained), and the specific relationship between GrimAge in young and middle-aged populations and age-related morbidity and mortality has yet to be determined. Nonetheless, this does not change the relative difference detected in AgeAccelGrim between our two age-matched groups. Finally, as previously mentioned, since most patients were on medication, our assessment of the effects of medication use on aging may not have been powerful enough, thus requiring future replication using an even larger sample of patients with and without medication use.

Overall, accelerated epigenetic aging, as measured by AgeAccelGrim, was increased in BD and significantly associated with smoking, blood cell proportions, and cognitive dysfunction. However, the exact biological mechanisms underlying the association between accelerated GrimAge and cognitive decline in patients still warrant further research. The anti-epigenetic aging effects of medications and other interventions to achieve functional recovery should also be explored, particularly focusing on residual symptoms, comorbid conditions, and neurocognitive deficits. Moreover, given the high prevalence of smoking in BD (Thomson et al. 2015), its association with worse cognitive functioning in BD (Depp et al. 2015), and its reported effects on epigenetic aging and lifespan in patients, smoking cessation should be viewed as a key goal in the management of BD.

Highlights.

• Bipolar disorder is associated with increased GrimAge acceleration

• GrimAge acceleration is associated with different blood cell proportions

• Higher GrimAge acceleration is associated with cognitive impairment in patients

• Smoking is associated with GrimAge acceleration in patients with bipolar disorder