A Comprehensive Characterization of the Phospholipid and Cholesterol Composition of the Uncinate Fasciculus in the Human Brain: Evidence of Age‐Related Alterations

ABSTRACT

The uncinate fasciculus (UF) is a long‐range association fiber tract connecting the anterior temporal lobe with the orbitofrontal cortex and has been linked to a multitude of physiological and pathophysiological conditions such as aging, epilepsy, and the vulnerability to psychopathology posed by a history of childhood abuse (CA). Since the myelin sheath is highly enriched in lipids, changes in white matter (WM) microstructure observed via neuroimaging may reflect alterations in the myelin lipid profile. Given that the UF does not exist in rodents, its molecular properties are highly understudied. Therefore, we sought to quantify the phospholipid FA and cholesterol quantities of the human postmortem UF and evaluate any lipid‐related or myelin‐constituent gene/protein changes associated with age and history of CA. UF samples were analyzed from individuals with depression who died by suicide with (DS‐CA) or without (DS) severe CA, and control individuals (CTRL), with an age span of 15 to 85 years. Phospholipids were separated by thin‐layer chromatography; FAs and nonderivatized cholesterol were quantified by gas chromatography‐flame ionization detection. The relative expression of myelin‐constituent genes and proteins was measured by RT‐qPCR and immunoblotting, respectively. We found no robust relationships between CA or depression and lipid measures or myelin‐constituent gene/protein levels. In contrast, phospholipids showed pronounced age effects that differed by fraction, with an overall trend of monounsaturated FAs increasing and long‐chain omega‐6 polyunsaturated FAs decreasing with age. The expression of most myelin‐constituent genes and proteins declined with age; PLP1 and MAG showed significant decreases. Therefore, changes in lipid composition and lipid‐protein interactions likely contribute to age‐related myelin deficits and may in part underlie age‐associated cognitive decline.

The uncinate fasciculus (UF), which serves to connect the anterior temporal lobe and the orbitofrontal cortex, is a highly understudied white matter fiber bundle. In this study, we characterized the fatty acid (FA) profile of myelin phospholipids in a fraction‐specific manner, as well as the cholesterol concentrations, in the human postmortem UF. In addition, we quantified myelin‐constituent genes/proteins and correlated these with the lipid metrics. We observed that each phospholipid fraction showed its own unique aging fingerprint, there was an overall pattern showing increased levels of monounsaturated FA and decreased levels of omega‐6 polyunsaturated FA with age. We did not identify any associations with childhood abuse, indicating a divergence from what has been previously observed in human cortical white matter. The UF graphic was adapted from Yeh et al. (2018), “Population‐averaged atlas of the macroscale human structural connectome and its network topology,” NeuroImage, 178, 57–68; http://brain.labsolver.org/, via Wikimedia Commons (https://commons.wikimedia.org/wiki/File:Uncinate_Fasciculus.jpg), CC BY‐SA 4.0. CerPCho, sphingomyelin; ChoGpl, choline glycerophospholipids; CNP, 2′,3′‐cyclic nucleotide 3′‐phosphodiesterase; EtnGpl, ethanolamine glycerophospholipids; MAG, myelin‐associated glycoprotein; MBP, myelin basic protein; MOBP, myelin‐associated oligodendrocyte basic protein; MOG, myelin oligodendrocyte glycoprotein; MUFA, monounsaturated fatty acid; PLLP, plasmolipin; PLP1, proteolipid protein; PtdIns, phosphatidylinositol; PtdSer, phosphatidylserine; TL, total lipid.

Abbreviations

CA

childhood abuse

CerPCho

sphingomyelin

ChoGpl

choline glycerophospholipids

CNP

2′, 3′‐cyclic nucleotide 3′‐phosphodiesterase

CTRL

control

DS

depressed suicides

DS‐CA

depressed suicides with a history of childhood abuse

EtnGpl

ethanolamine glycerophospholipids

MAG

myelin‐associated glycoprotein

MBP

myelin basic protein

MOBP

myelin‐associated oligodendrocyte basic protein

MOG

myelin oligodendrocyte glycoprotein

PLLP

plasmolipin

PLP1

proteolipid protein

PtdIns

phosphatidylinositol

PtdSer

phosphatidylserine

RRID

research resource identifiers

TL

total lipid

WM

white matter

1. Introduction

Approximately 5% of all human genes are devoted to lipid synthesis (van Meer et al. ²⁰⁰⁸) and when compared to nonneuronal tissues, the brain demonstrates a highly distinct lipid profile (Bozek et al. 2015). In the human neocortex, concentrations of brain‐enriched lipids have evolved three times faster when compared to chimpanzees (Bozek et al. 2015), pointing to an evolutionary significance for lipids in higher order cognitive function. White matter (WM) tracts, or fiber bundles, are collections of myelinated axons that transmit electric signals across distal brain regions and serve as “information highways”. The myelin sheath is comprised of 70%–85% lipids by dry weight (Williams and Deber 1993), and the specific composition and arrangement of lipids is known to influence its membrane properties and therefore its functional roles in brain circuits (Poitelon et al. 2020). For example, changes in myelin lipids such as the quantity of cholesterol and composition of fatty acids (FA) in phospholipids can impact membrane properties, including compactness, stability, and permeability (Naudí et al. 2015; Poitelon et al. ²⁰²⁰; Chrast et al. ²⁰¹¹). Myelin lipids are estimated to be comprised roughly of 40% phospholipids, 40% cholesterol, and 20% glycolipids (O'Brien ¹⁹⁶⁵).

Despite their importance in both health and disease, there is very little characterization of long‐range WM tracts across the lifespan in the human brain. One particularly interesting and understudied WM tract is the uncinate fasciculus (UF), a long‐range association fiber bundle that connects the anterior temporal lobe with the orbitofrontal cortex. Since this tract does not exist in rodents, its molecular properties have yet to be characterized (Lebel et al. 2012). Furthermore, the development of the UF is highly protracted and is one of the last tracts to fully mature with respect to its microstructural properties. Interestingly, it has been associated with both the underlying neurobiology of childhood trauma and the pathophysiology of depression via neuroimaging studies (Eluvathingal et al. 2006; Gur et al. ²⁰¹⁹; Hanson et al. ²⁰¹⁵; Ho et al. ²⁰¹⁷; Xu et al. ²⁰²³). Therefore, a comprehensive characterization of this tract is warranted.

Dysregulation in brain WM across a variety of limbic regions has been reported in both depression and in individuals with a history of childhood abuse (CA). However, it is unclear whether WM changes can be considered part of the pathophysiology of depression or whether they represent a biologically embedded risk factor for future psychopathology associated with experiencing CA. Notably, our previous work has shown that depressed suicides with a history of CA are associated with increased concentration of FAs in the arachidonic acid (ARA) synthesis pathway and corresponding gene expression changes in anterior cingulate cortex (ACC) gray matter (Perlman et al. 2021).

Importantly, WM tracts have recently been termed “hotspots” of aging due to the profound molecular changes observed in WM with advancing age, including alterations in gene expression, protein abundance, and crucially, lipid composition (Allen et al. 2023; Hahn et al. ²⁰²³; Yu et al. ²⁰²⁰; Ding et al. ²⁰²¹; Dimovasili et al. ²⁰²⁴; Wuerch and Yong ²⁰²⁵; Groh and Simons ²⁰²⁵). These molecular changes may in part underlie the cognitive deficits accrued with aging, as WM tracts regulate long distance communication in the brain (Coelho et al. 2021). In fact, there is neuroimaging evidence to suggest that the UF microstructure declines with age (Mendez Colmenares et al. 2024) and can mediate cognitive flexibility and performance in psychomotor tasks (Wolfe et al. 2024; Zahr et al. ²⁰⁰⁹).

Therefore, this study had 2 complementary aims. Firstly, to perform a fundamental, in‐depth characterization of phospholipid FA profiles and cholesterol composition of the UF. Secondly, with the addition of complementary measurements of key myelin‐constituent genes and proteins, to examine if CA‐, depression‐ or age‐related changes are present in the UF.

