We recruited participants who did not receive a psychiatric diagnosis and psychotropic medication, since either considerably affects the levels of immune/inflammatory markers and amygdala emotional reactivity (62, 63). The individuals’ risk of developing stress-related mental disorders was assessed by measuring the current stress-related symptoms (i.e., depression/anxiety) to examine their relationships with the variables of interest based on the perspective of the dimensional model for mental disorders (64). This model postulates that the manifestation of mental disorders can be dimensional, but not categorical, including individuals with various levels of correspondent mental symptoms from none to moderate (i.e., risk individuals) and severe (i.e., patients) (64).

Participants were recruited via websites, magazine advertisements, and billboards at Kitasato University from people in the community who dwelled in Tokyo and surrounding cities such as Kanagawa, Chiba, Saitama, and Ibaraki prefectures. The eligibility criteria were as follows: (1) 18–59 years old; (2) no current Axis-I psychiatric disorders or substance-abuse history as determined by the 4th edition of the Diagnostic and Statistical Manual of Mental Disorders by an experienced clinician (YH); (3) no major medical illnesses; (4) no regular intake of psychotropics, steroids, or opioids; (5) no metal nor medical appliance on/inside the body; (6) no history of brain injury or trauma with loss of consciousness over 10 min; and (7) no habitual intake of medicine for pain or fever (e.g., loxoprofen or acetaminophen) or excessive caffeine (> 400 mg/day) that could affect the blood-oxygen-level-dependent (BOLD) signal (65). Of the 134 recruited participants, 11 did not meet the inclusion criteria due to a history of brain trauma with loss of consciousness (> 10 min, n = 1), epilepsy (n = 1), orthodontic appliances (n = 2), immune-related diseases (n = 3, including 1 hyperlipidemia, 1 systemic lupus erythematosus, and 1 ulcerative colitis case), and regular intake of loxoprofen for migraines (n = 4). In total, 123 participants provided written informed consent.

Additionally, this study shared the sample of our previous studies as follows: 50.0% (66), 75.9% (67), 76.8% (68), and 91.7% (54), although there was no overlap with Izawa et al. (46) and Hakamata et al. (69). Studies by Hakamata et al. (66–68) aimed to examine the relationship between resting-state FC and memory biases, and the study by Hori et al. (54) examined the relationship between childhood abuse and IL-6 levels. However, the current study focused on the relationship between IL-6 levels, depression, and amygdala activity during an emotional task in terms of gene-stressor interactions. Thus, the focus of the current study was entirely different from those of the previous studies.

This study was approved by the Kitasato University Medical Ethics Organization (C17-126 and G18-02 for genome) and the National Center of Neurology and Psychiatry Ethics Committee (A2018-064) as part of a research project to examine cognitive bias and related neurobiological mechanisms; and was conducted in compliance with ethical guidelines issued by the National Ministry of Health, Labor, and Welfare and Declaration of Helsinki.

Participants underwent psychological assessment and MRI scans on the same day; saliva was collected five times daily on two consecutive weekdays within 2 weeks of the initial assessment day. Saliva was also collected from a subset of the participants (n = 73) for DNA extraction. The sample attrition was because the genetic part of this study (G18-02) was approved by the ethics committee in June 2018, which was about 8 months after the main part started (C17-126, approved in October 2017). Participants provided the following information on potential IL-6 confounders: age, sex, body mass index (BMI), daily caffeine intake, smoking habits, monthly alcohol consumption, and years of education. For women, menstrual status information was also obtained. Scoring details of BMI, daily caffeine intake, monthly alcohol consumption, and menstrual status are described in Supplementary material S1.1.

IL-6 levels were measured based on saliva collected five times daily over two consecutive typical weekdays: upon awakening (T1), 30 min after awakening (T2), midday (11:30–12:30) (T3), in the evening (17:30-18:30) (T4), and at bedtime (T5).

