Frequency-Specific Neural Signatures of Spontaneous Low-Frequency Resting State Fluctuations in Psychosis: Evidence From Bipolar-Schizophrenia Network on Intermediate Phenotypes (B-SNIP) Consortium

Abstract

Background:

We quantified frequency-specific, absolute, and fractional amplitude of low-frequency fluctuations (ALFF/fALFF) across the schizophrenia (SZ)-psychotic bipolar disorder (PBP) psychosis spectrum using resting functional magnetic resonance imaging data from the large BSNIP family study.

Methods:

We assessed 242 healthy controls (HC), 547 probands (180 PBP, 220 SZ, and 147 schizoaffective disorder—SAD), and 410 of their first-degree relatives (134 PBPR, 150SZR, and 126 SADR). Following standard preprocessing in statistical parametric mapping (SPM8), we computed absolute and fractional power (ALFF/fALFF) in 2 low-frequency bands: slow-5 (0.01–0.027 Hz) and slow-4 (0.027–0.073 Hz). We evaluated voxelwise post hoc differences across traditional Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition diagnostic categories.

Results:

Across ALFF/fALFF, in contrast to HC, BP/SAD showed hypoactivation in frontal/anterior brain regions in the slow-5 band and hypoactivation in posterior brain regions in the slow-4 band. SZ showed consistent hypoactivation in precuneus/cuneus and posterior cingulate across both bands and indices. Increased ALFF/fALFF was noted predominantly in deep subcortical and temporal structures across probands in both bands and indices. Across probands, spatial ALFF/fALFF differences in SAD resembled PBP more than SZ. None of these ALFF/fALFF differences were detected in relatives.

Conclusions:

Results suggest ALFF/fALFF is a putative biomarker rather than a familial endophenotype. Overall sensitivity to discriminate proband brain alteration was stronger for fALFF than ALFF. Patterns of differences noted in SAD were more similar to those observed in PBP. Differential effects were noted across the 2 frequency bands, more prominently for BP/SAD compared with SZ, suggesting frequency-sensitive physiologic mechanisms for the former.

Introduction

The traditional classification of major psychiatric disorders has been questioned recently because symptomatic boundaries not only overlap, but display rather limited biological validity.¹ Schizophrenia (SZ) and psychotic bipolar disorder (PBP) share multiple characteristics, including risk genes and abnormalities in cognition, neural function, and brain structure.^2–4 To overcome the traditional SZ/PBP dichotomy issue, “crosscutting” dimensional assessment of symptoms and clinical phenomena is advocated in the newly revised Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-V).⁵ Lacking external validation criteria for classification, psychiatry has seen a shift toward utilization of objective biological markers that could ultimately be used to develop targeted treatments. Such biological signatures could also be examined as candidate endophenotypes by testing these measures in unaffected relatives to ultimately improve understanding of underlying genetic/molecular risk mechanisms.

In contrast to task-related functional magnetic resonance imaging (fMRI) experiments where low-frequency signals are filtered out, in resting-state fMRI (rs-fMRI), low-frequency oscillations (LFOs; typically in the 0.01–0.08 Hz frequency band) are physiologically relevant and related to neuronal fluctuations in brain gray matter.^6,7 Such LFOs in general may represent an integration of neuronal firing effects with longer delays and larger variability and that recruit larger brain areas, compared with high-frequency oscillations that are characteristically more localized and incorporate synaptic events from spatially close regions.⁸ Several neuropsychiatric studies have evaluated LFO connectivity or coherence and demonstrated their usefulness as biomarkers and/or endophenotypes.^2,3,9,10 However, research on regional or local properties of the brain’s intrinsic functional dynamics is lacking. Conventionally, rs-LFOs have been examined in the 0.01–0.08 frequency band.^6,7,9 However, seminal studies by Penttonen, Buzsaki, and Draguhn suggest that neuronal oscillation classes are arrayed linearly when plotted on a logarithmic scale. They concluded that this regularity (and empirical data collected at higher frequencies) suggests that independent frequency bands are generated by distinct oscillators, each with specific properties and unique physiological functions.^8,11 They also asserted that neuronal oscillations in adjacent bands do not co-occur within a particular structure, by showing that major rest-/sleep-related oscillations (delta, alpha, etc.) are separated by active-state-related mechanisms. Zuo and colleagues were the first to extend this concept to rs-fMRI. They evaluated LFOs in 4 distinct bands—slow-5 (0.01–0.027 Hz), slow-4 (0.027–0.073), slow-3 (0.073–0.198 Hz), and slow-2 (0.198–0.25 Hz) and demonstrated differential and interactive amplitude effects within them in several brain regions.¹² Recent reports have further confirmed that integration of brain function occurs within multiple frequency bands and might have different neural manifestations, thus urging the scientific community to examine frequency-specific effects.^6,13,14 In the current study, we chose to adopt the Buzsaki framework, but only examine the slow-5 (0.01–0.027 Hz) and slow-4 (0.027–0.073) bands because these encompass most of the traditional 0.01–0.08 frequency spectrum and have minimal overlap with potential physiological noise frequency.⁷