2. Methods

2.1. Subject Information

This research received ethical approval from the Douglas Mental Health University Institute's Research Ethics Board (IUSMD‐18‐10) and was not preregistered given its exploratory nature. Group‐matched brain samples, acquired with informed consent from the donors' next of kin, were provided by the Douglas‐Bell Canada Brain Bank. Fresh‐frozen UF tissue was dissected from the left hemisphere. Data from the coroner's office such as cause of death, time of death, and blood toxicology were obtained along with medical records when available. This information was complemented with data collected during a standard psychological autopsy procedure, as described elsewhere (Dumais et al. 2005; Perlman et al. ²⁰²¹). Adapted versions of the Structured Clinical Interview for DSM‐IV (SCID) and Childhood Experience of Care and Abuse (CECA) questionnaires were administered as part of this procedure. A panel of clinicians, which includes a psychiatrist, synthesized all the evidence available to converge on an Axis I diagnosis and was used to classify brain donors into 3 groups: depressed suicides with history of CA (DS‐CA), depressed suicides without CA (DS), and controls without depression nor a history of CA (CTRL). Neurodegenerative disorders constituted an exclusion criterion for all groups.

For all the gene and protein data, the UF was dissected at the point of intersection between the frontal lobe and the temporal lobe, lateral to the amygdala. Due to tissue availability limitations, all the lipid data were generated from the UF dissected at Brodmann area 38 (temporal pole). These dissection sites were selected because the UF could be clearly distinguished from other concurrent tracts such as the inferior frontal occipital fasciculus. The samples were stored at −80°C until biochemical analyses.

2.2. Lipid Extraction and Phospholipid FA Quantification

We modified the protocol from our previous study (Perlman et al. 2021), to include the collection and measurement of all major phospholipid classes: choline glycerophospholipids (ChoGpl), ethanolamine glycerophospholipids (EtnGpl), phosphatidylserine (PtdSer), phosphatidylinositol (PtdIns), and sphingomyelin (CerPCho). Experimenters were unaware of subject groups while running the experiments. First, 50–70 mg of fresh‐frozen UF tissue was dissected per sample. An adapted Folch method was used to extract total lipids (TL) (Folch et al. 1957), at which point an internal standard cocktail containing di‐17:0 ChoGpl (Avanti Research; Catalog No: 850360), di‐17:0 EtnGpl (Avanti Research; Catalog No: 830756), and 5‐alpha cholestane (Sigma‐Aldrich; Catalog No: C8003) was added (Table S1). Then, 70 μL of the 500 μL total lipid extract (TL) was loaded into scored lanes of silica H‐plates to perform thin layer chromatography (TLC) (Miles Scientific; Catalog No: P10011). The TLC mobile phase was composed of 30:9:25:6:18 chloroform (Millipore Sigma; Catalog No: CX1055‐9): methanol (Sigma‐Aldrich; Catalog No: 179337): 2‐propanol (Sigma‐Aldrich; Catalog No: 190764): 0.25% KCl (Sigma‐Aldrich; Catalog No: P3911): triethylamine (Sigma‐Aldrich; Catalog No: T0886) (v/v). Phospholipid bands were visualized and marked under UV light after being sprayed with 0.1% (w/v) 8‐anilino‐1‐naphthalene sulfonic acid (Sigma‐Aldrich; Catalog No: A1028) and identified using a reference standard. Using a scraping tool, the silica gel within the respective marked bands (each containing a different fraction) was individually scraped onto weighing paper and then into glass test tubes. Specific amounts of 17:0 free fatty acid were added as internal standards to the tubes containing PtdIns, PtdSer, and CerPCho (Table S1) as purified, form‐specific internal standards were commercially unavailable. Each fraction was methylated with 14% BF₃‐MeOH (Sigma‐Aldrich; Catalog No: B1252), and all were incubated for 60 min, except for CerPCho which was incubated for 90 min, all at 100°C. Samples from each fraction, as well as the TL, were measured by gas chromatography‐flame ionization detection (GC‐FID) (Varian 430 gas chromatograph (Bruker, Billerica, MA, USA)). Fatty acid methyl esters were eluted on a DB‐FFAP column (30 m × 0.25 mm i.d. × 0.25 μm film thickness) (J&W Scientific, Agilent Technologies). Run program was set at initial temperature of 50°C (1 min), 130°C ramp at 30°C/min, 175°C ramp at 10°C/min, 230°C ramp at 5°C/min (9.5 min hold), and final 240°C at 5oC/min (11.13 min hold). The TL, ChoGpl, and EtnGpl fractions were run on split injection mode, while CerPCho, PtdSer, and PtdIns fractions were run on splitless injection mode.

Chromatograms were analyzed with the CompassCDS software and annotated based on the GLC 569B external reference standard (Nu‐Check Prep Inc.). Concentrations of each FA in μg/g were calculated based on peak comparison to the internal standard. Furthermore, relative percentages of each FA were calculated. Peaks that could not be reliably annotated were excluded—for example, C20:5n‐3 and C20:2n‐6 were detected at trace amounts in CerPCho but not reliably enough to quantify.

2.3. Cholesterol Quantification

Nonderivatized cholesterol was run on a HP‐5 ms capillary column (30 m × 0.25 mm i.d. × 0.25 μm d_f) (J&W Scientific, Agilent Technologies) using a Varian 430 GC‐FID. The injector and detector ports were set to 250°C and 300°C, respectively. Cholesterol was eluted using a temperature program set initially at 100°C for 1 min, followed by increases at 15°C/min for 17 min until reaching a final temperature of 280°C and completing the run of 30 min. The carrier gas was helium set at a constant flow rate of 1 mL/min. Cholesterol was quantified using peak comparison to the internal standard, 5α‐cholestane. Annotations were performed on CompassCDS.

2.4. Ratios and Indices

We also computed several relevant ratios and indices, including the ratio of omega‐6/omega‐3 FAs and the quantity of highly unsaturated FAs (HUFA; sum of FAs with 20 or more carbons and 3 or more double bonds). Then, we calculated a variety of indices, including the peroxidation index (Hulbert et al. 2007), which quantifies the susceptibility of a given lipid to oxidation, with the following formula:

$$\begin{array}{c}\text{Peroxidation index}=0.025\times \left(\%\text{monoenoics}\right)+1\times \left(\%\text{dienoics}\right)\hfill \\ +2\times \left(\%\text{trienoics}\right)+4\times \left(\%\text{tetraenoics}\right)\hfill \\ +6\times \left(\%\text{pentaenoics}\right)+8\times \left(\%\text{hexaenoics}\right)\hfill \end{array}$$

Furthermore, we calculated the unsaturation index (Hulbert et al. 2007) with the following formula:

$$\begin{array}{c}\text{Unsaturation index}=1\times \left(\%\text{monoenoics}\right)+2\times \left(\%\text{dienoics}\right)\hfill \\ +3\times \left(\%\text{trienoics}\right)+4\times \left(\%\text{tetraenoics}\right)\hfill \\ +5\times \left(\%\text{pentaenoics}\right)+6\times \left(\%\text{hexaenoics}\right)\hfill \end{array}$$

The final index calculated was the chain length index with the formula:

$$\begin{array}{c}\text{Chain length index}\hfill \\ =\sum \left(\text{Chain length}\times \%\text{fatty acid with that chain length}\right)\hfill \end{array}$$

2.5. Expression of Myelin‐Constituent Genes

RNA was extracted from homogenized UF tissue in Qiazol lysis buffer using the Qiagen RNEasy Lipid Tissue Mini Kit (Catalog No: 74804). DNA was removed with Qiagen RNase‐Free DNAse (Catalog No: 79254) to the extracted RNA. A standardized amount of purified RNA from each UF sample was reverse transcribed with MLV Reverse Transcriptase Kit (Thermo Fisher Scientific, Catalog No: 28025–013), yielding UF cDNA for each subject. A standard curve was made to determine the relative amount of cDNA amplification. The forward and reverse primers were diluted and mixed with Applied Biosystems PowerUp SYBR Green master mix (Catalog No: A25742). RT‐qPCR was carried out with the Applied Biosystems QuantStudio 6 Flex PCR machine. The output was analyzed with the QuantStudio Real‐Time PCR software. Replicates with an expression higher than ±0.3 standard deviation from the mean expression values were withdrawn from analysis. See Table S2 for details of the primers used in RT‐qPCR.

