Major depressive disorder and serum inflammatory biomarkers as predictors of reward processing dysfunction in an American Indian sample

Abstract

Introduction

American Indians (AI) experience chronic stressors which may be associated with disproportionate prevalence rates of major depressive disorder (MDD). Stress affects mental health through increased inflammatory processes and has been associated with increased risk of MDD and disruptions to reward processing. This study investigated the role of inflammation in reward processing disruptions among AI individuals with lifetime MDD; a population at heightened risk due to chronic stressors.

Method

Participants (N=73) completed a monetary incentive delay task during simultaneous electroencephalography (EEG) and functional magnetic resonance imaging (fMRI). Blood samples were analyzed for pro-inflammatory (tumor necrosis factor [TNF], interleukin-6 [IL-6], C-reactive protein [CRP]) and anti-inflammatory (interleukin-10 [IL-10]) biomarkers. Depression severity was assessed using the Patient-Reported Outcomes Measurement Information System (PROMIS) Depression scale. Covariates were included and assessed from self-report measures.

Results

Regression analyses revealed that elevated TNF concentrations and sex were associated with reduced activation across subregions of the basal ganglia during gain anticipation. Similarly, TNF and CRP concentrations, as well as medication, were associated with reduced activation within basal ganglia subregions across loss anticipation. IL-10, IL-6, and P300 showed limited predictive value for neural responses.

Conclusions

These findings suggest that inflammation may contribute to reward processing disruptions by impairing striatal function amongst a sample with lifetime MDD. The observed associations underscore the importance of inflammation’s potential role and association in the pathophysiology of MDD, particularly in contexts of chronic stress. This study highlights the need for addressing the disproportionate mental health burden among AI communities through a biopsychosocial approach.

Introduction:

Previous research has demonstrated an association between heightened inflammation, stress related events, and occurrence of major depressive disorder (MDD) [1], [2]. Amongst American Indian (AI)¹ populations there are chronic stressors rooted in historical traumas that disproportionately contribute to health disparities [3]. Symptoms of historical loss and more frequent thoughts about historical trauma amongst AI populations relate to disproportionate risk for depression and cardiovascular disease [4],[5]. Literature has also shown an association between adverse childhood events (ACEs) and inflammation, with levels of social connectedness to larger community moderating this relationship [6]. Specifically, individuals with high levels of ACEs who reported low levels of connection to the larger community had increased concentrations of interleukin-6 (IL-6) and C-reactive protein (CRP), both biomarkers of persistent low-grade physical inflammation [6]. Thus, AI individuals with repetitive and/or chronic stressors may experience compounded disruptions with inflammatory responses resulting in higher levels of pro-inflammatory biomarkers and associated deleterious effects on mental health.

Distinguishing possible diagnostic mechanisms is an important step in improving our understanding, prevention, and treatment efforts of MDD. Connections between peripheral inflammatory biomarkers, MDD, and disruptions to reward anticipation/consumption are well established throughout inflammatory literature [7], [8], [9]. For example, tumor necrosis factor (TNF)— a pro-inflammatory cytokine— is a mediator within the immune system and has been observed to be elevated within the context of MDD and associated with detrimental effects on neurocircuitry functions [10], [11], [12]. IL-6 is another pro-inflammatory cytokine that mediates the immune response, with higher concentrations contributing to adverse health outcomes and elevated concentrations reported in the context of MDD [13], [14]. Similarly, CRP, an acute phase protein, is secreted in response to the upregulation of inflammatory biomarkers and associated with elevated concentrations within subsamples of MDD [15], [16]. In contrast, cytokines with anti-inflammatory properties, such as IL-10 –responsible for tempering the immune response – has been measured in lower concentrations in MDD relative to healthy controls (HC) [17], [18]. Despite these associations, further work is needed to establish the directional relationship between elevated inflammation, MDD, and reward circuitry disruptions. Prior studies exploring inflammation as a pathophysiological pathway for reward circuity alterations found that following administration of exogenous inflammatory cytokines there were alterations in neural activity in brain regions associated with reward, with some specifically noting reductions in areas such as the ventral striatum [19], [20], [21], [22], [23]. Additionally, a study found decreased connectivity between the ventral striatum and ventral medial prefrontal cortex to be correlated with anhedonia following administration of inflammatory stimuli [22]. Specifically, CRPs effect on corticostriatal connectivity mediated the relationship between CRP and anhedonic behavior [22]. Together, this suggests dysregulation of pro and anti-inflammatory molecules are promising biological markers of MDD and its related symptoms [17], [24].