Local properties of spontaneous activity can be probed using various methods including regional coherence Regional Homogeneity, power spectrum analysis (absolute and fractional amplitude of low-frequency fluctuations [ALFF/fALFF], fractal dimension), and temporo-spatial clustering.¹⁵ In this study, we evaluated ALFF and fALFF because they have yielded the most promising results. ALFF is defined as the total power within a specific frequency range and thus indexes the strength or intensity of LFO. fALFF, on the other hand, measures the power within a specified band normalized to the power in the entire detectable frequency range, thus representing the relative contribution of a specific LFO to the whole frequency range. fALFF is thought to be an improvement over ALFF; however, each index has its own pros and cons.^16,17 For example, fALFF is previously reported to have higher specificity but lower reliability to gray matter signal, vs ALFF. Therefore, in order to maximize reliability across subjects while providing sufficient specificity to capture interindividual differences (as recommended by Zuo et al¹²), we report both metrics in the present study.

Although the exact origins and interacting mechanisms behind ALFF/fALFF are unknown, neurophysiologically, these measures have been proposed to be local intensity estimates of spontaneous brain activity^7,9,18 and have previously shown to be robustly sensitive to signals originating in gray matter.^7,16 Recent studies analyzing concurrent electroencephalogram (EEG) and rs-fMRI have helped shed more light on the underlying neurophysiology of these LFOs in rs-fMRI. Rodent studies suggest that rs-LFOs are correlated with synchronized delta EEG oscillations.¹⁹ Similarly, another study using simultaneous EEG-fMRI reported LFOs in monkey visual cortex were correlated with local field gamma band power.²⁰ In humans, increased alpha power has been linked to decreased rs-LFOs in multiple regions, including occipital, superior temporal, inferior frontal, and cingulate cortex.²¹ Additionally, several studies have shown task-related modulation of LFO amplitudes, especially in the realm of working memory and motor performance.^18,22 Rs-LFOs have also been related to arousal level, by demonstrating that sleep produces stage-dependent alterations in amplitude patterns.^23–25

Specific to SZ, a prior dual-band examination of ALFF/fALFF showed differential patterns between the slow-5 and slow-4 bands mainly in basal ganglia, midbrain, and ventromedial prefrontal cortex, suggesting that the 2 different frequency bands might be differentially sensitive in SZ.²⁶ Similar frequency-dependent ALFF changes have also been reported in elderly subjects with amnestic mild cognitive impairment.²⁷ Most ALFF/fALFF studies have used the conventional (0.01–0.08) band, with variability mainly due to sample size, medication effects, analytic techniques, and thresholding. However, regions showing the most consistent decreases in ALFF/fALFF in SZ across studies are medial prefrontal cortex (MPFC), cuneus/precuneus, posterior cingulate cortex (PCC), and medial occipital gyrus. Similarly, regions most consistently showing increased ALFF/fALFF in SZ were hippocampal/parahippocampal cortex and other subcortical structures like amygdala, thalamus, putamen, and caudate. Studies evaluating ALFF/fALFF in bipolar subjects are few, with only 2 reporting ALFF (0.01–0.08) differences; these showed decreased ALFF in postcentral, parahippocampal, lingual gyri, and cerebellum. Increased ALFF in bipolars was noted mainly in the dorsolateral prefrontal cortex, middle frontal, insula, orbito-frontal cortex, and anterior cingulate cortix (ACC).^28–30 As noted, only a few studies, in generally smallish samples, have evaluated this promising biological marker to compare SZ and PBP. More importantly, no study has yet evaluated ALFF/fALFF across different frequency bands across a broad psychosis spectrum.