2.6. Expression of Myelin‐Constituent Proteins

Protein lysates for each subject were generated by homogenizing UF tissue in RIPA buffer. Protein concentration of each lysate was measured using the BCA Protein Assay Kit (Thermo Fisher Scientific, Catalog No: 23227) and concentration was determined with a Tecan Spark10M plate reader. An equal quantity of protein for each subject was run on a BioRad Mini‐PROTEAN TGX Stain‐Free 4%–20% gel (Catalog No: 4568095) and separated by way of SDS‐PAGE. The wash buffer comprised of PBS (0.01 M) with 0.05% Tween‐20. The electrophoresis was run at 135 V for approximately 60 min at ambient temperature. Next, the proteins in the gel were transferred onto an Amersham Protran Premium 0.2 μm nitrocellulose blotting membrane (Amersham/GE healthcare [now Cytiva] Catalog No: 10600097) with the BioRad Trans‐Blot Turbo Transfer System (Catalog No: 1704150). This membrane was imaged in the BioRad Chemidoc Touch Imaging System (Catalog No: 12003153) to obtain a quantification of total protein expression.

The membrane was blocked for 1 h with 5% skim milk, then primary antibodies in 1% skim milk blocking solution were added and incubated overnight at 4°C. The antibody details are listed in Table S3. Then, HRP‐conjugated secondary antibodies (Amersham/GE Healthcare [now Cytiva] Catalog No: NA931) in 1% skim milk were incubated for 1 h at room temperature. After the incubations, the membranes were exposed to electrochemiluminescent (ECL) fluid (Catalog No: 1705060) for 2 min, and the chemilumiscent signal generated was captured with the BioRad Chemidoc Touch Imaging System imager and measured using the ImageLab software.

2.7. Statistics

Statistics were performed using R version 4.4.2. ANCOVA models were constructed with the following factors: group, age, sex, brain pH, and postmortem interval (PMI), and the presence/absence of a documented lipid‐related condition. For this study, specific attention was made to the donor's documented lipid‐related conditions including any listed history of elevated cholesterol, coronary artery disease, dyslipidemia, type 2 diabetes or medications to treat those conditions. This type of information is relevant when considering lipid quantities, as for example, statins lower brain cholesterol by way of reducing both de novo synthesis and turnover rate (Cibičková 2011). If the donor either had an antidepressant prescription in 3 months before their death or if they had antidepressants detected in their body at time of death, they were considered to be “positive” for antidepressants. No outlier tests were conducted in order to avoid artificially restricting naturally variable human brain data, but failed chromatograms were excluded (n = 1 for PtdIns, n = 7 for CerPCho). p‐values were corrected for multiple comparison within each predictor across FAs within in a phospholipid fraction using the Benjamini‐Hochberg (BH) method. Pearson's correlations were performed and represented in correlation plots (R corrplot package). p < 0.05 was selected as the threshold for nominal statistical significance across analyses. An LLM AI tool was implemented to refine a subset of the R code used for analysis and visualization.

3. Results

3.1. Fundamental Characterization of UF Lipids

Table S4 contains the subject information for the UF lipids cohort (n = 80). The mean total concentrations for each of the 5 main phospholipid fractions, cholesterol, and the TL are plotted in Figure 1A. Of the phospholipids, EtnGpl was the most abundant followed closely by ChoGpl while PtdIns was the least abundant phospholipid. The mean cholesterol concentration was 12611.59 ± 5989.21 μg/g, which is the most abundant individual lipid species measured. The distributions of FA classes of each phospholipid and the TL are illustrated in Figure 1B. The FA class distribution in each isolated phospholipid fraction matches the expected overall pattern based on the literature of WM (e.g., EtnGpl is enriched in omega‐6 FA and CerPCho is enriched in saturated FA [SFA]) (O'Brien et al. ¹⁹⁶⁴; Svennerholm ¹⁹⁶⁸; Taha et al. ²⁰¹³; Heipertz et al. ¹⁹⁷⁷; Igarashi et al. ²⁰¹¹; Söderberg et al. ¹⁹⁹⁰). The concentration data for each FA for each phospholipid fraction as well as the TL (represented as mean ± standard deviation) is available in Table 1. In the TL, the most abundant FA is C18:1n‐9 (OLA, 8468.14 ± 1925.87 μg/g) followed by C18:0 (STA, 6086.21 ± 821.10 μg/g), then C16:0 (PAM, 4974.11 ± 608.19 μg/g), C22:4n‐6 (AdA, 2891.96 ± 925.67 μg/g), and C20:4n‐6 (ARA, 2209.72 ± 495.80 μg/g). When performing a principal component analysis on the FAs detected in every phospholipid fraction, the fractions clearly separate along the first 2 principal components, reflective of their distinct FA “signatures”. Furthermore, we validated the coeluting peak of C24:0 and C22:6n‐3 in CerPCho, finding that 98.9% of the total area mapped to C24:0, and 1.1% to C22:6n‐3 (Supplementary Results, Table S2, Figure S2).

FIGURE 1.

Mean totals and fraction‐specific characterization of FA classes. (A) Bar plot showing the mean concentration ± standard error for the totals of all measured phospholipid fractions, cholesterol, and the TL in ascending order (n = 80). TL refers to the total FA, not just the FAs derived from phospholipids. Therefore, the FAs in the TL also come from glycolipids, diacylglycerols, and other lipid species. (B) Pie charts showing the mean relative percentage of MUFA (pink), omega‐6 (teal), omega‐3 (green), and SFA (purple) FAs for each phospholipid fraction as well as the TL (n = 80). CerPCho, Sphingomyelin; ChoGpl, Choline glycerophospholipids; EtnGpl, Ethanolamine glycerophospholipids; MUFA, Monounsaturated fatty acid; PtdIns, Phosphatidylinositol; PtdSer, Phosphatidylserine; SFA, Saturated fatty acid; TL, Total lipid.

TABLE 1.

Concentration (μg/g) data represented as mean ± standard deviation.