Functional magnetic resonance imaging (fMRI) facilitates the characterization of the pathophysiology of depression and related disruptions of mesocortico-limbic reward circuity [7], [25]. Key regions, like the striatum has demonstrated alterations in blood oxygenated-level-dependent (BOLD) signal activation and connectivity when processing emotional and/or rewarding stimuli within the context of MDD symptoms [26]. Regions such as the putamen (PUT) and caudate (Ca) have been associated with incentive-driven behaviors, in which activations in these brain regions are related to responses to monetary rewards and punishments during a monetary incentive delay (MID) task [22]. Activation of these key striatal regions reflected by fMRI are associated with reward processing at the anticipatory stage, in which the they have been shown to exhibit hypoactivity to reward anticipation, suggesting decreased motivation and incentives to attempt obtaining a reward [27], [28]. Similarly, blunted striatal activation and elevated cytokine concentrations are often found within the context of MDD [22], [29], [30], [31]. Deactivation within the nucleus accumbens (NAc)—a region associated with prediction detection error for incentive learning and motivated behavior—has been associated with elevated levels of IL-6 in response to stress-induced changes [9], [32]. In response to rewarding stimuli, studies have shown the caudate to be sensitive to reward anticipation and have a negative association with CRP levels in the context of depressive symptoms [22], [33]. Thus, fMRI techniques hold strong promise in delineating functional neural circuitry associated with inflammation and depression. Furthermore, the combination of electroencephalography (EEG), measured simultaneously with MRI, would enable a more precise characterization of the specific stages of cognitive processes impacted due to the complementary spatiotemporal strengths of concurrent EEG and fMRI recording.

The strong temporal resolution from EEG allows for delineation of specific stages of reward anticipation versus reward consumption, each of which have been associated with varying neurobiological mechanisms [34]. Specific event related potential (ERP) components, such as the P300, index attentional allocation in response to internal or external stimuli [35], updating in working memory [36], and is sensitive to salient stimuli and task difficulty [37]. Prior literature demonstrates an attenuated P300 amplitude and prolonged latency in individuals with MDD [38]. For example, a study found that participants with MDD exhibited a blunted P300 amplitude when compared to HC; furthermore, among individuals with MDD who enrolled in treatment, those who went on to complete treatment displayed larger P300 amplitudes at baseline relative to those who went on to drop out [39]. From the subsample of people who completed the treatment (n=39) there was a total of n=19 who responded to treatment; however, there were no significant differences in P300 amplitude across non-responders and responders who completed the treatment type [39]. Taken together, the literature suggest differences in reward anticipation of individuals with MDD compared to HC within EEG/ERP studies [40], [41]. While a study has explored spirituality as a protective factor with AI Peoples with generalized anxiety disorders and its relation to P300 amplitude [42], no ERP studies, to our knowledge, have examined the effects of inflammation with AI Peoples with MDD, and exclusively on P300 amplitude [43].

In a sample of AI adults with MDD, we investigated the effects of pro-inflammatory (TNF, IL-6, CRP) and anti-inflammatory (IL-10) biomarker concentrations and depression symptom severity on brain responses to monetary wins and losses within the MID task during simultaneous EEG-fMRI recording. We hypothesized that American Indians with lifetime MDD, elevated pro-inflammatory and attenuated anti-inflammatory biomarkers may experience disruptions in win/loss anticipation while performing an MID task, specifically, through reduced fMRI percent signal change (PSC) in the BOLD signal activation within the regions of interest of the striatum and exhibiting a blunted P300 amplitude.