The current study employed data from a large multisite study, Bipolar-Schizophrenia Network on Intermediate Phenotypes (B-SNIP),³¹ established to test the manifestations and distribution of multiple putative intermediate phenotypes including rs-fMRI across DSM-IV psychosis-centered diagnostic categories. We sought to answer 2 primary questions (1) to evaluate shared and unique characteristics of ALFF/fALFF across psychosis spectrum diagnoses across different frequency bands and (2) to test ALFF/fALFF as putative disease biomarkers and/or endophenotypes, evaluating these measures across psychosis probands and their biological relatives. Based on previous findings, we hypothesized (1) decreased ALFF/fALFF power in cuneus/precuneus, posterior cingulate, lingual gyrus, and occipital areas and increased power in striatal and deep brain regions in psychosis probands and (2) similar, albeit diminished ALFF/fALFF traits in relatives of probands. Furthermore, we predicted differential effects in probands across different bands, especially in basal, midbrain, and medial prefrontal regions.

Materials and Methods

Study Sample

Subjects passing all quality control (steps listed in methods) (N = 1199) and included in the final analysis comprised 242 healthy controls (HC), 547 probands (180 PBP, 220 SZ, and 147 schizoaffective disorder—SAD), and 410 first-degree relatives (134 PBPR, 150 SZR, and 126 SADR). All subjects were administered similar rs-fMRI protocol across 6 sites (Baltimore, Boston, Chicago, Dallas, Detroit, and Hartford). Probands/relatives Axis-I diagnoses were based on Structured Clinical Interview for DSM-IV TR Diagnosis (SCID-I/P).³¹ These diagnostic definitions (SZ, SAD, PBP) were established based on all available clinical information, followed by a formal diagnostic consensus discussion by at least 3 experienced clinicians (MD, PhD, or Master’s level), to improve interrater reliability of clinical diagnosis. As described previously, probands were stable, medicated outpatients.³² Relatives with lifetime psychiatric diagnoses were asymptomatic/mildly symptomatic during the imaging session.³² Relatives meeting criteria for Axis-I “proband-like” psychotic disorders (SZ, SAD, PBP) were regrouped into the corresponding proband category. The remaining biological relatives comprised SZR, SADR, and PBPR groups. We define “unaffected” status in relatives as absence of lifetime psychotic disorders. The study protocol was approved by each local site IRB. After complete description of the study to the participants, written informed consent was obtained. Complete demographic and clinical characteristics of the study sample are provided in table 1. Concomitant medications data are reported in supplementary table 1. Further, to evaluate current symptom severity, all probands were administered the Positive and Negative Syndrome Scale (PANSS).³³

Table 1.

Detailed Demographics and Clinical Characteristics of the Study Population

Demographics	Healthy Controls (N = 242)		Bipolar (N = 180)		Schizoaffective (N = 147)		Schizophrenia (N = 220)		Bipolar Relatives (N = 134)		Schizoaffective Relatives (N = 126)		Schizophrenia Relatives (N = 150)		Statistic
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	F	P Value
Age (years)	38.14	12.65	36.94	13.04	35.08	12.01	35.15	12.31	40.59	16.13	41.01	16.14	43.33	15.55	7.72a	<<.001
Clinical Scores
PANSS Positive	N/A	N/A	12.77	4.38	18.12	5.32	16.91	5.42	N/A	N/A	N/A	N/A	N/A	N/A	50.47b	<<.001
PANSS Negative	N/A	N/A	11.79	3.61	15.49	4.95	16.27	5.93	N/A	N/A	N/A	N/A	N/A	N/A	40.96c	<<.001
Schizo-Bipolar Scale	N/A	N/A	7.79	0.09	5.04	1.6	1.35	1.24	N/A	N/A	N/A	N/A	N/A	N/A	1006.21d	<<.001

	N	%	N	%	N	%	N	%	N	%	N	%	N	%	χ2	P value
Sex
Male	103	42.56	58	32.22	66	44.90	145	65.91	49	36.57	40	31.75	49	32.67	71.27	<<.001
Female	139	57.44	122	67.78	81	55.10	75	34.09	85	63.43	86	68.25	101	67.33
Ethnicity
Non-Hispanic	220	90.91	165	91.67	130	88.44	202	91.82	121	90.30	120	95.24	134	89.33	6.00	.560
Hispanic	22	9.09	15	8.33	17	11.56	18	8.18	13	9.70	6	4.76	16	10.67
Handedness
Left	30	12.40	27	15.00	12	8.16	28	12.73	15	11.19	14	11.11	18	12.00	18.00	0.460
Right	204	84.30	151	83.89	132	89.80	186	84.55	118	88.06	106	84.13	129	86.00
Ambidextrous	2	0.83	1	0.56	3	2.04	3	1.36	1	0.75	4	3.17	1	0.67
Missing	6	2.48	1	0.56	0	0.00	3	1.36	0	0.00	2	1.59	2	1.33
Sites															30.00	<<.001
Baltimore	51	21.07	35	19.44	30	20.41	71	32.27	30	22.39	24	19.05	39	26.00
Boston	9	3.72	4	2.22	1	0.68	5	2.27	4	2.99	3	2.38	7	4.67
Chicago	52	21.49	52	28.89	27	18.37	33	15.00	32	23.88	16	12.70	17	11.33
Dallas	57	23.55	28	15.56	40	27.21	34	15.45	12	8.96	29	23.02	21	14.00
Detroit	23	9.50	29	16.11	5	3.40	27	12.27	13	9.70	4	3.17	14	9.33
Hartford	50	20.66	32	17.78	44	29.93	50	22.73	43	32.09	50	39.68	52	34.67