Anatomical and functional MRIs were acquired using a 32-channel phased-array head coil, 3.0 T scanner (Discovery MR750; GE Healthcare, United States). Structural images were acquired using a 3D T1-weighted sequence [slice thickness without slice gap = 1.0 mm, field of view (FOV) = 256 mm, matrix = 256 × 256, repetition time (TR) = 6.4 ms, echo time (TE) = 2.6 ms, inversion time (TI) = 400 ms, and flip angle (FA) = 14°]. For functional images, data were acquired using fast-gradient echo-planar T2*-weighted imaging with five dummy volumes at the beginning of the session. Each functional volume comprised 38 transverse slices (slice thickness = 3.0 mm, slice gap = 1.0 mm, FOV = 192 mm, matrix = 64 × 64, TR = 3,000 ms, TE = 30 ms, and FA = 90°).

For preprocessing and analyses of brain activity and FC, we used SPM12¹ and CONN Functional Connectivity Toolbox, version 18b,² respectively; both were operated on the Matlab 2019b platform.³

For brain activity analysis, the following preprocessing procedures were performed using SPM12: realignment, slice timing correction, co-registration of the structural image to the functional images, segmentation of the T1 structural volume image, spatial normalization to the Montreal Neurological Institute (MNI) space, and Gaussian spatial smoothing [full width at half maximum (FWHM): 6 mm] (82). No excessive head motion (i.e., > 3 mm translation/rotation in any direction) was observed in the realignment.

For FC analysis, the aforementioned preprocessing procedures were similarly performed using a default pipeline with outlier detection (“scrubbing” to eliminate excessive head motion). In the outlier detection, three confounders were removed via principal-component-based noise-correction (“CompCor” method) (83): signal noise from the white matter and cerebrospinal fluid; within-subject covariates, including head-motion artifacts and scrubbing parameters; and the main condition effect convolved with a hemodynamic response function. The band-pass filter was set with a frequency window of 0.008–0.12 Hz. FWHM was 6 mm.

We employed a modified version of the face-house matching task (84), which is widely used to capture the amygdala activity in response to emotional stimuli (85), as previously reported (69). As illustrated in Figure 1, four frames are displayed for 2,000 ms, with either the two horizontal or vertical frames being highlighted in yellow. Then, after a fixation at the center of the screen presented for 1,000 ms, they were replaced by two faces (either neutral or fearful ones of the same individuals) in the location of the non-highlighted frames and two houses in the location of highlighted frames for 250 ms. Participants were instructed to always attend to the two highlighted frames and judge whether the subsequent houses are identical while faces are simultaneously appearing.

For face stimuli, we used eight different individuals’ faces with fearful and neutral expressions (four women and four men), selected from an established standard facial database for the Japanese population (DB99; ATR-Promotions, Inc., Kyoto, Japan). We used eight houses, as used in the original study (84). All pictures were black and white photographs (visual angle 18° × 13°) presented on a black background and projected through a mirror mounted onto the head coil. The stimulus presentation schedule was controlled by the E-prime 2.0 Professional software (Psychology Software Tools, Inc., Pittsburgh, PA) in synchronization with MR signals. The stimulus intervals varied randomly between 3.5 and 12.5 s for a total of 48 trials (mean: 7.5 s). The task duration was 8 min 51 s, including the five dummy scans. The positions and emotional valences of faces were randomly presented. To assess amygdala emotional reactivity, we compared responses to fearful vs. neutral faces.

For genotyping, saliva samples were collected from each participant with an Oragene DNA Kit (DNA Genotek Inc., Canada). We specifically studied the rs1800796 (IL6) and rs2228145 (IL6R) SNPs. The IL6 gene is located on chromosome 7p21, with the well-documented SNP, rs1800795 (−174G/C), located at its promoter region. However, the minor allele frequency of rs1800795 is known to be 0% in the Japanese population according to SNPedia (87). Therefore, we focused on rs1800796 (−572C/G) instead, which is also located in the promoter region of IL6 in strong linkage disequilibrium with rs1800795. Like rs1800795, rs1800796 is known to also affect circulating IL-6 levels (23). The IL6R gene is located on chromosome 1q21, and its missense variant, rs2228145 (Asp358Ala; A/C), located at exon 9, maps to the IL-6R cleavage site.