Fatty acid	ChoGpl	EtnGpl	PtdIns	PtdSer	CerPCho	TL
C 14:0	84.45 ± 14.19	13.37 ± 5.10	2.08 ± 1.55	3.16 ± 1.28	4.62 ± 1.90	98.38 ± 26.68
C 16:0; PAM	3491.83 ± 422.61	445.39 ± 71.65	95.88 ± 47.76	102.67 ± 32.28	129.10 ± 43.44	4974.11 ± 608.19
C 18:0; STA	1191.04 ± 163.53	1218.26 ± 356.44	166.84 ± 102.84	2491.91 ± 774.84	805.75 ± 247.94	6086.21 ± 821.10
C 20:0	7.69 ± 1.61	15.15 ± 4.37	1.05 ± 0.54	10.20 ± 2.60	33.90 ± 10.23	101.72 ± 36.97
C 22:0	2.34 ± 0.76	5.30 ± 1.54	0.81 ± 0.33	4.15 ± 1.20	41.16 ± 15.97	96.63 ± 67.19
C 23:0	ND	ND	0.61 ± 0.32	ND	54.08 ± 22.10	127.93 ± 65.78
C 24:0	6.90 ± 4.46	8.24 ± 5.62	2.16 ± 1.02	8.58 ± 34.39**	164.58 ± 73.57	2194.44 ± 827.68
Total SFAs	4784.24 ± 559.92	1705.71 ± 370.04	269.43 ± 142.71	2620.68 ± 798.89	1233.18 ± 373.98	13729.69 ± 1725.42
C 16:1n‐7	119.26 ± 25.55	44.55 ± 19.78	4.02 ± 2.04	7.75 ± 2.50	1.33 ± 0.52	1119.12 ± 402.35
C 18:1n‐7	617.46 ± 101.66	449.11 ± 120.35	36.62 ± 19.71	222.51 ± 70.62	4.66 ± 1.77	1562.66 ± 325.21
C 18:1n‐9 (OLA)	3327.61 ± 608.87	2678.00 ± 778.62	82.60 ± 50.78	1918.89 ± 653.95	27.30 ± 7.96	8468.14 ± 1925.87
C 20:1n‐9	67.07 ± 17.13	435.52 ± 157.47	5.98 ± 4.24	200.28 ± 79.58	1.61 ± 0.64	762.52 ± 285.36
C 22:1n‐9	10.28 ± 4.40	37.65 ± 13.24	5.27 ± 3.35	13.17 ± 4.57	9.37 ± 9.07	71.86 ± 29.88
C 24:1n‐9	10.64 ± 3.78	6.92 ± 2.50	0.98 ± 0.63	4.79 ± 1.68	767.18 ± 309.15	229.46 ± 111.62
Total MUFAs	4152.32 ± 731.78	3651.76 ± 1058.77	135.46 ± 75.62	2367.38 ± 796.84	811.43 ± 314.72	12433.13 ± 2947.73
C 18:2n‐6; LNA	65.58 ± 15.36	24.88 ± 6.12	4.24 ± 2.57	6.84 ± 2.28	1.97 ± 0.51	125.87 ± 29.38
C 20:2n‐6	8.87 ± 9.45	24.69 ± 11.53	1.37 ± 1.23	7.02 ± 3.15	ND	161.07 ± 45.94
C 20:3n‐6; DGLA	49.09 ± 11.60	134.87 ± 38.02	12.55 ± 9.49	40.34 ± 14.81	5.49 ± 2.11 * coeluted with C21:0	312.20 ± 86.98
C 20:4n‐6; ARA	341.26 ± 69.69	1257.35 ± 215.24	109.15 ± 79.98	122.63 ± 37.30	2.64 ± 1.54	2209.72 ± 495.80
C 22:2n‐6; DDA	6.68 ± 2.87	19.90 ± 7.14	1.08 ± 0.47	ND	1.81 ± 1.55	375.41 ± 171.98
C 22:4n‐6; AdA	73.79 ± 11.30	2053.30 ± 433.09	18.47 ± 14.45	230.68 ± 65.10	3.53 ± 1.68	2891.96 ± 925.67
C 22:5n‐6; DPAn‐6	10.10 ± 3.60	117.70 ± 41.52	1.18 ± 0.88	42.68 ± 20.14	0.40 ± 0.20	233.44 ± 87.76
Total N‐6	555.38 ± 95.31	3632.68 ± 572.61	148.03 ± 103.75	450.21 ± 125.32	10.35 ± 3.68	6310.38 ± 1570.11
C 20:5n‐3; EPA	ND	ND	ND	ND	ND	ND
C 22:5n‐3; DPAn‐3	5.78 ± 1.61	63.40 ± 18.04	0.72 ± 0.54	9.61 ± 2.88	ND	127.81 ± 54.15
C 22:6n‐3; DHA	97.76 ± 26.34	985.89 ± 381.64	11.86 ± 7.39	284.23 ± 142.19	ND**	1723.21 ± 828.39
Total N‐3	103.54 ± 27.19	1049.28 ± 385.77	12.58 ± 7.88	293.84 ± 143.59	ND**	1851.02 ± 855.90
Total FA	9595.48 ± 1243.32	10039.43 ± 1434.06	565.51 ± 319.92	5732.12 ± 1665.76	2060.45 ± 663.03	34324.21 ± 6025.21

Abbreviation: ND, nondetectable.

^*Coelution with C21:0.

^**Coelution with C24:0 (~1% of total represents DHA). As such, omega‐3 quantities in CerPCho are negligible.

Furthermore, we observed that the summary ratios and indices (e.g., chain length index, unsaturation index, peroxidation index, omega‐6/omega‐3 ratio, and HUFA) correspond with the known FA makeup of these phospholipids (O'Brien et al. ¹⁹⁶⁴; Svennerholm ¹⁹⁶⁸; Taha et al. ²⁰¹³; Heipertz et al. ¹⁹⁷⁷; Igarashi et al. ²⁰¹¹; Söderberg et al. ¹⁹⁹⁰). For example, CerPCho shows the highest chain length index and lowest unsaturated index of all the fractions, consistent with CerPCho being enriched with long FA tails, which are primarily saturated (Slotte 2016). The average values for the ratios and indices for each phospholipid fraction and the TL can be found in Table S7.

3.2. Group Differences in Characterization of UF Lipids

The distribution of phospholipid fractions between groups (i.e., the ratio of total FAs per fraction) are demonstrated in Figure S3. With respect to the phospholipids, the effect of group was nominally significant for PtdSer C20:3n‐6 relative percentage (p = 0.019) and the TL C20:3n‐6 relative percentage (p = 0.0087), but did not retain significance upon correction for multiple comparisons (Figure S4). Importantly, the changes were specific to the DS group (not both DS and DS‐CA), and for FA of very small quantities (< 1%), thus the biological relevance of these findings is unclear. We found no other changes in FA quantities (either concentration or relative percentage) associated with CA or with depression. Similarly, we did not find any differences between groups in cholesterol concentration (CTRL: 12242.92 ± 804.48 μg/g; DS: 12378.99 ± 1204.62 μg/g; DS‐CA: 13114.49 ± 1229.06 μg/g). When stratifying by violent versus nonviolent method of suicide, which has previously been shown to influence cholesterol concentrations peripherally and in gray matter (Aguglia et al. 2019; Alvarez et al. ²⁰⁰⁰; Golomb ¹⁹⁹⁸; Lalovic et al. ²⁰⁰⁷), we still did not identify group differences. Therefore, cholesterol concentration in the UF is not associated with CA or depression. All statistical reports for the group predictor can be found in Tables S8–S20.

3.3. Age Relationships

Based on the literature, we hypothesized that the relationships between age and lipid levels may be nonlinear, and so for the age‐specific analyses, a multi‐step approach was implemented. First, regressions were fitted for linear, quadratic, cubic, and logarithm relationships. Then, following the approach described in (Perlman et al. 2025), whichever model minimized the Bayesian Information Criteria (BIC) was considered to be the best fitting. However, since a BIC value difference less than 2 does not meaningfully inform the selection of one model over another (Berchtold 2010), we selected the linear model as the “default” due to its interpretability, and only chose a nonlinear model when BIC_linear—BIC_other ≥ 2.