Methods:

1. Participants

A sample of n=73 self-identifying AI participants met diagnostic criteria for lifetime MDD, had quality EEG (n=63; n=54 met a score ≥ 52.5 for probable current depression) and fMRI data (n=73; n=60 met a score ≥ 52.5 for probable current depression), and quantified inflammatory biomarkers were drawn from baseline assessment of two parent studies: the Tulsa 1000 (T-1000) study and the Neuroscience-Based Mental Health Assessment and Prediction (NeuroMAP) Center of Biomedical Research Excellence (CoBRE) Core project. See [44], [45] for more information on the study design and protocol. See Table 1–2 for demographics.

Table 1:

Descriptive table of final sample (n=73) for fMRI analysis.

Variables	M	SD	SE	Variance	Frequency	Percent
PROMIS	59.73	7.47	0.87	55.74
Log TNF	0.03	0.30	0.04	0.05
Log IL-6	−0.29	0.30	0.04	0.09
Log CRP	6.22	0.55	0.07	0.31
Log IL-10	−0.60	0.30	0.03	0.09
Age	34.67	10.39	1.22	107.95
BMI	29.03	5.11	0.60	26.09
Female					48	65.73
Male					25	34.23
Medicated					42	57.53
Unmedicated					31	42.47
Comorbid					61	83.56
Not Comorbid					12	16.44

Note: PROMIS= Patient-Reported Outcomes Measurement Information System, TNF= Tumor Necrosis Factor, IL-6= Interleukin-6, CRP= C-Reactive Protein, IL-10= Interleukin-10. All inflammatory biomarkers are log transformed.

Table 2:

Descriptive table of final sample (n=63) for P300 analysis.

Variables	M	SD	SE	Variance	Frequency	Percent
PROMIS	60.85	7.32	0.04	53.55
Log TNF	0.26	0.022	0.03	0.05
Log IL-6	−0.28	0.28	0.04	0.08
Log CRP	6.23	0.55	0.07	0.31
Log IL-10	−0.62	0.29	0.04	0.08
Age	34.27	10.64	1.34	112.3
BMI	29.15	4.85	0.61	23.5
Female					41	65.08
Male					22	34.92
Medicated					35	55.56
Unmedicated					28	44.44
Comorbid					54	85.71
Not comorbid					9	14.28

Note: PROMIS= Patient-Reported Outcomes Measurement Information System, TNF= Tumor Necrosis Factor, IL-6= Interleukin-6, CRP= C-Reactive Protein, IL-10= Interleukin-10. All inflammatory biomarkers are log transformed.

1. Measures

1.1. MID Task

Concurrent EEG-fMRI measurement data were recorded as participants completed an MID task. Trials (90 total; duration=562 seconds/trial) were split into two runs with 6 conditions each (n=15 trials per condition). A variable inter trial interval with a mean of 4s that ranged from 2–6s was used [46]. Each trial began with a fixation circle for 2–6s prior to receiving a cue for potential gain (circle), loss (square), or no gain (circle or square). Each cue had a horizontal line that indicated the degree of gain/loss: top of the cue +/$5, middle of the cue +/−$1, and bottom of the cue +/−$0. A 2s fixation circle was presented between the cue and target, and participants pressed a button within a specific time window. Trials were calibrated based on the subject’s reaction time prior to the scan and adjusted for a success rate of two-thirds of trials, following the onset of a white triangle. Once the response was made, a 2s blank screen appeared prior to feedback presentation (duration: 2s) indicating results (+/−$5,+/−$1,+/− $0). Trials were pseudo-randomized, and the order of runs was kept consistent for all participants. See Figure 1 for the MID task.

Figure 1.

Overview of the MID task performed while simultaneous EEG-fMRI data was collected. Sample trial sequence of gain and loss (top) and type of trial cues (bottom).

1.2. Patient-Reported Outcomes Measurement Information System (PROMIS) Depression Scale

The PROMIS Depression scale v1.0 was utilized to measure depression symptomology (e.g., sadness), such as negative mood and disruptions in decision making to assess perceived changes in symptoms through self-report [47]. Higher PROMIS Depression scores indicate greater depressive and negative symptoms associated with MDD. Cutoff scores were utilized, with a cutoff score of 52.5 indicating current mild depressive symptoms, a score of 57.5 indicating moderate symptoms, and a score of 72.5 indicating severe depressive symptoms [48], [49].