Genomic DNA was purified from centrifuged saliva using the ethanol precipitation protocol and the prepIT.L2P manual purification protocol (DNA Genotek Inc., Canada). IL6 rs1800796 (assay ID: C___11326893_10) and IL6R rs2228145 (assay ID: C___16170664_10) were genotyped using the TaqMan SNP Genotyping Assays. The polymerase chain reaction was carried out using GeneAce Probe qPCR Mixα (Nippon Gene, Japan) under the following conditions: 1 cycle at 95°C for 10 min followed by 45 cycles at 95°C for 15 s and 60°C for 1 min. The allele-specific fluorescence was measured with an ABI PRISM 7900 Sequence Detection System (Applied Biosystems, United States).

The numbers of participants with rs1800796 C/C, C/G, and G/G genotypes were 41 (56.2%), 24 (32.9%), and 8 (11.0%), respectively; and those with rs2228145 A/A, A/C, and C/C genotypes were 26 (35.6%), 37 (50.7%), and 10 (13.7%), respectively. The genotype frequency for these two SNPs did not deviate from Hardy–Weinberg equilibrium: χ²(1) = 2.20, p = 0.14, and χ²(1) = 0.31, p = 0.58, respectively. Due to the low frequency of minor alleles, participants were assigned to the following groups: “C/C” or “C/G + G/G” for rs1800796 and “C/C” or “A/A + A/C” for rs2228145.

Of the 123 enrolled participants, two did not provide saliva samples and one had an MRI data writing error. For the IL-6 assay, the data whose absorbance values could not be measured with an absorptiometer were excluded from the analysis. Consequently, 1,119 (93.7%) of the total 1,200 IL-6 measurements (i.e., the remaining 120 individuals with 10 measurements each) were included. When IL-6 data was available for only one of the 2 days (i.e., data missing from any time point for either day), the two-day average data was substituted by this single-day data. Of the 120 participants, 12 were excluded from the entire analysis because their IL-6 data were not available for one or more corresponding time points for both days. Thus, we analyzed the data of 108 participants as a final data set (Table 1). Of these participants, two women regularly took oral contraceptives and none were engaged in shift work. Although no participant met the criteria for a current MDD episode, 12 of them had significant depression levels defined by the BDI-II cut-off score, indicating that they were at a high-risk state for MDD. These individuals had moderate levels of depression (BDI-II: Mean ± SD = 22.6 ± 8.5), and were not significantly different from non-depressive individuals in terms of age and sex (age: t = 0.34, p = 0.74, Mean ± SD = 26.8 ± 11.9 and 27.9 ± 11.4, respectively and sex: χ² = 0.68, p = 0.54, 66.7% and 54.2%, respectively, for female ratios).

Saliva collection time at T1, T2, T3, T4, and T5 was as follows: 7:09 (± 1:16), 7:46 (± 1:17), 12:06 (± 0:20), 18:03 (± 0:19), and 0:20 (± 1:09), respectively. Average and standard errors of raw IL-6 values at each time point are shown in Table 1 and Figure 3. The observed IL-6 diurnal pattern was similar to that reported in our previous study (46), confirming that the salivary IL-6 levels were properly measured in this study. Since IL-6 diurnal index and total output did not follow a normal distribution, they were square-root transformed (W = 0.65, p < 0.001; W = 0.71, p < 0.001, respectively).

Likewise, the balanced impact score of LES did not follow a normal distribution (W = 0.95, p < 0.001), and thus participants were classified into those who experienced NLC and those who did not.

For fMRI data, 11 participants were unable to respond by the time the next screen was presented in more than 30% of the total trials since the timing of stimulus presentation was synchronized with that of each MR signal. The overall hit ratio was 85.3 ± 11.4% in the 97 participants, indicating that they concentrated well on the task. In addition, the time of fMRI scans was 13:39 (± 1:22). The scanning time did not affect amygdala activity (see Supplementary material S2.3).