Note: ^aAge post hoc differences: SADR, SZR > HC; SZR, BPR, SADR > BP; HC, BPR, SADR, SZR > SZ; BPR, SADR, SZR > SAD.

^bPANSS_Positive post hoc differences: SZ > BP; SAD > SZ, BP.

^cPANSS_Negative post hoc differences: SZ > BP, SAD; SAD > BP.

^dSchizo-Bipolar Scale post hoc differences: BP > SAD > SZ.

Data Acquisition

Imaging Data.

All subjects underwent a single 5-min run of resting state fMRI and a 3D T1-weighted structural scan on a 3T scanner at each site. Participants were instructed to keep their eyes open, focus on a crosshair displayed on a monitor, and remain still during the entire scan. In addition, head motion was restricted with a custom-built head-coil cushion. Alertness during the scan was confirmed immediately afterward. If necessary, the scan was repeated. Differences in scanning parameters and sites were handled appropriately during both preprocessing and statistical analysis (see supplementary table 2 for scanning parameters).

Data Processing

Preprocessing and ALFF/fALFF Computation.

Prepro cessing of fMRI, T1-weighted and computation of ALFF/fALFF images were conducted using the REST toolbox.³⁴ The initial 6 images, during which T2-effects stabilized, were discarded. Images were then realigned and corrected for slice timing differences, acquisition parameters such as repetition time, slice acquisition direction etc. The above pipeline was adapted individually for each site due to differences in scanning parameters. At this stage, subjects with excessive motion (>3mm of motion and/or 3° rotation) were dropped from further analysis. Next, structural and functional images were co-registered to each other. Six motion parameters, signal from the cerebro spinal fluid (CSF), and white matter were used as nuisance covariates to reduce effects of head motion and non-neuronal BOLD fluctuations. Structural images were normalized using the DARTEL technique.³⁵ Normalization was checked visually and any anomalous data were discarded. For the remaining subjects, normalization parameters obtained during the previous step were then applied to the functional images to bring them into a common DARTEL-MNI space. Also, linear detrending was performed before computation of ALFF/fALFF.

Following the above preprocessing, ALFF images were computed by extracting power spectra via a Fast Fourier Transform and computing the sum of amplitudes in 2 separate low-frequency bands: slow-5 (0.01–0.027 Hz) and slow-4 (0.027–0.073 Hz). The ALFF measure at each voxel represents the averaged square root of the power in the above frequency windows normalized by the mean within-brain ALFF value for that subject. For fractional ALFF, the measure was scaled by total power across all available frequencies. Finally, prior to conducting statistical analysis, ALFF/fALFF images were smoothed by an 8-mm full-width half maximum Gaussian kernel.

Statistical Analyses

Voxelwise group differences were examined using an ANCOVA model using a permutation-based approach in conjunction with the threshold-free cluster enhancement (TFCE) option³⁶ implemented within the TFCE toolbox in SPM8 software adjusted for age, sex, and site. Post hoc pairwise t-contrasts were generated to visualize group differences in (1) respective proband groups with respect to controls and (2) between proband groups. If significant differences were noted in any proband group compared with controls, then we sought to see if these regional differences were present in their respective unaffected relative groups. These analyses were repeated for both ALFF/fALFF bands, thresholded at P < .01 family wise error (FWE) adjusted for multiple comparisons across the whole brain and visualized using xjview (http://www.alivelearn.net/xjview8/). Supplementary voxelwise association analyses with PANSS scores were computed for both ALFF/fALFF bands. In addition, we also conducted association analyses with the Schizo-Bipolar Scale (SBS)³⁷ to see if ALFF/fALFF measures were associated with psychosis symptomology on a dimensional scale. However, for these supplementary analyses because PANSS and SBS were significantly associated with site, due to varying proportions of DSM diagnoses recruited at each, we chose not to include site as a covariate.