1.3. Blood Collection: Inflammatory Biomarkers

Venous blood was collected via BD Vacutainer serum collection tubes by a trained phlebotomist. The serum was isolated by centrifuging at 1,300g for 10 minutes at room temperature, and then aliquoted and stored at −80°C until biomarkers analysis. IL-6, TNF, CRP, and IL-10 from baseline samples were measured with the Neuroinflammation Panel 1 Human Kit (Meso Scale Diagnostics, Maryland, USA) and processed using the Meso Quickplex SQ120 instrument (MesoScale Diagnostics, Maryland, USA) [29], [50]. All samples were run in duplicate.

1.4. Covariates: Medication Status, Age, Sex, Comorbidities, and Body Mass Index (BMI)

Medication status was defined as having taken any psychotropic medications in the past 6 weeks [29]. Medication status was collected through an interview of medication and medical history by a trained masters or nurse level assistant while supervised by a clinical psychologist, see Supplemental material Table S5 for the medication breakdown [44], [45]. Age and sex were collected through a self-report demographics and psychosocial form, with participants 18-years old or older being eligible to participate. Lastly, BMI was collected during the biomarker collection session, in which all physical measurements were collected [44], [45].

Comorbidity status was defined as meeting any additional criteria for a DSM-V disorder along with lifetime MDD. A Mini International Neuropsychiatric Interview (MINI V.6.0) was conducted by study personnel to determine if participants meet criteria of DSM-V disorders. All participants who met criteria for psychotic disorders, bipolar and related disorders, obsessive-compulsive disorder, or neurocognitive disorders were ineligible [44], [45]. See Supplemental Material Table S6–S7 for the comorbidity’s breakdown.

1.5. Neuroimaging data collection

1.5.1. fMRI

fMRI data were acquired with two identical GE MR750 3T scanners (axial slices, TR/TE=2000/27ms, flip angle=78°, sampling bandwidth=250 kHz, FOV/slice thickness=240/2.9mm, 128×128 matrix producing 1.875×1.875×2.9mm voxels, 256 volumes, R/L frequency encoding direction, sensitivity encoding (SENSE) acceleration factor=2) [51]. T-1000 data used a 3D axial T1-weighted magnetization-prepared rapid acquisition with gradient echo sequence to obtain high resolution structural images (TR/TE=5/2.012ms, FOV/slice thickness= 240×192/0.9mm, flip angle=8°, sampling bandwidth=31.25 kHz, 256×256 matrix producing 0.938×0.938×0.9mm voxels, 186 axial slices, SENSE acceleration factor=2) [29], [51]. CoBRE-CORE data used a 3D axial T1-weighted magnetization-prepared rapid acquisition with gradient echo sequence to obtain high resolution structural images (TR/TE=6/2.92ms, FOV/slice thickness=256×256/1.0mm, flip angle=8°, sampling bandwidth=31.25 kHz, 256×256 matrix producing 1mm isotropic voxels, 208 sagittal slices, SENSE acceleration factor=2) [45].

1.5.2. EEG

EEG data were concurrently collected with fMRI data, utilizing a 31-electrode cap, attached to an MRI-compatible BrainAmp MR Plus amplifier [44]. Attached to the subject’s back was an additional electrode to monitor electrocardiographic (ECG) signal. All electrodes were referenced to the FCz position, with a ground electrode at the AFz position [44].