Furthermore, there was no significant difference between the whole sample (n = 108) and the DNA subsample (n = 73) in age, sex, IL-6 diurnal index, or total IL-6 output (t = 1.10, df = 179, p = 0.27; X² (1, 181) = 4.2, p = 0.55; t = 0.58, df = 179, p = 0.57; t = −0.36, df = 179, p = 0.72; respectively).

Whole-brain analysis revealed that IL-6 diurnal index had a significantly positive association with left BLA activity to fearful (vs. neutral) faces (MNI: −30 −7 −22, t = 3.67, FWE-corrected p = 0.003, 54 mm³; Figure 4), such that the more the IL-6 diurnal index diminished, the more the BLA activity decreased. No other significant cluster was found in the right BLA, CEM, and other brain regions. For IL-6 total output, no significant correlation was found with activity in any brain region.

Additional correlation analysis showed that the IL-6 diurnal index was associated with BDI-II depressive symptoms only, with increased blunting of the IL-6 diurnal pattern corresponding with greater depressive symptoms, although this correlation failed to reach the Bonferroni-corrected statistical threshold (r = −0.24, p = 0.018). The IL-6 diurnal index did not correlate with HSCL anxiety and depressive symptoms (Table 2). Despite the distinct correlation with the IL-6 diurnal index (r = 0.30, p = 0.002) as observed in the whole-brain analysis, the left BLA activity was uncorrelated with BDI-II depressive symptoms (r = 0.04, p = 0.67), although it had a relative correlation with HSCL anxiety symptoms (r = 0.20, p = 0.04).

Although the linear relationship between BDI-II depression and IL-6 diurnal index slightly failed to survive the Bonferroni correction, we also compared depressive individuals defined by the BDI-II cut-off score (n = 12) and non-depressive ones (n = 96) with ANCOVA controlling for age, sex, sleep duration, and BMI, in consideration of a possible non-linearity of biomarkers as suggested in the allostatic load model (1). As a result, we found a significant difference between the two groups (F[1,102] = 14.8, p < 0.001, partial η² = 0.13). The depressive group had a significantly blunted IL-6 diurnal pattern (mean ± SD = −0.08 ± 2.98) as compared to the non-depressive group (mean ± SD = 2.13 ± 2.00; Figure 5).

Three-way ANCOVA demonstrated a significant rs1800796 × NLC interaction effect [F(1,61) = 15.1, p < 0.001, partial η² = 0.20], with a significant main effect detected for NLC [F(1,61) = 4.26, p = 0.04, partial η² = 0.07]. No other significant effects were observed: F (1,61) = 0.68, p = 0.41 for IL6; F(1,61) = 0.44, p = 0.51 for IL6R; F(1,61) = 0.00, p = 0.99 for NLC × IL6R; F(1,61) = 0.05, p = 0.82 for IL6 × IL6R; F(1,61) = 0.69, p = 0.41 for NLC × IL6 × IL6R.

When the significant rs1800796 × NLC effect was parsed into a simple main effect, a significant difference was found in the C/C group [n = 41, F(1,35) = 19.71, p < 0.001], but not in the G/C + G/G group [n = 32, F(1,26) = 1.22, p = 0.28]. Specifically, in the C/C group, the IL-6 diurnal index was significantly diminished under the presence of NLC (n = 16, mean ± SD = −0.32 ± 3.22, −0.80 ± 1.11 in Z score) but not under its absence (n = 25, mean ± SD = 2.80 ± 1.56, 0.22 ± 0.81 in Z score; Figure 6).

Additionally, our sample size for DNA analysis was observed to have a power of 0.99 to detect the observed effect size for the IL6 × NLC interaction (i.e., partial η² = 0.20) at an α = 0.05.

Further, a multiple regression analysis where the LES balanced impact score was included as a continuous variable similarly found the rs1800796 × LES interaction effect (β = 0.82, t = 2.66, p = 0.010; see Supplementary material S2.4), confirming the IL6 SNP-psychosocial stressor interaction effect on the IL-6 diurnal index.