We also implemented an ANCOVA model to test main effects of site across controls only and a full-factorial model across all subjects to test for any significant diagnosis × site effects. For the latter model, all probands and relatives were clustered together. Site effects were noted for both indices across both bands throughout the cortex. In general, fALFF was more robust against site effects compared with ALFF. Main effect of site across controls is visualized in supplementary figure 1. Apart from a few minimal clusters (< 30 voxels), no significant diagnoses-by-site (D × S) interactions were seen (supplementary figure 2). For fALFF, only the slow-5 band showed interactions, mainly in cerebellum. ALFF interaction in both bands was in the temporal pole. We therefore suggest interpreting regional differences across groups keeping these interactions in mind.

Relative Risk Measures

To further validate the endophenotype status of ALFF/fALFF measures, relative risk scores were calculated on all unaffected relatives. The average fALFF/ALFF values for each post hoc analysis were extracted using the REX toolbox (http://web.mit.edu/swg/software.htm) and used for analysis. Significant relative risk (λ_1.5) scores were assessed for relatives by measuring the proportion of relatives and controls over/under 1.5 SDs from control mean and computing a Fisher’s exact P value for significance based on chi-square analyses.

Medication-Related Analyses

In order to reveal any medication effects in our study, we performed 2 types of supplementary analyses. First, we converted all available antipsychotic data to their respective chlorpromazine (CPZ) dosage equivalents for all probands, as prescribed by Andreasen et al.³⁸ We then computed a linear regression analysis within SPM8 to find associations with ALFF/fALFF signal in both bands. For the second analysis, we recoded available medication data into a binary “on”/“off” format for 4 main drug classes (antipsychotics, antidepressants, mood stabilizers [w/o lithium], and lithium). We then incorporated these into a multi-regression framework to assess their associations with ALFF/fALFF across each band separately. All medication data were based upon current status. Significance thresholds and covariates used for the above analyses were all similar to the main analysis reported above.

Results

In regards to head movement, values of maximum motion in the 3 translational and rotational directions were not significantly different between the control and proband groups. Also, average root mean square values of motion were not significantly different across these groups.

Controls vs Proband Differences

Slow-5 (0.01–0.027 Hz) ALFF.

We observed both reduced and increased regional ALFF in all 3 proband groups compared with HC. SZ demonstrated significantly lower regional ALFF in posterior regions including precuneus/cuneus and PCC. In contrast, SAD and PBP had decreased power more anteriorly in medial frontal gyrus and ACC. Increased power was observed in striatal and temporal structures including inferior/middle temporal gyrus, uncus, and parahippocampus (PHG) in all groups. In general, increased regional ALFF was more widespread in SZ followed by SAD and then PBP. The above effects were not noted in respective relatives.

Slow-4 (0.027–0.073 Hz) ALFF.

In contrast to slow-5 ALFF, all 3 proband groups showed reduced regional power in posterior regions including precuneus/cuneus, cingulate gyrus, and PCC. Also, reduced power in pre-/postcentral gyri was seen in both PBP and SZ, but not SAD. Similarly, increased slow-4 regional ALFF in proband groups was seen in temporal cortex, parahippocampal/hippocampal, and subcortical structures such as uncus, thalamus, and regions. In addition to the above, hyperactivity in SZ seemed be more widespread including superior/middle and inferior frontal gyri. Compared to controls, none of the relative groups showed significant slow-4 ALFF differences.

Slow-5 (0.01–0.027 Hz) fALFF

Hypo-fALFF was noted among all probands. Spatially, a similar pattern of anterior vs posterior shift as in slow-5 ALFF was noted. However, the anterior differences in SAD and PBP were more robust and covered more regions throughout lateral and medial frontal areas. SAD and PBP showed substantial overlap in terms of their frontal hypo-power. Regional power increases were limited to SZ and SAD and in similar regions as reported in ALFF, albeit to a lesser spatial extent. Again, no relative group showed a significant difference for fALFF slow-5 compared with HC.

Slow-4 (0.027–0.073 Hz) fALFF

fALFF slow-4 showed reduced power in all 3 proband groups, mainly in posterior regions comprising precuneus and inferior/superior parietal cortex. Increased slow-4 fALFF was more widespread than other indices and bands, spanning previously mentioned regions plus several additional frontal structures including ACC and superior and inferior frontal gyri. Again, relative groups showed no alterations compared with HC.