1.6. Data processing

1.6.1. fMRI

fMRI processing was completed using the Analysis of Functional Neuroimaging software package [52]. Standard preprocessing for fMRI included slice timing correction, signal scaling, spatial smoothing, and motion correction according to the recommendations of [53], [54]. Data were registered to MNI space via affine transformation and followed by general linear modeling using 4-second block regressors convolved with a canonical hemodynamic response function for each of the six anticipatory task conditions. Volumes with excessive motion (Euclidean norm of the derivatives of the six motion parameters) or greater than 10% outlier fraction were censored at the regression step. Regions of interest (ROI) included the following bilaterally: NAc (ROIs:223, 224), Globus Pallidus (GP; [ROIs: 21, 222]), Ca (ventral [ROIs:219, 220] and dorsal [ROIs:27, 228]), and ventral medial (ROIs:225, 226) and dorsal lateral (ROIs:229, 230) PUT (ventral medial [ROIs: 225, 226] and dorsal lateral [ROIs:229, 230]) [25], [29], [33]. ROIs were defined by the Brainnetome Atlas, in which allows the correlation of brain anatomy and psychological/cognitive functions [55].Striatal activity was indexed by PSC in BOLD response to the MID trial cue phase across gain and loss conditions.

1.6.2. EEG

EEG data were processed using Brain Vision Analyzer (BVA) 2.1 (Brain Products GmbH, Munich, Germany). Processing of EEG data included: re-referencing offline to linked mastoids, bandpass filtered with a 2^nd-order Butterworth bandpass filter to 0.01–30 Hz half amplitude cut off and down sampled to 250 Hz. One Hz notch filters were applied to remove fMRI gradient artifacts at slice timing (19.5 Hz and harmonics) and 60 Hz line noise. Ballistocardiogram (BCG) artifacts were reduced from the ECG channel using the BVA software semi-automated average artifact subtraction routine to detect R-peaks using that channel and create a BCG correction template [56]. The BCG residual artifacts and ocular artifacts were corrected by independent components analysis utilizing an extended-Infomax algorithm [57]. Artifact rejection routines were done to exclude bad segments with >50 μV change between 2 data points and absolute fluctuations >200 μV within any 200 ms interval or changes < .05μV within a 100 ms period [58].

The grand average ERP waveforms were stimulus locked to the cue onset for all participants across the gain (+$1,+$5 trials) and loss (-$1,-$5) trials. Measurement window for P300 were 300ms-500ms at the POz electrode site [59].

2. Statistical Analysis

Using R Software (Version 4.4.2), a Kaiser-Meyer-Olkin test was conducted to explore the data fit for an Exploratory Factor Analysis. A parallel analysis was then used to obtain the number of factors for the data analysis. Lastly, a feasible solutions algorithm (FSA) was conducted to obtain the optimal model with the lowest Bayesian Information Criterion of all covariates and their 2-way interaction to conduct linear regression models. Each model investigated inflammatory biomarker’s concentrations (TNF, IL-6, CRP, IL-10), PROMIS depression scores (PROMISDep), and relevant confounders (BMI, age, sex, comorbidity, and medication status) as potential predictors of fMRI PSC in the BOLD signal activation in the bilateral ROIs for gain (+$1,+$5) and loss (-$1,-$5) conditions independently. Similar models were computed to explore the predictability of inflammation, PROMISDep, and relevant confounders (BMI, age, sex, comorbidity, and medication status) on P300 amplitude across all conditions (gain and losses). Because of the non-significant results, all P300 analysis can be found in the Supplemental Material, see Table S1. Inflammatory biomarkers failed the normality of distribution assumption; therefore, they were all log-transformed to adjust normality. Additionally, an exploratory analysis was conducted to investigate potential differences across lateralized regions of the Ca, PUT, NAc, and GP and P300 amplitude across the midline, see Supplemental Materials Tables S2–S4.

Results:

1. Neuroimaging results

An FSA was conducted to obtain the optimal models and their 2-way interaction from the covariates (inflammatory biomarkers [TNF, IL-6, CRP, or IL-10], PROMISDep, age, BMI, sex, comorbidity, and medication status). Linear regression models were then employed to test the optimal model and their 2-way interaction to predict fMRI BOLD signal PSC throughout the basal ganglia to wins and losses. See Table 3 for fMRI results.