Since the correlation of IL-6 diurnal index with BDI-II depression was still relatively significant, we proceeded to an exploratory path analysis on possible pathways from BLA activity to the IL-6 diurnal index and BDI-II depression under the rs1800796 × NLC interaction.

Based on the significant rs1800796 × NLC effect on IL-6 diurnal index and the relative correlation, Path analysis demonstrated that the diminished IL-6 diurnal index specifically explained greater depressive symptoms (β = −0.40, p < 0.001; respectively), being affected by the left BLA activity and the rs1800796 × NLC interaction (β = 0.36, p < 0.001; β = −0.41, p < 0.001; respectively; Figure 7).

Additionally, the path coefficients in the DNA subsample were comparable to those in the whole sample where the gene-stressor interaction was not included (BLA–IL-6: β = 0.29, p = 0.002; IL-6–depression: β = −0.28, p = 0.004; BLA–depression: β = 0.16, p = 0.10), supporting the observed relationships.

Whole-brain FC analysis revealed that the IL-6 diurnal index was significantly associated with increased FC between the CEM and middle frontal gyrus (MFG) on the right side (t = 3.92, FDR-corrected p = 0.012), as well as other cortical regions (Figure 8). All the significant FC values are reported in Table 3.

To summarize, the diminished IL-6 diurnal pattern was associated with the hypoactivation of the left BLA to fearful (vs. neutral) faces, and was predominantly observed in individuals with the C-allele homozygotes of IL6 and NLC in the past year. When considered comprehensively, the blunted diurnal pattern predicted greater depressive symptoms, modulated by the amygdala hypoactivity and rs1800796-NLC interactions.

Several limitations should be considered when interpreting these results. First, our sample for the SNP-related analysis was not very large, although the sample size was sufficiently powered to detect the observed significant interaction effect according to post hoc power analysis. Still, future research with larger sample sizes is necessary. Second, information on oral hygiene of the participants was unavailable in this study. However, even if the IL-6 levels at each time point were affected by inflammation due to oral hygiene, it would not significantly undermine the quality of IL-6 diurnal index data as it was calculated based on differences between two-time points, conceivably eliminating the systematic error due to such inflammation. In future research, information on oral hygiene should be obtained. Third, this was a cross-sectional study and did not reveal potential causal relationships between IL-6 levels, amygdala emotional reactivity, and depressive symptoms. Future research should elucidate these causal relationships by simultaneously performing fMRI and other imaging techniques with a higher temporal resolution, such as electroencephalography. Furthermore, the development of indices to assess individuals’ regulatory functions of the immune system beyond acute inflammation (i.e., not only phasic but also tonic changes of the immune system) might enable the early detection of high-risk individuals whose psychopathology is yet to be manifested and the effective prevention for them by ameliorating their immune system dysregulation. To capture such baseline regulatory functions, consecutive measurements of unstimulated biological markers on a daily basis, such as IL-6 diurnal patterns measured in this study and spontaneous low-frequency fluctuations measured by resting-state fMRI, might bring a new insight to psychiatry research.

We showed that blunting of the IL-6 diurnal pattern was associated with both amygdala (especially BLA) hypoactivity in response to fearful (vs. neutral) faces and greater depressive symptoms in a sizeable community sample with various levels of depression. Disrupted IL-6 diurnal rhythm was explained by the interaction between IL6 polymorphism and stressful life events. These findings represent the first evidence that diminished IL-6 diurnal rhythm could affect depression, modulated by amygdala emotional hyporeactivity and the interaction between the individual’s genetic predisposition and the intensity of environmental stress. Developing an understanding of this blunted diurnal IL-6 rhythm, a possible precipitant in stress-related mental disorders such as MDD, is expected to facilitate their early detection, prevention, and treatment such as pharmacotherapy to ameliorate the disrupted secretion of inflammatory markers.