A snapshot of control vs proband findings across bands and metrics is provided in table 2.

Table 2.

Snapshot of Group Differences Across Different Indices and Bands

Note: Decreases in regional ALFF/fALFF are denoted by red arrows and increases by blue. The number of arrows is an approximate representation of the extent and strength of the difference noted with respect to the control group.

Between-Proband Differences

Between-proband group differences were only noted for slow-4 fALFF, with SZ showing increased power relative to BP in superior, medial, middle, and inferior frontal regions. Regional differences in probands relative to controls across both ALFF and fALFF indices and power bands are depicted in figures 1 and 2, respectively. Detailed differences are presented in supplementary tables 3 and 4.

Fig. 1.

Regional absolute and fractional amplitude of low-frequency fluctuations (fALFF/ALFF) differences noted in the slow-5 frequency band across traditional Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition diagnostic groups. Differences shown are with respect to controls and thresholded at P < .01 FWE corrected. Red areas correspond to differences noted in schizophrenia, yellow corresponds to psychotic bipolar disorder and green corresponds to SAD. Note the shared and unique differences across different diagnostic groups.

Fig. 2.

Regional absolute and fractional amplitude of low-frequency fluctuation (fALFF/ALFF) differences noted in the slow-4 frequency band across traditional Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition diagnostic groups. Differences shown are with respect to controls and thresholded at P < .01 FWE corrected. Red areas correspond to differences noted in schizophrenia, yellow corresponds to psychotic bipolar disorder and green corresponds to SAD. Note the shared and unique differences across the different diagnostic groups.

Supplementary voxelwise correlations between fALFF/ALFF measures and PANSS clinical scores yielded no significant results at FWE P < .01. However, significant SBS correlations with ALFF/fALFF were noted in both bands. Primarily larger SBS scores were associated with decreased ALFF/fALFF in regions including the cuneus/precuneus, inferior parietal lobe, and posterior cingulate and similarly positive SBS-ALFF/fALFF correlations were noted in frontal, temporal, and subcortical regions. Regions showing significant SBS correlations are shown in supplementary figure 3 and outlined in supplementary tables 5 and 6.

Relative Risk Scores

Supporting our imaging findings, no computed RR_1.5 (relative risk ratio) values were significant.

Medication Effects

No significant associations were noted in ALFF/fALFF with CPZ equivalents for either band. For binary coded associations, we saw no significant effects of medication classes for slow-5 ALFF/fALFF and slow-4 fALFF. Minimal effects (<30 total voxels) of medication were noted for antipsychotics for slow-4 ALFF, with results showing antipsychotics related to reduced ALFF in lingual and cuneus/precuneus regions.

Discussion

The goal of this large-scale multisite study was to evaluate for the first time whether local spontaneous fluctuations measured under different frequency bands using ALFF/fALFF are a valid psychosis biomarker and/or endophenotype in a multisite sample. As seen from Table 3 which summarizes regional findings across prior studies, most ALFF/fALFF studies in SZ or BP investigated the traditional frequency range of 0.01–0.08 Hz.⁷ Pentonnen and Buzsaki observed that neural oscillations in adjacent frequency bands within the same structure do not co-occur and that signals in such frequency bands might be potentially influenced by different physiological mechanisms.^8,11,39 In line with the above evidence, we adopted Buzsaki’s nomenclature to subdivide and analyze frequency-specific effects in the slow-5 (0.01–0.027 Hz) and slow-4 (0.027–0.073 Hz) rs-fMRI bands in psychosis for the first time. Importantly, we chose to excluded higher frequency signal (>0.1 Hz) that might contain more physiological noise contamination.⁷

Table 3.

Summary Table of Comparable Studies and Their Findings That Have Investigated ALFF/fALFF in the Realm of Psychosis

Note: fALFF/ALFF, absolute and fractional amplitude of low-frequency fluctuations; ACC, anterior cingulate cortices; OFC, orbito-frontal cortex; PCC, posterior cingulate cortex; SZ, schizophrenia; MFG, medial frontal gyrus; MOG, medial occipital gyrus; DLPFC, dorsolateral prefrontal cortex; SPM, statistical parametric mapping; FG, frontal gyrus.