We observed a series of pronounced patterns between FA quantities and age, spanning a wide age range of 15 to 85 years old. Significant relationships between concentration and age as well as between relative percentage and age are plotted in Figure 2A,B, respectively. PtdSer and EtnGpl showed the most age‐related changes, while CerPCho showed none. The significant changes are summarized in table format (Figure 2C). For ChoGpl, directly proportional relationships with age were observed for the concentration and relative percentage of C16:1n‐7, C18:1n‐7, and C22:5n‐3, while inversely proportional relationships were observed for concentration and relative percentage of C22:4n‐6, relative percentage of C18:0, and omega‐6/omega‐3 ratio (Tables S8 and S9). For EtnGpl, directly proportional relationships with age were observed for the concentration and relative percentage of C22:5n‐3 and relative percentage of C20:1n‐9, while inversely proportional relationships were observed for the concentration and relative percentage of C22:4n‐6, omega‐6, as well as the relative percentage of C20:3n‐6, PUFA, HUFA, and chain length index (Tables S10 and S11). For PtdSer, directly proportional relationships with age were observed for the relative percentage of C20:1n‐9, C24:1n‐9, and C22:0, while inversely proportional relationships were observed for the concentration and relative percentage of C20:4n‐6, C22:4n‐6, omega‐6, as well as the relative percentage of C18:0, SFA, and C20:3n‐6 (Tables S12 and S13). No age‐related changes were observed for any FA in CerPCho (either concentration or relative percentage, Tables S14 and S15). For PtdIns, a directly proportional relationship with age was observed for the concentration of C23:0, while an inversely proportional relationship was observed for the relative percentage of C22:4n‐6 (Tables S16 and S17). For the TL, directly proportional relationships with age were observed for the concentration and relative percentage of C16:1n‐7 and the relative percentage of C18:1n‐7 (Tables S18 and S19). The age term for C18:0 ChoGpl concentration was best fit by a logarithmic model, while the other significant age terms were best fit by linear models. No significant age relationships were observed for cholesterol (Table S20). The FAs that survived correction for multiple comparison (BH corrected p < 0.05) are bolded in Figure 2C.

FIGURE 2.

UF phospholipids show no robust effect of group but pronounced effect of age. (A) Scatter plots with regression line showing significant relationships between age and concentration of FAs in ChoGpl (trend line in blue, n = 80), EtnGpl (trend line in purple, n = 80), PtdSer (trend line in green, n = 80), PtdIns (trend line in coral, n = 79), and TL (trend line in brown, n = 80). No significant age‐FA concentration relationships were identified for CerPCho (n = 73). (B) Scatter plots with regression line showing significant relationships between age and relative percentage of FAs in ChoGpl (trend line in blue, n = 80), EtnGpl (trend line in purple, n = 80), PtdSer (trend line in green, n = 80), PtdIns (trend line in coral, n = 79), and TL (trend line in brown, n = 80). No significant age‐FA relative percentage relationships were identified for CerPCho (n = 73). (C) Summary table showing age related changes in lipids separated by phospholipid fraction and TL. Arrows pointing upward indicate a positive coefficient for age, and arrows pointing downwards indicate a negative coefficient for age. FAs or summary metrics in bold indicate statistical significance after correction for multiple comparisons (BH‐corrected p < 0.05). CerPCho, Sphingomyelin; ChoGpl, Choline glycerophospholipids; EtnGpl, Ethanolamine glycerophospholipids; MUFA, Monounsaturated fatty acid; PtdIns, Phosphatidylinositol; PtdSer, Phosphatidylserine; PUFA, Polyunsaturated fatty acid; SFA, Saturated fatty acid; TL: Total lipid.

3.4. Regional Comparison of ChoGpl

We previously published data from the ChoGpl fraction of the human ACC white matter (Perlman et al. 2021). Thus, to understand whether we observe differences in the same phospholipid fraction across brain regions, we compared the concentrations of the ACC and UF ChoGpl fractions of the overlapping FAs in overlapping subjects between the two studies (n = 70, Table 2). All FAs, except certain trace FAs (< 1% of total FA) overlapped. The correlation between the average ACC and UF ChoGpl FA concentrations was nearly perfect (r = 0.993). Despite this near‐perfect correlation, the two regions still show clear separation across a PCA plot (Figure S5). While batch effects cannot be ruled out, there are differences in the FA composition that can explain the separation. For example, in the UF, the concentrations of PAM (C16:0, 3490.12 ± 424.73 μg/g) and OLA (C18:1n‐9, 3400.93 ± 578.75 μg/g) are roughly equal, but in the ACC, there is more OLA (4066.80 ± 690.47 μg/g) than PAM (3397.46 ± 614.38 μg/g). When comparing the age patterns of ChoGpl FAs across both regions, we observe both similarities and differences (Figure S6). In ChoGpl in ACC WM, the total FA concentration is decreasing with age in addition to most other FAs, with no FAs showing increasing concentrations (Figure S6A). In UF, the changes observed in this subset of the cohort (n = 70) are similar to those observed in the full cohort (excluding those for C22:5n‐3 and omega‐6/omega‐3 ratio which did not quite reach significance) and are not reflective of overall lower ChoGpl content. In other words, the changes in ChoGpl FA concentration with age in ACC are indicative of a more global effect (i.e., significant decreases in most FAs), while the changes in the UF concentration are more specific (C16:1n‐7, C18:1n‐7 increasing, decrease in AdA). Across both regions, we observed marked significant decreases of AdA concentration and relative percentage metrics with age, and significant increases in C16:1n‐7 and C18:1n‐7 relative percentage with age (Figure S6B). It is worth noting that AdA was the FA most strongly associated with aging in the ACC ChoGpl by absolute correlation coefficient (Perlman et al. 2021), and AdA also decreases in UF ChoGpl, in addition to UF EtnGpl, PtdSer, and PtdIns. Thus, AdA appears be more heavily influenced by the aging process across brain regions.

TABLE 2.

FA concentrations of ChoGpl for the UF and the ACC. Data are demonstrated as mean ± standard deviation.

Fatty acid	UF ChoGpl (μg/g)	ACC ChoGpl (μg/g)
C 14:0	85.49 ± 13.47	86.96 ± 17.51
C 16:0; PAM	3490.12 ± 424.73	3397.46 ± 614.38
C 18:0; STA	1206.27 ± 158.77	1401.18 ± 248.57
C 20:0	7.89 ± 1.58	10.44 ± 2.40
C 22:0	2.43 ± 0.77	9.23 ± 2.77
C 24:0	7.13 ± 4.48	16.11 ± 6.85
C 16:1n‐7	121.88 ± 23.82	124.73 ± 22.35
C 18:1n‐7	628.78 ± 95.08	747.65 ± 122.35
C 18:1n‐9 (OLA)	3400.93 ± 578.75	4066.80 ± 690.47
C 20:1n‐9	69.11 ± 16.80	97.99 ± 24.52
C 22:1n‐9	10.62 ± 4.30	10.24 ± 3.76
C 24:1n‐9	11.19 ± 3.52	24.69 ± 8.95
C 18:2n‐6; LNA	64.89 ± 15.62	66.12 ± 17.59
C 20:2n‐6	8.79 ± 10.10	10.84 ± 3.13
C 20:3n‐6; DGLA	48.33 ± 11.36	43.58 ± 11.78
C 20:4n‐6; ARA	333.40 ± 65.70	233.76 ± 51.65
C 22:4n‐6; AdA	74.19 ± 11.77	67.43 ± 16.14
C 22:5n‐6; DPAn‐6	10.31 ± 3.77	9.68 ± 4.77
C 22:6n‐3; DHA	95.30 ± 25.07	66.25 ± 19.41
Total FA	9689.42 ± 1219.34	10566.85 ± 1754.43

3.5. Myelin‐Constituent Genes and Proteins

Next, we sought to examine the key myelin‐constituent proteins via western blot (MAG, MBP, MOG, PLP) and the genes that code for them via RT‐qPCR (CNP, MAG, MBP, MOBP, MOG, PLLP, PLP1). The subject information for qPCR cohort can be found in Table S5 and the western blot subject information can be found in Table S6, though these cohorts are largely overlapping. No statistically significant differences between groups were observed for any of the genes (Figure 3A) or proteins (Figure 3B) measured, indicating no robust association with CA or depression. With respect to age, PLP1 gene expression was significantly associated with age (p = 0.032), though this did not retain significance after correction for multiple comparisons (p = 0.081). Nevertheless, all genes measured (except MOBP) show a decrease across the age span (Figure 3C). Of the proteins measured, MAG was significantly associated with age (p = 0.022), though this did not retain significance after correction for multiple comparisons (p = 0.11). Descriptively, however, MAG, MBP, and MOG show decreases across the age span, while PLP protein appears steady (Figure 3D). Overall, the expression of myelin constituent‐gene and protein expression declines with age. The statistical reports for the group and age predictors can be found in Table S21 (qPCR) and Table S22 (western blot).