Table 3.

fMRI Model Results

Condition	ROI	R 2	F	p
					β	t	p
Gain	Ca	0.33	12.91	<.001
	Log TNF				−0.31	−5.23	<.001
	Sex (Male)				−0.18	−3.66	<.001
	LogTNF*Sex (Male)				0.39	2.64	.010
	NAc	0.29	11.15	< .001
	Log TNF				−0.35	−4.87	<.001
	Sex (Male)				−0.22	−3.71	<.001
	Log TNF*Sex (Male)				0.511	2.87	.005
	GP	0.33	13.07	<.001
	Log TNF				−0.25	−5.33	<.001
	Sex (Male)				−0.14	−3.57	<.001
	Log TNF*Sex (Male)				0.30	2.59	.012
	PUT	−0.22	−4.66	<.001
	Log TNF				−0.22	−4.66	<.001
	Sex (Male)				−0.15	−3.97	<.001
	Log TNF*Sex (Male)				0.34	3.03	.003
Loss	Ca	0.13	2.49	0.03
	Log CRP				0.06	2.36	.021
	Log TNF*Log CRP				−0.17	−2.02	0.031
	CRP Low				−0.18	−3.01	<.001
	CRP High				−0.38	−5.32	<.001
	NAc	0.41	17.82	<.001
	Log TNF				−0.32	−6.26	<.001
	Medication (Yes)				−0.09	−3.75	<.001
	Log TNF*Medication (Yes)				0.23	3.13	.002
	GP	0.28	10.77	<.001
	Lof TNF				0.05	2.59	.011
	Log TNF*Log CRP				−0.15	−2.48	.015
	CRP Low				−0.11	−2.49	<.001
	CRP High				−0.28	−5.24	<.001
	PUT	0.30	11.76	<.001
	Log TNF				0.86	2.12	.037
	Log TNF				0.06	2.83	.005
	Log TNF*Log CRP				−0.17	−2.63	.010
	CRP Low				−0.12	−2.55	.010
	CRP High				−0.32	5.48	<.001

Note: Ca= Caudate, NAc= Nucleus Accumbens, GP= Globus Pallidus, PUT= Putamen, CRP Low= Simple slopes at – 1 SD from CRP mean, CRP High= Simple slope at +1 SD from CRP mean.

1.1. MID Gain Condition Results

FSA showed TNF concentration, sex, and their interaction to be the optimal model predicting activity within the subregions of the basal ganglia (Ca, NAc, GP, & PUT) across the gain condition. The linear regression model significantly varied in BOLD signal activation, F(3,69)=11.12–13.07, p’s<.001, R²=0.297–0.335. Simple effects displayed that TNF concentrations (β=−0.221 to −0.358, t(69)=−4.669 to −5.339, p’s<.001) and sex (Male; β=−0.143 to −0.227, t(69)=−3.573 to −3.714, p’s<.001) predicted and were negatively associated with BOLD signal activation across the subregions of the basal ganglia. Additionally, their interaction between TNF concentration*Sex(Male; β=0.302–0.511, t(69)=2.596–3.032, p’s=.005-.032) predicting BOLD signal activation across the subregions of the basal ganglia. See Figure 2 for all subregion results across gain conditions.

Figure 2.

Interaction of log transformed TNF and Sex on fMRI percent signal BOLD signal activation within the subregions of the basal ganglia (top right [Ca], top left [NAc], bottom right [GP], bottom right [PUT]) across MID gain conditions, separated by sex (Red= female; Blue= male).

Note. Ca= Caudate, NAc= Nucleus Accumbens, GP= Globus Pallidus, PUT= Putamen.