Neuronal oscillations have been reported to bias various functions, including short- and long-term information consolidation, input selection, synaptic plasticity, psychological representation, and learning.¹¹ Even though the exact molecular mechanisms driving resting state spontaneous fluctuations are still unknown, more recent concurrent EEG-fMRI studies in animals and humans have presented intriguing evidence linking rs-LFOs directly to gamma, delta, and alpha EEG power.^19–21,25 Concurrent EEG-rsfMRI connectivity based studies have shown that a subset of rs-fMRI networks are positively modulated by delta/theta and upper beta/lower gamma EEG responses and negatively modulated by EEG alpha.¹⁴ These data could thus help elucidate the origins and mechanisms of LFOs. Synchronization of gamma power has been shown to influence object recognition and short-term memory.⁴⁰ Alpha waves have been recently proposed to play an active role in network coordination and communication.⁴¹ Similarly, delta waves are suggested to aid in the formation of declarative and explicit memory formation.⁴² Given the above EEG-derived relationships and our current results, it is possible that ALFF/fALFF alterations indicate deficiencies related to object encoding, memory, and general network coordination leading to psychoses.

Recent evidence also suggests that 2 major personality domains (neuroticism and extraversion) are associated with slow-4 and slow-5 ALFF in key emotional regulatory regions, including hippocampus, superior temporal gyrus, and MPFC.⁴³ Interestingly, these same personality traits have also been linked to increased risk of psychosis.⁴⁴ In the context of our current results, it might be possible that abnormal modulation of ALFF/fALFF via the fronto-striatal loop in SAD/PBP contributes to abnormal personality and emotional regulation in these disorders.

Generally, our results replicated the most consistent prior results of studies evaluating ALFF/fALFF in SZ, showing reduced power in brain regions including medial prefrontal, lingual, middle occipital, cuneus/precuneus, fusiform, PCC, and cerebellum and increased ALFF/fALFF mainly in limbic/subcortical regions including amygdala, PHG, putamen, and thalamus.^26,45,46 Reduced ALFF slow-4 power in the pre-/postcentral gyri in PBP was consistent with findings from Liu et al^28,29 that investigated the traditional band. We also noted increased ALFF power in PHG in PBP consistent with elevated limbic ALFF power as noted in the 2 studies of BP.^28,30 Although we did not evaluate resting state functional connectivity per se, our results demonstrated more abnormal regional ALFF/fALFF activation in fronto-striatal regions in PBP and SAD compared with SZ. This observation is in line with the much-replicated fronto-striatal loop dysfunction implicated in bipolar disorder.^3,47–50 Also, for the first time, we explored differences in SAD using dual-band ALFF/fALFF, where regional differences in ALFF and fALFF in SAD resembled PBP more closely than SZ. Interestingly, we found significant associations with SBS but not PANSS, suggesting that abnormalities in several of the regions reported in the study tracked consistently with degree of proximity to prototypic SZ and PBP when assessed in a dimensional manner.

Proband Differences Across Bands (Slow-5 and Slow-4)

Across bands, in SZ, both slow-4 ALFF and slow-5 ALFF captured predominantly posterior precuneus/cuneus, PCC differences. However, in SAD and PBP, slow-4 ALFF elucidated more anterior hypo-ALFF vs slow-5 which showed a pattern of more posterior hypo-ALFF. Hyper-ALFF was mostly consistent across bands and in subcortical and temporal regions for all proband groups. Likewise, a similar albeit more marked pattern of anterior vs posterior shift in fALFF reduction was seen for SAD and PBP across slow-5 vs slow-4 bands, respectively. Consistent with prior observations, we also noticed increased basal ganglia LFO magnitude differences in slow-4 compared with slow-5 in psychosis probands.¹² Interestingly, similar slow-4 neuronal fluctuation recordings in the basal ganglia from awake, anesthetized rats have shown to be selectively modulated by low doses of dopaminergic drugs such as those used to treat psychiatric disorders.^51,52

Proband Differences Across Indices (ALFF and fALFF)

Overall, fALFF was more sensitive in capturing group differences in probands relative to controls. fALFF differences were also more widespread compared with ALFF across all proband groups, perhaps because fALFF measures the fractional power of a particular frequency band and might thus be more sensitive indicator of the relative strength of LFOs than ALFF. Our findings of fALFF being more sensitive are consistent with previous studies that reported both metrics.^16,46,53

Differences Between Proband Groups

Most between-proband group differences did not survive multiple comparison correction, except for slow-4 fALFF, where SZ relative to PBP, had significantly higher fALFF power in the superior, medial, middle, and inferior/orbital frontal cortex (Brodmann Areas 10, 11, 47), suggesting that specific alterations in physiological mechanisms supporting very low-frequency oscillations might impact SZ. Similar regions differentiated SZ and PBP in some of our previous large-scale connectivity studies that utilized a subset of subjects from the current study.^2,3 Chai and colleagues⁵⁴ reported differential connectivity patterns between MPFC and ventral prefrontal area that distinguished bipolars from SZ. Similarly, Liu et al⁵⁵ reported a dorsal vs ventral prefrontal cortex dissociation in amygdala-prefrontal neural system abnormalities between SZ and PBP.