FIGURE 3.

Myelin‐constituent genes and proteins differ by age, not by group. (A) Box and whisker plots showing gene expression values (normalized to GAPDH) for CNP, MAG, MBP, MOBP, MOG, PLLP, and PLP1 across all 3 groups with no statistically significant differences (n = 69). (B) Box and whisker showing protein expression values (normalized to total protein quantity) for MAG, MBP, MOBP, MOG, and PLP across all 3 groups no statistically significant differences (n = 67). Both box and whisker plots (A and B) show the distribution of the expression data, in which the black line in the middle of each box shows the median, while bottom and top of the box represent the first and third quantiles, respectively. The lower line (whisker) extends to the value of the first quartile minus 1.5 times the interquartile range and the upper line (whisker) extends to the third quartile plus 1.5 times interquartile range. The black dots represent data points that fall outside these ranges. (C) Scatter plot with trend line (blue) showing age plotted against gene expression values (normalized to GAPDH), with PLP1 showing a significant negative relationship (p = 0.032) (n = 69). (D) Scatter plot with trend line (blue) showing age plotted against protein expression values (normalized to total protein quantity), with MAG showing a significant negative relationship (p = 0.022) (n = 67). CTRL, Control; DS, Depressed suicides; DS‐CA, Depressed suicides with a history of CA; GAPDH, Glyceraldehyde 3‐phosphate dehydrogenase; CNP, 2′,3′‐cyclic nucleotide 3′‐phosphodiesterase; MAG, Myelin‐associated glycoprotein; MBP, Myelin basic protein; MOBP, Myelin‐associated oligodendrocyte basic protein; MOG, Myelin oligodendrocyte glycoprotein; PLLP, Plasmolipin; PLP1, Proteolipid protein.

We then examined the correlations between summary lipid metrics (e.g., SFAs, MUFAs, etc.) and gene or protein expression for subjects that were overlapping between the datasets (n = 44). Due to the exploratory nature of the correlation analyses, correlation test p‐values were not corrected for multiple comparisons. Broadly, the quantities of all lipids were positively correlated with all myelin protein expression (Figure 4a). If a greater concentration of lipids is indicative of more myelin overall, then it follows that more protein would be required as well to maintain the proper stoichiometric relationship. This broadly positive correlation did not hold when assessing lipid concentration and gene expression, in which there was a mix of positive and negative correlations (Figure 4b). In the relative percentage metric, MUFA percentage was largely negatively correlated with myelin gene expression, while SFA percentage was largely positively correlated (Figure S8a). The most prominent observation was that the myelin proteins (Figures S7b and S8b), but not the genes (Figures S7a and S8a), appeared to be strongly correlated with PtdIns quantities both in concentration and relative percentage, as seen by the consistent statistical significance and strong correlation coefficients. Interestingly, PtdIns MUFA relative percentage was significantly negatively correlated with myelin proteins, in contrast to all other PtdIns lipid classes which were significantly positively correlated (Figure S8b).

FIGURE 4.

Correlations between total lipids and myelin‐constituent proteins/genes. Correlation plots showing the relationships between total concentration (μg/g) of each lipid class (phospholipid fractions, the TL, and cholesterol) and (A) myelin‐constituent genes (n = 43 overlapping subjects) and (B) proteins (n = 42 overlapping subjects). Due to the exploratory nature of this analysis, p‐values were not corrected for multiple comparisons. The color bar represents Pearson's correlation coefficient between −1 and 1, where negative coefficients are blue in color and positive coefficients are red in color. ***p < 0.001, **p < 0.01, *p < 0.05. CerPCho, Sphingomyelin; ChoGpl, Choline glycerophospholipids; CNP, 2′,3′‐cyclic nucleotide 3′‐phosphodiesterase; EtnGpl, Ethanolamine glycerophospholipids; MAG, Myelin‐associated glycoprotein; MBP, Myelin basic protein; MOBP, Myelin‐associated oligodendrocyte basic protein; MOG, Myelin oligodendrocyte glycoprotein; PLLP, Plasmolipin; PLP1, Proteolipid protein; PtdIns, Phosphatidylinositol; PtdSer, Phosphatidylserine; TL, Total lipid.

4. Discussion

In this study, we provide, to the best of our knowledge, the first characterization of the UF phospholipid FA and cholesterol profile, paired with measures of myelin‐constituent genes and proteins, to examine relationships with CA, depression, and age. The importance of region‐specific lipid characterization is becoming increasingly apparent. A lipidomic study of 75 regions of the adult human brain recently showed that the composition of 93% of the lipids measured was variable across brain regions (Osetrova et al. 2024), lending further support to the notion of regional specificity (Naudí et al. 2015). A recent spatial lipidomic study even described differences in lipid composition across WM tracts (Zavolskova et al. 2025). As such, we compared the UF ChoGpl profile to the existing data of the ACC ChoGpl profile. There is a paucity of human postmortem brain datasets for most WM fiber bundles, and the current characterization of UF lipid composition and comparison of the UF ChoGpl and ACC WM ChoGpl should constitute a useful resource to the field.

The most notable findings from this data concern the wide array of age‐related changes observed in the UF phospholipids. While each phospholipid fraction had its own aging signature or “fingerprint”, we observed a pattern in which MUFA (especially C16:1n‐7 and C18:1n‐7) increased with age and long chain omega‐6 FA (especially AdA) decreased with age. In contrast, the UF had a remarkably steady concentration of cholesterol across a wide age range. A previous study of the human postmortem hippocampus suggested that while the de novo cholesterol synthesis rate may decrease upon aging, the absolute cholesterol content remains stable (perhaps due to a reduction in 24S‐hydroxycholesterol levels, limiting cholesterol efflux from the brain) (Thelen et al. 2006). This stability of cholesterol content has been replicated in the corpus callosum of rhesus monkeys (Dimovasili et al. 2024). Though age‐related decreases in brain cholesterol have been reported as well (Stommel et al. 1989), it is likely that cholesterol relationships with age are region‐specific (Söderberg et al. 1990).

Previous aging studies of both animal and postmortem human tissue are in general agreement with the FA results presented here (Carver et al. 2001; Furber et al. ²⁰²²; Mota‐Martorell et al. ²⁰²²; Söderberg et al. ¹⁹⁹⁰) (40–43) (Carver et al. 2001; Furber et al. ²⁰²²; Mota‐Martorell et al. ²⁰²²; Söderberg et al. ¹⁹⁹⁰). A study of the orbitofrontal cortex (which sends and receives projections via the UF) found that C20:4n‐6 decreases with age, and a study of the frontal cortex shows that both ARA and AdA (along with many other PUFAs) also decrease with age (Carver et al. 2001). Interestingly, our findings recapitulate those of the anterior corpus callosum in aging mouse brain as investigated with a completely different method, Fourier transform infrared spectroscopic imaging (Furber et al. 2022). These authors found that age was associated with an increase in MUFA in tandem with a decrease in PUFA, which parallels our results (Furber et al. 2022). It is interesting to consider the notion that the overall pattern of aging in FA classes may be conserved across species.

While most FA supplementation studies focus on omega‐3s, there is some evidence that omega‐6 supplementation may be beneficial in aging. In a double‐blind trial of ARA supplementation among 25 elderly men, it was observed that 1 month of ARA supplementation (as compared to olive oil placebo) decreased P300 event related potential (ERP) latency and increased P300 ERP amplitude, indicating improved cognition (Ishikura et al. 2009). Studies in rats have shown that (1) aged rats have lower membrane ARA concentration, (2) lower ARA concentration is associated with lower hippocampal long‐term potentiation, and (3) supplementing with ARA can reverse these hippocampal long term potentiation deficits and improve cognitive abilities (Bethlehem et al. 2022; McGahon et al. ¹⁹⁹⁷; Kotani et al. ²⁰⁰³; Okaichi et al. ²⁰⁰⁵). Together, the evidence suggests that further research on omega‐6 supplementation in aging is warranted.