1.2. MID Loss Condition Results

1.2.1. Loss Caudate

FSA showed TNF concentration, CRP concentration, and their interaction to be the optimal model predicting activity within the Ca, GP, and PUT across the loss condition. The linear regression model significantly varied in BOLD signal activation, F(3,69)=10.77–11.89, p’s<.001, R²=0.289–0.312. Main effects within the Ca and GP displayed that CRP concentrations (β=0.069, 0.058, t(69)=2.361, 2.598, p=.021, .011) and the interaction between CRP concentration*TNF concentration (β=−0.179, −0.154, t(69)=−2.202, −2.484, p=.031, .015) predicted BOLD signal activation across the Ca and GP, respectively. Main effects within the PUT displayed that CRP concentrations (β= 0.068, t(69)=2.838 p=.006) and TNF concentrations (β= 0.861, t(69)=2.125 p=.0371) and the interaction between CRP concentration*TNF concentration (β=−0.174, t(69)=−2.635 p=.010) predicted BOLD signal activation across the PUT. Simple slopes analyses indicted that at low levels (β=−0.11 to −0.18, t(69)=−2.49 to −3.01 p’s= .001-.002) and high levels (β=−0.28 to −0.38, t(69)=−5.32 to −5.66, p’s< .001) of CRP concentrations there was a negative relation between TNF concentrations and Ca, GP, and PUT activity, respectively. See Figure 3.

Figure 3.

Interaction of TNF concentration and CRP concentration on (AREAS) across MID loss conditions, split by low (−1 SD from mean) and high (+1 SD of mean) CRP concentrations.

Note. Ca= Caudate, GP= Globus Pallidus, PUT= Putamen. High CRP concentration= + 1 SD of log CRP mean, Low CRP concentration= −1 SD of log CRP mean.

Additionally, FSA showed TNF concentration, medication status, and their interaction to be the optimal model predicting NAc activity across the loss condition. The linear regression model significantly varied in BOLD signal activation, F(3,69)=17.82, p<.001, R²=0.412. Simple effects displayed that TNF concentrations (β=−0.328, (69)t=−6.260 p<.001) and medicated (Yes; β=−0.010, t(69)=−3.753 p<.001) predicted and were negatively associated with BOLD signal activation across the NAc. Additionally, their interaction between TNF concentration*Medicated (Yes; β=0.240, t(69)=3.136 p=.003) predicted BOLD signal activation across the NAc. See Figure 4.

Figure 4.

Interaction of log-transformed TNF concentration and medication on fMRI percent signal BOLD signal activation within the NAc across MID loss conditions, separated by medication status (Red= Medicated; Blue= Non-medicated).

Note. NAc= Nucleus Accumbens.

Discussion:

In this study, we aimed to explore pro (TNF, IL-6, CRP) and anti-inflammatory biomarkers (IL-10), PROMISDep, and their interaction on brain responses [BOLD signal activity (activation within the basal ganglia) and ERP amplitude (P300)] amongst an AI sample. Results displayed that TNF concentration, Sex(Male) and their interaction were negatively associated with activation within regions of the Ca, NAc, GP, and PUT across gain conditions. Similarly, TNF concentration, medication status(Yes), and their interaction were negatively associated with activation of the NAc across loss conditions. Lastly, at low and high levels of CRP concentrations, TNF concentration was negatively associated within the Ca, GP, and PUT across loss conditions. The current study did not show covariates significantly predicting P300 amplitude across all conditions. Together, this suggests that inflammation is associated with disruptions to reward anticipation and may play a role in the pathophysiology of MDD. The current results underscore inflammation as a physiological component in subsamples of MDD and its related cognitive disruptions with future work needed to delineate and contextualize findings, especially within vulnerable communities with disproportionate burdens of chronic stressors.

Results were consistent with literature with elevated pro-inflammatory concentrations relating to disruptions in striatum activity in response to anticipation of a monetary reward [21]. However, within the current study, the hypothesis was partially supported, with results showing pro-inflammatory biomarkers (TNF, CRP), sex, and medication status being associated with a blunted fMRI BOLD PSC activation within areas of the Ca, NAc, GP, & PUT to the anticipation of a monetary win and loss, but not IL-10 which has anti-inflammatory properties. This suggests that when elevated pro-inflammatory biomarkers are circulating there may be disruptions to not just reward gains, but reward losses as well. Mounting evidence suggests the potential role of TNF and CRP on the pathophysiology of MDD and related disruptions to the reward circuitry [60], consistent with the current findings. Furthermore, previous neuroimaging findings have suggested the possible adverse effect of pro-inflammatory cytokines on dopamine function in the ventral striatal regions—key structures of motivation regulation—in which motivation is commonly affected in those with psychiatric disorders, such as MDD, with the results of the current study supporting these prior findings [20]. Regarding homogenous symptoms of MDD (e.g., anhedonia), it has been suggested that disruptions to reward anticipation may in part be due to a reduced incentive to exert effort to either gain a reward or avoid a loss [61], [62], [63]. In line with those prior studies, the current findings may suggest that within a subsample of AI individuals experiencing elevated TNF and CRP concentrations may be less sensitive to the potential gain or loss of a monetary reward.