Differences in Unaffected Relatives of Probands

Neither metric showed differences in any relative group after multiple comparison correction. Further, secondary analysis also looked at differences whole-brain differences in relatives (not just limited to regions affected in probands) finding no such differences. Additionally, in support of our imaging findings, relative risk scores were not significant. Together, our results suggest that ALFF/fALFF are illness markers rather than psychosis endophenotypes.

Site Effects and Interactions

Similar to a recent large-scale multisite study by Turner et al, we noted widespread ALFF/fALFF site/scanner effects. However, Turner et al⁴⁶ concluded that ALFF/fALFF effects were robust to multisite variability. More importantly, we only noted minimal D × S interactions, further suggesting that group differences reported in our study were robust to site/scanner variations.

Previous Studies Using Subsets of the Current Data Set

Given the richness of the data set, it is important to note that 4 previous studies have used subsets of the current data set to probe different rs-fMRI metrics. Meda et al³ and Khadka et al² used data from the Hartford B-SNIP site to evaluate independent component analysis-related between- and within-network functional connectivity across SZ, PBP, and their relatives. Unschuld et al⁵⁶ used data from the Baltimore cohort to evaluate seed-based rs-fMRI connectivity across SZ and their relatives in the context of cognitive dysfunction. It is important to note that the above studies probed global functional connectivity^2,3,49,57 while the present study evaluated a local voxelwise metric derived in the frequency domain that has shown promise as a biomarker in a variety of neuropsychiatric disorders.^27,45,58 The most comparable analysis to the current study was by Lui et al that utilized data from the Chicago B-SNIP site. However, that analysis was performed only on ALFF and in the traditional frequency band. In contrast, the current study uses a much larger sample and more importantly explored frequency-specific alterations across both ALFF and fALFF. Notably, ALFF alterations noted in Lui et al were similar to the current study but encompassed far fewer regions (possibly due to lack of power and overall clumping of frequency bands). Also, similar to the current report, Lui et al⁵⁹ found no ALFF differences in relatives.

Our study had numerous advantages: (1) to our knowledge, this is the largest study to date to study ALFF/fALFF in psychosis probands and relatives simultaneously, (2) this was the first study to explore both ALFF/fALFF and slow-4/slow-5 bands to show differential spatial patterns and effects in a large population, (3) we analyzed SAD subjects as a separate diagnostic group in the context of ALFF/fALFF evaluation for the first time, and (4) we used a nonparametric statistical approach coupled with a recently developed thresholding technique that removes potential bias on employing cluster level thresholds. The study also had caveats: (1) we noted minimal but significant medication effects for slow-4 ALFF. Despite our seeing no medication correlates of ALFF/fALFF within probands in other bands, these findings could represent, in part, an effect of dopamine antagonist medication on brain, especially because these drugs have been shown to have regional effects on slow-4 neuronal fluctuations in basal ganglia and (2) subtle but significant regional D × S effects were noted.

Conclusion

In summary, our data illustrate that ALFF and fALFF data capture both shared and unique effects across PBP, SAD, and SZ. We were able to show that ALFF/fALFF measures are strong biomarkers but not endophenotypes for psychosis. Using a dimensional approach focused on psychosis, we showed SAD to be more similar to PBP than SZ in the targeted biomarker measures. Also, differential sensitivity to psychosis was noted across different frequency bands, most likely due to differences in underlying physiological mechanisms at play. We also pinpointed specific frequency-based neural substrates of psychosis that could be potential “psychosis signatures” and targeted for therapeutic interventions in these debilitating disorders. Future resting state studies should therefore strongly consider presenting data separated by different frequency bands, especially when examining BP and SAD groups.

Supplementary Material

Supplementary material is available at http://schizophreniabulletin.oxfordjournals.org.

Funding

National Institute of Mental Health (NIMH) (MH077851, MH078113, MH077945, MH077852, MH077862); National Institute of General Medical Sciences (NIGMS) (P20GM103472); Sunovion (to M.S.K.); Janssen, Takeda, Bristol-Myers Squibb (BMS), Roche, and Lilly (to J.A.S.).