Animal studies support our finding that myelin‐constituent genes and proteins tend to decline with age (Xie et al. 2013; Ximerakis et al. ²⁰¹⁹). The biophysical and biochemical interactions between lipids and proteins in the myelin sheath are absolutely essential to the specialized molecular organization of the membrane (Ozgen et al. 2016). For example, microdomains such as lipid rafts serve as platforms for myelin protein trafficking and hubs for cell signaling (Dupree and Pomicter 2010). Myelin proteins interact with the myelin lipids to create and maintain structural integrity of the sheath (e.g., PLP anchors the external leaflets, MBP compacting the cytoplasmic leaflets, and MAG works as an adhesion molecule between the axon‐myelin interface by binding with gangliosides) (Ozgen et al. 2016; Pronker et al. ²⁰¹⁶). There are many complex lipid‐protein interactions that are key to myelin biosynthesis, maintenance, and integrity over the long term (Barnes‐Vélez et al. ²⁰²³; Ozgen et al. ^{2014, 2016}). Correspondingly, our findings here show that several lipids, especially the signaling lipid PtdIns, are correlated with myelin‐constituent proteins. This strong myelin protein‐PtdIns correlation is possibly due to the key signaling role of phosphorylated PtdIns in trafficking myelin proteins to their proper location in the sheath (Baron et al. 2015; Baskin et al. ²⁰¹⁶; De Craene et al. ²⁰¹⁷; Mironova et al. ²⁰¹⁶; Nawaz et al. ²⁰⁰⁹; Snaidero et al. ²⁰¹⁴). Age‐related declines in myelin integrity have been associated with age‐related declines in cognition in the human brain (Coelho et al. 2021; Mendez Colmenares et al. ²⁰²⁴), including in the UF proper (Zahr et al. 2009). Thus, we speculate that age‐related changes in myelin lipid and protein levels, as observed in this study, might impact lipid‐lipid and lipid‐protein interactions such that microstructural integrity of the myelin is impaired sufficiently to affect information transfer along axons.

The absence of robust significant changes of UF FA or cholesterol in depression or CA concurs with the absence of changes found in myelin‐consistent gene expression or protein levels. Taken together, it appears that the myelin in the UF does not display any evident long‐lasting impacts of depression or CA. In the ACC WM, we observed DS‐CA group‐specific increases of ChoGpl FAs that are part of the ARA synthesis pathway, and corresponding gene expression changes in ARA metabolism genes (Perlman et al. 2021). Furthermore, in the ACC, myelin‐constituent genes were heavily downregulated, painting an overall picture of myelin dysregulation in this region (Lutz et al. 2017). As such, different patterns of CA‐associated neurobiology likely exist across regions, even within WM regions, and may be circuit‐specific. We speculate that the UF neuroimaging findings of CA might represent transient, dynamic changes in maturation rate, rather than long‐lasting neurobiological correlates that can be observed postmortem, especially given that most studies are conducted in adolescents and young adults (Eluvathingal et al. 2006; Granger et al. ²⁰²¹; Gur et al. ²⁰¹⁹; Ho et al. ²⁰¹⁷). Nonetheless, it is also possible that the effect size of CA‐ or depression‐associated changes in the UF may be smaller than other regions, and thus obfuscated by the inherent high variability of human postmortem tissue. This notion is supported by the recent lipidomic study of schizophrenia (Senko et al. 2024), in which the UF showed the less pronounced schizophrenia‐related changes, as compared to other WM regions.

It is worth noting the limitations of this study. Firstly, due to the low number of females in our cohort, consistent with the higher rates of suicide in males, we were likely unable to reliably detect sex differences (Turecki et al. 2019). Secondly, for the annotation of FAs, we only considered FAs that were present on the GLC569 external standard (with the exception of C24:1n‐9), suggesting that we could be missing changes in FAs that we did not consider (e.g., longer FAs). Third, we do not separate the ether phospholipids from the ester phospholipids. It is estimated that over 80% of EtnGpl in myelin is in plasmalogen form, thus we can assume that most of the EtnGpl measured here is ethanolamine plasmalogen (PlsEtn) (Barnes‐Vélez et al. ²⁰²³). It seems as though choline plasmalogens are enriched in WM fractions as well, though less so than PlsEtn (Osetrova et al. 2024). Finally, we cannot ascertain whether alterations in FA are due to changes in synthesis, transport, turnover, oxidation, and/or dietary intake. We do not have information on diet or other lifestyle factors that could influence the brain lipidome, particularly PUFAs which are taken up from the periphery (Bazinet and Layé 2014). In the future, it would be interesting to probe the dietary origins of PUFAs using natural abundance carbon isotope ratio analysis (¹³Carbon/¹²Carbon ratio) (Lacombe et al. 2023).

A key future direction would be to probe the glycolipid composition of the UF and other WM tracts. Glycolipids, especially the cerebroside galactosylceramidase and its sulfated form (sulfatide), are known to be highly enriched in the myelin sheath and to play a key role in long‐term myelin and axonal stability (Marcus et al. 2006; Barnes‐Vélez et al. ²⁰²³). It would also be tremendously valuable to have both central and peripheral (e.g., serum, plasma) lipid measurements to assess to what degree the blood might reflect lipid changes in the brain and whether such peripheral measures could serve as biomarkers of health and disease.

Newer methods for quantification (e.g., flow cytometry assisted single‐cell lipidomics (Hancock et al. 2023)) and visualization (e.g., MALDI‐imaging mass spectrometry (Sugiura et al. 2009)) of lipids in individual cells will provide better resolution to answer questions regarding cell type specificity and distribution of any given lipid species. In conclusion, this large comprehensive dataset of the postmortem human UF phospholipids and cholesterol will serve as a starting point for future research into this unique WM tract and how its molecular properties are impacted in the aging brain.

Author Contributions

John Kim: methodology, formal analysis, investigation. Kelly Perlman: conceptualization, methodology, formal analysis, investigation, writing – original draft. Naguib Mechawar: conceptualization, project administration, resources, writing – review and editing, funding acquisition, supervision. Gustavo Turecki: resources, writing – review and editing. Mackenzie E. Smith: methodology, investigation, writing – review and editing.

Funding

This research was funded by a CIHR Project grant to NM and a CIHR doctoral research award to KP. MES holds an NSERC postgraduate scholarship (PGS‐D). The Douglas‐Bell Canada Brain Bank is supported by platform support grants from the FRQS, Healthy Brains for Healthy Lives (CFREF), and Brain Canada.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1:: Internal standard information for lipid processing.

Table S2: Primer information for RT‐qPCR experiments.

Table S3: Antibody (Ab) information for immunoblotting experiments with Research Resource Identifiers (RRID).

Table S4: Subject information for lipid experiments. Data are demonstrated as mean ± standard error of the mean. p‐values are derived from one‐way anovas.

Table S5: Subject information for immunoblotting experiments. Data are demonstrated as mean ± standard error of the mean. p‐values are derived from one‐way anovas.

Table S6: Subject information for RT‐qPCR experiments. Data are demonstrated as mean ± standard error of the mean. p‐values are derived from one‐way anovas.

Table S7: FA special metrics and indices for all phospholipid fractions and the TL. Data are demonstrated as mean ± standard deviation. *Not calculated due to low concentration of omega‐3 FAs.

Table S8: ChoGpl concentration ANCOVA statistical report for age and group predictors. Nominally statistically significant predictors are highlighted in blue.

Table S9: ChoGpl relative percentage ANCOVA statistical report for age and group predictors * C18:0 age model was best fit by a logarithmic model. Nominally statistically significant predictors are highlighted in blue.

Table S10: EtnGpl concentration ANCOVA statistical report for age and group predictors. Nominally statistically significant predictors are highlighted in blue.