Importantly, the current results displayed some differences across sex and medication status. The literature has established sex differences across immune responses and inflammatory biomarkers. A meta-analysis of 23 studies found elevated concentrations of IL-6 and CRP within the MDD female group in comparison to the HC female group [64]. However, mixed results have been seen when exploring sex differences across TNF concentrations. Therefore, additional work is needed to explore potential sex differences across other inflammatory biomarkers, such as TNF in the context of MDD. Similarly, differences were seen across those taking medications versus those who went unmedicated.

Contrary to the original hypothesis, PROMIS Depression score was not a significant predictor within the current study. This may suggest that the disruptions in immune responses may play a significant role in predicting the development and/or maintenance of MDD symptoms, as opposed to only self-report measures. The heterogeneity of MDD presentation and comorbidities challenge precisely attributing pathophysiological mechanisms when using heterogenous diagnostic grouping [65], [66], [67]. Therefore, future investigations may benefit from studying specific symptom(s) dynamics in relation to underlying mechanistic relationships. Particularly, anhedonia has been shown to be associated with both reward circuitry disruptions as well as treatment resistant MDD. Similarly, the current results did not show inflammation to be a significant predictor of P300 amplitude differing from prior studies that have investigated differences in P300 amplitude amongst samples with depression [68], [69]. However, it is important to note that to our knowledge this is the first study specifically exploring the relationship between inflammation and reward anticipation in an AI sample. Future studies are needed to further delineate this relation.

The current findings need to be considered in the context of their limitations. All analyses conducted come from secondary data, which was not designed to answer the question posed by the current study and had limited power because of the small sample size [44], [45]. Additionally, the P300 was the only ERP used because the MID task was designed to accommodate fMRI signal acquisition, which required longer individual trials and lead to fewer overall trial numbers. This limited the ERPs that could be included in the current study because a larger number of trails are needed to have reliable ERP signal across conditions and stimulus type. Another limitation was the inability to have an HC group to compare with the MDD sample because of the limited sample of HC individuals in the parent studies who met the criteria for the current study. Furthermore, when working with AI communities, broader historical and contemporary factors may play a critical role in understanding underlying factors related to psychopathology [70].This is especially important, as prior studies have found when research studies have incorporated historical (e.g., historical trauma) or contemporary (e.g., discrimination) cultural stressors, disparate prevalence rates have either 1) vanished and are similar to the prevalence rates of other populations or 2) are lower than other populations [52]. Therefore, the current study is limited in its ability to interpret the current findings given the inability to include these contextualizing factors.

Conclusion

Future research would benefit from investigating the associations between depression symptomology and pro/anti-inflammatory cytokines to further delineate the relationship between these cytokines and neural activity in response to reward anticipation. Also, the inclusion of resilience factors discerned in previous literature would allow for contextualization of the current and future findings [71]. This would further our understanding of the risk factors that place AI communities at disproportionate burden of mental health disparities, and what resiliency factors can be effective tools to inform preventative care and treatment for AI communities. Specifically, aspects of cultural engagement and connectedness to Native communities may serve as a protective factor [42], [72]. Prior findings suggest intervention efforts could benefit from the inclusion of culturally relevant aspects early on and facilitate treatment success, especially for treatment-resistant MDD [73], [74], [75]. Although the current study was unable to measure these protective factors, this aspect of AIs health research, along with the implication of inflammation on cognitive function, could inform future studies when examining protective factors to improve both prevention and treatment efforts for MDD. Overall, the current study suggest that inflammation is associated with reward circuitry disruptions and may play a role in the pathophysiology of MDD; however, future work is needed to further contextualize the findings to inform intervention efforts that may be beneficial for AI populations.