Table S11: EtnGpl relative percentage ANCOVA statistical report for age and group predictors. Nominally statistically significant predictors are highlighted in blue.

Table S12: PtdSer concentration ANCOVA statistical report for age and group predictors. Nominally statistically significant predictors are highlighted in blue.

Table S13: PtdSer relative percentage ANCOVA statistical report for age and group predictors. Nominally statistically significant predictors are highlighted in blue.

Table S14: CerPCho concentration ANCOVA statistical report for age and group predictors.

Table S15: CerPCho relative percentage ANCOVA statistical report for age and group predictors.

Table S16: PtdIns concentration ANCOVA statistical report for age and group predictors. Nominally statistically significant predictors are highlighted in blue.

Table S17: PtdIns relative percentage ANCOVA statistical report for age and group predictors. Nominally statistically significant predictors are highlighted in blue.

Table S18: TL concentration ANCOVA statistical report for age and group predictors. Nominally statistically significant predictors are highlighted in blue.

Table S19: TL relative percentage ANCOVA statistical report for age and group predictors. Nominally statistically significant predictors are highlighted in blue.

Table S20: Cholesterol concentration ANCOVA statistical report for age and group predictors.

Table S21: qPCR ANCOVA statistical report for age and group predictors. Nominally statistically significant predictors are highlighted in blue.

Table S22: Western Blot ANCOVA statistical report for age and group predictors. Nominally statistically significant predictors are highlighted in blue.

Figure S1: PCA plot colored by phospholipid fractions. The x‐axis represents the first principal component, and the y‐axis represents the second principal component. ChoGpl, EtnGpl, PtdSer n = 80; PtdIns n = 79; CerPCho n = 73. CerPCho, sphingomyelin; ChoGpl, choline glycerophospholipids; EtnGpl, ethanolamine glycerophospholipids; PtdIns, phosphatidylinositol; PtdSer, phosphatidylserine.

Figure S2: Sphingomyelin (CerPCho) coelution of DHA and C24:0 validation performed by gas chromatography–mass spectrometry (GC–MS) analysis. Supplementary analysis to determine the relative contribution of the coeluting peak between C24:0 and DHA (C22:6n‐3). (A) Chromatograms for the coeluting peak (top), DHA (middle), and C24:0 (bottom). The yellow circles highlight the scale of the y‐axis, demonstrating the comparative abundance of C24:0 compared to DHA in CerPCho. (B) Table with 6 test subjects with area and percentage values for C24:0 and DHA. On average, the coeluting peak is composed of 98.9% C24:0 and 1.1% DHA. CerPCho, sphingomyelin; DHA, Docosahexaenoic acid.

Figure S3: Stacked bar charts show the proportion of the total of each phospholipid class by group, as measured by total FA for each class (n = 72). CerPCho, sphingomyelin; ChoGpl, choline glycerophospholipids; EtnGpl, ethanolamine glycerophospholipids; PtdIns, phosphatidylinositol; PtdSer, phosphatidylserine.

Figure S4: Bar plots for FA quantities that show nominally significance between groups. Significance stars indicate nominal p < 0.05. Data are represented as mean ± standard error. None of these FA retain significance after correction for multiple comparisons. DS‐CA n = 31, DS n = 29, CTRL n = 20. CTRL, control subjects without depression nor a history of CA; DS‐CA, depressed suicides with history of CA; DS, depressed suicides without CA.

Figure S5: PCA plot of ChoGpl FA profiles colored by region. The x‐axis represents the first principal component, and the y‐axis represents the second principal component. Dots in pink correspond to the ACC and dots in teal correspond to UF (n = 70 per region). ACC, anterior cingulate cortex; UF, uncinate fasciculus.

Figure S6: Scatter plots with regression line showing significant relationships between age and ACC (A) concentration and (B) relative percentage. A blue regression line indicates the model coefficient for age is negative, and a magenta line indicates the model coefficient for age is positive. All significant models were best fit by a linear age term. Nominal p‐values for the plotted FAs: C14:0 concentration p = 0.0008; C16:0 concentration p = 0.0011; C 18:0 concentration p = 0.0001; C 24:0 concentration p = 0.033; C 16:1n‐7 relative percentage p = 0.0002; C18:1n‐7 concentration p = 0.045, relative percentage p = 0.022; C18:1n‐9 concentration p = 0.0008; C 20:1n‐9 concentration p = 0.045, C18:2n‐6 concentration p = 0.020; C20:2n‐6 concentration p = 0.018; C20:3n‐6 concentration p = 0.0011; C20:4n‐6 concentration p = 0.0001; C22:4n‐6 concentration p < 0.00001, relative percentage p = 0.0008; C22:5n‐6 concentration p = 0.032; C22:6n‐3 concentration p = 0.047; total concentration p = 0.0005. All of these values survive correction for multiple comparisons (BH‐corrected value p < 0.05) except for C18:1n‐7 relative percentage, C22:5n‐6 concentration, C24:0 concentration, C 18:1n‐7 concentration, C20:1n‐9 concentration, and C22:6n‐3 concentration.

Figure S7: Correlation plots showing the relationships between summary concentration metrics (ug/g) for all FA classes and (A) myelin‐constituent genes (n = 43 overlapping subjects) and (B) proteins (n = 42 overlapping subjects). Due to the exploratory nature of this analysis, p‐values were not corrected for multiple comparisons. The color bar represents Pearson's correlation coefficient between −1 and 1, where negative coefficients are blue in color and positive coefficients are red in color. ***p < 0.001, **p < 0.01, *p < 0.05. CerPCho, sphingomyelin; ChoGpl, choline glycerophospholipids; CNP, 2′,3′‐cyclic nucleotide 3′‐phosphodiesterase; EtnGpl, ethanolamine glycerophospholipids; HUFA, highly unsaturated fatty acid; MAG, myelin‐associated glycoprotein; MBP, myelin basic protein; MOBP, myelin‐associated oligodendrocyte basic protein; MOG, myelin oligodendrocyte glycoprotein; MUFA, monounsaturated fatty acid; N.3, omega‐3, N.6: omega‐6; PLLP, plasmolipin; PLP1, proteolipid protein; PtdIns, phosphatidylinositol; PtdSer, phosphatidylserine; PUFA, polyunsaturated fatty acid; SFA, saturated fatty acids; TL, total lipid.

Figure S8: Correlation plots showing the relationships between summary relative percentage metrics (%) for all FA classes and (A) myelin‐constituent genes (n = 43 overlapping subjects) and (B) proteins (n = 42 overlapping subjects). Due to the exploratory nature of this analysis, p‐values were not corrected for multiple comparisons. The color bar represents Pearson's correlation coefficient between −1 and 1, where negative coefficients are blue in color and positive coefficients are red in color. ***p < 0.001, **p < 0.01, *p < 0.05. CerPCho, sphingomyelin; ChoGpl, choline glycerophospholipids; CNP, 2′,3′‐cyclic nucleotide 3′‐phosphodiesterase; EtnGpl, ethanolamine glycerophospholipids; HUFA, highly unsaturated fatty acid; MAG, myelin‐associated glycoprotein; MBP, myelin basic protein; MOBP, myelin‐associated oligodendrocyte basic protein; MOG, myelin oligodendrocyte glycoprotein; MUFA, monounsaturated fatty acid; N.3, omega‐3; N.6, omega‐6; PLLP, plasmolipin; PLP1, proteolipid protein; PtdIns, phosphatidylinositol; PtdSer, phosphatidylserine; PUFA, polyunsaturated fatty acid; SFA, saturated fatty acids; TL, total lipid.

Data Availability Statement

Raw chromatograms are available at this link https://osf.io/w97ap/overview?view_only=46295308bde8480589c1e4c7f6ac2ff2 and can be viewed on the CompassCDS software. Western blot and qPCR raw data are available upon request. A preprint of this article was posted to bioRxiv on 10/13/2025 (https://doi.org/10.1101/2025.10.12.681536).